eel 22497

Partitioning the relative fitness effects of diet and

trophic morphology in the threespine stickleback

Daniel I. Bolnick

and Márcio S. Araújo

Howard Hughes Medical Institute; Section of Integrative Biology, University of Texas at Austin,

Austin, Texas, USA and

Departamento de Ecologia, Instituto de Biociências,

Universidade Estadual Paulista, Rio Claro, SP, Brazil

ABSTRACT

Background: Numerous models show that if morphology and diet are correlated, frequency-

dependent competition will lead to fitness differences among phenotypically dissimilar

individuals within a species.

Hypothesis: Selection acts primarily on diet, and only indirectly on morphology via its

correlation with diet.

Field sites and organism: British Columbia, Canada; 340 individual threespine stickleback

(Gasterosteus aculeatus) from McNair Lake and 430 individuals from First Lake.

Measurements: Stable isotopes (δ

C and δ

N; a proxy for diet); trophic morphology

(quantitative traits and geometric shape variables); and growth rates (RNA/DNA ratios; a

proxy for the component of fitness arising from competitive or foraging ability).

Analysis: Linear and quadratic regression of growth rate on stable isotopes and

morphological variables to calculate the relationship between growth (a fitness proxy) and diet

and/or morphology. When both morphology and isotopes affected growth rates, we used a path

analysis to separate their effects.

Conclusions: In the McNair Lake population, growth was dependent primarily on diet type

and only indirectly on trophic morphology. In a second population, path analysis found

that isotopes and body shape separately explain variation in growth rates. We infer that, in

stickleback, selection on trophic morphology is often a correlated side-effect of selection on diet

composition, rather than direct fitness effects of morphology per se.

Keywords: directional selection, frequency-dependent selection, fitness landscape,

function-valued trait, Gasterosteus aculeatus, stabilizing selection, stable isotopes,

trophic morphology.

INTRODUCTION

A number of adaptive dynamic and quantitative genetic models suggest that frequency-

dependent competition for resources can generate stabilizing or disruptive selection

(Roughgarden, 1972; Rosenzweig, 1978; Slatkin, 1980, 1984; Taper and Case, 1985; Dieckmann and Doebeli, 1999; Doebeli and

Correspondence: D.I. Bolnick, Section of Integrative Biology, University of Texas at Austin, One University

Station C0930, Austin, TX 78712, USA. e-mail: [email protected].edu

Consult the copyright statement on the inside front cover for non-commercial copying policies.

Evolutionary Ecology Research, 2011, 13: 439–459

Dieckmann, 2003; Ackermann and Doebeli, 2004; Doebeli et al., 2007). Models of frequency-dependent

competition typically assume that diet has a one-to-one relationship with a measurable

morphological character (e.g. body size or gape width). In natural populations, morpho-

logical characters are indeed correlated with among-individual variation in resource use

(Smith, 1987; Robinson et al., 1996; Bolnick et al., 2003; Svanbäck and Eklöv, 2003, 2004; Martin and Pfennig, 2009).

For such normally distributed trophic traits, average phenotypes are more abundant and

thus likely to experience higher competition, but may be adapted to the most abundant

resources. The mode of selection (stabilizing or disruptive) therefore depends on the

distribution of the trophic trait, the resource diversity, and the strength of competition.

Strong intraspecific competition can generate disruptive selection on trophic morphology

by disproportionately reducing the abundance of resources for the most abundant

phenotypes in the population (Smith, 1987, 1993; Bolnick, 2004a; Bolnick and Lau, 2008; Martin and Pfennig, 2009;

Svanbäck and Persson, 2009)

, particularly if resources are diverse (e.g. bimodal). In contrast,

stabilizing selection could arise if the phenotypic variance is greater than the optimal

variance determined by the resource distribution, for instance, if there is a single resource.

Models of competition-induced selection thus assume that selection acts directly on

individuals’ diets, and only indirectly generates selection on morphology. This indirect

action of selection on morphology is often overlooked, because models assume a one-to-one

relationship between diet and morphology. However, in most cases trophic morphology is

only moderately or weakly correlated with diet [typically <0.5 (Price, 1987; Robinson, 2000; Bolnick

and Paull, 2009)

]. Consequently, it may be more appropriate to measure selection or variation in

diet itself, rather than a morphological proxy (e.g. Bolnick et al., 2007). Following the terminology

introduced by Arnold (1983), we posit that there is a performance function in which

morphology influences diet, and a fitness function relating diet to fitness, which results in an

indirect morphology–fitness correlation. We thus view diet as an emergent trait that, much

like other ‘performance’ traits, is determined by a combination of underlying phenotypes

(e.g. morphology, behaviour, experience).

A weak correlation between diet and morphology should allow one to statistically

separate the effects of selection on diet and morphology to test the hypothesis that selection

acts directly on diet and indirectly on morphology. The alternative hypothesis, contrary

to existing models of selection resulting from frequency-dependent competition, is that

selection acts on morphology via some performance trait other than diet type. For instance,

phenotypically different individuals might adopt the same diet but have different rates of

prey capture or handling.

Here, we test these alternatives using quadratic regression and path analysis to evaluate

the partial correlations between each phenotypic metric (morphology or diet) and a proxy

for a component of selection that reflects competitive or foraging ability (hereafter,

‘fitness’). Based on models of competition-induced selection, we expect that diet is the

best predictor of our fitness proxy, with no additional contribution of morphology. If so,

measures of selection on diet and morphology should yield similar qualitative trends. For

instance, if diet is subject to disruptive selection, morphology should be under disruptive

selection as well, but have no separate partial correlation with fitness. Note that there are

reasonable scenarios under which these predictions can be falsified. While it may appear

obvious that diet affects fitness more directly than does morphology, if fitness variation

arises primarily via foraging efficiency rather than prey identity, then morphology is subject

to selection for reasons independent of diet composition. In other words, the performance

function relevant to selection may entail performance metrics other than diet.

Bolnick and Araújo440

We tested the above predictions by estimating selection gradients simultaneously for both

resource use and morphology, in the threespine stickleback, Gasterosteus aculeatus, which

has previously been shown to experience selection arising from frequency-dependent

competition in some populations (Bolnick, 2004a; Bolnick and Lau, 2008). Revisiting two populations

previously shown to experience disruptive selection, we instead found evidence for

stabilizing and directional selection. This selection acted on both resource use and

morphology, but in one lake the selection on morphology was simply a result of the

diet–morphology correlation as predicted.

Study system

Threespine stickleback is a suitable model organism to test the theoretical assumptions

regarding frequency-dependent competition. First, solitary natural populations show

substantial diet variation among co-occurring individuals, which tend to specialize on either

pelagic or benthic resources (Robinson, 2000; Nosil and Reimchen, 2005; Bolnick and Paull, 2009; Bolnick et al.,

2010; Matthews et al., 2010)

. Second, this diet variation is correlated with morphology (Robinson, 2000;

Bolnick et al., 2008; Snowberg and Bolnick, 2008; Bolnick and Paull, 2009; Matthews et al., 2010)

. As a result, within

a single lake population, diet similarity between individuals declines as a function of

morphological distance (Bolnick and Paull, 2009), as required for competition to be frequency-

dependent. Third, there is empirical evidence that stickleback trophic morphology is subject

to natural selection, ranging from disruptive in lakes with benthic/limnetic generalists, to

stabilizing in very large lakes, and directional in lakes with large migration load (Schluter, 1994,

1995, 2003; Bolnick, 2004a; Bolnick and Lau, 2008; Bolnick et al., 2008)

. Finally, diet and morphology are only

moderately correlated (Robinson, 2000; Bolnick, 2004a; Bolnick et al., 2008; Snowberg and Bolnick, 2008), making

it possible to statistically separate selection on diet and morphology, to test the predictions

listed above.

MATERIALS AND METHODS

Collection

In May 2008, we collected adult stickleback from McNair Lake and First Lake (n = 340 and

430, respectively) on northern Vancouver Island, British Columbia, Canada. The two lakes

contain solitary stickleback populations, morphologically intermediate between the

classic benthic and limnetic species pair forms. Within these solitary populations,

individuals utilize various combinations of littoral and pelagic prey. We chose these two

lakes because their stickleback populations were previously shown to be subject to natural

selection on trophic morphology (Bolnick and Lau, 2008).

In each lake, fish were collected using about 50 unbaited minnow traps placed for ∼4 h in

a variety of microhabitats along ∼0.5 km of shoreline in depths ranging from 0.1 to 10 m.

The sampling scheme was intended to capture a representative sample of the phenotypic

variation within lakes, since both benthic and pelagic phenotypes nest close to shore. The

vast majority (>90%) of morphological and dietary variance within a lake occurs within

small-scale samples [<10 m radius (L.K. Snowberg and D.I. Bolnick, unpublished results)], so we are

confident that our samples reflect the variance within each lake. Any sampling bias due to

selective trapping might affect trait means or variances, but should not bias our selection

estimates.

Relative fitness effects of diet and trophic morphology 441

Collected specimens were euthanized in MS 222 and kept on ice in the field after a caudal

muscle tissue clip (∼0.5 mg) was removed and preserved in 0.5 mL of RNALater (Ambion).

Fish were quickly transferred to a −20⬚C freezer and the RNALater-preserved samples to a

4⬚C refrigerator for ∼5 weeks. All samples were stored at −80⬚C thereafter. Collection

and euthanasia were carried out in accordance with University of Texas institutional guide-

lines for the care of vertebrate animals (Institutional Animal Care and Use Committee

03120501).

Diet

Carbon and nitrogen stable isotope ratios are widely used in ecology as a measure of long-

term feeding history (Peterson and Fry, 1987; Post, 2002; Dalerum and Angerbjörn, 2005; Newsome et al.,

2007)

. Within a lacustrine population of stickleback, stable isotope values are moderately to

weakly correlated with individuals’ trophic morphology, use of benthic versus limnetic prey

(gut contents), and foraging microhabitat (Bolnick et al., 2008; Matthews et al., 2010). We dissected a

piece of caudal muscle tissue from each stickleback. Samples were dried at 60⬚C for at least

48 h, ground with a mortar and pestle, and put into tin capsules (∼1 mg of material). These

were measured for δ

C and δ

N at the University of California at Davis Stable Isotope

Facility. We did not perform any lipid extraction/normalization techniques, because lipid

content in our samples was low, as indicated by the C:N ratios [MacNair Lake: 3.45 ± 0.88;

First Lake: 3.27 ± 0.08; mean ± .. (Post et al., 2007)], and not correlated with isotope

signatures.

In stickleback, δ

C reflects the proportion of benthic versus pelagic carbon in their diet,

and δ

N reflects relative trophic position (Vander Zanden and Rasmussen, 2001; Post, 2002). These

interpretations can be made explicit by standardizing fish isotope ratios by the isotopes of

basal benthic and limnetic consumers [snails and mussels, respectively (Post, 2002)]. However,

in both lakes nearly half of stickleback had δ

C exceeding snail δ

C (maximum δ

C for

stickleback and snails was −19

Ⲑ

C and −25

Ⲑ

, respectively). This can occur if snails

consume phytoplankton detritus, making their isotopes less than fully benthic. The more

extreme values in fish leads to nonsensical estimates of percent benthic carbon (e.g. 250%

benthic carbon) and relative trophic position. Consequently, we report raw isotope values,

which we typically find are highly correlated with percent benthic carbon and trophic

position within a given population [r > 0.98 (L.K. Snowberg and D.I. Bolnick unpublished)].

Morphology

We collected two types of morphological data, which were analysed separately. The first

dataset was composed of univariate measurements including body mass (blotted dry and

weighed to 0.001g), standard length, length of the lower jaw, body width at the pectoral fins,

length of the two largest gill rakers (GRL1 and GRL2), and gill raker number (GRN).

Body shape measurements were taken with a digital calliper accurate to 0.01 mm. Gill

rakers were measured with an ocular micrometer on a dissecting microscope. Sex was

determined by gonad inspection. Gill rakers were only measured on 244 of the 340 fish from

McNair Lake. All measurements were performed by M.S. Araújo.

A second dataset of geometric morphometric shape was obtained using tpsDig2 (Rohlf,

2008)

to digitize 23 homologous landmarks from the left side of each fish. We aligned the

landmarks in tpsRelw (Rohlf, 2005). To measure overall body shape variation, we calculated

Bolnick and Araújo442

relative warps from the 23 photograph landmarks, using tpsRelw. The first two relative

warps (RW

and RW

; see Fig. A1 at evolutionary-ecology.com/data/2657Appendix.pdf)

were used as measures of body shape in subsequent analyses. In McNair Lake, RW

and

explained 34.2% and 16.3% of the variance in position, respectively, of the 23 aligned

landmarks (35.9% and 16.4%, respectively, in First Lake).

Selection estimate

Survivorship and lifetime fecundity are impossible to measure in natural populations of

stickleback. Moreover, we are more interested in the component of selection arising from

resource competition and foraging success, rather than total fitness variation. We therefore

use growth rate as a measure of foraging success and thus a proxy for the component of

fitness arising from competition. Growth rate is widely used as a fitness proxy in stickleback

(Schluter, 1994, 1995, 2003; Hatfield and Schluter, 1999; Hendry et al., 2002; Bolnick, 2004a; Bolnick and Lau, 2008; but see

Taylor et al., 2011)

. Better foragers (or competitors) grow faster (Wootton, 1976, 1994), and

consequently are more fecund [number of eggs (Wootton, 1973, 1977; Ali and Wootton, 1999; Huntingford

et al., 2001; Bolnick, 2004a)

] and may survive better (Reimchen, 1991); although some studies argue to

the contrary (Bell and Haglund, 1978; Litvak and Leggett, 1992; Barber et al., 2001). We acknowledge that

measuring adult growth does not reveal selection on juveniles, measure other sources of

selection (e.g. predation, parasitism, sexual selection), and that selection may not always

favour the fastest growth rate.

To measure growth rate, we used the ratio of RNA to DNA concentration in muscle

tissue. The RNA/DNA ratio is highly correlated with growth rate in stickleback [r = 0.92

(Ali and Wootton, 2003)] and other fishes (Caldarone et al., 2001; Dahlhoff, 2004), because fish that grow

faster synthesize more protein and have higher RNA titre as a result (particularly ribosomal

RNA), whereas the amount of DNA remains constant in the cell. In a field experiment,

increased intraspecific competition depressed growth rate as measured by the RNA/DNA

ratio because of reduced prey availability (Svanbäck and Bolnick, 2007). This reduction in the

RNA/DNA ratio occurred over 2 weeks, whereas brief periods of starvation (e.g. 24 h) have

no effect on the ratio. Thus, it appears that the RNA/DNA ratio reveals growth variation

among individuals arising from differences in energy income over the preceding several days

to weeks. The RNA/DNA ratio has previously successfully revealed disruptive selection on

gill raker length and number in stickleback in the lakes studied here (Bolnick and Lau, 2008). We

followed Bolnick and Lau’s (2008) RNA/DNA quantitation protocol.

Statistical analyses

For each of the two lakes, we used the following analyses to test our hypothesis that

selection acts most directly on diet but is indirectly translated into selection on trophic

morphology. We estimated the fitness landscape within each lake using linear and quadratic

regression of our fitness proxy (ln-transformed RNA/DNA) against morphology

(quantitative traits and geometric shape variables) and diets (δ

C and δ

N) (Lande and Arnold,

1983)

. For the quantitative traits, we focused on size-corrected gill raker length and gill raker

number, which are most strongly correlated with diet here and in previous studies (Schluter,

1995; Robinson, 2000; Bolnick, 2004a; Bolnick et al., 2008; Snowberg and Bolnick, 2008)

and which have previously

been implicated as a target of selection (Bolnick, 2004a; Bolnick and Lau, 2008; Bolnick et al., 2008). For body

shape, we focused on RW

and RW

. For diet, we examined the separate effects of δ

C and

Relative fitness effects of diet and trophic morphology 443

N, which are only weakly correlated (First Lake: r =−0.14, P = 0.003; McNair: r =−0.27,

P < 0.001).

We regressed our fitness proxy (ln-transformed RNA/DNA) on three sets of variables:

(1) size-adjusted gill raker length and number; (2) body shape (RW

and RW

); and (3) diet

(measured as δ

C and δ

N). We size-corrected morphological traits by calculating the

residuals from regressions of the ln-transformed traits on an independent metric of overall

body size (centroid size computed from the 23 digitized landmarks). All independent and

dependent variables were transformed to have standard normal distributions (unit standard

deviation, mean of zero) to allow for comparison of selection gradients with other studies

(Lande and Arnold, 1983; Kingsolver et al., 2001).

Inspection of residual plots indicated linearity of the models and normal distribution of

errors in all cases. Applying a PCA-based size standardization yielded equivalent results

to our size-standardized measures. We therefore focus on the analyses of univariate size-

standardized traits. Several studies of selection on trophic morphology used the PCA

size standardization (Bolnick, 2004a; Bolnick and Lau, 2008; Bolnick et al., 2008), so for comparison with

those previous studies we also provide the results of selection gradient estimates for

PC axes, presented in the Appendix (Tables A1–A4; evolutionary-ecology.com/data/

2657Appendix.pdf). Finally, we undertook cubic spline analysis for all quadratic regressions

to evaluate whether a quadratic model was appropriate (Schluter, 1988). In all cases where

quadratic regression terms were significant, cubic spline confirmed that a quadratic model

was an appropriate fit to the data.

We ran separate quadratic multiple regressions for each of the three sets of traits (gill

raker traits, body shape, and diet). For each multiple regression, we included sex as a factor

but excluded sex × trait interactions, which were never supported by F-tests or AIC. Sex was

removed from the models when non-significant (McNair Lake), but were retained when it

had a significant effect on growth (First Lake). Also, we found no significant selection on

correlations among morphological traits (consistent with Bolnick and Lau, 2008) or δ

C and δ

and so do not discuss those analyses here. To test for possible multicollinearity between

independent variables, we calculated variance inflation factors (VIFs) and found no

evidence of multicollinearity in any of the models (all VIFs < 2.5).

Whenever we found simultaneous significant selection gradients on diet and morphology

within a population, we used quadratic path analysis (Scheiner et al., 2000) to partition the linear

and quadratic effects of morphology (gill raker length) and diet on growth. The path model

is presented in the Results section. A more complex path model with gill raker number

and/or body shape yielded no additional significant path terms and did not change

the qualitative results, so we focus on gill raker length alone. A multiple regression model-

fitting approach using Akaike’s Information Criterion yielded equivalent inferences as the

path analysis, so we focus our presentation on the conclusions arising from the path

analysis.

Effects of diet or trait disparity on growth

Quadratic regression works well when disruptive selection acts along a single univariate trait

axis. Alternatively, frequency-dependent competition may favour any individual who

departs from the population mean regardless of the trait(s). To evaluate this possibility, we

took an alternative approach, in which we tested whether individuals’ distance (‘disparity’)

from the population centroid (in any direction in multivariate trait space) conferred higher

fitness (RNA/DNA). First, we calculated the morphological (quantitative traits and relative

Bolnick and Araújo444

warps) and isotopic Mahalanobis distance between each individual and the population

centroid. With this metric, two individuals could differ from the population mean by the

same amount but along very different phenotypic axes. We then regressed standardized

ln(RNA/DNA) onto trait disparity (normalized to unit standard deviation), with sex

included where significant. We ran three separate disparity regressions, one for all

quantitative traits, one for the relative warps, and one for isotopes. A positive slope would

indicate that phenotypically peripheral individuals have higher growth rate, consistent with

disruptive selection, while a negative slope would indicate stabilizing selection. This

approach requires no a priori assumption as to which trait(s) or trait combinations are

subject to selection. As with the quadratic regressions described above, whenever both

morphology and isotopes showed significant fitness gradients, we used a path analysis

to partition the relative effects of morphological and dietary disparity on growth rates

(see Results for the model). All analyses were carried out in the R environment (R Development

Core Team, 2007.)

RESULTS

McNair Lake

There was no significant linear or quadratic relationship between growth rate and δ

C in

McNair Lake (Fig. 1A). In contrast, we found a negative linear relationship between growth

and δ

N (Table 1; Fig. 1B), suggesting there may be a directional component to selection

favouring individuals who feed on lower trophic position prey. There was no quadratic

relationship between growth and δ

N (Table 1) or correlation between growth and diet

disparity (Table 2).

We found evidence of directional selection on trophic morphology, paralleling the

selection on diet. Individuals with shorter gill rakers exhibited higher growth rates (Table 1;

Fig. 2A). Growth rates were also positively correlated with RW

(Fig. 2C), although the

relationship was only marginally significant (Table 1). Gill raker number and RW2 were

not correlated with growth rate (Fig. 2B, D). None of the morphological axes exhibited

quadratic relationships with growth. Consistent with the absence of quadratic selection

gradients, there was no relationship between growth rate and multivariate disparity in body

shape, morphology or stable isotopes (Fig. 3A–C).

Gill raker length was positively correlated with trophic position (δ

N; see Table A5 at

evolutionary-ecology.com/data/2657Appendix.pdf). Consequently, the inferred directional

selection for lower trophic position is consistent with the directional selection for shorter gill

rakers. To isolate the relative contributions of morphology and diet to variation in growth

rates, we constructed a path analysis including the effect of gill raker length (linear and

quadratic) on diet (measured by linear and quadratic δ

N and δ

C), and the effects of gill

rakers and diet on growth (Fig. 4). The path analysis supports a model in which gill raker

length affects diet (both δ

N and δ

C, which are correlated with each other). However,

only relative trophic position (δ

N) influences growth rate. Thus, the directional selection

on gill raker length, noted above, is entirely explained by indirect selection in which gill

raker length affects trophic position, which affects growth rates. In the terminology of

Arnold (1983), there is a performance function relating morphology to diet, and a fitness

function relating diet to growth.

Relative fitness effects of diet and trophic morphology 445

First Lake

In First Lake, there was a negative quadratic relationship between growth and δ

C (Table 1;

Fig. 1C), implying a stabilizing component to selection. In this lake, growth rate was

maximized for individuals with average values of δ

C, representing a roughly intermediate

mixture of benthic and limnetic resources. Note that both quadratic regression (Fig. 1C)

and cubic spline indicate that the fitness maximum is within the observed range of trait

values. There was no significant association between growth and δ

N in First Lake

(Fig. 1D), unlike McNair. Finally, regression of growth on isotopic disparity revealed a

significant negative association, implying that individuals far from the isotopic centroid

Fig. 1. Relationships between growth rate (RNA/DNA ratio) and two weakly inter-correlated meas-

ures of diet (δ

C in panels A and C; δ

N in panels B and D) for McNair Lake (A, B) and First Lake

(C, D). Isotope ratios and the RNA/DNA ratio were transformed to a standard normal distribution.

Quadratic (or linear) regression results are plotted as a solid line with dashed lines representing

95% confidence intervals for the slope. Significant relationships are plotted as thick trend lines,

non-significant relationships as thin lines. Sexes are plotted with separate symbols (solid = male,

open = female). A single trend line is provided for both sexes because trends are the same with or

without sex.

Bolnick and Araújo446

Table 1. Linear and quadratic relationships between growth and morphology or isotopes, by lake.

Estimates of linear and quadratic selection gradients (β and 2γ) for size-standardized log gill raker

length (GRL), size-standardized gill raker number (GRN), geometric shape variables (relative warps;

and RW

), and δ

C and δ

N stable isotopes. All traits were transformed into standard normal

distributions to provide gradient estimates comparable to other studies of selection

Lake and trait β .. (β) tP2γ 2 .. (γ) tP

McNair Lake

GRL −

−

0.079 0.036 −

−

2.224 0.027 −0.009 0.055 −0.160 0.873

GRN −0.002 0.036 −0.065 0.948 −0.054 0.055 −0.982 0.327

0.057 0.030 1.912 0.057 −0.027 0.042 −0.640 0.522

0.036 0.030 1.220 0.223 0.044 0.042 1.043 0.298

C 0.008 0.031 0.266 0.790 0.032 0.043 0.734 0.463

N −

−

0.074 0.031 −

−

2.423 0.016 −0.029 0.029 −0.997 0.320

First Lake

GRL 0.095 0.029 3.240 0.001 −0.011 0.041 −0.278 0.781

GRN 0.001 0.025 0.045 0.964 0.037 0.031 1.180 0.239

−0.027 0.025 −1.089 0.277 0.560 0.037 −1.516 0.130

0.024 0.025 0.931 0.352 −0.041 0.037 −1.128 0.260

C 0.029 0.028 1.021 0.308 −

−

0.099 0.038 −

−

2.589 0.010

N −0.006 0.025 −0.255 0.799 0.011 0.027 0.392 0.695

Note: GRL is the standardized log transformed mean length of the two longest gill rakers. For each trait in each

lake we provide the least squares slope estimate, its standard error, and statistical significance. Values in bold

indicate regression terms that are significantly different from 0 at α = 0.05. Linear and quadratic gradients were

estimated from different linear models. A sex effect was included in each model initially, but was dropped when not

supported by AIC. Sex had no significant effect on growth rate in McNair Lake, so was not included in the models

presented here. In First Lake, sex had a significant effect on growth rate (P < 0.001) in all the regression models and

so were retained, but we only present the regression slopes for the traits putatively under directional selection.

Table 2. Linear correlations between growth rates and isotopic and morphological disparity, by lake.

Morphological disparity (Mahalanobis distance from population centroid) was based on quantitative

traits (Morphological disparity

) and geometric shape variables (Morphological disparity

geo

Isotopic disparity was based on δ

C and δ

N stable isotopes. The P-value for the sex effect on growth

is listed in the final column

Lake and trait β .. (β) tPSex

McNair Lake

Morphological disparity

−0.056 0.038 −1.479 0.140 0.027

Morphological disparity

geo

0.015 0.033 0.485 0.628 0.244

Isotopic disparity −0.010 0.032 −0.313 0.754 0.210

First Lake

Morphological disparity

−0.048 0.026 −1.877 0.061 << 0.001

Morphological disparity

geo

−0.079 0.025 −4.389 0.002 << 0.001

Isotopic disparity −

−

0.076 0.025 −

−

3.024 0.003 << 0.001

Note: Quantitative traits were body mass, standard length, body width, lower jaw length, gill raker length, and gill

raker number. For each trait in each lake we provide the least squares slope estimate, its standard error, and

statistical significance. Values in bold indicate regression terms that are significantly different from 0 at α = 0.05.

Relative fitness effects of diet and trophic morphology 447

Fig. 2. Relationships between growth rate and four morphological traits (relative gill raker length in panels A and E; gill raker number in panels B and

F; and body shape relative warp 1 in panels C and G, relative warp 2 in panels D and H) for McNair Lake (A–D) and First Lake (E–H). Gill raker

length was log-transformed and standardized by body size. All traits and growth rate were transformed to standard normal distributions. Quadratic

(or linear) regression results are plotted as a solid line with dashed lines representing 95% confidence intervals for the slope. Significant linear and/or

quadratic relationships are plotted as thick trend lines, non-significant relationships as thin lines. Sexes are plotted with separate symbols (solid

= male, open = female). With one exception, a single trend line is provided for both sexes because quadratic and linear terms are similar whether sex is

included or excluded. The exception is panel E, where a linear effect is only detected when sex is included in the model. For panel E we therefore plot

the trends for the sexes separately; in females there is a positive but non-significant effect of gill raker length on growth (thin line), while in males there

is a positive significant effect of gill raker length on growth (bold line). The linear gradient does not differ significantly between the sexes (no

sex × raker length interaction).

Fig. 3. Correlations between growth rate and three measures of trait disparity (quantitative trait morphology in panels A and D; geometric shape in

panels B and E; isotope ratios in panels C and F) in McNair Lake (A–C) and First Lake (D–F). Linear regression trends are plotted as a solid line with

dashed lines representing 95% confidence intervals for the slope. Significant linear relationships are plotted as thick trend lines, non-significant

relationships as thin lines. Sexes are plotted with separate symbols (solid = male, open = female). A single trend line is provided for both sexes because

trends are the same with or without sex.

Fig. 4. The results of quadratic path analysis, for each of the two lakes, estimating the relationships

between trophic morphology (gill raker length), diet (δ

C and δ

N), and a measure of growth rate

(RNA/DNA ratio; a fitness proxy). Morphology (rectangles) is assumed to determine individuals’

diets (a performance trait, indicated by ovals), which affect fitness (diamond). Quadratic terms are

symbolized by shapes with dashed-line margins. Significant partial regressions are indicated by solid

arrows, non-significant regressions by dashed arrows. Correlations (no direction of causation

assumed) are indicated by double-headed arrows. Correlation coefficients and P-values are presented

for each element of the path (except the trait variances, and correlations between linear and quadratic

terms for a given trait). Path models with gill raker number yield identical patterns, but with no

significant effects of gill raker number on diet or growth; consequently, we do not present those

components of the path analysis.

grew poorly (Table 2; Fig. 3F). Both the quadratic and disparity approaches suggest that

ecologically generalist intermediate individuals had highest growth.

We found no significant quadratic selection gradients on gill raker traits or geometric

shape variables (Table 1; Fig. 2E–H). However, in a model omitting the quadratic effect

we did find a significant linear relationship between gill raker length and growth rate

(Table 1; Fig. 2E). Thus, the stabilizing selection acting on δ

C is not translated into

stabilizing selection on gill raker length, in spite of the correlation between gill raker length

and δ

Path analysis clarified the relationship between the stabilizing selection on diet and the

concurrent directional selection on gill raker length (Fig. 4B). Much like McNair Lake,

in First Lake there was a correlation between gill raker length and both δ

C and δ

However, in First Lake growth rate exhibited both a linear and quadratic relationship with

C, but had no independent relationship with gill raker length. The path analysis thus

reveals that the inferred directional selection on gill raker length is an artefact of the linear

association between gill raker length and diet, together with the linear component of the

relationship between diet and growth. As with McNair Lake, we therefore conclude that gill

raker morphology only indirectly affects growth via their shared relationship with diet.

Consistent with our inference of stabilizing selection on diet, multivariate disparity in

body shape was negatively related with growth rate (Table 2; Fig. 3E). Thus, there appears to

be a component of selection acting to reduce isotopic and body shape disparity in First

Lake. We observed no such effect on quantitative trophic morphology traits (Fig. 3D).

A multiple regression model including both isotopic and shape disparity suggested that

both terms are independently and significantly related to growth rate (isotopic disparity:

t =−2.825, P = 0.005; morphological disparity: t =−2.967, P = 0.003). A path analysis

further clarifies this simultaneous trait–growth and diet–growth association (Fig. 5). The

path analysis confirms that body shape disparity and isotopic disparity both have a

significant direct negative effect on growth rate. However, the path analysis also reveals that

shape and isotopic disparity are not significantly correlated with each other. Thus, in First

Lake body shape must be linked to growth rate via some other, unmeasured performance

function. In contrast, quantitative trait disparity is positively correlated with isotopic

disparity, and thereby indirectly associated with growth rate. However, that indirect

correlation between quantitative trait disparity and growth is too weak to reach significance

when measured in a simple regression (P = 0.061; Table 2).

DISCUSSION

Several models of quantitative genetics and adaptive dynamics assume that selection on

individuals’ resource use (a performance trait) leads to correlated selection on trophic

morphology (Roughgarden, 1972; Rosenzweig, 1978; Taper and Case, 1985; Abrams et al., 1993; Dieckmann and Doebeli,

1999; Bolnick and Doebeli, 2003; Ackermann and Doebeli, 2004)

. This indirect effect of morphology on

fitness can be tested by partitioning the effects of morphology and resource use on growth

(Arnold, 1983). In the present study, individuals with different diets exhibited different current

growth rates. In McNair Lake, individuals feeding at lower trophic position exhibited higher

growth rates; in First Lake, growth was higher for ecological generalists with an inter-

mediate carbon signature representing a mixture of benthic and limnetic prey. To the extent

that current growth is a proxy for fitness, we conclude that selection does act on

sticklebacks’ resource use.

Relative fitness effects of diet and trophic morphology 451

The selection on diet indirectly generates correlations between growth and morphology.

In McNair Lake, growth was highest for individuals who both feed at lower trophic levels

and have shorter gill rakers. In First Lake, both gill raker length and benthic/limnetic

carbon use were correlated with growth and with each other. In both cases, path analysis

revealed no independent morphology–growth correlation. Instead, morphology was

correlated with diet, which was correlated with growth. Thus, selection on gill raker

morphology in these systems conforms to a model of trait–performance–fitness mapping

(Arnold, 1983), in which diet is an emergent (performance) trait that is most relevant for

selection.

The exception to this pattern of indirect selection on morphology was seen for body

shape and isotopic disparity. Positive or negative relationships between growth and disparity

should generally conform to patterns of disruptive or stabilizing selection, respectively. The

difference in these approaches is that disparity measures multivariate differences between

individuals, and thus is not restricted to a pre-defined (or even a single) trait axis. Consistent

with our inference of stabilizing selection on δ

C, we found a negative relationship between

isotopic disparity and growth. Notably, we found a similar negative trend for body shape

disparity, even though the first two body shape axes (RW1–2) were not under stabilizing

selection. Given the simultaneous selection against shape and diet disparity, we might have

expected another case of indirect selection on morphology via diet. However, in this

Fig. 5. The results of path analysis, for each of the two lakes, estimating the relationships between

morphological disparity (in trophic morphology or body shape), isotopic disparity, and a measure of

growth rate (RNA/DNA ratio; a fitness proxy). The meanings of various line and box shapes are as

described for Fig. 4.

Bolnick and Araújo452

instance diet and shape disparity have independent effects on growth, and are not noticeably

correlated with each other.

It is not immediately clear why body shape and diet have separate effects on growth.

We speculate that body shape might affect growth primarily via locomotion (and associated

energetic costs), or prey-capture success rates, rather than diet composition. Another

possible explanation could be that body shape affects aspects of diet that are not effectively

measured using carbon and nitrogen stable isotopes. Carbon and nitrogen isotopes have

been very effectively used to identify basic benthic/pelagic differences within stickleback

(Snowberg and Bolnick, 2008; Bolnick and Paull, 2009; Matthews et al., 2010), but additional axes of diet

variation do occur and are correlated with morphology (Araújo et al., 2008). If these other

axes of diet variation entail isotopically similar prey, our measures would fail to detect that

diet variation. In this case, selection acting on undetected diet variation would appear to

directly affect morphology. Finally, there is a potential mismatch between the time-scale of

our fitness and diet metrics. The RNA/DNA ratios reveal growth rate variation that can

respond to adult feeding success within ∼2 weeks (Svanbäck and Bolnick, 2007). In contrast,

isotopes are retained in muscle tissue for months. Consequently, any recent diet switch could

generate variation in growth rate (our fitness metric) that has not yet had time to affect

muscle isotopes.

One important caveat regarding our results is our use of growth rate as a proxy for fitness.

The justification for this proxy is described in more detail in Bolnick and Lau (2008), but

comparisons of growth versus diet raise a few novel challenges. An obvious concern is that

different prey may inherently differ in energy content or capture rate, and thus confer

different growth rates. However, inasmuch as these differences in growth rate lead to differ-

ences in survival or reproductive success, it is still appropriate to view growth rate variation

as indicative of selection. That is, selection favours individuals that specialize on higher-

value resources and thus achieve higher growth rates. This scenario might explain the

directional selection in McNair Lake favouring individuals with lower trophic position,

although we cannot be sure without detailed analyses of prey energy content and

availability. Similarly, the stabilizing selection on δ

C observed in First Lake might reflect

a higher overall prey encounter rate for generalized individuals, because they consume both

benthic and limnetic invertebrates. It is not clear why the lakes differ in their selection

regimes, especially because in a 2005 sample they exhibited similar fitness landscapes

(see below).

We emphasize that our selection estimates only reflect variation in foraging success –

additional selective sources (predators, parasites, mate choice) might generate additional

selection gradients that counterbalance or even overwhelm the partial selection observed

here. However, we were specifically interested in testing the often-modelled association

between morphology, diet, and the component of selection arising from variation in

foraging success. As a final caveat, we acknowledge that in certain environments there

may be selection against fast growth (Barber et al., 2001), contrary to the general trend that larger

stickleback have larger clutch sizes (Wootton, 1973, 1977; Ali and Wootton, 1999). For instance, fast

growth may coincide with a weakened immune system or greater visibility to predators.

Evolutionary implications

Direct selection on diet preferences may have important evolutionary implications for cor-

related phenotypic traits. In our study, gill raker length and number explained between 6%

Relative fitness effects of diet and trophic morphology 453

and 34% of variation in stable isotopes (see Table A4: evolutionary-ecology.com/data/

2657Appendix.pdf). While the unexplained variation may be noise, it is possible that

diet variation also depends on variation in unmeasured morphological, behavioural or

physiological traits. For example, among-individual variation in microhabitat use is not only

correlated with gill raker length but also with jaw lever ratios that are not typically measured

in stickleback (D.I. Bolnick, unpublished results). Direct selection on diet could thus drive evolution

not just of individual morphological structures such as gill rakers, but suites of traits that

might even include important but unmeasured behavioural or physiological traits for which

there is substantial genetic variation (Gibbons et al., 2005; Latshaw and Smith, 2005). Conversely, weak

correlations between diet and these various phenotypic traits may mean that strong

frequency-dependent selection is nevertheless ineffective at driving adaptive diversification

on commonly measured trophic phenotypes.

A second implication of our results is more specific to stickleback. Namely, a previous

study found that diet variation, as measured by isotopes, is subject to positive assortative

mating (Snowberg and Bolnick, 2008). We here show that the same diet metric is subject to natural

selection. Several models of sympatric speciation predict that, in sexual populations,

evolutionary divergence depends on a combination of ecological disruptive selection and

assortative mating (Dieckmann and Doebeli, 1999; Bolnick, 2004b; Doebeli et al., 2007). In these models,

speciation is greatly facilitated when selection and assortative mating act on the same trait

[‘magic traits’ (Gavrilets, 2004; Servedio et al., 2011)]. Our results, combined with those of Bolnick and

Lau (2008) and Snowberg and Bolnick (2008), suggest that in stickleback, diet may represent an

instance of a magic trait, as isotopic measures of diet are correlated with fitness proxies

(implying they are a target of natural selection), and are subject to assortative mating. Of

course, this does not guarantee that speciation will ensue, especially because (1) both

assortative mating and selection on diet are very weak in stickleback (Bolnick, 2011), and

(2) selection is not consistently disruptive.

The temporal dynamics of selection in stickleback

In a previous study on these two populations based on samples collected in 2005 (First

Lake) and 2006 (McNair), Bolnick and Lau (2008) found support for significant disruptive

selection on gill raker number in both lakes. Using the same phenotypes and fitness proxies,

the 2008 samples analysed here revealed no significant disruptive selection on this trait in

either lake [the tables in the Appendix present quadratic selection estimates on principal

component measures of diet and morphology that are directly comparable to Bolnick and

Lau (2008); see evolutionary-ecology.com/data/2657Appendix.pdf]. In the more recent

sample, we found significant directional selection on δ

N and gill raker length (McNair

Lake), and stabilizing selection on δ

C (First Lake). The selection gradients observed in

2008 differ significantly from those of 2006. For example, quadratic regression coefficients

of growth on gill raker number in McNair Lake differ significantly between years (t = 2.25,

P < 0.03; 2008: γ =−0.026 ± 0.0275; 2006: γ = 0.074 ± 0.035). We therefore conclude

that the fitness landscape in lake stickleback varies across years. Our results add to the

literature showing that fitness landscapes can be rather dynamic and change over short

time-scales in natural populations (Siepielski et al., 2009, 2011; Kingsolver and Diamond, 2011; Morrissey

and Hadfield, 2011)

. Some of the temporal heterogeneity in selection may simply be due to

sampling error, although here we find statistically significant differences in selection

between years.

Bolnick and Araújo454

Changes in the form and direction of selection can arise from a variety of sources, which

we cannot distinguish among at present. First, inter-annual or seasonal variation in

resource availability may give rise to fluctuating selection gradients (Reimchen and Nosil, 2004).

Second, selection landscapes may change as a function of population density and the

degree of intraspecific competition (Svanbäck and Bolnick, 2005; Abrams et al., 2008), a prediction

that has recently received observational support in Eurasian perch, Perca fluviatilis (Svanbäck

and Persson, 2009)

, and experimental support in stickleback (Bolnick, 2004a). Under low com-

petition, selection is either directional or stabilizing, favouring phenotypes functionally

associated with benthic resources; as population density increases and competition for these

preferred resources becomes stronger, the use of pelagic resources becomes increasingly

advantageous, and selection turns disruptive (Svanbäck and Persson, 2009). Although we do

not have rigorous population density estimates, natural stickleback populations do cycle in

a density-dependent fashion (Wootton et al., 2005; Wootton, 2008). We therefore speculate that

fluctuations in either stickleback and/or resource densities may lead to fluctuations in the

sticklebacks’ fitness landscape, varying between stabilizing and disruptive selection. Long-

term monitoring of fitness landscapes and lake ecology is required to test this prediction.

Such fluctuating selection would likely undermine any progress towards speciation arising

from the joint action of disruptive selection and assortative mating on a magic trait.

CONCLUSIONS

In the present study, we provide evidence for an implicit but largely untested assumption of

models of frequency-dependent competition, by showing that resource use may be a

primary target for natural selection that is indirectly transferred to trophic morphology. For

most (but not all) selection gradients found here, morphology had no independent effect on

growth rates except via a performance function relating morphology to diet. This finding

suggests that natural selection acting on diets may be an important evolutionary force on

traits correlated with foraging in natural populations. On the other hand, the same finding

suggests that weak correlations between diet and phenotypic traits may mean that strong

frequency-dependent selection is nevertheless ineffective at driving adaptive diversification

on commonly measured trophic phenotypes. Theoretical models, therefore, should be

re-evaluated in light of the general weak morphology–diet correlations in empirical systems

and relax the assumption of one-to-one mapping between morphology and diet. Models

of competition-induced disruptive selection would benefit from recognizing that trait–

performance–fitness mapping may be complex and noisy.

ACKNOWLEDGEMENTS

Financial support was provided by the David and Lucille Packard Foundation. M.S.A. thanks

CAPES, FAPESP, and NSF for additional support. W. Stutz, L.K. Snowberg, T. Ingram, O.L. Lau,

J. Paull, and R. Carlson provided invaluable help during field/lab work. We thank S.F. Reis, T. Ingram,

R. Carlson, and R. Svanbäck for useful comments on a previous version of the manuscript.

REFERENCES

Abrams, P.A., Matsuda, H. and Harada, Y. 1993. Evolutionary unstable fitness maxima and stable

fitness minima of continuous traits. Evol. Ecol., 7: 465–487.

Relative fitness effects of diet and trophic morphology 455

Abrams, P.A., Rueffler, C. and Kim, G. 2008. Determinants of the strength of disruptive and/or

divergent selection arising from resource competition. Evolution, 62: 1571–1586.

Ackermann, M. and Doebeli, M. 2004. Evolution of niche width and adaptive diversification.

Evolution, 58: 2599–2612.

Ali, M. and Wootton, R.J. 1999. Effect of variable food levels on reproductive performance of

breeding female three-spined sticklebacks. J. Fish Biol., 22: 1040–1053.

Ali, M. and Wootton, R.J. 2003. Correlates of growth in juvenile three-spined sticklebacks: potential

predictors of growth rates in natural populations. Ecol. Freshw. Fish, 12: 87–92.

Araújo, M.S., Guimarães, P.R.J., Svanbäck, R., Pinheiro, A., Guimarães, P., Reis, S.F.d. et al. 2008.

Network analysis reveals contrasting effects of intraspecific competition on individual versus

population diets. Ecology, 89: 1981–1993.

Arnold, S.J. 1983. Morphology, performance, and fitness. Am. Zool., 23: 347–361.

Barber, I., Arnott, S.A., Braithwaite, V.A., Andrew, J. and Huntingford, F.A. 2001. Indirect fitness

consequences of mate choice in sticklebacks: offspring of brighter males grow slowly but resist

parasitic infections. Proc. R. Soc. Lond. B, 268: 71–76.

Bell, M.A. and Haglund, T.R. 1978. Selective predation of threespine sticklebacks (Gasterosteus

aculeatus) by garter snakes. Evolution, 32: 304–319.

Bolnick, D.I. 2004a. Can intraspecific competition drive disruptive selection? An experimental test in

natural populations of sticklebacks. Evolution, 58: 608–618.

Bolnick, D.I. 2004b. Waiting for sympatric speciation. Evolution, 58: 895–899.

Bolnick, D.I. 2011. Sympatric speciation in threespine stickleback: why not? International Journal of

Ecology (DOI: 10.1155/2011/942847).

Bolnick, D.I. and Doebeli, M. 2003. Sexual dimorphism and adaptive speciation: two sides of the

same ecological coin. Evolution, 57: 2433–2449.

Bolnick, D.I. and Lau, O.L. 2008. Predictable patterns of disruptive selection in stickleback in

postglacial lakes. Am. Nat., 172: 1–11.

Bolnick, D.I. and Paull, J. 2009. Morphological and dietary differences between individuals are

weakly but positively correlated within a population of threespine stickleback. Evol. Ecol. Res.,

11: 1217–1233.

Bolnick, D.I., Svanbäck, R., Fordyce, J.A., Yang, L.H., Davis, J.M., Hulsey, C.D. et al. 2003. The

ecology of individuals: incidence and implications of individual specialization. American

Naturalist, 161: 1–28.

Bolnick, D.I., Svanbäck, R., Araújo, M.S. and Persson, L. 2007. Comparative support for the niche

variation hypothesis that more generalized populations are also more heterogeneous. Proc. Natl.

Acad. Sci. USA, 104: 10075–10079.

Bolnick, D.I., Caldera, E. and Matthews, B. 2008. Evidence for asymmetric migration load in a pair

of ecologically divergent stickleback populations. Biol. J. Linn. Soc., 94: 273–287.

Bolnick, D.I., Ingram, T., Stutz, W.E., Snowberg, L., Lau, O.L. and Paull, J. 2010. Ecological release

from interspecific competition leads to decoupled changes in population and individual niche

width. Proc. R. Soc. Lond B, 277: 1789–1797.

Caldarone, E.M., Wagner, M., Ogner-Burns, J.S. and Buckley, L.J. 2001. Protocol and guide for

estimating nucleic acids in larval fish using a fluorescence microplate reader. Northeast Fisheries

Science Center Reference Document. Woods Hole, MA: Northeast Fisheries Science Center.

Dahlhoff, E.P. 2004. Biochemical indicators of stress and metabolism: applications for marine

ecological studies. Annu. Rev. Physiol., 66: 183–207.

Dalerum, F. and Angerbjörn, A. 2005. Resolving temporal variation in vertebrate diets using

naturally occuring stable isotopes. Oecologia, 144: 647–658.

Dieckmann, U. and Doebeli, M. 1999. On the origin of species by sympatric speciation. Nature, 400:

354–357.

Doebeli, M. and Dieckmann, U. 2003. Speciation along environmental gradients. Nature, 421:

259–264.

Bolnick and Araújo456

Doebeli, M., Blok, H.J., Leimar, O. and Dieckmann, U. 2007. Multimodal pattern formation in

phenotype distributions of sexual populations. Proc. R. Soc. Lond. B, 274: 347–357.

Gavrilets, S. 2004. Fitness Landscapes and the Origin of Species, Vol. 41. Princeton, NJ: Princeton

University Press.

Gibbons, M.E., Ferguson, A.M. and Lee, D.R. 2005. Both learning and heritability affect foraging

behaviour of red-backed salamanders, Plethodon cinereus. Anim. Behav., 69: 721–732.

Hatfield, T. and Schluter, D. 1999. Ecological speciation in sticklebacks: environment-dependent

hybrid fitness. Evolution, 53: 866–873.

Hendry, A.P., Taylor, E.B. and McPhail, J.D. 2002. Adaptive divergence and the balance between

selection and gene flow: lake and stream stickleback in the Misty system. Evolution, 56:

1199–1216.

Huntingford, F.A., Chellappa, S., Taylor, A.C. and Strang, R.H.C. 2001. Energy reserves and

reproductive investment in male three-spined sticklebacks, Gasterosteus aculeatus. Ecol. Freshw.

Fish, 10: 111–117.

Kingsolver, J.G. and Diamond, S.E. 2011. Phenotypic selection in natural populations: what limits

directional selection? Am. Nat., 177: 346–357.

Kingsolver, J.G., Hoekstra, H.E., Hoekstra, J.M., Berrigan, D., Vignieri, S.N., Hill, C.E. et al.

2001. The strength of phenotypic selection in natural populations. Am. Nat., 157: 245–261.

Lande, R. and Arnold, S.J. 1983. The measurement of selection on correlated characters. Evolution,

37: 1210–1226.

Latshaw, J.S. and Smith, B.H. 2005. Heritable variation in learning performance affects foraging

preferences in the honey bee (Apis mellifera). Behav. Ecol. Sociobiol., 58: 200–207.

Litvak, M.K. and Leggett, W.C. 1992. Age- and size-selective predation on larval fishes: the

bigger-is-better hypothesis revisited. Mar. Ecol. Progr. Ser., 81: 13–24.

Martin, R.A. and Pfennig, D.W. 2009. Disruptive selection in natural populations: the roles of

ecological specialization and resource competition. Am. Nat., 174: 268–281.

Matthews, B., Marchinko, K.B., Bolnick, D.I. and Mazumder, A. 2010. Specialization of

trophic position and habitat use by sticklebacks in an adaptive radiation. Ecology, 91:

1025–1034.

Morrissey, M.B. and Hadfield, J.D. 2011. Directional selection in temporally replicated studies is

remarkably consistent. Evolution (DOI: 10.1111/j.1558–5646.2011.01444.x).

Newsome, S.D., del Rio, C.M., Bearhop, S. and Phillips, D.L. 2007. A niche for isotopic ecology.

Front. Ecol. Environ., 5: 429–436.

Nosil, P. and Reimchen, T.E. 2005. Ecological opportunity and levels of morphological variance

within freshwater stickleback populations. Biol. J. Linn. Soc., 86: 297–308.

Peterson, B.J. and Fry, B. 1987. Stable isotopes in ecosystem studies. Annu. Rev. Ecol. Evol. Syst.,

18: 293–320.

Post, D.M. 2002. Using stable isotopes to estimate trophic position: models, methods, and

assumptions. Ecology, 83: 703–718.

Post, D.M., Layman, C.A., Arrington, D.A., Takimoto, G., Quattrochi, J.P. and Montaña,

C.G. 2007. Getting to the fat of the matter: models, methods and assumptions for dealing with

lipids in stable isotope analyses. Oecologia, 152: 179–189.

Price, T. 1987. Diet variation in a population of Darwin’s finches. Ecology, 68: 1015–1028.

R Development Core Team. 2007. R: A Language and Environment for Statistical Computing.

Vienna, Austria: R Foundation for Statistical Computing.

Reimchen, T.E. 1991. Trout foraging failures and the evolution of body size in stickleback. Copeia,

1991: 1098–1104.

Reimchen, T.E. and Nosil, P. 2004. Variable predation regimes predict the evolution of sexual

dimorphism in a population of threespine stickleback. Evolution, 58: 1274–1281.

Robinson, B.W. 2000. Trade offs in habitat-specific foraging efficiency and the nascent adaptive

divergence of sticklebacks in lakes. Behaviour, 137: 865–888.

Relative fitness effects of diet and trophic morphology 457

Robinson, B.W., Wilson, D.S. and Shea, G.O. 1996. Trade-offs of ecological specialization: an

intraspecific comparison of pumpkinseed sunfish phenotypes. Ecology, 77: 170–178.

Rohlf, F.J. 2005. TpsRelw (available at http://life.bio.sunysb.edu/morph).

Rohlf, F.J. 2008. TpsDig2 (available at http://life.bio.sunysb.edu/morph).

Rosenzweig, M.L. 1978. Competitive speciation. Biol. J.Linn. Soc., 10: 275–289.

Roughgarden, J. 1972. Evolution of niche width. Am. Nat., 106: 683–718.

Scheiner, S.M., Mitchell, R.J. and Callahan, H.S. 2000. Using path analysis to measure natural

selection. J. Evol. Biol., 13: 423–433.

Schluter, D. 1988. Estimating the form of natural selection on a quantitative trait. Evolution, 42:

849–861.

Schluter, D. 1994. Experimental evidence that competition promotes divergence in adaptive

radiation. Science, 266: 798–801.

Schluter, D. 1995. Adaptive radiation in sticklebacks: trade-offs in feeding performance and growth.

Ecology, 76: 82–90.

Schluter, D. 2003. Frequency dependent natural selection during character displacement in

sticklebacks. Evolution, 57: 1142–1150.

Servedio, M.R., Van Doorn, G.S., Kopp, M., Frame, A.M. and Nosil, P. 2011. Magic traits in

speciation: ‘magic’ but not rare? Trends Ecol. Evol., 26: 389–397.

Siepielski, A.M., DiBattista, J.D. and Carlson, S.M. 2009. It’s about time: the temporal dynamics of

phenotypic selection in the wild. Ecol.Lett., 12: 1261–1276.

Siepielski, A.M., DiBattista, J.D., Evans, J.A. and Carlson, S.M. 2011. Differences in the temporal

dynamics of phenotypic selection among fitness components in the wild. Proc. R. Soc. Lond. B,

278: 1572–1580.

Slatkin, M. 1980. Ecological character displacement. Ecology, 61: 163–177.

Slatkin, M. 1984. Ecological causes of sexual dimorphism. Evolution, 38: 622–630.

Smith, T.B. 1987. Bill size polymorphism and intraspecific niche utilization in an African finch.

Nature, 329: 717–719.

Smith, T.B. 1993. Disruptive selection and the genetic basis of bill size polymorphism in the African

finch Pyrenestes. Nature, 363: 618–620.

Snowberg, L.K. and Bolnick, D.I. 2008. Assortative mating by diet in a phenotypically unimodal but

ecologically variable population of stickleback. Am. Nat., 172: 733–739.

Svanbäck, R. and Bolnick, D.I. 2005. Intraspecific competition affects the strength of individual

specialization: an optimal diet theory model. Evol. Ecol. Res., 7: 993–1012.

Svanbäck, R. and Bolnick, D.I. 2007. Intraspecific competition drives increased resource use

diversity within a natural population. Proc. R. Soc. Lond. B, 274: 839–844.

Svanbäck, R. and Eklöv, P. 2003. Morphology dependent foraging efficiency in perch: a trade-off for

ecological specialization? Oikos, 102: 273–284.

Svanbäck, R. and Eklöv, P. 2004. Morphology in perch affects habitat specific feeding efficiency.

Funct. Ecol., 18: 503–510.

Svanbäck, R. and Persson, L. 2009. Population density fluctuations change the selection gradient

in Eurasian perch. Am. Nat., 173: 507–516.

Taper, M.L. and Case, T.J. 1985. Quantitative genetic models for the coevolution of character

displacement. Ecology, 66: 355–371.

Taylor, E.B., Gerlinsky, C., Farrell, N. and Gow, J.L. 2011 A test of hybrid growth disadvantage

in wild, free-ranging species pairs of threespine stickleback (Gasterosteus aculeatus) and its

implications for ecological speciation. Evolution (DOI: 10.1111/j.1558-5646.2011.01439.×).

Vander Zanden, M.J. and Rasmussen, J.B. 2001. Variation in δ

N and δ

C trophic fractionation:

implications for aquatic food web studies. Limnol. Oceanogr., 46: 2061–2066.

Wootton, R.J. 1973. Fecundity of the three-spined stickleback, Gasterosteus aculeatus L. J. Fish

Biol., 5: 683–688.

Wootton, R.J. 1976. The Biology of the Sticklebacks. New York: Academic Press.

Bolnick and Araújo458

Wootton, R.J. 1977. Effect of food limitation during the breeding season on the size, body

components and egg production of female sticklebacks (Gasterosteus aculeatus). J. Anim. Ecol.,

46: 823–834.

Wootton, R.J. 1994. Energy allocation in the threespine stickleback. In The Evolutionary Biology of

the Threespine Stickleback (M.A. Bell and S.A. Foster, eds.), pp. 114–143. New York: Oxford

Science Publications.

Wootton, R.J. 2008. Demography of the threespine stickleback, Gasterosteus aculeatus, in the Afon

Rheidol 1972–1998: search for mechanisms, using simulation models. Behaviour, 145: 603–623.

Wootton, R.J., Adams, C.E. and Attrill, M.J. 2005. Empirical modelling of the population dynamics

of a small population of the threespine stickleback, Gasterosteus aculeatus. Environ. Biol. Fish.,

74: 151–161.

Relative fitness effects of diet and trophic morphology 459