EVOLUTION OF RESISTANCE TO A MULTIPLEHERBIVORE …

ORIGINAL ARTICLE

doi:10.1111/evo.12061

EVOLUTION OF RESISTANCE TO AMULTIPLE-HERBIVORE COMMUNITY: GENETICCORRELATIONS, DIFFUSE COEVOLUTION, ANDCONSTRAINTS ON THE PLANT’S RESPONSE TOSELECTIONMichael J. Wise1,2,3 and Mark D. Rausher1

1Department of Biology, Box 90338, Duke University, Durham, North Carolina 277082E-mail: [email protected]

3Department of Biology, Roanoke College, 221 College Lane, Salem, Virginia 24153

Received September 7, 2012

Accepted January 7, 2013

Data Archived: Dryad doi: 10.5061/dryad.h3t86

Although plants are generally attacked by a community of several species of herbivores, relatively little is known about the strength

of natural selection for resistance in multiple-herbivore communities—particularly how the strength of selection differs among

herbivores that feed on different plant organs or how strongly genetic correlations in resistance affect the evolutionary responses

of the plant. Here, we report on a field study measuring natural selection for resistance in a diverse community of herbivores of

Solanum carolinense. Using linear phenotypic-selection analyses, we found that directional selection acted to increase resistance to

seven species. Selection was strongest to increase resistance to fruit feeders, followed by flower feeders, then leaf feeders. Selection

favored a decrease in resistance to a stem borer. Bootstrapping analyses showed that the plant population contained significant

genetic variation for each of 14 measured resistance traits and significant covariances in one-third of the pairwise combinations of

resistance traits. These genetic covariances reduced the plant’s overall predicted evolutionary response for resistance against the

herbivore community by about 60%. Diffuse (co)evolution was widespread in this community, and the diffuse interactions had an

overwhelmingly constraining (rather than facilitative) effect on the plant’s evolution of resistance.

KEY WORDS: Diffuse coevolution, directional selection, G-matrix, horsenettle, multiple herbivores, resistance to herbivory,

selection analysis, Solanum carolinense.

Most plants are subjected to feeding by multiple species of her-

bivores that may collectively attack all parts of the plant. The

presence of multiple herbivores can influence the evolutionary dy-

namics of resistance to herbivory in ways that are not predictable

from investigations of two-species plant-herbivore interactions

(Fox 1981; Gould 1988; Thompson 1994; Strauss and Irwin 2004;

Agrawal 2005; Strauss et al. 2005). To distinguish between situa-

tions in which multiple-herbivore dynamics are different from

single-herbivore dynamics, investigators have used the terms

“pairwise” and “diffuse” (Janzen 1980; Fox 1981). Diffuse evolu-

tion (or coevolution) is generally considered to occur if the pres-

ence of a third species alters the nature of selection or the response

to selection in a pair of interacting species (Hougen-Eitzman and

Rausher 1994; Iwao and Rausher 1997; Stinchcombe and Rausher

2001). If the (co)evolutionary interactions between the pair are not

influenced by the presence of a third species, then the interactions

are considered pairwise, regardless of how many species may be

involved.

1C© 2013 The Author(s).Evolution

M. J. WISE AND M. D. RAUSHER

Although pairwise (co)evolution is relatively straightfor-

ward, diffuseness can have a variety of implications for the direc-

tion or rate of evolution of defense to herbivory (Thompson 1982;

Pilson 1996; Stinchcombe and Rausher 2001). On one hand, dif-

fuse interactions can facilitate the evolution of resistance, perhaps

enabling a plant to evolve resistance to a species that it could not

have if other species were not also present. On the other hand, dif-

fuse interactions can constrain the evolution of resistance, perhaps

even enabling the continued coexistence of multiple herbivores

that the plant might otherwise be able to eliminate. Therefore,

rather than being a mere semantic distinction, the characterization

of plant-herbivore interactions as pairwise versus diffuse can offer

substantial insight into the outcome of (co)evolution in multiple-

herbivore communities (Rausher 1996; Juenger and Bergelson

1998; Inouye and Stinchcombe 2001; Anderson and Paige 2003;

Lankau and Strauss 2008; Wise 2010).

A major factor potentially contributing to diffuse (co)evolu-

tion is genetic correlations among plant’s resistances to differ-

ent herbivores. In the few studies that have measured genetic

covariances in multiple-herbivore communities, significant cor-

relations between resistances have tended to be uncommon, gen-

erally occurring in fewer than about one-quarterof the species

pairs (Simms and Rausher 1989; Maddox and Root 1990; Roche

and Fritz 1997; Agrawal 2005; Johnson and Agrawal 2007). The

small number of significant correlations may be partially due to

low statistical power resulting from the relatively low number

of plant families (<24) involved in most of these studies. Nev-

ertheless, even if genetic correlations are infrequent, they may

still substantially influence whether coevolution between a plant

and its herbivore community is pairwise or diffuse (Agrawal and

Stinchcombe 2009).

Although genetic correlations in resistances indicate the po-

tential for diffuse coevolution in a plant-herbivore community,

whether the correlations actually influence evolutionary response

depends on the directions and magnitudes of the selection gradi-

ents. Specifically, genetic correlations may act to constrain, fa-

cilitate, or have little effect on evolutionary change in resistance,

depending on the orientation of the selection gradient relative to

the principal components of the variance–covariance matrix of re-

sistances (Blows and Hoffmann 2005; Smith and Rausher 2008;

Agrawal and Stinchcombe 2009). Analysis of the combined ef-

fect of selection and genetic-correlation structure on response to

selection thus has the potential to advance our understanding of

the degree to which the evolution of resistances to multiple her-

bivores is diffuse. The primary objective of the research reported

here was to determine the degree to which the genetic architecture

of resistance interacts with the pattern of selection on resistance to

either constrain or facilitate the evolution of resistance in a plant

species with multiple herbivores.

A secondary objective was to characterize more precisely

the nature of any diffuse evolutionary interactions in the commu-

nity. In particular, because damage to different plant tissues often

does not reduce plant fitness to the same degree (Hendrix 1988;

Crawley 1989, 1999), it might be expected that herbivores attack-

ing different tissues may exert different intensities of selection. In

turn, it would be expected that herbivores exerting the strongest

selection would contribute most to diffuse selection and evolu-

tion by having the greatest indirect effects on the evolution of

resistances to other herbivores. Thus, a goal of this study was to

identify which types of herbivores were responsible for driving

diffuse interactions and which types were most affected by diffuse

interactions.

Materials and MethodsSTUDY SYSTEM

Solanum carolinense L. (Solanaceae), or horsenettle, is a common

herbaceous weed native to the southeastern United States (Bassett

and Munro 1986). Although it spreads clonally within old fields,

pastures, and disturbed areas through rapidly growing perennial

roots, sexual reproduction is required for dispersal to new sites

(Ilnicki et al. 1962). In Virginia, horsenettle can be in flower from

late May to October (Wise 2007b). Its white-to-purple flowers

mature acropetally on racemes that can bear from a single flower

to more than 20 flowers. Individual stems may have zero to over

a dozen racemes. Horsenettle fruits are roughly spherical, yellow

berries, generally from 1 to 2 cm in diameter (Nichols et al. 1991;

Wise 2007b), that are fed upon by a variety of birds and mammals

(Martin et al. 1951; Bassett and Munro 1986; Cipollini and Levey

1997). A typical fruit contains on the order of ∼100 seeds, but

fruits range from just a few to over 250 seeds (Wise and Sacchi

1996).

With its array of thorns, spines, stellate trichomes, and toxic

alkaloids, horsenettle would appear to be well defended from her-

bivory (Ilnicki et al. 1962; Bassett and Munro 1986; Thacker et

al. 1990; Nichols et al. 1992; Cipollini and Levey 1997). Never-

theless, horsenettle is heavily attacked by a diverse assemblage

of mostly native herbivores (Burke 1963; Bailey and Kok 1978;

Nichols et al. 1992; Wise 2007b). In this study, we measured

damage by 11 of the most abundant and conspicuous herbi-

vores, including 10 insect species and one mammal (Table 1).

These include all of the common feeders of leaves, flowers, fruits,

and stems. Importantly, the population of horsenettle used in

this study has been found to possess significant genetic vari-

ation for resistance to all 11 of these herbivores, and thus it

has the potential to evolve in response to selection for resistance

(Wise 2007a).

2 EVOLUTION 2013

MULTIPLE HERBIVORES CONSTRAIN HOST-PLANT EVOLUTION

Table 1. Herbivores included in the study, how they feed, and how their damage was quantified.

Order, species (family) Common name Feeding mode and site Damage measurement units

HemipteraGargaphia solani

Heidemann (Tingidae)Eggplant lace bug Sucks leaf mesophyll cells Presence of eggs and nymphs

ColeopteraEpitrix fuscula Crotch

(Chrysomelidae)Eggplant flea beetle Adults chew leaves Relative leaf area eaten (grid)

Leptinotarsa juncta (Say)(Chrysomelidae)

False potato beetle Chews leaves, flowers, and fruits (1) Percent leaf area eaten(2) No. of flowers destroyed(3) No. of fruits destroyed

Gratiana pallidula(Boheman)(Chrysomelidae)

Eggplant tortoise beetle Chews leaves No. of leaves with damage

Trichobaris trinotata (Say)(Curculionidae)

Potato stalk borer Larvae bore stems Presence of borer or damage

Anthonomus nigrinusBoheman (Curculionidae)

Potato bud weevil Larvae feed inside flower buds No. of flower buds destroyed

DipteraProdiplosis longifila Gagne

(Cecidomyiidae)Citrus gall midge Larvae feed in rolled leaves No. of leaves with damage

Zonosemata electa (Say)(Tephritidae)

Pepper maggot Larvae feed on fruit pulp No. of fruits infested

LepidopteraFrumenta nundinella

(Zeller) (Gelechiidae)N/A Larvae eat ovules inside fruits No. of flowers destroyed

Tildenia inconspicuella(Murtfeldt) (Gelechiidae)

Eggplant leafminer Larvae mine leaves No. of leaves with damage

RodentiaMicrotus pennsylvanicus

(Cricetidae)Meadow vole Chews flowers and fruits (1) No. of flowers destroyed

(2) No. of fruits destroyed

EXPERIMENTAL DESIGN

The horsenettle plants used in this experiment were derived from

roots of a subset of 120 genets (genetic individuals) collected in

the spring of 1997 from four populations within a 13-km radius of

Blandy Experimental Farm (in Clarke County, VA; 39◦N, 78◦W).

The plants were grown in a semiprotected area outdoors in 18.9-L

pots in commercial growing media (Wesco Growing Media III,

Wetsel Seed Company, Harrisonburg, VA), starting with new root

growth each year, for 4 years prior to this study. This process

served to generate root material to propagate numerous clonal

replicates (ramets) of each genet, and to purge the genets of po-

tential maternal, carryover effects from microhabitat differences

in their source fields (Roach and Wulff 1987).

In early spring of 2001, we chose 10 genets from each source

population and cut their roots into segments 2 mL in volume, as

measured by water displacement in a 100-mL graduated cylinder.

We planted 38 root segments from each of the 40 genets individu-

ally in Wesco growing media in 3.8-L plastic pots. We randomly

chose 24 ramets for each genet and arranged the pots in three

randomized blocks (eight pots per genet per block) on wooden

pallets in a semiprotected outdoor area. The 14 remaining ramets

per genet were also moved outdoors and served as alternates. Pots

in which ramets did not emerge, or in which the ramet was dam-

aged, were replaced with alternates of the appropriate genet prior

to transplantation into the field. If more than one ramet emerged

in a pot, all but the largest were removed.

In the spring of 2001, we plotted a grid of three blocks in a

37-ha abandoned agricultural field that had a sizable horsenettle

population. Each block consisted of 10 rows (2 m apart) of 32

planting positions (1.5 m apart). Between 28 June and 2 July, we

transplanted the 960 horsenettle ramets into the field, using the

same randomized positions that the pots occupied on the pallets.

We covered the growing medium from the transplants with field

soil and covered this soil with dead grass from the field, which

made the transplanted ramets blend extremely well with resident

horsenettle ramets, at least to the human eye. The ratio of resident

to transplanted ramets within the blocks was about 30:1 (Wise

2007a).

EVOLUTION 2013 3


From the date of transplanting through ramet senescence

in October, we took data on herbivore damage for 11 species

(Table 1). Nine of these species fed mainly or solely on one type

of organ, whereas one species (Microtus) fed on both flowers

and fruits, and one species (Leptinotarsa) fed on leaves, flowers,

and fruits. Thus, there were 14 different damage measurements

rather than 11. Details on the natural history of these species and

the characteristics of their feeding damage can be found in Wise

(2007b). Importantly, each of these species leaves characteristic

evidence of its feeding, such that the damage could unambigu-

ously be assigned to a particular herbivore.

We collected fruits from each ramet as the fruits began to

mature (before they could be dispersed). We measured the di-

ameter (d) of each fruit and estimated the number of seeds us-

ing a previously reported (Wise and Cummins 2007) prediction

equation:

seeds = 70.1 – 23.0d + 2.18d2 – 0.0415d3 (P < 0.0001;

r2 = 0.90).

DAMAGE MEASUREMENTS

Damage measurements for the different herbivores took a variety

of forms, depending on the feeding mode and phenology of the

herbivore (Wise 2007b). For two (Gargaphia and Trichobaris),

damage was simply scored as either present (1) or absent (0) from

the ramet. For two others, damage was based on leaf area. Specif-

ically, Leptinotarsa damage was quantified as the proportion of a

ramet’s total leaf area that was consumed, which was estimated

by visual quantification of up to 10 main-stem leaves. Damage

by Epitrix was estimated by placing a piece of transparent plastic

with a 94.7 mm2 grid of 25 squares over five main-stem leaves

and counting the squares that encompassed a feeding hole. The

other 10 damage measurements were based on the proportion of

tissues damaged: three were based on the proportion of leaves

damaged, four on the proportion of flowers destroyed, and three

on the proportion of fruits destroyed. More details on the methods

used to quantify damage can be found in Wise (2007b).

RELATIVE STRENGTHS OF DIRECTIONAL SELECTION

AMONG HERBIVORES

To quantify the magnitude and direction of natural selection act-

ing on resistance to each herbivore in the horsenettle popula-

tion, we performed a phenotypic-selection analysis (Lande 1979;

Lande and Arnold 1983). This analysis consisted of a multiple

linear regression of relative plant fitness (described below) on

the 14 herbivore-damage measurements. Prior to the analysis, the

damage measurements for each herbivore were standardized to

a mean of zero and standard deviation of one (by subtracting

the mean from each measurement and dividing by the standard

deviation). The partial-regression coefficients (i.e., the selection

gradients) obtained from this standardized selection analysis rep-

resent the strength and direction of selection acting directly on

resistance to each herbivore in units (standard deviations) that are

directly comparable among the herbivores. A significant neg-

ative gradient for a herbivore means that its damage signifi-

cantly decreased horsenettle’s fitness, and thus that horsenettle

experienced directional selection for increased resistance to that

herbivore.

We also performed a second phenotypic-selection analysis

identical to the first, but including two covariates—the number

of leaves and the number of flower buds per ramet. Including

these covariates not only provided selection gradients for these

plant traits, but it allowed estimates of selection gradients on

resistance that were independent of the relationship between her-

bivory and leaf number and between herbivory and flower-bud

number. As such, to the extent that resistance and leaf or flower

number were correlated, the analysis with the covariates pro-

vided cleaner estimates of selection acting directly on resistance.

Another way of looking at this is that the covariates have re-

moved, from the selection gradients for resistance, the indirect

effect that herbivore damage may have on plant fitness through

an effect on leaf or flower production. Although such a cleaner

estimate may be desirable, it may also lead to an underestimate

of the impact of herbivore damage on plant fitness (e.g., if a

folivore depresses fitness indirectly by decreasing flower produc-

tion). Because there are both advantages and disadvantages of

including the covariates, we present the selection analyses with

and without them. For selection gradients that differ between

the two analyses, we examined more closely the relationship be-

tween herbivory and leaf and flower number to shed more insight

into inferences regarding selection for resistance against these

herbivores.

For fitness in the selection analyses, we used the estimate

of the number of seeds produced per ramet, which represents

the maternal contribution to fitness. We considered fruits infested

with pepper maggots (Zonosemata) to have zero viable seeds.

When a fruit is infested with a pepper maggot, seeds are still

present, but the fruits spoil and tend to remain undispersed, and

the seeds tend to be less viable (Wise 2007b). We converted the

seed-production estimates per ramet into relative fitness by di-

viding each by the mean seed production for all the ramets in

the population. Because horsenettle is a perennial plant, the fit-

ness measurements we used only represent selection acting in one

growing season. We have no reason to believe that this episode

of selection on resistance was unusual in any way. Similarly, we

have no reason to believe that selection on the paternal compo-

nent of fitness (through pollen) would have negated the pattern

of selection acting on the maternal component. Therefore, while

we acknowledge that the ideal fitness variable would be the life-

time genetic contribution by both male and female routes, we

consider our data to be a valid, representative sample of selection

4 EVOLUTION 2013


similar to that acting over the plants’ lifetimes. These selection

analyses were performed with JMP-IN 4.0.4 (SAS Institute, Cary,

NC).

GENETIC VARIANCE–COVARIANCE (G) MATRIX

Genetic variances of the resistance traits were estimated from the

family (genet) component of an analysis of variance in which the

trait was the dependent variable and family was the independent

variable. The significance of the genetic variation was determined

by the significance of the family effect in the ANOVA. Genetic

covariances were calculated from the covariances of the genet

means. Using this procedure with clonal replicates slightly over-

estimates the genetic variances, but because the number of indi-

viduals per genet was large (mean ± SD = 21 ± 3), the estimates

for the genetic covariances should be close to the true genetic

covariances (Via 1984). Thegenetic variances and covariances

were used to construct the G-matrix, a square matrix with genetic

variances along the diagonal and covariances in the off-diagonal

elements.

To determine whether covariances were significantly differ-

ent from 0, we used bootstrapping. For each of the 40 genets,

we calculated the mean for each resistance trait. This yielded

a set of 40 vectors (one per genet) with 14 elements represent-

ing the genet means for the 14 resistance traits. Next, we took a

bootstrap sample by randomly sampling (with replacement) 40

of these vectors. The genetic covariances of the resistance traits

for that sample were then calculated from the covariances of the

genet means. We repeated the bootstrapping procedure for 5000

samples. The significance of each off-diagonal element of the

G-matrix was calculated using a two-tailed test. In this test, we

determined the proportion of the 5000 bootstrap values for that

element that were greater than 0. If more than 4875 of the 5000

covariance estimates between two resistance traits were greater

than 0, we concluded that the covariance was significantly posi-

tive. By contrast, if fewer than 125 were greater than 0, then we

concluded that the covariance between two resistance traits was

significantly negative. This procedure yielded an overall signif-

icance level of α = 0.05 for each correlation. (Adjustments for

multiple comparisons are described in the Results section.)

EVOLUTIONARY FACILITATION AND CONSTRAINT

To assess the overall magnitude of constraint or facilitation among

the 14 resistance traits due to the genetic correlation structure,

we employed the statistical approaches described by Smith and

Rausher (2008) and Agrawal and Stinchcombe (2009). Briefly,

5000 bootstrap samples were created by randomly resampling the

resistance data for horsenettle individuals from each family, with

the number of individuals sampled being equal to the number

of individuals in that family in the original data set. Each boot-

strap sample was used to calculate both a directional phenotypic-

selection gradient (β) and a G-matrix. Elements of β were esti-

mated by multiple linear regression of fitness on the traits (damage

by different herbivores) (Lande and Arnold 1983). Elements of the

G-matrix were calculated from the variance (diagonal elements)

or covariance (off-diagonal elements) of family means. Predicted

responses to selection were then calculated by multiplying the

G-matrix by β for each bootstrap sample.

For each bootstrap sample, we calculated two different vec-

tors of predicted responses. For one (Response Vector 1, or

“RV1”), we multiplied the full G-matrix by β. For the other

(Response Vector 0, or “RV0”), we mathematically eliminated

the genetic correlations by replacing the off-diagonal elements

of the G-matrix with zeroes before multiplying by β. Thus, RV1

estimates the evolutionary responses for the 14 resistance traits

considering the genetic correlations among them, whereas RV0

estimates what the evolutionary responses would have been if

there had been no genetic correlations among them (Smith and

Rausher 2008; Agrawal and Stinchcombe 2009).

We used these two response vectors in two ways. First, we

asked whether the genetic correlation structure tended to be an

overall constraint or facilitator on the response to selection. To

do this, we calculated the ratio of the length of RV1 to the length

of RV0 for each bootstrap sample. If more than 97.5% of these

ratios were >1, then the correlations were significantly facilitative

(i.e., the correlations in resistance had increased the evolutionary

response for resistance). However, if more than 97.5% of the

ratios were <1, then the correlations significantly constrained the

evolution of resistance to the herbivore community.

Second, we then asked which of the 14 resistance traits were

individually constrained or facilitated by the genetic-correlation

structure. To do this, we calculated the difference in the two types

of response vectors (RV1-RV0) for each element (i.e., each re-

sistance trait) for each bootstrap sample. Recall that a negative

RV element means an evolutionary decrease in the amount of

damage by a herbivore (i.e., an increase in resistance to that her-

bivore). An RV0 that is less than (or “more negative” than) the

corresponding RV1 will result in a difference (RV1-RV0) that is

greater than zero, which would indicate that genetic covariances

reduced, or constrained, the expected evolutionary increase in re-

sistance. A negative difference, in contrast, would suggest that the

covariances intensified, or facilitated, the evolutionary increase in

resistance. We determined the proportion of the 5000 bootstrap

differences that were <0, and inferred a significant constraint at

the level α if this proportion was less than 1/2 α, and we inferred

a significant facilitation if this proportion was greater than 1- 1/2α (two-tailed test). (Adjustments for multiple comparisons were

conducted as described in the Results section.)

Finally, we determined which herbivores in the horsenettle

community tended to be the most important in terms of produc-

ing constraint or facilitation on the evolution of resistance to the

EVOLUTION 2013 5


other herbivores. To do this, we “computationally removed” each

herbivore individually and performed an analysis that addressed

how responses to selection would change in the absence of that

herbivore. Specifically, for each herbivore in turn, we set the corre-

sponding element of the selection gradient, β to zero, which would

be its value in the absence of the herbivore. (Because Leptinotarsa

and Microtus fed on multiple tissue types, when removing these

herbivores, we set multiple elements to zero simultaneously.) We

then determined the expected response to selection by multiplying

the full G-matrix by the modified β. This response vector was then

compared to the response vector obtained using the unmodified β.

The significance of the differences between the response vectors

for each herbivore was calculated using bootstrapping procedures

as described above.

ResultsDAMAGE AND SELECTION ON RESISTANCE

The six leaf-feeding herbivores in this study each attacked more

than 5% of the horsenettle ramets (see also Wise 2007b). Her-

bivory by Epitrix was the most widespread, with no ramet escap-

ing its damage. The stem borer, Trichobaris, damaged 73% of

the ramets. Just over half of the flowers were destroyed by four

species of herbivores, and nearly three-fourths of the fruits were

destroyed by three species of herbivores (Table 2).

Directional selection acted on resistance to seven of the

11 species, with the selection gradients for nine of the 14 dif-

ferent damage measurements being significant in at least one

of the two selection analyses (i.e., with or without covariates;

Table 3A). Except for the stem feeder Trichobaris, all signif-

icant selection gradients were negative, indicating selection to

increase resistance. Directional selection also acted to increase

both the number of leaves and the number of flower buds per plant

(Table 3A).

When herbivores were grouped into categories based on the

type of tissue consumed, there was substantial variation among

the categories in the magnitude of selection (Table 3B). In par-

ticular, selection for resistance to florivores was approximately

six times stronger than selection for resistance to folivores; selec-

tion for resistance to frugivores was in turn approximately three

times stronger than for resistance to florivores. These differences

were highly significant in the models with and without covariates

(Table 3C).

The correlation of selection gradients between the selection

models with and without the leaf- and flower-number covariates

was exceptionally strong (r = 0.97), suggesting that herbivore

damages are only weakly genetically correlated with either of the

covariates: The average genetic correlation between resistances

and covariates was 0.27. The major difference in results from

the two models involves the stem borer, Trichobaris. In the model

without covariates, the selection gradient for resistance to this her-

bivore was significantly positive, indicating selection to decrease

resistance, whereas in the model with covariates, the selection gra-

dient was essentially zero. This difference is likely due in large

part to the relatively strong genetic correlation (r = 0.42) between

this resistance and number of leaves. Because selection acted to

increase leaf number, this correlation caused indirect selection to

increase damage by (i.e., decrease resistance to) Trichobaris. In

the model with covariates, this indirect component was removed

from the resistance selection gradient, reducing the gradient to

approximately zero.

EVOLUTIONARY FACILITATION AND CONSTRAINT

Genetic variance–covariance matrixAnalysis of the genetic variance–covariance matrix for resistances

to different herbivores reveals substantial potential for both con-

straint and facilitation in the evolution of resistance. The genetic

variances for all 14 resistance traits were significantly greater

than 0 (Table 4). At an overall significance level of 0.05, at most

two would be expected to be significant by chance (i.e., Type I

error). There thus appears to be detectable genetic variation for

most, if not all, of the resistance traits.

Of the 91 covariances calculated between resistance traits,

31 were nominally significant at the P < 0.05 level (Table 4). At

an overall significance level of 0.05, at most eight of these would

be expected to found significant by chance alone. Consequently,

at least 22 of the 91 covariances are truly significant (not counting

covariances found to be nonsignificant due to Type II error). Of

the 31 nominally significant covariances, 18 were negative and

13 were positive.

Effects of genetic covariances on expected response toselectionThe following analyses were performed in two ways—with and

without leaf- and flower-number covariates. For simplicity, we

present below only results of the analyses not including the co-

variates, noting the instances in which the two models produced

different results. The detailed results of both sets of analyses can

be found in the Supporting Information (Tables S1–S4).

To evaluate the degree to which these covariances are likely

to facilitate or to constrain the evolution of resistances to different

herbivores, we first compared the lengths of the expected response

vectors considering the actual genetic covariances (RV1) to the

lengths with covariances set to zero (RV0). In each of the 5000

bootstrap samples, RV1 was shorter than RV0; thus, the lengths

of the response vectors are different at P = 0.0002. On average,

RV1 was only 41.2% as long as RV0, indicating that covariances

among resistance traits would reduce (i.e., constrain) the overall

response to selection for resistance by nearly 60%.

6 EVOLUTION 2013


Table 2. Mean and variation in damage levels by each herbivore. Of the 960 ramets transplanted, 935 were available for determining

the percentages of leaves damaged by the first three folivores, 930 and 931 were available for Leptinotarsa and Epitrix leaf damage,

respectively, 951 were available for Gargaphia and Trichobaris assessments, 931 flowered, and 891 set fruit. The florivory and frugivory

percentages are based on the total number of potentially successful flowers or fruits, excluding the 6% of flower buds and 15% of fruits

that aborted unrelated to herbivore damage or the 10% of fruits that were killed by frost before maturing (Wise 2007b).

Damage Mean StandardHerbivore metric damage deviation

FolivoresTildenia inconspicuella % leaves damaged 2.6% 4.5%Gratiana pallidula % leaves damaged 4.6% 8.3%Prodiplosis longifila % leaves damaged 0.26% 1.2%Leptinotarsa juncta % leaf area eaten 5.6% 7.3%Epitrix fuscula Relative area index 77 26Gargaphia solani Presence of eggs and nymphs 5.3% –

Stem borerTrichobaris trinotata Presence of borer 73.3% –

FlorivoresAnthonomus nigrinus % buds destroyed 30.9% 17.9%Leptinotarsa juncta % buds destroyed 11.7% 15.7%Frumenta nundinella % buds destroyed 2.7% 8.3%Microtus pennsylvanicus % buds destroyed 5.4% 12.8%

FrugivoresLeptinotarsa juncta % fruits destroyed 21.2% 27.2%Zonosemata electa % fruits destroyed 35.5% 29.8%Microtus pennsylvanicus % fruits destroyed 15.5% 26.7%

Genetic covariances significantly affected the predicted re-

sponse to selection for five of the 14 resistance traits at a nominal

significance level (P < 0.05) (Fig. 1; Tables S1 and S2). The

effects of covariances on three herbivores (Leptinotarsa, Zonose-

mata, and Microtus frugivory) were significant after a Bonferroni

correction. For these three herbivores, the predicted response to

selection was reduced by an average of 74% (Table S1). The effect

of genetic covariances on the predicted response to selection was

nominally significant for two additional herbivores—Trichobaris

(facilitative), and Frumenta (constraining) (Fig. 1). At an over-

all significance level of P < 0.05, one would expect only 0.7 of

the 14 resistances to be significant by chance alone. The prob-

ability of more than two being significant by chance alone is

0.03. Thus, the effect for at least one of these two additional

herbivores was probably significant. Genetic covariances had a

marginally significant effect on the response to selection for re-

sistance to Epitrix. Without covariances, the plants are predicted to

increase resistance by more than 0.5 units, while with covariances,

evolution of resistance to Epitrix is predicted to be negligible

(Fig. 1).

In the analyses in which each herbivore in turn was com-

putationally removed to assess its influence on the evolutionary

response for resistance to the other species, seven species were

found to affect horsenettle’s evolutionary response to at least two

other species each (Fig. 2, Tables S3 and S4). These species in-

cluded leaf, flower, fruit, and stem feeders. Microtus was the most

influential herbivore in this sense, significantly reducing the re-

sponse to selection on resistance to five species and facilitating

selection on one other species. Overall, constraining effects out-

numbered facilitative effects 17-to-6 in the model without covari-

ates (Fig. 2) and 23-to-4 in the model with covariates (Tables S3

and S4). Thegreatest difference between the two models involved

the effects of Epitrix. In the model without covariates, Epitrix did

not significantly affect the predicted evolutionary response for any

of the resistance traits (though many were marginally significant;

Table S3). In the model with covariates, Epitrix significantly con-

strained the evolutionary response for resistance to seven other

herbivores, while facilitating the evolution of resistance to fru-

givory by Microtus (Table S4).

In the same analyses, the predicted evolutionary responses

for 10 of the 14 resistance traits were found to be significantly in-

fluenced by genetic covariances with at least one other resistance

trait (Fig. 2; Tables S3 and S4). These resistance traits included

the five that exhibited an overall effect of the herbivore commu-

nity on response to selection. They also included five resistances

for which positive and negative effects of other species appeared

to cancel, such that there was neither overall constraint nor facil-

itation of the evolutionary response.

EVOLUTION 2013 7


Table 3. Results of standardized, linear phenotypic-selection analyses on resistance. (A) Directional selection gradients (β) for resistance

to different herbivores and for covariates. A negative directional-selection gradient means that damage decreased plant fitness, and thus

selection acted to increase resistance. Selection gradients in boldface are significant at P < 0.05. (B) Mean gradients for herbivore guilds

(i.e., tissue fed upon). (C) ANOVA contrasts for differences among herbivore guilds.

Model without covariates Model with covariates

Herbivore (feeding site) β SE P-value β SE P-value

(A) Selection gradients for different herbivoresCovariate: leaves – – – 0.230 0.025 <0.0001Covariate: flower buds – – – 0.479 0.029 <0.0001Tildenia (leaves) −0.057 0.032 0.08 0.014 0.028 0.60Gratiana (leaves) 0.055 0.035 0.09 0.033 0.026 0.21Prodiplosis (leaves) 0.003 0.032 0.92 0.007 0.026 0.79Leptinotarsa (leaves) −0.051 0.034 0.13 −0.041 0.027 0.13Epitrix (leaves) −0.059 0.035 0.09 −0.104 0.029 0.0004Gargaphia (leaves) −0.041 0.033 0.22 0.045 0.026 0.08Trichobaris (stem) 0.087 0.032 0.006 0.01 0.026 0.68Anthonomus (flowers) −0.196 0.035 <0.0001 −0.192 0.028 <0.0001Leptinotarsa (flowers) −0.142 0.036 0.0001 −0.164 0.029 <0.0001Frumenta (flowers) −0.164 0.032 <0.0001 −0.117 0.026 <0.0001Microtus (flowers) −0.06 0.035 0.09 −0.139 0.028 <0.0001Leptinotarsa (fruits) −0.362 0.04 <0.0001 −0.386 0.032 <0.0001Zonosemata (fruits) −0.456 0.04 <0.0001 −0.487 0.032 <0.0001Microtus (fruits) −0.403 0.04 <0.0001 −0.416 0.032 <0.0001

β SE P-value β SE P-value(B) Mean gradients for herbivore guildsFolivores −0.148 0.089 0.097 −0.14 0.071 0.057Stem feeder 0.528 0.192 0.006 0.065 0.153 0.67Florivores −0.844 0.132 <0.0001 −0.918 0.105 <0.0001Frugivores −2.436 0.064 <0.0001 −2.573 0.153 <0.0001

df F P-value df F P-value(C) Contrasts between guildsAmong four guilds 3 49.3 <0.0001 3 78.5 <0.0001Stem versus folivores 1 10.25 0.0014 1 1.4 0.24Folivores versus florivores 1 17.81 <0.0001 1 35.7 <0.0001Florivores versus frugivores 1 44.49 <0.0001 1 76.28 <0.0001

DiscussionCONSTRAINT, FACILITATION, AND DIFFUSE

SELECTION

Evolutionary interactions between host plants and multiple

species of herbivores can be either pairwise or diffuse. In pairwise

interactions, the selection exerted by a given herbivore on host re-

sistance, as well as the response to that selection, is independent

of the effects of other herbivore species. By contrast, in diffuse

interactions, either selection by a given herbivore or response

to that selection (or both) depends on what other species are

present, as well as the magnitude of selection exerted by the other

species. Although several investigations have demonstrated the

potential for diffuse interactions—for example, by the presence of

genetic correlations among resistances (Simms and Rausher 1989;

Maddox and Root 1990; Roche and Fritz 1997; Agrawal 2005;

Johnson and Agrawal 2007)—there have been no previous at-

tempts to determine whether the resistance-correlation struc-

ture translates into effects on the evolutionary response to

selection for resistance. By combining field estimates of se-

lection gradients with the genetic variance–covariance matrix

for 14 resistance traits in the host plant S. carolinense, we

found ample evidence that the presence of certain herbivores

can alter the evolutionary response for resistance to other her-

bivores. That is, the evolutionary interactions in the herbivore

community of this plant include plenty of diffuse as well as pair-

wise elements.

8 EVOLUTION 2013


Ta

ble

4.

Gen

etic

vari

ance

–co

vari

ance

mat

rix

(G-m

atri

x)fo

rre

sist

ance

.G

enet

icva

rian

ces

are

on

the

mai

nd

iag

on

alin

ital

ics;

cova

rian

ces

are

off

-dia

go

nal

elem

ents

.A

llg

enet

ic

vari

ance

sar

est

atis

tica

llysi

gn

ifica

nt

atP

<0.

05.G

enet

icco

vari

ance

ssi

gn

ifica

nt

atP

<0.

05in

dic

ated

inb

old

Her

bivo

reTi

lden

iaG

rati

ana

Pro

dipl

osis

Lep

tino

tars

aE

pitr

ixG

arga

phia

Tric

hoba

ris

Ant

hono

mus

Lep

tino

tars

aF

rum

enta

Mic

rotu

sL

epti

nota

rsa

Zon

osem

ata

Mic

rotu

s(o

rgan

fed

on)

(lea

ves)

(lea

ves)

(lea

ves)

(lea

ves)

(lea

ves)

(lea

ves)

(ste

m)

(flo

wer

s)(f

low

ers)

(flo

wer

s)(f

low

ers)

(fru

its)

(fru

its)

(fru

its)

Tild

enia

(lea

ves)

0.96

5−0

.199

0.08

3−0

.499

−3.4

46−0

.112

1.75

8−0

.307

1.25

40.

489

0.60

7−0

.041

1.91

50.

004

Gra

tian

a(l

eave

s)8.

597

0.30

70.

395

−18.

825

11.7

9111

.439

1.44

62.

473

1.24

1−0

.839

7.04

2−1

.865

−4.9

89

Pro

dipl

osis

(lea

ves)

0.12

2−0

.061

−1.5

24−0

.198

−0.2

36−0

.152

0.30

9−0

.052

−0.0

720.

729

−0.2

39−0

.504

Lep

tino

tars

a(l

eave

s)4.

441

6.02

6−2

.163

−8.4

25−2

.825

1.67

10.

110

−0.6

18−2

.347

−3.1

085.

376

Epi

trix

(lea

ves)

156.

725

−56.

851

−68.

021

−19.

649

−7.0

94−1

0.53

49.

178

−48.

712

−32.

535

65.9

01

Gar

gaph

ia(l

eave

s)62

.382

17.1

145.

800

−0.5

363.

223

−3.4

9522

.563

1.89

2−2

6.63

8

Tric

hoba

ris

(ste

m)

211.

456

23.6

559.

035

2.61

84.

077

14.0

0454

.295

−23.

873

Ant

hono

mus

(flo

wer

s)24

.172

−3.9

022.

460

−4.7

396.

826

15.4

66−1

9.76

4

Lep

tino

tars

a(f

low

ers)

15.7

90−0

.492

−3.9

337.

254

4.03

3−7

.841

Fru

men

ta(f

low

ers)

6.61

9−4

.121

1.01

82.

016

−9.1

78

Mic

rotu

s(f

low

ers)

13.1

05−5

.046

−7.5

8921

.174

Lep

tino

tars

a(f

ruits

)49

.339

1.39

8−3

8.97

2

Zon

osem

ata

(fru

its)

77.6

65−3

9.40

2

Mic

rotu

s(f

ruits

)92

.684

EVOLUTION 2013 9


-1

-0.5

0

0.5

1

1.5

2

2.5

-1 -0.5 0 0.5 1 1.5 2 2.5

Microtus

Zonosemata

Lep�notarsaTrichobaris

Frumenta

Epitrix

Resp

onse

with

full

G

Response with G0

Figure 1. Predicted evolutionary response for 14 resistance traits

in the presence of genetic covariances (full G matrix, Y-axis) ver-

sus the absence of genetic covariances in resistance (G0, X-axis).

Responses that are significantly different in the two scenarios are

indicated by filled circles: red for responses that were reduced and

blue for responses that were increased by genetic covariances.

The dashed line is a boundary on which the predicted response

is equivalent with and without covariances. Points above the line

indicate that covariances would facilitate the evolution of resis-

tance, whereas points below the line indicate that covariances

would constrain the evolution of resistance.

Tildenia

Gra�ana

Prodiplosis

Lep�notarsa

Epitrix

Gargaphia

Trichobaris

Anthonomus

Frumenta

Microtus

Zonosemata

Tildenia

Gra�ana

Prodiplosis

Lep�notarsa (leaves)

Epitrix

Gargaphia

Trichobaris

Anthonomus

Lep�notarsa (flowers)

Frumenta

Microtus (flowers)

Lep�notarsa (fruits)

Zonosemata

Microtus (fruits)

Herbivore:

Resistance to:

Effect on:

Figure 2. Interaction diagram for the 11 herbivore species and 14

types of resistance. Names of herbivores are in the left column;

types of resistance are in the right column. Arrows indicate signif-

icant indirect effects of one herbivore on the expected response

to selection for resistance to another species. Thin lines indicate a

nominal significance level of 0.001 < P < 0.05, and thick lines indi-

cate P < 0.001. Dashed, red lines indicate that the indirect effects

would constrain, and solid, black lines indicate that the indirect

effects would facilitate the evolution of resistance.

The expected responses to selection for about one-third of

the resistances were influenced by whether there were nonzero

genetic covariances among the resistances. This influence was

overwhelmingly constraining—that is, the presence of genetic co-

variances acted to slow or prevent the plants from evolving greater

resistance against much of the herbivore community. In partic-

ular, the magnitude of the multidimensional selection-response

vector under the estimated correlation structure was only 41% of

that when genetic covariances were eliminated. In addition, four

of the five resistance traits exhibiting significant effects of ge-

netic covariances in resistance experienced substantially smaller

responses to selection when genetic covariances were nonzero

(namely, resistance to Microtus, Zonosemata, Leptinotarsa, and

Frumenta). For the fifth (the stem-borer Trichobaris), genetic

covariances significantly facilitated horsenettle’s predicted evolu-

tionary response. In other words, genetic covariances in resistance

only acted to increase the evolution of resistance against one of

the 11 species of herbivores.

Although the genetic correlation structure did not signifi-

cantly affect the expected responses to selection for the resistances

to the remaining species, these species do exhibit some elements

of diffuse interactions. In particular, the evolutionary responses

of 10 resistance traits, involving nine species, were influenced

by whether one or more other herbivore species were present, as

simulated by setting the selection gradient associated with those

species to zero. For five of these resistance traits (involving five

different species), there was no net effect of the correlation struc-

ture, primarily because the positive effects of some herbivores

on the expected response to selection of a focal herbivore can-

celed the negative effects of other herbivores. For instance, while

four species acted to constrain horsenettle’s evolution of resis-

tance to Epitrix, three other species acted to facilitate its evolu-

tion. These findings suggest that whether the coexisting herbivore

community influences the evolutionary interactions between a

particular herbivore and the host plant may vary from one loca-

tion to another if the herbivore-species composition differs among

locations.

The analyses in which one herbivore was mathematically

removed at a time may have underestimated the true effect that

removing a herbivore from a community would have in nature.

Specifically, the presence or absence of a particular herbivore

might affect the abundance of other herbivores due to ecologi-

cal interactions. Such effects have been demonstrated for several

members of the herbivore community of horsenettle (Wise and

Weinberg 2002; Wise 2009, 2010). Consider the predictions for

resistance to the frugivores in this study. Even with the constrain-

ing effects of negative genetic correlations, evolution is predicted

to increase resistance to frugivory by Leptinotarsa, Zonosemata,

and Microtus. However, the presence of competitive effects among

the three species suggests that reducing the damage by any one

1 0 EVOLUTION 2013


of the three will increase the amount of damage by the other

two. Although the current analysis did not take into account such

ecological effects, these effects represent an additional diffuse

element to the interactions, suggesting that our analysis is conser-

vative in detecting diffuse interactions.

Another potential source of diffuse coevolution not exam-

ined in these analyses is nonadditive fitness impact of different

herbivores. Specifically, the combined effect of damage on the

fitness of a host plant by different species of herbivores may not

equal the sum of the individual impacts if the herbivores were

feeding alone (Hougen-Eitzman and Rausher 1994; Morris et al.

2007). Such nonadditivity in impact may be evidenced by sig-

nificant correlational-selection gradients in a quadratic-selection

analysis (Pilson 1996; Juenger and Bergelson 1998; Wise 2003).

However, analyses of nonadditivity were beyond the scope of the

current paper, which focused strictly on directional selection.

The great majority of the direct selection favored in-

creased resistance (e.g., eight of the nine statistically significant

directional-selection gradients in these analyses were in the direc-

tion of increasing resistance.) At the same time, from 71 to 85%

(depending on the model) of the nominally significant effects of

individual herbivores on the predicted responses to selection by

other individual herbivores were constraining (i.e., they reduced

the selection for resistance). This general pattern of constraint

in the presence of direct selection to increase resistance resulted

primarily from the negative genetic correlations among the re-

sistance traits that would have been selected for in the absence

of other herbivores. In our study, only 17% of the 91 possible

pairwise correlations were significantly negative, a result that is

in line with those reported by other investigators in other sys-

tems (Simms and Rausher 1989; Maddox and Root 1990; Roche

and Fritz 1997; Agrawal 2005; Johnson and Agrawal 2007). Two

inferences may be drawn from these patterns. First, substantial

diffuse constraints can arise even with relatively few negative

genetic correlations. And second, diffuse constraints mediated

through negative genetic correlations may be relatively common

across plant-herbivore communities.

DIFFERENCES AMONG FEEDING GUILDS

A second objective of our investigation was to determine whether

differences in selection imposed by herbivores, as well as the

response to selection, differed among feeding guilds. A num-

ber of patterns are evident when the herbivores of S. carolinense

are categorized into four different guilds: folivores, stem feeders,

florivores, and frugivores. First, the magnitude of selection differs

substantially among the guilds, with folivores and stem feeders

generally imposing weak selection, florivores imposing interme-

diate levels of selection, and frugivores imposing the strongest

selection. This pattern generally corresponds to the “value” of

the different tissues in contributing to plant fitness. By feeding

on fruits, the frugivores exert the strongest selection presum-

ably because their damage directly destroys developing seeds.

The fitness reduction they cause per unit of tissue consumed is

thus presumably greater than for the other guilds. Furthermore,

plants have more time and more options for compensating for

and thus tolerating damage to leaves and flowers than damage to

fruits (Wise and Cummins 2006).

One surprising finding was significant directional selection

to decrease resistance to the stem borer, T. trinotata. When the

leaf- and flower-number covariates were included in the selec-

tion model, selection for decreased resistance to Trichobaris dis-

appeared, indicating that this selection was mediated by selec-

tion on leaf and/or flower production. Indeed, the likelihood of a

ramet being bored by Trichobaris was positively correlated with

the number of leaves and flower buds on the ramet. This find-

ing could be explained either by borers preferring larger, more

vigorous ramets, or borers having a beneficial effect on ramet

growth and fitness. Although the latter explanation seems less

likely, this study did not involve manipulations of plant size

or herbivory, and thus we cannot distinguish between the two

alternatives.

Similarly, a positive correlation between leaf damage by the

flea beetle E. fuscula and flower-bud production resulted in a dis-

crepancy between the selection gradients calculated for Epitrix

resistance in the two models. Epitrix damage has previously been

shown in a manipulative study to be detrimental to horsenet-

tle reproduction (Wise and Sacchi 1996). Therefore, it seems

more likely that vigorous, abundantly flowering plants may attract

Epitrix, rather than damage by the beetles causing an increase in

flowering. An important evolutionary implication of this relation-

ship is that direct selection to increase flower production would

entail indirect selection to decrease resistance to Epitrix. There-

fore, a preference of Epitrix for more vigorous plants may help

explain why Epitrix remains one of the most abundant and dam-

aging herbivores in horsenettle populations, even in the midst of

genetic potential for increased resistance.

Compared to the other guilds, the interaction between fru-

givores and horsenettle appears to be more highly diffuse and

more strongly constrained by the presence of other herbivores.

In particular, the only three species showing significant effects of

correlation structure after a Bonferroni correction were the three

frugivores. In contrast, resistance to none of the six folivores and

only one of the three florivores were highly significantly con-

strained by the presence of other herbivores. Thus, it appears that

the more strongly a guild exerts direct selection for resistance,

the more susceptible the plant may be to evolutionary constraints

caused by diffuse interactions that result from genetic correlations

in resistance.

EVOLUTION 2013 1 1


ConclusionAmple genetic variation in this population of S. carolinense in the

face of directional selection for increased resistance suggests that

the plants should be able to evolve greater levels of resistance.

Still, the plants suffer high levels of damage to all types of tis-

sues. The picture that is emerging of the evolutionary-ecology of

herbivore communities that share a host plant is that having to cope

with multiple herbivores makes it harder for a plant to evolve re-

sistance to any one species of herbivore. Diffuse phenomena such

as competitive interactions among herbivores, nonadditivity of

fitness impact of different herbivores, and negative genetic corre-

lations in resistance can act to diminish the advantage of increased

resistance or to limit a plant’s ability to evolve increased resistance

in response to selection. As such, the presence of one herbivore

may facilitate the persistence of other herbivores that the host plant

might otherwise have been able to eliminate. Through diffuse

coevolution, multiple-herbivore communities may enjoy greater

long-term ecological stability than single-plant/single-herbivore

relationships.

ACKNOWLEDGMENTSWe thank J. A. Leachman and J. B. Hebert for technical support withthe field study and C. P. Blair for helpful comments on the manuscript.The University of Virginia’s Blandy Experimental Farm provided finan-cial and logistical support through a graduate student fellowship to MJW.Financial support was also provided by a U.S. EPA STAR fellowship toMJW (Number U-915654) and an National Science Foundation Disserta-tion Improvement Grant (DEB-00–73176) to MDR and MJW. MJW wasalso supported during the writing of the paper by National Science Foun-dation grant DEB-0614395/0745562 to D. E. Carr and M. D. Eubanks.Any opinions, findings, and conclusions expressed in this material arethose of the authors and do not necessarily reflect the views of the U.S.Environmental Protection Agency or the National Science Foundation.

LITERATURE CITEDAgrawal, A. A. 2005. Natural selection on common milkweed (Asclepias

syriaca) by a community of specialized insect herbivores. Evol. Ecol.Res. 7:651–667.

Agrawal, A. F., and J. R. Stinchcombe. 2009. How much do genetic covari-ances alter the rate of adaptation? Proc. R. Soc. Lond. B 276:1183–1191.

Anderson, L. L., and K. N. Paige. 2003. Multiple herbivores and coevolution-ary interactions in an Ipomopsis hybrid swarm. Evol. Ecol. 17:139–156.

Bailey, T. E., and L. T. Kok. 1978. Insects associated with horsenettle (Solanumcarolinense L.) in southwest Virginia. VA J. Sci. 29:37.

Bassett, I. J., and D. B. Munro. 1986. The biology of Canadian weeds. 78.Solanum carolinense L. and Solanum rostratum Dunal. Can. J. Plant Sci.66:977–991.

Blows, M. W., and A. A. Hoffmann. 2005. A reassessment of genetic limitsto evolutionary change. Ecology 86:1371–1384.

Burke, H. R. 1963. Coleoptera associated with three species of Solanum inTexas. Southwest. Nat. 8:53–56.

Cipollini, M. L., and D. J. Levey. 1997. Why are some fruits toxic? Glycoalka-loids in Solanum and fruit choice by vertebrates. Ecology 78:782–798.

Crawley, M. J. 1989. Insect herbivores and plant population dynamics. Annu.Rev. Entomol. 34:531–564.

———. 1999. Herbivory. Pp. 199–217 in M. C. Press, J. D. Scholes, andM. G. Barker, eds. Physiological plant ecology. Blackwell Science Ltd.,Oxford, U.K.

Fox, L. R. 1981. Defense and dynamics in plant-herbivore systems. Am. Zool.21:853–864.

Gould, F. 1988. Genetics of pairwise and multispecies plant-herbivore coevo-lution. Pp. 13–55 in K. C. Spencer, ed. Chemical mediation of coevolu-tion. Academic Press, Inc., San Diego.

Hendrix, S. D. 1988. Herbivory and its impact on plant reproduction.Pp. 246–263 in J. Lovett Doust, and L. Lovett Doust, eds. Plant re-productive ecology: patterns and strategies. Oxford Univ. Press, NewYork.

Hougen-Eitzman, D., and M. D. Rausher. 1994. Interactions between herbiv-orous insects and plant-insect coevolution. Am. Nat. 143:677–697.

Ilnicki, R. D., T. F. Tisdell, S. N. Fertig, and A. H. Furrer, Jr. 1962. Lifehistory studies as related to weed control in the Northeast—horse nettle.Agricultural experimental station, University of Rhode Island, Kingston,RI.

Inouye, B., and J. R. Stinchcombe. 2001. Relationships between ecological in-teraction modifications and diffuse coevolution: similarities, differences,and causal links. Oikos 95:353–360.

Iwao, K., and M. D. Rausher. 1997. Evolution of plant resistance to multipleherbivores: quantifying diffuse coevolution. Am. Nat. 149:316–335.

Janzen, D. H. 1980. When is it coevolution? Evolution 4:611–612.Johnson, M. T. J., and A. A. Agrawal. 2007. Covariation and composi-

tion of arthropod species across plant genotypes of evening primrose,Oenothera biennis. Oikos 116:941–956.

Juenger, T., and J. Bergelson. 1998. Pairwise versus diffuse natural selec-tion and the multiple herbivores of scarlet gilia, Ipomopsis aggregata.Evolution 52:1583–1592.

Lande, R. 1979. Quantitative genetic analysis of multivariate evolution, ap-plied to brain:body size allometry. Evolution 33:402–416.

Lande, R., and S. J. Arnold. 1983. The measurement of selection on correlatedcharacters. Evolution 37:1210–1226.

Lankau, R. A., and S. Y. Strauss. 2008. Community complexity drives patternsof natural selection on a chemical defense of Brassica nigra. Am. Nat.171:150–161.

Maddox, G. D., and R. B. Root. 1990. Structure of the encounter betweengoldenrod (Solidago altissima) and its diverse insect fauna. Ecology71:2115–2124.

Martin, A. C., H. S. Zim, and A. L. Nelson. 1951. American wildlife andplants. McGraw-Hill Book Company, New York.

Morris, W. F., R. A. Hufbauer, A. A. Agrawal, J. D. Bever, V. A. Borowicz,G. S. Gilbert, J. L. Maron, C. E. Mitchell, I. M. Parker, A. G. Power,et al. 2007. Direct and interactive effects of enemies and mutualists onplant performance: a meta-analysis. Ecology 88:1021–1029.

Nichols, R. L., J. Cardina, and T. P. Gaines. 1991. Growth, reproductionand chemical composition of horsenettle (Solanum carolinense). WeedTechnol. 5:513–520.

Nichols, R. L., J. Cardina, R. L. Lynch, N. A. Minton, and H. D. Wells. 1992.Insects, nematodes, and pathogens associated with horsenettle (Solanum

carolinense) in Bermudagrass (Cynodon dactylon) pastures. Weed Sci.40:320–325.

Pilson, D. 1996. Two herbivores and constraints on selection for resistance inBrassica rapa. Evolution 50:1492–1500.

Rausher, M. D. 1996. Genetic analysis of coevolution between plants and theirnatural enemies. Trends Genet. 12:212–217.

Roach, D. A., and R. D. Wulff. 1987. Maternal effects in plants. Annu. Rev.Ecol. Syst. 18:209–235.

Roche, B. M., and R. S. Fritz. 1997. Genetics of resistance of Salix sericea toa diverse community of herbivores. Evolution 51:1490–1498.

1 2 EVOLUTION 2013


Simms, E. L., and M. D. Rausher. 1989. The evolution of resistance to her-bivory in Ipomoea purpurea. II. Natural selection by insects and costsof resistance. Evolution 43:573–585.

Smith, R. A., and M. D. Rausher. 2008. Selection for character displacementis constrained by the genetic architecture of floral traits in the ivyleafmorning glory. Evolution 62:2829–2841.

Stinchcombe, J. R., and M. D. Rausher. 2001. Diffuse selection on resistanceto deer herbivory in the ivyleaf morning glory, Ipomoea hederacea. Am.Nat. 158:376–388.

Strauss, S. Y., and R. E. Irwin. 2004. Ecological and evolutionary consequen-ces of multispecies plant-animal interactions. Annu. Rev. Ecol. Evol.Syst. 35:435–466.

Strauss, S. Y., H. F. Sahli, and J. K. Conner. 2005. Toward a moretrait-centered approach to diffuse (co)evolution. New Phytol. 165:81–90.

Thacker, J. D., J. Bordner, and C. Bumgardner. 1990. Carolinoside: a phytos-teroidal glycoside from Solanum carolinense. Phytochemistry 29:2965–2970.

Thompson, J. N. 1982. Interaction and coevolution. John Wiley & Sons, NewYork.

———. 1994. The coevolutionary process. The University of Chicago Press,Chicago.

Via, S. 1984. The quantitative genetics of polyphagy in an insect herbivore.II. Genetic correlations in larval performance within and among hostplants. Evolution 38L:896–905.

Wise, M. J. 2003. The ecological genetics of plant resistance to herbivory:evolutionary constraints imposed by a multiple-herbivore community.

Department of Biology. Duke University, Durham, North Carolina. P.169.

———. 2007a. Evolutionary ecology of resistance to herbivory: an investiga-tion of potential genetic constraints in the multiple-herbivore communityof Solanum carolinense. New Phytol. 175:773–784.

———. 2007b. The herbivores of horsenettle, Solanum carolinense, in north-ern Virginia: natural history and damage assessment. Southeast. Nat.6:505–522.

———. 2009. Competition among herbivores of Solanum carolinense as aconstraint on the evolution of host-plant resistance. Evol. Ecol. 23:347–361.

———. 2010. Diffuse interactions between two herbivores and constraintson the evolution of resistance in horsenettle (Solanum carolinense).Arthropod Plant Interact. 4:159–164.

Wise, M. J., and J. J. Cummins. 2006. Strategies of Solanum carolinense forregulating maternal investment in response to foliar and floral herbivory.J. Ecol. 94:629–636.

———. 2007. Herbivory as an agent of natural selection for floral-sex ratioin horsenettle (Solanum carolinense). Evol. Ecol. Res. 9:1319–1328.

Wise, M. J., and C. F. Sacchi. 1996. Impact of two specialist insect herbi-vores on reproduction of horse nettle, Solanum carolinense. Oecologia108:328–337.

Wise, M. J., and A. M. Weinberg. 2002. Prior flea beetle herbivory affectsoviposition preference and larval performance of a potato beetle on theirshared host plant. Ecol. Entomol. 27:115–122.

Associate Editor: T. Craig

Supporting InformationAdditional Supporting information may be found in the online version of this article at the publisher’s website:

Table S1. Statistics for the bootstrap analysis of the effects of genetic covariances on responses to selection (in a model not

including leaf number and flower number as covariates).

Table S2. Statistics for the bootstrap analysis of the effects of genetic covariances on responses to selection (in a model including

leaf number and flower number as covariates).

Table S3. Statistics for the bootstrap analyses of the effects of individual herbivores on responses to selection for resistance to the

other herbivores (in a model not including leaf number and flower number as covariates).

Table S4. Statistics for the bootstrap analyses of the effects of individual herbivores on responses to selection for resistance to the

other herbivores (in a model including leaf number and flower number as covariates).

EVOLUTION 2013 1 3

Documents

EVOLUTION OF RESISTANCE TO A MULTIPLEHERBIVORE …