Community Complexity Drives Patterns of Natural Selection on a

vol. 171, no. 2

the american naturalist



february 2008

Community Complexity Drives Patterns of Natural Selection on a Chemical Defense of Brassica nigra

Richard A. Lankau* and Sharon Y. Strauss†

Section of Evolution and Ecology, University of California, Davis, California 95616 Submitted April 11, 2007; Accepted August 16, 2007; Electronically published December 14, 2007 Online enhancements: appendix, tables, figures.

abstract: Plants interact with many different species throughout their life cycle. Recent work has shown that the ecological effects of multispecies interactions are often not predictable from studies of the component pairwise interactions. Little is known about how multispecies interactions affect the evolution of ecologically important traits. We tested the direct and interactive effects of inter- and intraspecific competition, as well as of two abundant herbivore species (a generalist folivore and a specialist aphid), on the selective value of a defensive chemical compound in Brassica nigra. We found that investment in chemical defense was favored in interspecific competition but disfavored in intraspecific competition and that this pattern of selection was dependent on the presence of both herbivores, suggesting that selection will depend on the rarity or commonness of these species. These results show that the selective value of ecologically important traits depends on the complicated web of interactions present in diverse natural communities and that fluctuations in community composition may maintain genetic variation in such traits. Keywords: glucosinolate, competition, herbivory, specialist, generalist.

The roles that interactions between species play in population and community dynamics, as well as the evolutionary dynamics within species, have long been a central focus of ecology. While the complexity of the natural world is well recognized, the bulk of the research in this area has * Present address: Illinois Natural History Survey, 1816 South Oak Street, Champaign, Illinois 61802; e-mail: [email protected]. †

E-mail: [email protected].

Am. Nat. 2008. Vol. 171, pp. 150–161. 䉷 2007 by The University of Chicago. 0003-0147/2008/17102-42536$15.00. All rights reserved. DOI: 10.1086/524959

focused on pairwise interactions between species. However, a growing body of literature attests to the importance of multispecies interactions for both ecological and evolutionary processes (Holt and Lawton 1994; Hamback and Beckerman 2003; Agrawal 2004; Strauss and Irwin 2004). A key finding in many of these studies is that the outcome of multispecies interactions often is not predictable from the study of pairwise interactions in isolation (Strauss and Irwin 2004). Plants in natural communities interact with a multitude of different species throughout their life cycle, including herbivores, pathogens, mutualists, and plant neighbors. Consequently, plant traits that reduce the impacts of antagonists or enhance the impact of mutualists are likely to be selectively favored (Mauricio and Rausher 1997; Pilson 2000; Shonle and Bergelson 2000; Stinchcombe 2002; Agrawal 2005). Nevertheless, high levels of additive genetic variation remain in these traits, suggesting that selection is weak or inconsistent, or that some form of balancing selection is acting on populations (Berenbaum and Zangerl 1992; Fritz and Simms 1992; Mauricio and Rausher 1997; Rausher and Simms 1989). This variation may be partly due to trade-offs that occur when traits affect multiple interactions simultaneously (Stinchcombe and Rausher 2002; Strauss et al. 2002; Tiffin 2002; reviewed in Strauss and Irwin 2004). In this study, we focused on the interactions between competing plants and two dominant herbivores in an attempt to understand how these multiway interactions affect the magnitude and direction of selection on a chemical defense trait. Secondary compounds are often studied as defenses against herbivores and pathogens; however, they may affect competitive interactions through allocational costs or direct allelopathic effects (Callaway and Aschehoug 2000; Weis and Hochberg 2000; Siemens et al. 2002). Additionally, competitors and herbivores may have interactive effects on plant fitness and selection if competitors alter the amount of herbivory a plant receives or the per capita effect of that herbivory (Hamback and Beckerman 2003; Agrawal 2004; Haag et al. 2004). Plant neighbors have been shown to both increase (as-

Community Complexity and Natural Selection sociational susceptibility) and decrease (associational resistance) the level of herbivory on target plants (Tahvanainen and Root 1972; Futuyma and Wasserman 1980; Hay 1986; Andow 1991; White and Whitham 2000; Gomez et al. 2001). Associational susceptibility often occurs when herbivores build up to high numbers on preferred hosts and then spill over onto less preferred ones (Futuyma and Wasserman 1980; White and Whitham 2000) or when herbivores are attracted to one species for other reasons (nectar, predator refuges) and then feed on nearby plants of other species (Karban 1997; Agrawal 2004). Associational resistance often occurs when herbivores avoid areas with unpalatable plants, resulting in lower herbivory rates on palatable plants growing nearby (Tahvanainen and Root 1972; Hay 1986; Gomez et al. 2001). Additionally, the identity of plant neighbors can affect the community of herbivores attracted to a patch, with generalists favoring diverse patches while specialists favor monocultures of their host plant (Root 1973; Kareiva 1982; Andow 1991; Banks 1998). While a substantial amount of research has focused on the ecological consequences of multiple interactions, the evolutionary consequences are much less understood (Iwao and Rausher 1997; Stinchcombe and Rausher 2001; Strauss et al. 2005). Plant neighbors could affect selection on defense traits by increasing the abundance of herbivores or the per capita effect of herbivory. Additionally, plant defenses, especially secondary compounds, often affect different herbivores in different ways. The few communitywide studies find a mix of positive, negative, and neutral correlations in resistances to herbivores of different diet specializations or feeding guilds (Maddox and Root 1990; Roche and Fritz 1997; Leimu and Koricheva 2006). For instance, while generalists are often deterred by these compounds, coevolved specialists may be unaffected, or even use the chemicals as cues for host location or feeding and oviposition stimulants (Da Costa 1971; Siemens and Mitchell-Olds 1996; Berenbaum and Zangerl 1998; Kliebenstein et al. 2002; Macel and Vrieling 2003). Van der Meijden (1996) suggested that the net selection on chemical defenses would be dependent on the ratio of specialist and generalist herbivores in a community. Lankau (2007) showed this to be true for Brassica nigra and its dominant generalist and specialist herbivores within its introduced range in California. Therefore, plant neighbors, aside from any direct interactions, may also indirectly affect selection on defenses by changing the composition of the herbivore community attracted to an area. Because the relative abundance of specialist and generalist herbivores (as well as other plant associates) changes with plant community composition, we expect plants to experience a range of selective pressures when the species goes from rare to common across local patches or com-

151

munities. Competitive interactions change from mostly interspecific to mostly intraspecific across this gradient, and herbivore communities change from generalist to specialist dominated. Shifting plant community composition over space or time could thus provide a powerful stabilizing effect, maintaining genetic diversity in plant traits through direct or indirect interactions with shifting community members. To explore the complex ways through which herbivores and plant competitors could interact to affect selection on a plant defense, we performed a multifactorial experiment with B. nigra in which we manipulated the intensity of competition and the identity of neighboring species and estimated selection on sinigrin concentration (the primary secondary compound produced by this species). We also manipulated the presence and absence of generalist mollusks and specialist aphids, the two most abundant herbivore groups at our site. We used artificial selection on field-collected plants to create B. nigra families with a range of variation in sinigrin concentration. Previous research has shown that the molluskan folivores select for increased sinigrin concentrations, while the specialist aphid selects for decreased levels of the trait (Lankau 2007). We therefore predicted that sinigrin would be selectively favored when plants grew with heterospecific neighbors (generalist herbivore dominated) but disfavored when grown with conspecifics (specialist dominated). By independently manipulating multiple species interactions, we were able to disentangle their direct and indirect effects as selective agents on sinigrin concentration. Methods Study System Brassica nigra is an introduced Eurasian annual in the central valley of California and is fed on by introduced and native generalists as well as by several introduced oligophagous species specialized on the Brassicaceae. In these areas, B. nigra occurs both in dense monospecific stands and as individual plants surrounded by heterospecific competitors (R. A. Lankau, personal observation). Plants in the Brassicaceae all produce glucosinolates, a class of amino acid derived from secondary compounds. In B. nigra, sinigrin (allyl-glucosinolate) represents 90%– 99% of the total glucosinolate concentration and has a heritable basis (Traw 2002; table A1 in the online edition of the American Naturalist). When combined with the myrosinase enzyme, glucosinolates break down into a number of toxic by-products involved in herbivore and disease resistance (e.g., Raybould and Moyes 2001; Agrawal and Kurashige 2003; Kliebenstein 2004).

152 The American Naturalist This study was conducted in an organically managed old field, left uncultivated for 5 years, at the University of California, Davis, student farms. The plant community was dominated by introduced and native annuals, such as the forbs Amsinckia menziesii, Malva parviflora, and Sonchus oleraceus, and the grasses Avena fatua and Bromus diandrus. Brassica nigra occurs naturally at the site but is not abundant. By far the most numerous herbivores observed were the generalist slug Deroceras reticulatus (Muller; the gray garden slug) and the specialist cabbage aphid Brevicoryne brassicae (L.). Brevicoryne brassicae individuals were present from April until June (when leaves senesced), appearing mainly on stems, flowers, and developing fruits.

the presence or absence of plant competitors, and the identity of the competing species (inter- or intraspecific competitors). We will refer to these treatments as ⳲMol, ⳲAph, ⳲComp, and Het or Con (heterospecific or conspecific neighbor). The target plant in each plot was a B. nigra individual from one of the 32 artificially selected families mentioned above. We grew one individual from each family in each treatment combination. With 32 families and 16 treatment combinations, this resulted in 512 experimental plots.

Plant Source

Neighbor Identity Treatment

In summer 2003, we collected B. nigra seeds from six sites around the Sacramento Valley. From these collections, we grew in the greenhouse three seeds from each of 256 maternal half-sib families. Based on the average sinigrin concentration, we split families into high and low sinigrin lines (128 families each). Within each line, we randomly assigned families to mating pairs (using the most extreme individual per family). This process was repeated for three successive generations, resulting in 32 full sibships that expressed increased variance in the trait due to assortative mating and within-family selection; final sinigrin concentrations were still representative of natural levels (45th and 66th percentiles in natural populations for low and high sibships, respectively). No families were ever reused in the matings; thus, all families were equally independent of the others, regardless of the selection line. Experimental Design To determine how species’ interactions directly and indirectly affect selection on sinigrin concentration, we performed a target-neighbor experiment. Plants were grown from November 2004 to June 2005 in pairs in a variety of experimental conditions. Our aim was to manipulate both herbivore community and plant neighbors. At the field site during this season, the most abundant herbivores, overwhelmingly, were generalist mollusks and B. brassicae specialist aphids. Our treatments in fact manipulated all the specialist and generalist herbivores at the site, but since these groups were each dominated by a single group, they are most conservatively viewed as mollusk and aphid manipulations. The degree to which we can generalize from mollusk and aphid impacts to impacts of generalists and specialists as a whole is discussed later. In total, we performed a 2 # 2 # 2 # 2 factorial experiment, factorially manipulating the presence or absence of mollusks, the presence or absence of B. brassicae aphids,

The target plant was grown with either a conspecific or heterospecific neighbor. Conspecific neighbors were B. nigra individuals grown from locally collected seeds not involved in the selection scheme. To avoid idiosyncratic effects of any single heterospecific species, the heterospecific neighbors were individuals of one of three co-occurring annual species from different plant families: A. menziesii (Boraginaceae), M. parviflora (Malvaceae), or S. oleraceus (Asteraceae). These three species were chosen because, although phylogenetically diverse, they had similar life histories and phenologies as B. nigra (all annuals that germinate in November–January and senesce by midsummer) and were common in the experimental field. The purpose of this manipulation was to contrast the effects of intraand interspecific competition on selection; therefore, the experiment was balanced with respect to con- and heterospecific neighbors. Thus, in one-third of heterospecific competition plots, B. nigra target plants had a Sonchus neighbor, in one-third they had an Amsinckia neighbor, and in the remaining third they had a Malva neighbor (85, 85, and 86 replicates per species, respectively). This assignment was stratified with respect to the predicted (midparent) sinigrin concentration of the target plant so that each neighbor species experienced a similar range of sinigrin concentrations. We planted six seeds of the appropriate species (neighbors) and family (targets) approximately 12 cm apart in each plot. After 2 weeks, plots were thinned to one plant per position (target or neighbor), and positions with no germinants were replanted with greenhouse-germinated seedlings (34 out of 512 plots). We included germination time and the source of the seedling (field or greenhouse germinated) as terms in subsequent analyses to control for these differences. While both terms were highly significant, they did not interact with any experimental treatments (F ! 2.4, P ! .12 for all interactions).

Community Complexity and Natural Selection Competition Treatment The intensity of competition was manipulated with below ground partitions and above ground cages. Before planting, a 46-cm-deep # 7.62-cm-wide trench was cut into every plot with a Ditch Witch trencher (Charles Machine Works). In half of the plots, a 45-cm-deep # 61 cmlong # 3-mm-thick plastic sheet was inserted into the trench, and then all trenches were refilled with the displaced soil. In the same plots with the root partitions, two 31-cm-high # 61-cm-long sections of chicken wire were used to separate plant shoots—the chicken wire sections were shaped into opposing semicircles to guide the plants’ growth away from each other. These treatments were effective in reducing competitive effects of neighbors. When barriers were present, target and neighbor plant size were not correlated (r p 0.082, P p .22, df p 252); however, they were significantly, albeit weakly, negatively correlated in the absence of barriers (r p ⫺0.176, P ! .01, df p 252), suggesting that either competition or other negative effects of neighbors (e.g., allelopathy, microbially mediated interactions, etc.) were reduced by the physical barriers. At the end of the season, we confirmed that no roots had penetrated the barriers. Herbivore Removal Treatments We manipulated the two dominant herbivore groups at our site; aphids were manually removed (using a paintbrush) from target plants (and the neighbor if it was also B. nigra) in the aphid removal treatment every 2 weeks. We surrounded all target and neighbor plants with a 16oz plastic cup (with the bottom removed), and for those in the generalist absent treatment only, we placed a ∼6.5cm-wide copper strip (Snail-Barr, Custom Copper) around the top of the cup. The copper barriers reduced percent leaf area removed by fourfold at the peak of mollusk activity (1.64% vs. 6.59%, F p 90.28, df p 1, 492, P ! .0001), while the aphid removals resulted in a more than 20-fold decrease in peak aphid abundance per plant (92 vs. 1,967 aphids/plant, F p 41.10, df p 1, 468, P ! .0001). Measurements On target plants, we measured damage levels, aphid population size, sinigrin content, plant biomass, and seed and fruit production. Generalist damage was estimated visually several times throughout the experiment; the percentage of leaf area missing from each leaf was averaged over all leaves to estimate damage. We calibrated nondestructive visual estimates by scanning 50 damaged B. nigra leaves, we digitally estimated the area removed (Scion Image), and then we took images of known damage levels to the

153

field. Specialist aphid population sizes on all B. nigra plants were estimated once, in late May, when populations were near their peak. When plants had 10 leaves, after generalist damage but before aphids colonized plants (i.e., before the specialist treatments were imposed), we collected approximately 5 mg of leaf tissue (five hole punches) from the seventh leaf of each B. nigra plant in order to measure sinigrin concentration using HPLC (see Kliebenstein et al. 2001; Lankau 2007 for details). Generalist mollusk damage begins immediately after seedling emergence (at the cotyledon stage), so it was not possible to measure sinigrin before mollusk damage. Basal stem diameter (a proxy for biomass) was measured on all plants in April, before the emergence of the specialist aphids. When plants had completely senesced in late June, all reproductive branches (those with fruits directly attached) and attached fruits were collected and weighed. This measure (mass of reproductive branches) explained over 98% of the variance in fruit number (R 2 p 0.985, n p 27). The average number of seeds per fruit and the average mass per seed were determined from 15 fruits randomly selected from each plant. These three fitness components were multiplied to get a final estimate of female fitness, equal to the total biomass of seeds produced, that combines seed quantity and mass. This method does not include male fitness, which may respond differently to the experimental treatments. Selection Analysis To estimate the magnitude and direction of selection acting on sinigrin concentration in the various experimental treatments, we performed a modified genotypic selection analysis. Selection differentials are estimated as the covariance between a trait value and relative fitness. If traits and fitness are measured on the same individual, selection analyses may be biased due to environmental covariances that may affect both quantities simultaneously (Rausher 1992; appendix in the online edition of the American Naturalist). For this reason, we measured the fitness of the target plant but estimated that plant’s sinigrin content (trait value) from full siblings in other plots. Our experimental design had only one individual per family per treatment. Therefore, it was necessary to combine across treatments to get the required within-family replication for trait estimates. This approach would only be justified if at least one treatment did not affect the expression of sinigrin concentration. Recall that the specialist aphid removal treatment was randomly assigned to individuals after sinigrin concentrations were measured, so it was unbiased with respect to sinigrin concentration, and, accordingly, there were no differences in sinigrin concen-

154 The American Naturalist tration in the aphid presence/absence treatments (F p 0.04, df p 1, 31, P p .83). Additionally, an ANOVA on sinigrin concentration revealed that the neighbor identity (NI) treatment also had no significant effect on sinigrin expression (P p .51 for main effect, P 1 .20 for all interactions; table A1), nor were there any interactions between NI treatment and family on sinigrin content (table A1). We tested this latter effect by including family and all family-by-treatment interactions as random effects in a mixed-model ANOVA using the lmer package in the R statistical language (table A1; Bates and Sarkar 2006). In contrast, the competition and mollusk removal treatments did affect sinigrin concentrations (table A1). Therefore, it would be inappropriate to average across the competition or mollusk treatments. Because sinigrin expression was affected by four treatment combinations (⫺Mol, ⫺Comp; ⫺Mol, ⫹Comp; ⫹Mol, ⫺Comp; ⫹Mol, ⫹Comp), we estimated sinigrin content of a target plant by averaging the contents of its three full siblings in the treatments in which sinigrin was unaffected—the aphid and NI treatments. This approach is best understood from table A2 in the online edition of the American Naturalist, but we verbally walk through one example here. To estimate sinigrin of a target plant in family 1 in the ⫺Mol, ⫺Comp, ⫺Aph, Con treatment, we averaged sinigrin content for its three full siblings across the different aphid and neighbor identity treatments that do not affect sinigrin but within the same mollusk and competition treatments that do affect sinigrin concentration (in this case, ⫺Mol, ⫺Comp, ⫺Aph, Het; ⫺Mol, ⫺Comp, ⫹Aph, Con; ⫺Mol, ⫺Comp, ⫹Aph, Het). The same process was repeated independently for each combination of mollusk and competition treatment. This estimate was then used as the explanatory variable to estimate selection. This process produced a unique estimate of sinigrin concentration for each individual that was free of any environmental covariance with the fitness of that individual. Therefore, any covariance between fitness and our measure of sinigrin concentration can only occur through a genetic link; any environmental effects on sinigrin concentration (such as induction by herbivores or volatile emissions of neighboring plants) only add unexplained noise to our analysis, making our results very conservative. To ensure that comparisons of selection between treatments were unaffected by different mean sinigrin concentrations, we standardized sinigrin concentrations within each treatment combination (16 total) to have a mean of 0 and a standard deviation of 1. Our measure of selection is then the covariance of estimated sinigrin concentration and the relative fitness of an individual plant within that experimental treatment combination (Price 1970). Relative fitness was calculated separately for each treatment com-

bination as the absolute fitness of an individual divided by the mean fitness in that treatment combination. An analytic and simulation study of our modified method as compared with traditional phenotypic and genotypic selection analyses showed that our approach eliminated any environmental bias and always produced a conservative estimate of selection, with the explanatory power dependent on the heritability of the trait in question (appendix). The estimated heritabilities and coefficients of genetic variation for sinigrin concentration, as well as the significance of any gene-by-treatment interactions, within the four combinations of mollusk and competition treatments are presented in table A3 in the online edition of the American Naturalist. Estimated heritability in one group (⫹Mol, ⫺Comp) was low, was not statistically significant, and should be interpreted cautiously; however, the resulting selection estimates are still very conservative. To determine whether selection on sinigrin is affected by community composition, we performed a fully factorial ANCOVA with the four treatments (competition, mollusk removal, aphid removal, and neighbor identity), the standardized sinigrin concentration, and all interactions as explanatory variables, and relative fitness as the dependent variable. A significant interaction involving sinigrin concentration and an experimental treatment implies that selection (i.e., the slope of the regression between sinigrin concentration and relative fitness) is significantly different in the two levels. With standardized trait values, the regression slope between trait and fitness is mathematically equivalent to the covariance (Price 1970). Selection analysis, combined with experimental manipulations, provides a powerful method to determine the selective agents acting on a trait, even if the exact mechanism is not known. For instance, aphids may exert selection on leaf sinigrin levels through a diversity of mechanisms even though they do not feed on leaves. Our analysis seeks to determine the most important agents creating selection pressures on sinigrin concentration by testing whether (1) there is a significant genetic covariance between sinigrin concentration and fitness and (2) this covariance differs in the presence versus the absence of a given agent. In this analysis, we focus on selection differentials, which combine direct selection on a specific trait and indirect selection on correlated traits. Because they incorporate both direct and indirect selection, selection differentials provide the most accurate prediction of how trait values are expected to change across generations. Because the data violated assumptions of normality and homogeneity of variances and transformations can complicate the interpretation of selection analyses (Lande and Arnold 1983; Stanton and Thiede 2005), a stratified bootstrap procedure was used to determine 95% confidence intervals (CIs) on the estimated ANCOVA parameters (Ef-

Community Complexity and Natural Selection ron and Tibshirani 1993; Ryan 1997; Power and Moser 1999). If these CIs did not include 0, the effect was considered significant. We used accelerated bootstrap CIs as determined by the Boot library in the R statistical language (Canty 2005). The bootstrap was stratified within each treatment combination, meaning that resampling took place separately within each treatment combination to ensure that sample sizes for all resamples matched the original data set. These results agreed with those from a parametric analysis in all cases. Our ANCOVA analysis provides a direct test of whether a particular species interaction (herbivory, competition, etc.) alters the selection pressures acting on sinigrin but does not provide information on the mechanisms behind these effects. To help elucidate these mechanisms, we used structural equation modeling to explore the intercorrelations between sinigrin concentration, generalist damage, midseason plant size for targets and neighbors, specialist aphid loads on targets, target plant seed number, and average seed mass. We developed a biologically realistic a priori model structure determined by the phenology of events in the field; for instance, generalist damage was allowed to affect specialist loads, but not vice versa, because specialist aphids did not emerge until after the vast majority of mollusk damage had occurred. Damage and aphid population size were transformed to meet assumptions of normality. In all cases, we used the estimated values of sinigrin concentration to avoid bias due to environmental covariance. We used the SEM package in R to determine whether our a piori model structure provided a reasonable fit to the data in each experimental treatment (based on x2 statistics and goodness-of-fit indexes; Fox 2006). Because the intensity-of-competition treatment did not affect selection pressures (table 1), for simplicity we performed the SEM only on those treatments experiencing natural levels of competition (controls). We then tested whether the estimated path coefficients differed between our experimental treatments by constraining (a) all paths or (b) one path at a time to be equal between two experimental groups and then testing whether this produced a significant decrease in model fit compared with the model with freely estimated parameters (Grace 2006). Because the number of potential comparisons between all eight treatment combinations is huge, we focused on differences between (1) con- versus heterospecific plots with all herbivores present and (2) all herbivores present versus treatments with only the mollusks, only the aphids, or no herbivores, separately for con- and heterospecific treatments. Results The experimental treatments had little overall effect on the mean fitness of the target plants, with no treatments sig-

155

Table 1: Analysis of selection among experimental treatments Effect

Estimate

Lower 95% CI

Upper 95% CI

NI Generalist mollusks Specialist aphids Competition intensity Sinigrin concentration NI : Mol NI : Aph Mol : Aph NI : Comp Mol : Comp Aph : Comp NI : Sinigrin Mol : Sinigrin Aph : Sinigrin Comp : Sinigrin NI : Mol : Aph NI : Mol : Comp NI : Aph : Comp Mol : Aph : Comp NI : Mol : Sinigrin NI : Aph : Sinigrin Mol : Aph : Sinigrin NI : Comp : Sinigrin Mol : Comp : Sinigrin Aph : Comp : Sinigrin NI : Mol : Spec : Comp NI : Mol : Aph : Sinigrin NI : Mol : Comp : Sinigrin NI : Aph : Comp : Sinigrin Mol : Aph : Comp : Sinigrin NI : Mol : Aph : Comp : Sinigrin Germination date Seedling source

⫺.0004 .0092 .0370 ⫺.0158 .0784 ⫺.0353 ⫺.0124 ⫺.0280 .0107 ⫺.0063 .0435 .0029 ⫺.0115 ⫺.0428 ⫺.0396 .0047 .0029 ⫺.0207 ⫺.0273 .0942 .0039 ⫺.0522 ⫺.0586 .0529 .0116 .0384 .1257 .0430 .0017 .0002 ⫺.0124 ⫺.0424 .4564

⫺.0953 ⫺.0717 ⫺.0523 ⫺.1047 ⫺.0132 ⫺.1178 ⫺.0971 ⫺.1094 ⫺.0708 .0924 ⫺.0435 ⫺.0964 ⫺.0987 ⫺.1343 ⫺.1269 ⫺.0825 ⫺.0853 ⫺.1110 ⫺.1122 .0004 ⫺.0911 ⫺.1419 ⫺.1533 ⫺.0437 ⫺.0866 ⫺.0486 .0351 ⫺.0563 ⫺.0986 ⫺.0969 ⫺.1145 ⫺.0527 .2631

.0811 .1009 .1239 .0689 .1721 .0561 .0697 .0611 .1029 .0819 .1309 .0947 .0881 .0505 .0591 .0894 .9270 .0616 .0549 .1916 .0990 .0449 .0284 .1486 .1058 .1214 .2260 .1302 .0937 .0976 .0833 ⫺.0329 .6611

Note: ANCOVA results for the factorial field experiment. Relative fitness is the dependent variable, and significant interactions between the trait (sinigrin concentration) and experimental treatments show that selection on sinigrin differs significantly among those treatments. Estimates refer to the estimated parameter values for a given term in a general linear model with effect coding, so that a binary treatment is represented by 1 (yes) or ⫺1 (no). Lower and upper 95% confidence intervals (CIs) for each estimate were calculated from 2,000 bootstrapped resamples. Bold text refers to effects whose bootstrapped 95% CIs do not overlap 0. All effects deemed significant by the bootstrap analysis were also significant in a parametric test, and vice versa. NI p neighbor identity treatment, Mol p mollusk removal treatment, Aph p aphid removal treatment, Comp p intensity-of-competition treatment. Germination date and seedling source refer to the date of germination of the target plant (0 p first germinant of the experiment) and whether the target plant germinated naturally in the field or was replaced by a greenhousegerminated seedling.

nificantly affecting mean fitness singly or in combination with any others (see fig. A2 in the online edition of the American Naturalist). Despite the lack of strong treatment effects on average fitness, the pattern of natural selection

156 The American Naturalist did differ among the treatments (table 1). In fact, the lack of strong treatment effects probably resulted from the differential response of high and low sinigrin genotypes to the treatments. ANCOVA revealed that both the direction and intensity of selection on sinigrin was affected by the herbivore and plant community composition, as evidenced by the highly significant three-way interaction between generalist mollusks, specialist aphids, and the identity (con- or heterospecific) of the neighboring plant (table 1). In contrast, the intensity of competition did not appear to affect selection either directly or through interactions with any of the other treatments. We therefore present only the results from plants without experimental barriers (fig. 1; for full results, see table A4 in the online edition of the American Naturalist). Our very conservative method for removing environmental covariance sacrificed the amount of variance explained by the model in order to ensure that there was no bias in the estimates of selection (see “Methods” and the appendix). Thus, while highly significant, the three-way interaction explained a modest amount of variation (!5% of the total), and the confidence intervals around the individual selection differentials were quite large. However, the main goal of this experiment was not to determine whether the selection differentials in any given treatment combination differed from 0 but rather to test whether the pattern of selection differed between treatments. Nevertheless, in the most ecologically relevant conditions (with competition and all herbivores present, i.e., all community members), the estimated selection differentials were quite strong (fig. 1; table A4).

The selective impact of generalist mollusks and specialist aphids depended on the presence of the other herbivore and the species of neighboring plant (table 1). For instance, when plants were grown with conspecific neighbors, aphids favored increased sinigrin concentrations in the absence of mollusks but not in the presence of mollusks. This pattern of selection by aphids is reversed, however, when plants were grown with heterospecific neighbors (fig. 1). The removal of any of the interacting species weakens selection to the point that it is no longer detectable, given the statistical power of the experiment. No evidence for stabilizing or disruptive selection was found in any situation, although this design probably did not have sufficient power to detect weak quadratic selection gradients. However, our results indicate that both the direction and the intensity of selection on sinigrin depend on the community composition of the surrounding vegetation and the relative abundance of the two dominant herbivores. Generalist mollusks are most abundant during the cool, wet winter months while specialist aphids emerge in the hotter, dry months in late spring and early summer. This natural history suggests a mechanism underlying selection in which generalist damage may affect aphid specialists, but not vice versa, in this monocarpic annual. This natural history was used to generate a structural equation model, which was fit to each treatment combination shown to affect selection. Note that because our experimental herbivore removals were not perfect, a small number of aphids were present on some plants in the aphid removal treatment, and a small amount of damage occurred on some

Figure 1: Selection differentials (the covariance between standardized sinigrin concentration and relative fitness) in each of the eight treatment combinations, excluding treatments with no competition. Error bars refer to bootstrapped 99.3% confidence limits, to control for the eight implied tests of whether the selection differential differs from 0.

Community Complexity and Natural Selection

157

Figure 2: Structural equation models for selection on sinigrin for the eight treatments with competition. Variables proceed from left to right according to the timing of events in the field. The a priori model structure included only paths from a given variable to other variables to its right on the diagram (i.e., a variable can affect later events but not previous ones). For clarity, paths with correlations !0.2 (P 1 .2 ) were omitted from the figure. A–D, Conspecific neighbor. E–H, Heterospecific neighbor. A, E, Treatments with full herbivore communities. B, F, Mollusks but not aphids. C, G, Aphids but not mollusks. D, H, Aphids and mollusks removed. Solid lines refer to a positive estimate, dashed lines to a negative one. Line width corresponds to the magnitude of the effect. Asterisks in A and E refer to paths that were significantly different between the two full herbivore treatments.

plants in the mollusk removal treatment. Therefore, there are some paths present from these interactions even in treatments where herbivores were experimentally reduced. The intercorrelations among sinigrin concentration, leaf damage levels (due primarily to mollusks), midseason sizes

of both target and neighbor plants, and specialist aphid loads on the target plant were used to explain the two components of target plant fitness: seed quantity and quality (mass). With conspecific neighbors and the full complement of herbivores present (fig. 2A), high sinigrin con-

158 The American Naturalist centration was negatively correlated with leaf damage on the target Brassica nigra (fig. 2A, path a) and positively correlated with damage on the neighbor plant (fig. 2A, path b). This differential damage between target and conspecific neighbor appeared to give high sinigrin target plants a competitive edge over neighbors, as strong differences in plant size occurred by midseason. High sinigrin targets were much larger than their neighbors, whereas low sinigrin targets were much smaller than their neighbors (fig. 2A, paths a and e and paths b and f). However, the advantage gained by large plants was lessened with the arrival of specialist aphids. Aphids tended to colonize the larger of the two host plants in a plot, so high sinigrin targets suffered from high specialist loads (positive correlation between target plant size and target aphid load, and an independent negative correlation between neighbor plant size and target aphid load, fig. 2A path h and i). These aphids, in turn, had a strong negative effect on seed mass, although they did not affect seed number (fig. 2A, path l). So, by the end of the season, increased sinigrin concentration led to lower fitness (see fig. 1), primarily through increases in specialist loads. This model provided a reasonably accurate fit to the data (x 2 p 11.05, df p 13, P p .61, goodness-of-fit index p 0.93), although low sample sizes make the nonsignificant x2 statistic suspect. When mollusks (fig. 2B), aphids (fig. 2C), or both (fig. 2D) were experimentally removed, this chain of events broke down. All three models were significantly different from the model fit to the control (all herbivores present) treatment (P ! .02 for all; see fig. 2 for statistical comparisons of specific paths between treatments). As evidenced by the selection analysis, the interactions between sinigrin concentration and the two herbivore species played out differently when the target plants were grown with heterospecific neighbors. Applying the same a priori model structure to the treatment with all herbivores and heterospecific neighbors, we found that the overall model structure differed significantly from the conspecifics case (x 2 p 65.5, df p 28, P ! .0001). The positive pathway in which sinigrin reduces leaf damage on targets, leading to larger plants and more seeds, was qualitatively similar to the conspecific case. However, pathways through the neighboring plant were qualitatively different, with these correlations being essentially 0 in the heterospecific case (fig. 2E, paths b, f, g, and i). These differences probably occurred because (1) the mollusks had preferences for certain species such that their decision to forage on the neighbor plants was not strongly influenced by the defensive status of the target and (2) the colonization decisions of the specialist aphids were not influenced by the size of the neighbor plant because the neighbor was not a potential host plant in the heterospecific plots. Thus, the aphid load on a particular target plant was independent

of plant size, and the aphids were not a strong selective force in this case. Therefore, selection favored high sinigrin plants (see fig. 1) because these plants did not face the high specialist aphid loads that their siblings did in the conspecific treatment. The major paths connecting sinigrin to fitness are similar when specialists are removed (because specialists were not selective agents in this treatment; fig. 2F) but are significantly weaker in the absence of generalists (fig. 2G, 2H). Discussion Plants must interact with numerous species throughout their life cycles, but the ecological and, especially, evolutionary consequences of multispecies interactions are still poorly understood. In this study, we found that the selection pressures acting on sinigrin concentration in Brassica nigra depended on the precise mix of plant and herbivore species present in the community. These results highlight the importance of considering the community context in which natural selection operates and the myriad direct and indirect ways in which the ecological effects of traits such as secondary chemistry reverberate through communities. Sinigrin concentration had different selective values in different experimental communities, being positively, negatively, or not correlated with fitness depending on the other species present in the treatment. This experiment was designed to test whether the pattern of selection differed among treatments, and for this purpose the experiment had sufficient statistical power. Unfortunately, the large number of factorially crossed treatments necessitated small sample sizes within each treatment combination (32 individuals) even though overall sample size for the experiment was quite large (512 plots). In addition, our analytic method of estimating sinigrin content sacrificed resolution in favor of avoiding bias, which constrained our ability to test higher-order selection gradients or whether any given selection differential differed significantly from 0. Despite these drawbacks, we were able to show that community composition is an important source of variable selection on sinigrin in B. nigra. The selective impact of each herbivore species depended strongly on the presence of the other herbivore and the species identity of the neighboring plant. Together, the mollusks and aphids acted to negate each other’s selective impact when plants were grown with conspecific neighbors, suggesting that spatial and temporal variation in the relative abundance of the two herbivores could help maintain genetic variation in sinigrin concentration in dense stands of B. nigra. However, this process broke down when plants were grown with heterospecific neighbors, where net selection favored higher sinigrin concentrations. Thus,

Community Complexity and Natural Selection when B. nigra is rare in a community and growing primarily with heterospecifics, sinigrin concentrations are expected to rise over generations. The fact that neighbor identity strongly affected selection on sinigrin but direct manipulations of competition did not suggests that the neighbor effect acted indirectly, by altering other species interactions. Our experimental treatments and structural equation modeling suggest that this indirect effect was mediated through the two herbivore species. However, other indirect interactions—via pollinators, predators, or microbes—cannot be ruled out and deserve further attention. Despite the complex pattern of selection pressures observed in our experiment, path analyses connecting sinigrin concentration to molluskan leaf damage, specialist aphid loads, and target and neighbor plant size show that sinigrin concentration may only directly affect mollusk feeding. Subsequent differences between treatments arise because differential rates of damage affect plant growth and vigor, which, in turn, affect the host choice of the specialist aphid later in the season. Thus, genes with seemingly simple phenotypic effects could experience different evolutionary outcomes when embedded in complex interconnected ecological communities. The major finding of our study, that the selective value of sinigrin concentration depends on the species identity of plants growing nearby, resulted from the differential responses of the two dominant herbivores in our system. These two species differed in many ways, including diet breadth (broad generalist vs. Brassicaceae specialist), feeding mode (folivore vs. phloem feeder), and phenology (early vs. late season). Thus, we cannot conclusively determine what aspects of their biology drive their differential responses. However, two lines of evidence suggest that the difference in diet breadth played a major role in determining selective impacts on sinigrin concentration. First, the generalist mollusks were deterred by high sinigrin concentrations in target plants while the specialist aphids seemed unaffected by sinigrin concentrations, a pattern commonly observed for generalist versus specialist herbivores (Da Costa 1971; Siemens and Mitchell-Olds 1996; van der Meijden 1996; Berenbaum and Zangerl 1998; Kliebenstein et al. 2002; Macel and Vrieling 2003). Second, the foraging behavior of the generalist mollusks was qualitatively similar in conspecific and heterospecific plots because both plants represented potential hosts in both cases. However, the specialist aphids showed quite different hostchoice behavior in the two situations, finely discriminating between the two potential hosts in the conspecific case based on relative size but colonizing hosts in heterospecific plots independent of plant size because no other hosts were available. Thus, while the two herbivore species differ in many aspects of their biology, the differences in diet

159

breadth seem to drive their different selective impacts on sinigrin concentration. Ecological research has long focused on the role of species interactions in determining the distribution and abundance of populations. Herbivory and competition have repeatedly been found to have strong impacts on plant populations and community structure (Harper 1977; Crawley 1983; Gurevitch et al. 2000). Recently, several studies have investigated how these two important interactions act on one another to have greater or lesser influence on average plant fitness than would be predicted from either interaction acting alone (Hamback and Beckerman 2003; Agrawal 2004; Haag et al. 2004). This study extends these ecologically focused studies by asking how interactions with multiple species affect the evolutionary trajectory of plant traits. While many studies have examined the way multispecies interactions affect average fitness, few studies have also measured specific plant traits and asked how the relationship between traits and fitness varies with the presence or absence of multiple interacting species. Strauss and Irwin (2004) reviewed these studies and found that five out of five studies found nonadditive affects on selection. For instance, Gomez (2003) found that pollinators selected for taller plants in the absence of browsers but not in their presence. Juenger and Bergelson (1998) found stabilizing selection for flowering time in Ipomopsis aggregata in the presence of three herbivores (simulated browsing, a caterpillar, and a seed fly) but linear selection for decreased flowering time if any of the three were removed. However, in the only study to date testing the simultaneous effects of herbivory and competition on selection, Tiffin (2002) found that the presence or absence of intraspecific competitors had no effect on selection for either herbivore resistance or tolerance in Ipomoea purpurea. Similarly, in our study, the intensity of competition had no effect on the pattern of selection. However, the identity of the competing species (con- or heterospecific) had dramatic effects on selection in the presence of herbivores. An especially interesting result from our study is that when the full herbivore community was present, high sinigrin concentrations led to high fitness when plants grew with interspecific neighbors but low fitness when they grew with intraspecific neighbors. This result suggests that selection on sinigrin may be density dependent, as at high densities individuals will mostly interact with conspecifics, while at low densities they will interact mostly with heterospecifics. This study suggests that microevolutionary changes may occur as populations grow or decline (becoming common or rare) and that these changes could potentially alter the rate and direction of population dynamics (Lawlor and Smith 1976; Pease 1984). Understanding the complexity of natural selection for traits involved

160 The American Naturalist in multispecies interactions is inherently challenging but is a necessary step toward understanding evolution in natural communities. Acknowledgments We thank R. Karban, M. L. Stanton, and E. Wheeler for comments on the manuscript. D. Kliebenstein assisted with and provided equipment for chemical analysis. M. van Horn provided space at the University of California, Davis, Student Farms. This work was funded by a Doctoral Dissertation Improvement Grant to S.Y.S. and R.A.L. (DEB 0508753) and the Univervsity of California, Davis. R.A.L. was supported by a National Science Foundation graduate research fellowship. S.Y.S. was partially supported by the University of California Agricultural Experiment Station. Literature Cited Agrawal, A. A. 2004. Resistance and susceptibility of milkweed: competition, root herbivory, and plant genetic variation. Ecology 85: 2118–2133. ———. 2005. Natural selection on common milkweed (Asclepias syriaca) by a community of specialized insect herbivores. Evolutionary Ecology Research 7:651–667. Agrawal, A. A., and N. S. Kurashige. 2003. A role for isothiocyanates in plant resistance against the specialist herbivore Pieris rapae. Journal of Chemical Ecology 29:1403–1415. Andow, D. A. 1991. Vegetational diversity and arthropod population response. Annual Review of Entomology 36:561–586. Banks, J. E. 1998. The scale of landscape fragmentation affects herbivore response to vegetation heterogeneity. Oecologia (Berlin) 117:239–246. Bates, D., and D. Sarkar. 2006. lme4: linear mixed-effects models using S4 classes, R package version 0.995-2. Berenbaum, M. R., and A. R. Zangerl. 1992. Genetics of secondary metabolism and herbivore resistance in plants. Pages 415–438 in G. A. Rosenthal and M. R. Berenbaum, eds. Herbivores: their interactions with secondary plant metabolites. Vol. 2. Ecological and evolutionary processes. Academic Press, San Diego, CA. ———. 1998. Chemical phenotype matching between a plant and its insect herbivore. Proceedings of the National Academy of Sciences of the USA 95:13743–13748. Callaway, R. M., and E. T. Aschehoug. 2000. Invasive plants versus their new and old neighbors: a mechanism for exotic invasion. Science 290:521–523. Canty, A. 2005. Boot: bootstrap R (S-plus) functions. R port by B. Ripley, version 1.2-24. Crawley, M. J. 1983. Herbivory: the dynamics of animal-plant interactions. Blackwell Scientific, Oxford. Da Costa, C. P., and C. M. Jones. 1971. Cucumber beetle resistance and mite susceptibility controlled by the bitter gene in Cucumis sativus L. Science 172:1145–1146. Efron, B., and R. Tibshirani. 1993. An introduction to the bootstrap. Chapman & Hall, New York. Fox, J. 2006. Sem: structural equation models, R package version 0.9-2. Fritz, R. S., and E. L. Simms. 1992. Plant resistance to herbivores and pathogens. University of Chicago Press, Chicago.

Futuyma, D. J., and S. S. Wasserman. 1980. Resource concentration and herbivory in oak forests. Science 210:920–922. Gomez, J. M. 2003. Herbivory reduces the strength of pollinatormediated selection in the Mediterranean herb Erysimum mediohispanicum: consequences for plant specialization. American Naturalist 162:242–256. Gomez, J. M., J. A. Hodar, R. Zamora, J. Castro, and D. Garcia. 2001. Ungulate damage on Scots pines in Mediterranean environments: effects of association with shrubs. Canadian Journal of Botany–Revue Canadienne de Botanique 79:739–746. Grace, J. B. 2006. Structural equation modelling and natural systems. Cambridge University Press, Cambridge. Gurevitch, J., J. A. Morrison, and L. V. Hedges. 2000. The interaction between competition and predation: a meta-analysis of field experiments. American Naturalist 155:435–453. Haag, J. J., M. D. Coupe, and J. F. Cahill. 2004. Antagonistic interactions between competition and insect herbivory on plant growth. Journal of Ecology 92:156–167. Hamback, P. A., and A. P. Beckerman. 2003. Herbivory and plant resource competition: a review of two interacting interactions. Oikos 101:26–37. Harper, J. L. 1977. Population biology of plants. Academic Press, London. Hay, M. E. 1986. Associational plant defenses and the maintenance of species diversity: turning competitors into accomplices. American Naturalist 128:617–641. Hereford, J., T. F. Hansen, and D. Houle. 2004. Comparing strengths of directional selection: how strong is strong? Evolution 58:2133– 2143. Holt, R. D., and J. H. Lawton. 1994. The ecological consequences of shared natural enemies. Annual Review of Ecology and Systematics 25:495–520. Iwao, K., and M. D. Rausher. 1997. Evolution of plant resistance to multiple herbivores: quantifying diffuse coevolution. American Naturalist 149:316–335. Juenger, T., and J. Bergelson. 1998. Pairwise versus diffuse natural selection and the multiple herbivores of scarlet gilia, Ipomopsis aggregata. Evolution 52:1583–1592. Karban, R. 1997. Neighbourhood affects a plant’s risk of herbivory and subsequent success. Ecological Entomology 22:433–439. Kareiva, P. 1982. Experimental and mathematical analyses of herbivore movement: quantifying the influence of plant spacing and quality on foraging discrimination. Ecological Monographs 52: 261–282. Kliebenstein, D., D. Pedersen, B. Barker, and T. Mitchell-Olds. 2002. Comparative analysis of quantitative trait loci controlling glucosinolates, myrosinase and insect resistance in Arabidopsis thaliana. Genetics 161:325–332. Kliebenstein, D. J. 2004. Secondary metabolites and plant/environment interactions: a view through Arabidopsis thaliana tinged glasses. Plant Cell and Environment 27:675–684. Kliebenstein, D. J., J. Kroymann, P. Brown, A. Figuth, D. Pedersen, J. Gershenzon, and T. Mitchell-Olds. 2001. Genetic control of natural variation in Arabidopsis glucosinolate accumulation. Plant Physiology 126:811–825. Lande, R., and S. J. Arnold. 1983. The measurement of selection on correlated characters. Evolution 37:1210–1226. Lankau, R. A. 2007. Specialist and generalist herbivores exert opposing selection on a chemical defense. New Phytologist 175:176– 184.

Community Complexity and Natural Selection Lawlor, L. R., and J. M. Smith. 1976. Co-evolution and stability of competing species. American Naturalist 110:79–99. Leimu, R., and J. Koricheva. 2006. A meta-analysis of genetic correlations between plant resistances to multiple enemies. American Naturalist 168:E15–E37. Macel, M., and K. Vrieling. 2003. Pyrrolizidine alkaloids as oviposition stimulants for the cinnabar moth, Tyria jacobaeae. Journal of Chemical Ecology 29:1435–1446. Maddox, G. D., and R. B. Root. 1990. Structure of the encounter between goldenrod (Solidago altissima) and its diverse insect fauna. Ecology 71:2115–2124. Mauricio, R., and M. D. Rausher. 1997. Experimental manipulation of putative selective agents provides evidence for the role of natural enemies in the evolution of plant defense. Evolution 51:1435–1444. Pease, C. M. 1984. On the evolutionary reversal of competitive dominance. Evolution 38:1099–1115. Pilson, D. 2000. The evolution of plant response to herbivory: simultaneously considering resistance and tolerance in Brassica rapa. Evolutionary Ecology 14:457–489. Power, J. H., and E. B. Moser. 1999. Linear model analysis of net catch data using the negative binomial distribution. Canadian Journal of Fisheries and Aquatic Sciences 56:191–200. Price, G. R. 1970. Selection and covariance. Nature 227:520–521. Rausher, M. D. 1992. The measurement of selection on quantitative traits: biases due to environmental covariances between traits and fitness. Evolution 46:616–626. Rausher, M. D., and E. L. Simms. 1989. The evolution of resistance to herbivory in Ipomoea purpurea. 1. Attempts to detect selection. Evolution 43:563–572. Raybould, A. F., and C. L. Moyes. 2001. The ecological genetics of aliphatic glucosinolates. Heredity 87:383–391. Roche, B. M., and R. S. Fritz. 1997. Genetics of resistance of Salix sericea to a diverse community of herbivores. Evolution 51:1490– 1498. Root, R. B. 1973. Organization of a plant-arthropod association in simple and diverse habitats: the fauna of collards (Brassica oleracea). Ecological Monographs 43:95–124. Ryan, T. P. 1997. Modern regression methods. Wiley, New York. Shonle, I., and J. Bergelson. 2000. Evolutionary ecology of the tropane alkaloids of Datura stramonium L. (Solanaceae). Evolution 54:778– 788. Siemens, D. H., and T. Mitchell-Olds. 1996. Glucosinolates and herbivory by specialists (Coleoptera: Chrysomelidae, Lepidoptera: Plutellidae): consequences of concentration and induced resistance. Environmental Entomology 25:1344–1353.

161

Siemens, D. H., S. H. Garner, T. Mitchell-Olds, and R. M. Callaway. 2002. Cost of defense in the context of plant competition: Brassica rapa may grow and defend. Ecology 83:505–517. Stanton, M. L., and D. A. Thiede. 2005. Statistical convenience vs biological insight: consequences of data transformation for the analysis of fitness variation in heterogeneous environments. New Phytologist 166:319–337. Stinchcombe, J. R. 2002. Environmental dependency in the expression of costs of tolerance to deer herbivory. Evolution 56:1063– 1067. Stinchcombe, J. R., and M. D. Rausher. 2001. Diffuse selection on resistance to deer herbivory in the ivyleaf morning glory, Ipomoea hederacea. American Naturalist 158:376–388. ———. 2002. The evolution of tolerance to deer herbivory: modifications caused by the abundance of insect herbivores. Proceedings of the Royal Society B: Biological Sciences 269:1241–1246. Strauss, S. Y., and R. E. Irwin. 2004. Ecological and evolutionary consequences of multispecies plant-animal interactions. Annual Review of Ecology, Evolution, and Systematics 35:435–466. Strauss, S. Y., J. A. Rudgers, J. A. Lau, and R. E. Irwin. 2002. Direct and ecological costs of resistance to herbivory. Trends in Ecology & Evolution 17:278–284. Strauss, S. Y., H. Sahli, and J. K. Conner. 2005. Toward a more traitcentered approach to diffuse (co)evolution. New Phytologist 165: 81–89. Tahvanainen, J. O., and R. B. Root. 1972. The influence of vegetational diversity on the population ecology of a specialized herbivore, Phyllotreta cruciferae (Coleoptera: Chrysomelidae). Oecologia (Berlin) 10:321–346. Tiffin, P. 2002. Competition and time of damage affect the pattern of selection acting on plant defense against herbivores. Ecology 83:1981–1990. Traw, M. B. 2002. Is induction response negatively correlated with constitutive resistance in black mustard? Evolution 56:2196–2205. van der Meijden, E. 1996. Plant defence, an evolutionary dilemma: contrasting effects of (specialist and generalist) herbivores and natural enemies. Entomologia Experimentalis et Applicata 90:307– 310. Weis, A. E., and M. E. Hochberg. 2000. The diverse effects of intraspecific competition on the selective advantage to resistance: a model and its predictions. American Naturalist 156:276–292. White, J. A., and T. G. Whitham. 2000. Associational susceptibility of cottonwood to a box elder herbivore. Ecology 81:1795–1803. Associate Editor and Editor: Monica A. Geber