Specialist and generalist herbivores exert opposing selection on a chemical defense

Summary

• • 
  Plant defense traits often show high levels of genetic variation, despite clear impacts on plant fitness. This variation may be partly maintained by trade-offs in the defense against multiple herbivore species, for example between generalists and coevolved specialists. Despite a long-standing discussion in the literature on the subject, no study to date has specifically manipulated specialist and generalist herbivores independently of one another to determine whether the two guilds exert opposing selection pressures on specific defensive traits.
• • 
  In two separate experiments, the dominant specialist and generalist herbivores of Brassica nigra were independently manipulated to test whether the composition of the herbivore community altered the direction of selection on a major defensive trait of the plant, sinigrin concentration.
• • 
  It was found that generalist damage was negatively correlated but specialist loads were positively correlated with increasing sinigrin concentrations; and sinigrin concentration was favored when specialists were removed, disfavored (past an intermediate point) when generalists were removed and selectively neutral when plants faced both generalists and specialists.

  These results suggest that specialist and generalist herbivores can exert opposing selection pressures on chemical defenses, and thus that changes in herbivore community composition can alter the net selective value of defensive traits.

Introduction

Plants have evolved a diverse array of adaptations to reduce the amount of damage suffered from herbivores. These defenses include toxic secondary compounds as well as various structural defenses, such as spines, thorns or trichomes (Ehrlich & Raven, 1964; Fritz & Simms, 1992; Mauricio & Rausher, 1997). As herbivore damage generally decreases plant fitness, natural selection is expected to favor high levels of these defensive traits. However, plant populations exist within complex communities, and ecologically important traits such as secondary compounds may be under conflicting selection pressures from multiple interacting species. Numerous studies of different plant species and populations have found intermediate levels of defense traits with high levels of additive genetic variation, suggesting that selection is often weak or inconsistent (Rausher & Simms, 1989; Berenbaum & Zangerl, 1992; Fritz & Simms, 1992; Mauricio & Rausher, 1997). Selection on defense traits may be inconsistent in direction and magnitude between sites or years because of changes in the community context in which the plant population is embedded (Strauss et al., 2002; Strauss & Irwin, 2004).

Plant species must often defend themselves against a diverse array of herbivore species, including generalists that feed on many different species and specialists that feed on a restricted set of related species. While generalists are often deterred by secondary compounds, numerous studies have suggested that many specialists have evolved effective countermeasures to the chemical defenses of their hosts (van Dam et al., 1995; Siemens & Mitchell-Olds, 1996; Berenbaum & Zangerl, 1998; Kliebenstein et al., 2002). In addition, since specialists must be able to locate host plants growing in a matrix of nonhosts, many species have evolved to use the unique secondary chemicals of their host plants as host-finding, feeding and oviposition cues (Da Costa & Jones, 1971; Raybould & Moyes, 2001; Macel & Vrieling, 2003; Nieminen et al., 2003). Thus, there may be a trade-off between defense against generalists and attractiveness to specialists (Da Costa & Jones, 1971; van der Meijden, 1996). If the ratio of specialist to generalist herbivores varies through space and time, this variation could potentially lead to fluctuating selection pressures and the maintenance of genetic variation in defensive traits (Gillespie & Langley, 1974; Ludwig & Levin, 1991; Ellner & Hairston, 1994). In addition, if specialist herbivores are absent, as is the case with some introduced plant species, selection may lead to rapid increases in the levels of chemical defense (Müller-Schärer et al., 2004).

Many studies have documented a preference by specialist herbivores for host plants with higher concentrations of characteristic secondary compounds. However, it is not clear from these studies if these preferences translate into selection for reduced levels of secondary compounds in host plants. Typically, studies have measured selection on plants protected from all herbivores through insecticide sprays, and generally found reduced selection for defense traits (Mauricio & Rausher, 1997; Mauricio, 1998; Juenger & Bergelson, 2000), although not always (Mauricio, 2000; Shonle & Bergelson, 2000). While many studies have studied the interactive effects of multiple herbivores on plant fitness, relatively few have determined whether selection pressures on specific traits differed depending on the presence or absence of specific herbivore species (reviewed in Strauss & Irwin, 2004). Juenger & Bergelson (1998) found selection on flowering time to differ depending on the presence or absence of artificial clipping, a seed fly, and a caterpillar. Other studies have examined how interactions between herbivores and pollinators (Gomez, 2003) or competitors (Tiffin, 2002) affect selection for floral traits and resistance and tolerance to herbivores, respectively. However, only three studies to date have independently manipulated multiple herbivore species and/or guilds and determined their impacts on selection for resistance traits. Pilson (1996) independently manipulated Phylotretta cruciferae and Plutella xylostella, two specialist herbivores of Brassica rapa, and found that selection tended to favor increased resistance against P. cruciferae in the absence, but decreased resistance in the presence, of the other herbivore species. Stinchcombe & Rausher (2001) found that selection favored resistance to deer herbivory more in the presence vs absence of insect herbivores, while Juenger et al. (2005) found selection against a secondary compound (curcurbitacins) in the presence of seed flies with or without artificial clipping to simulate deer browsing. Despite a long-standing discussion in the literature on the subject, no study to date has specifically manipulated specialist herbivores independently of generalists (and vice versa) to determine whether the two guilds exert opposing selection pressures on specific defensive traits.

To evaluate selection imposed by generalist and specialist herbivores on defensive traits, two field experiments were performed with an introduced annual plant, Brassica nigra, a major Brassicaceae family specialist, Brevicoryne brassicae (the cabbage aphid, family Aphididae), and the most abundant generalist guild in this system, mollusks. Glucosinolates, the major chemical defenses of brassicaceous plants, have been shown to be deterrent to many generalists like slugs (Giamoustaris & Mithen, 1995), but attractive to Brassicaceae specialists (including Brevicoryne brassicea) for either feeding or oviposition (Giamoustaris & Mithen, 1995; Cole, 1997; Raybould & Moyes, 2001; Kliebenstein, 2004).

In this study, the presence of the specialist in a background of ambient generalist damage was experimentally manipulated, and the next year generalist damage was manipulated in a background of ambient specialist densities. These experiments were designed to answer two related questions: how does genetic variation in a chemical defense affect damage from generalist and specialist herbivores of B. nigra, and does altering the balance of specialist and generalist herbivores in the herbivore community lead to different patterns of selection on this defense?

Based on previous research on how generalist and specialist herbivores respond to glucosinolates, it was predicted that: (1) selection will favor increased sinigrin concentration in the presence of generalists but not specialists; (2) selection will favor decreased sinigrin concentrations in the presence of specialists, but not generalists; (3) in the presence of both herbivore guilds, selection will favor an intermediate sinigrin concentration (stabilizing selection), or alternatively, sinigrin concentration may be selectively neutral.

Materials and Methods

Study system

Brassica nigra L. is an introduced Eurasian annual that commonly occurs in disturbed sites throughout the USA. In its introduced range, B. nigra is fed on by introduced and native generalists, as well as several introduced oligophagous species specialized on the Brassicaceae.

Plants in the Brassicaceae all produce glucosinolates, a class of secondary compounds derived from several amino acid sources. In B. nigra, sinigrin (allyl-glucosinolate) represents 90–99% of the total glucosinolate concentration and has a heritable basis (Traw, 2002, and see the Results section). When combined with the myrosinase enzyme, glucosinolates break down into a number of toxic byproducts (such as isothiocyanates), which contribute to herbivore and disease resistance (Raybould & Moyes, 2001; Kliebenstein, 2004). Sinigrin concentration, rather than its breakdown product allyl-isothiocyanate, was used as the trait studied since it is highly correlated with allyl-isothiocyanate concentrations (R. Lankau unpublished) but is easier to quantify.

The two field experiments were conducted in an uncultivated area of the organically managed University of California, Davis Student Farms. The area used for this study had not been tilled or cultivated for 5 yr and was dominated by native and introduced annual species; B. nigra occurs naturally at the site but is not abundant. The most numerous generalist herbivore observed was Agriliomax reticulates (Muller), the gray garden slug, and although several Brassicaceae specialists have been observed at this site, only the cabbage aphid, Brevicoryne brassicae L., was abundant during the active stages of B. nigra growth. B. brassicae individuals were present from April until June (when plants senesced), feeding mainly on reproductive organs.

Plant source

In summer 2003, B. nigra seeds were collected from six sites around the Sacramento Valley. From these collections, two seeds were grown from 256 maternal half-sib families in the glasshouse. Based on the average glucosinolate concentration, families were split into high- and low-sinigrin lines. Within each line, families were randomly assigned 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. There were no effects of inbreeding in this design, as all matings were between families that never shared ancestors across the three generations. Seeds were used from the first generation of selection for the specialist removal experiment (2003–04), and seeds from the third generation for the generalist removal experiment (2004–05). Note that this selection scheme altered the distribution of sinigrin values (more extreme and fewer intermediate families), but did not result in an expansion of the overall range in sinigrin values. Thus the use of the third-generation selected lines increased ability to detect effects, but did not present herbivores with unusually high or low levels of sinigrin concentration.

Specialist removal experiment

Two individuals were grown per family for 32 families in each of two experimental treatments; specialists removed or specialists present (n = 128 plants). In November 2004, seedlings (2–3 d post germination) were planted into the field, protected by plastic cups to reduce (but not eliminate) damage by early season herbivores during establishment. Leaf number and damage was assessed four times throughout the experiment by estimating the per cent leaf area removed on each leaf (to within 5%), and then averaging damage levels over all leaves on a plant. These visual estimations were calibrated by scanning 50 leaves from naturally occurring B. nigra individuals and determining the per cent leaf area removed with imaging software. Plant height and basal diameter were measured on April 4, just before the emergence of the B. brassicae.

Following aphid emergence in early April, B. brassicae populations were censused weekly on each plant. For the specialist present treatment, aphids were counted but otherwise not manipulated. For the specialist removal treatment, all aphids were counted and then removed with a paintbrush. This removal continued until a plant completely senesced. Other Brassicacea specialists (Pieris rapae and Plutella xylostella) were very rare on the experimental plants; however, any eggs or larvae of these species were removed whenever encountered. The total number of fruits on each plant were then counted and 30 fruits sampled uniformly from each plant to determine the average number of seeds per fruit and average seed mass per plant. If a plant produced fewer than 30 fruits, all were used. Fruits were sampled uniformly, rather than randomly, because of a consistent trend for fruits further along a branch to produce fewer seeds.

Phenotypic selection analyses based on individuals may be biased if environmental conditions affect both the trait and fitness simultaneously (Rausher, 1992). Genotypic selection analysis avoids this bias by using trait values averaged over replicate individuals within genetic families. Since this design had a low degree of within-family replication, trait values for each family were instead determined by measuring sinigrin concentrations (as described later) in 10 full-siblings per family grown concurrently in a glasshouse. Field experiments in 2005–06 using descendants of these lines confirmed significantly positive additive genetic correlations between glasshouse and field measures for sinigrin concentration (r = 0.46 ± 0.38, 95% CI, t_1,31 = 2.278, P < 0.05). Thus, while using glasshouse measurements for analysis may introduce some error, it should not introduce any additional bias, thus making the conclusions more conservative. Importantly, since the traits and fitness were measured on different individuals in different environments, there is no bias introduced by environmental covariance (Rausher, 1992).

For all plants sampled in this study, glucosinolates were quantified by high-pressure liquid chromatography (HPLC) following a high though-put extraction and purification procedure developed by Kliebenstein et al. (2001). Specific glucosinolates were identified by comparison of retention times and UV absorbance spectra with previously published standards. Absorption data were converted to µmol g⁻¹ DW using previously determined response factors (D. Kliebenstein pers. comm.)

Generalist removal experiment

At the same site in the following year, a similar experiment was performed but the presence or absence of the dominant generalists (snails and slugs) was manipulated, leaving the specialists at ambient levels. Six individuals from each of 32 families were planted into each treatment (generalists absent or present) in November, 2005. To manipulate generalist damage, each plant was surrounded with a 16 oz (473 ml) plastic cup (with the bottom removed), and for those in the generalist absent treatment, a c. 6.5 cm wide copper strip (Snail-Barr; Custom Copper, Inc., Fairfield, NJ, USA) was placed around the cup. While other generalists (mostly the earwig, Forficula auricularia, and various rabbits and rodents) occasionally fed on the experimental plants, these species were not affected by the experimental barriers. Thus, any differences between the generalist treatments were driven by differential mollusk herbivory. For each plot, six seeds were planted, and 2 wk after germination began plots were thinned to one seedling. Leaf number and damage, stem height and diameter were measured as in the previous experiment. Aphid population sizes were surveyed on all plants once, in mid-May, when populations were near their maximum. Unfortunately, because of a data collection error, aphid loads were not collected on one family in the generalists-present treatment. Final plant fitness was estimated as in the previous experiment.

When plants had 10 leaves (March 2006), approx. 5 mg of leaf tissue (five hole-punches) were collected from the seventh leaf of each plant to measure glucosinolate concentration (quantified as already described). In a closely related crucifer, Raphanus raphanistrum, there was no significant induction of glucosinolates in the remaining leaves after mechanical leaf area removal (Agrawal et al., 1999). Sinigrin concentrations were measured directly from experimental plants in this experiment, rather than glasshouse-grown plants as in the previous year, because there was a greater level of within family replication. Measuring sinigrin concentrations differently between the specialist and generalist manipulation experiments may make comparisons between the two problematic. Nonetheless, sinigrin levels in glasshouse-grown plants are significantly genetically correlated with those in field-grown descendants (see earlier). Because glasshouse and field measures of sinigrin concentration were positively genetically correlated, it is statistically justifiable to use glasshouse measures as a proxy of field measures (as they largely explain the same variation), although the glasshouse measures should not be taken as a direct measure of the sinigrin levels in field grown plants. Thus the calculated relationships between sinigrin concentration and either herbivore damage or plant fitness should be at least qualitatively comparable across the two experiments.

Analyses

Specialist and generalist herbivore responses to sinigrin  A series of linear regressions were performed to explore how herbivores responded to sinigrin concentration. Since multiple measures of leaf damage and aphid populations on individual plants were not independent, for analysis only the final measurement was used for each. For both leaf damage and specialist aphid loads, the final measure (in late March and May, respectively) tended to be the highest. Both measures were averaged within families and then regressed against family-averaged sinigrin concentration. Leaf damage was arcsin-square root and aphid population size natural log-transformed before analysis. ancovas were used to compare the slope of the regression of leaf damage or aphid population size on sinigrin concentration between herbivore removal treatments (within a given experiment). Since plant size had a strong effect on aphid colonization, it was included as a covariate in the analysis of aphid population size, as well as any interactions with sinigrin concentration and experimental treatments (see the Results section).

Selection on sinigrin from specialists and generalists  Absolute fitness was calculated by multiplying the average number of seeds per fruit times the total number of fruits. To quantify selection acting on sinigrin concentration in the specialist and generalist removal experiments, a modified genotypic selection analysis was performed (Rausher, 1992). As is traditional, linear selection differentials were calculated as the covariance between the variance standardized family mean trait value and family mean relative fitness for each treatment (Rausher, 1992). Selection differentials measure the net selection acting on a trait, including both direct selection and indirect selection acting through correlated traits. Note that in the specialist removal experiment, family sinigrin concentrations were determined by 10 full-siblings grown concurrently in the glasshouse, while in the generalist removal experiment sinigrin concentration was determined from the experimental plants themselves (possibly due to the higher within-family replication in that experiment). Standardized trait values were determined by subtracting each family mean value from the overall population mean, and dividing by the population standard deviation, to produce a mean of 0 and standard deviation of 1. Relative fitness was determined by dividing family mean fitness by the population mean within each herbivore treatment, to compare selection between treatments independent of differences in mean fitness. The presence of disruptive or stabilizing selection was tested with the partial regression coefficients from a model including both linear and quadratic terms. In addition to significant curvature, true stabilizing (or disruptive) selection requires a fitness maximum (or minimum) within the range of observed trait values (Mitchell-Olds & Shaw, 1987). To test this, the sinigrin concentration that yielded the maximum (or minimum) fitness value based on the best fitting quadratic equation estimated from 5000 bootstrapped resamples of the data was calculated and the accelerated bootstrap method used to determine the 95% confidence interval for the maximizing (or minimizing) trait value. If the estimated confidence interval contained neither the upper nor lower limit of observed sinigrin concentrations for a given treatment, one can conclude that stabilizing (or disruptive) selection is acting. All analyses were performed separately for all treatments, and whether the regression coefficients differed between treatments within a given experiment was tested using ancova. In this case, a significant trait × treatment interaction indicates that the selection pressures differed between treatments. All plants were used for this analysis, and no additional covariates (such as mid-season size) were included to ensure that the analysis accurately reflects selection across the entire life-cycle of the plant. Since the fitness data were not normally distributed, and transformations can complicate the interpretation of selection differentials (Stanton & Thiede, 2005), the accelerated bootstrap method was used to calculate 95% confidence intervals on individual selection differentials, as well as on the estimated ancova parameters (boot library, R statistical language; Canty, 2005). Since the two experiments took place in different years and had some methodological differences, statistical comparisons were made only between treatments within an experiment. However, because the two experiments shared the same control treatment (ambient levels of both specialists and generalists) and the majority of measurements were taken identically, interesting qualitative comparisons can still be drawn between the two sets of results.

Results

Specialist and generalist herbivore responses to sinigrin

Brassica nigra individuals from families that produced high concentrations of sinigrin received lower levels of damage from generalist folivores (Fig. 1a,d). Sinigrin concentration explained c. 13% of the variation in damage in the first year, with similar effects in the presence or absence of specialist aphids (Table 1a). In the second year, sinigrin concentration was negatively correlated with leaf damage in the presence, but not absence, of generalist, since plants in the generalist removal treatment received very little damage regardless of sinigrin concentration (Table 1b). Sinigrin levels were significantly higher in the generalist present vs generalist absent treatment in the second experiment (mean = 7.58 vs 4.81 µmol g⁻¹, respectively, F_1,32 = 28.62, P < 0.0001); however, since there was no family × treatment interaction (F_31,258 = 0.650, P = 0.92), the ranking of families was largely consistent across the two treatments despite large differences in mean values.

Figure 1. Relationship between standardized family mean sinigrin concentration and (a,d) generalist foliar damage (b,e) specialist aphid (Brevicoryne brassicae) load (only top 50% of families based on midseason size presented in (b) (see text for explanation) and (c,f) relative fitness (individual seed biomass divided by mean seed biomass within treatments) in the specialist (a–c) and generalist (d–f) removal experiments. Closed symbols, specialist (a–c) or generalist (d–f) removal treatment; open symbols, specialist (d–f) or generalist (a–c) present treatment; circles, specialist removal experiment; diamonds, generalist removal experiment.

Table 1.  Analysis of specialist aphid (Brevicoryne brassicae) population sizes in (a) the specialist removal experiment and (b) the generalist removal experiment

Source	F	P
Size refers to the stem diameter of Brassica nigra plants measured before aphid emergence. Aphid population sizes were natural log transformed.
(a) Specialist removal experiment
Specialist treatment	2.18	0.145
Sinigrin concentration	5.29	0.025
Specialist treatment × sinigrin	5.27	0.026
Size	68.03	< 0.001
Specialist treatment × size	8.13	0.006
Sinigrin × size	16.55	< 0.001
Specialist treatment × sinigrin × size	0.37	0.544
(b) Generalist removal experiment
Generalist treatment	1.60	0.212
Sinigrin concentration	4.15	0.047
Generalist treatment × sinigrin	0.47	0.495
Size	0.08	0.783
Generalist treatment × size	0.00	0.972
Sinigrin × size	0.55	0.462
Generalist treatment × sinigrin × size	0.65	0.422

Sinigrin concentration and plant size jointly determined a plant's specialist aphid load. Consistent with the plant vigor hypothesis (Price, 1991), it was found that larger plants were colonized to a much greater extent than small ones. The aphid removal treatment thus had larger effects on large plants than on small ones, resulting in a treatment × plant size interaction (Table 2a).Therefore, midseason plant size and the sinigrin concentration × plant size interaction were included in the model relating log-transformed aphid population size to sinigrin concentration. It was found that sinigrin concentration had an increasingly large effect on aphid feeding intensity as plant size increased, since small plants were rarely colonized by aphids (significant sinigrin concentration × size interaction, Table 2a). When considering only large plants (top 50% based on midseason stem diameter), sinigrin concentration had a significant positive effect on specialist feeding intensity in both the specialist present and removal treatments, explaining 30% and 25% of the variation, respectively (specialists present, P < 0.05, specialists absent: P < 0.05, Fig. 1b). Results were similar in the generalist removal experiment, with specialist population sizes increasing with increasing sinigrin concentration in both the presence and absence of generalist mollusks (Table 2b), although plant size explained much less of the variation in aphid number, partly owing to reduced variation in family averaged plant size because of higher within-family replication.

Table 2.  Analysis of leaf damage from generalist mollusks in (a) the specialist removal experiment and (b) the generalist removal experiment

Source	F	P
Per cent leaf area removed was arcsin-square root transformed before analysis.
(a) Specialist removal experiment
Specialist treatment	0.32	0.575
Sinigrin concentration	5.37	0.024
Specialist treatment × sinigrin	0.15	0.697
(b) Generalist removal experiment
Generalist treatment	14.54	< 0.001
Sinigrin concentration	2.45	0.123
Generalist treatment × sinigrin	4.79	0.033

Selection on sinigrin from specialists and generalists

The direction of selection on sinigrin depended on the presence or absence of the specialist aphids and generalist mollusks. In the absence of specialists, higher fitness was associated with higher levels of sinigrin concentration (Fig. 1c), with linear selection differentials significantly greater than zero (Table 3). However, in the presence of specialists, there was no evidence for selection on sinigrin concentration, owing to the opposing effects of generalists and specialists (Fig. 1c, Table 3). These selection differentials were significantly different (Table 4a). These results did not change when the extreme (high relative fitness and sinigrin concentration) data point in the specialist removal treatment was excluded from the analysis. There was no evidence for disruptive or stabilizing selection on sinigrin concentration in either the specialist present or removed treatments (Tables 3 and 4a). This study likely did not have the necessary power to detect moderate levels of disruptive or stabilizing selection. However, there was no indication of an intermediate minimum or maximum, which would be required for disruptive or stabilizing selection to act; for example, in the absence of specialists, the lower 95% confidence limit of the minimizing sinigrin concentration from the best fitting quadratic function (–2.40) is far lower than the lowest observed sinigrin concentration (–2.16).

Table 3.  Linear and quadratic selection differentials with bootstrapped confidence intervals for the specialist removal (a) and generalist removal (b) experiments

Trait	Estimate	Lower95% CI	Upper95% CI	Estimate	Lower95% CI	Upper95% CI
Estimates in italic type are significantly different from zero. Those in bold type differ significantly between treatments within one experiment.
(a) Specialists present				Specialists absent
Sinigrin	–0.001	–0.991	0.247	0.967	0.251	1.88
Sinigrin^2	0.263	–0.996	1.368	1.08	–0.295	2.521
(b) Generalists present				Generalists absent
Sinigrin	0.153	–0.173	0.489	0.200	–0.039	0.537
Sinigrin^2	–0.096	–0.277	0.209	–0.411	–0.588	–0.251

Table 4.  Analysis of covariance testing whether linear and quadratic selection differentials differ between specialist (a) and generalist (b) removal treatments

Source	Estimate	Lower 95% CI	Upper 95% CI
Estimates involving linear terms come from models without quadratic terms. Lower and upper 95% confidence intervals were determined by bootstrap resampling. Estimates significantly different from zero are in bold. Results were qualitatively similar using a parametric analysis.
(a) Specialist removal experiment
Specialist treatment	0.298	–0.542	1.167
Sinigrin concentration	1.049	0.369	2.332
Sinigrin^2	0.512	–0.136	1.545
Specialist treatment × sinigrin	–1.026	–2.320	–0.108
Specialist treatment × sinigrin^2	–0.297	–1.437	0.643
(b) Generalist removal experiment
Generalist treatment	–0.118	–0.282	0.096
Sinigrin concentration	0.137	–0.045	0.324
Sinigrin^2	–0.254	–0.378	–0.085
Generalist treatment × sinigrin	0.144	–0.014	0.347
Generalist treatment × sinigrin^2	0.157	0.004	0.310

As in the specialist removal experiment, sinigrin concentration was selectively neutral in the presence of both generalists and specialists in the generalist manipulation experiment (Fig. 1f, Table 3). In the absence of generalists, fitness initially increased with increasing sinigrin concentration, but then decreased sharply as high sinigrin families accumulated high specialist loads. This resulted in significant stabilizing selection on sinigrin, with intermediate genotypes having the highest relative fitness (Fig. 1f, Table 3). The 95% confidence interval for the estimated fitness maxima fell well within the range of observed sinigrin concentrations (estimated maximum = –0.008, lower 95% CI = –0.588, upper 95% CI = 0.359). The linear selection differentials did not differ between generalist treatments, but the quadratic term was significantly greater in the absence vs presence of generalists, reflecting the different shape of the adaptive landscape in the two treatments (Table 4b).

Discussion

In this study, it was found that the major chemical defense of B. nigra effectively deterred generalist herbivores, but led to increased loads of a specialist aphid, B. brassicae. The resulting selection favored higher concentrations of sinigrin in the absence of the specialist, but had neutral effects on sinigrin concentrations in the presence of the specialist. By contrast, in the absence of the dominant generalist, stabilizing selection for intermediate sinigrin concentrations was found. Thus, the net selective value of sinigrin may depend crucially on the specific herbivore community in which plants grow.

Many studies have found that generalist herbivores are deterred, while specialists are either unaffected or attracted to the secondary compounds of their host plants (Da Costa & Jones, 1971; Raybould & Moyes, 2001; Macel & Vrieling, 2003). It is less clear what the consequences of these contrasting herbivore adaptations are for the selective regime acting on chemical traits in plants. Many authors have suggested that plants face a trade-off in defense against the two guilds that could constrain the evolution of high defenses, and potentially help maintain genetic variation in defensive traits (van der Meijden, 1996; Müller-Schärer et al., 2004; Strauss & Irwin, 2004). Therefore, one would predict that: selection will favor increased sinigrin concentration in the presence of generalists but not specialists; selection will favor decreased sinigrin concentrations in the presence of specialists but not generalists; and finally, in the presence of both herbivore guilds, selection will favor an intermediate sinigrin concentration (stabilizing selection), or alternatively, sinigrin concentration may be selectively neutral.

The results supported predictions (1) and (3), but only partly supported prediction (2). Rather than finding directional selection for decreased sinigrin concentration in the absence of generalists, significant stabilizing selection was found, with fitness initially increasing with increasing sinigrin concentration, but then decreasing as sinigrin concentrations increased past the mean value. Thus, extremely low sinigrin concentrations were selected against even with experimentally reduced levels of generalist folivory. In addition to herbivore defense, glucosinolates may play a role in pathogen resistance and allelopathy (Kliebenstein et al., 2001; Raybould & Moyes, 2001). So even without generalist herbivores present, individuals with very low sinigrin concentrations may have reduced fitness because of higher pathogen loads or increased competition. In addition, since specialist aphid populations tended to increase exponentially (rather than linearly) with sinigrin concentration, one would expect the per capita indirect cost of sinigrin production via increased aphid loads to increase as sinigrin concentration increases. Thus the observed pattern of stabilizing selection may be the result of a balance between a slight benefit to sinigrin production even in the absence of generalist herbivory, and a higher per capita cost of production at higher levels of sinigrin concentration.

Unfortunately, this study did not manipulate specialists and generalists factorially in the same year, but rather manipulated each guild separately in two successive years. Therefore, it was not possible to make direct statistical comparisons of selection in the presence of specialists only vs generalists only. However, several lines of evidence suggest that performing the experiments in successive years did not greatly affect the results or their interpretation. First, the same control treatment (ambient levels of specialists and generalists) was used in both experiments, and it was found that sinigrin concentration was selectively neutral in both cases. Second, the responses of the specialist and generalist herbivores were strikingly consistent in both years, with generalists deterred by high sinigrin concentrations, and specialists attracted to both large plants and high sinigrin concentrations. Thus while it would have been ideal to measure selection in all situations in the same year, performing successive experiments has the alternative advantage of demonstrating consistency in the results across years.

A growing literature has emerged investigating the role of complex herbivore communities in selection on plant traits. Selection on secondary compounds has been found to differ when all herbivores are experimentally removed (Mauricio & Rausher, 1997; Shonle & Bergelson, 2000). Additionally, some studies have found that different herbivore species and/or guilds can exert differing selection pressures, and can have interactive effects on selection (Pilson, 1996, 2000; Juenger & Bergelson, 1998; Stinchcombe & Rausher, 2001, 2002; Strauss & Irwin, 2004). For example, Stinchcombe & Rausher (2001) found that selection for resistance to deer herbivory was stronger in the presence vs absence of insects. Similarly, in this study, selection on a chemical defense depended strongly on the specific herbivore community (including both generalists and specialist or only one of the guilds).

Since herbivore populations are highly variable in space and time, selection on chemical defensive traits may fluctuate, being favored in years or sites with high generalist loads and disfavored (past an intermediate point) in years or sites with high specialist loads. This could potentially lead to the maintenance of genetic variation in these traits (Gillespie & Langley, 1974; Ludwig & Levin, 1991; Ellner & Hairston, 1994). Alternatively, the permanent absence of certain herbivores could allow for the evolution of unusually high or low levels of chemical defense. This may be the case for some invasive species, which are often introduced without coevolved specialists, but suffer generalist damage comparable to their native ranges (Memmott et al., 2000). Müller-Schärer et al. (2004) suggest that if specialists commonly constrain the evolution of qualitative defenses, one might expect increased levels of such defenses in introduced vs native ranges. Two recent studies have found significantly higher levels of chemical defenses in introduced populations; pyrrolizidone alkaloids in Senecio jacobea (Joshi & Vrieling, 2005) and glucosinolates in Lepidium draba (Muller & Martens, 2005). Thus, altered selection regimes could facilitate invasions by allowing invaders access to phenotypic space (extremely high levels of defense) not accessible to native plants that face a full complement of herbivores.

Populations may respond to fluctuating selection pressures by maintaining high levels of genetic variation in fixed traits, but under some circumstances fluctuating selection may lead to the evolution of phenotypic plasticity instead. In this study, it was found that sinigrin concentration was strongly inducible by generalist damage. The natural history of this system, in which the dominant specialist does not emerge until after the majority of generalist damage has occurred, suggests that an inducible strategy is not likely to lead to lower specialist loads, since the majority of plants are fully induced when the specialist aphids begin colonizing. However, in systems where specialists and generalists are active concurrently, inducible defenses may allow plants to benefit from lower attractiveness to specialists when generalists are rare, but then increase defenses when generalists are abundant.

This study has shown that a dominant specialist of B. nigra was attracted to and achieved higher population sizes on host plants with higher levels of sinigrin, the primary chemical defense of the plant, while generalist damage was negatively correlated with the same compound. Additionally, these herbivore responses had evolutionary consequences, since sinigrin concentration was favored when specialists were removed, disfavored (past an intermediate point) when generalists were removed, and selectively neutral when both specialists and generalists were present. This is the first experimental evidence that specialist and generalist herbivores can create divergent selection pressures against a secondary compound, thus supporting the hypothesis that plants face a trade-off between resistance against generalists and attraction of coevolved specialists. Therefore the net selective value of defensive traits may vary with the particular community context in which plants grow, and thus lead to the maintenance of genetic variation in the traits.

Acknowledgements

The author thanks S. Y. Strauss, E. Wheeler, M. Rausher and two anonymous reviewers for comments on the manuscript, D. Kliebenstein for assistance with chemical analysis, and Z. Costa and O. Ervin for assistance with field work. Funding was received from the NSF Graduate Research Fellowship and a UC Davis Jastro-Shields grant.