Herbivores can select for mixed defensive strategies in plants


Author for correspondence:

Juan Fornoni

Tel: +52 (55) 56229039

Email: jfornoni@ecologia.unam.mx


  • Resistance and tolerance are the most important defense mechanisms against herbivores. Initial theoretical studies considered both mechanisms functionally redundant, but more recent empirical studies suggest that these mechanisms may complement each other, favoring the presence of mixed defense patterns. However, the expectation of redundancy between tolerance and resistance remains unsupported.
  • In this study, we tested this assumption following an ecological genetics field experiment in which the presence/absence of two herbivores (Lema daturaphila and Epitrix parvula) of Datura stramonium were manipulated. In each of three treatments, genotypic selection analyses were performed and selection patterns compared.
  • Our results indicated that selection on resistance and tolerance was significantly different between the two folivores. Tolerance and resistance are not redundant defense strategies in D. stramonium but instead functioned as complementary defenses against both beetle species, favoring the evolution of a mixed defense strategy. Although each herbivore was selected for different defense strategies, the observed average tolerance and resistance were closer to the adaptive peak predicted against E. parvula and both beetles together.
  • In our experimental population, natural selection imposed by herbivores can favor the evolution of mixed defense strategies in plants, accounting for the presence of intermediate levels of tolerance and resistance.


Plants simultaneously allocate resources to two general defense mechanisms against herbivores and pathogens: resistance (the ability to avoid or reduce the probability of being eaten or infected) and tolerance (the ability to reduce the negative fitness effect of damage or infection once it occurs; Rosenthal & Kotanen, 1994; Strauss & Agrawal, 1999). Initial studies conceived both strategies as functionally redundant alternative defense mechanisms (van der Meijden et al., 1988; Fineblum & Rausher, 1995), because the same fitness benefits can theoretically be obtained by each defense strategy alone. In turn, if both strategies are associated with fitness costs, a genotype should be either highly tolerant or resistant but not both (Abrahamson & Weis, 1997; Mauricio et al., 1997). This would result in an adaptive surface with two alternative fitness peaks, one of maximum tolerance and minimum resistance and the other of maximum resistance and minimum tolerance (Fig. 1a). Although theoretical work has been shaped by researchers assuming defense strategies are redundant, redundancy has only rarely been supported in previous studies (Mauricio et al., 1997; Tiffin & Rausher, 1999; Mauricio, 2000; Pilson, 2000).

Figure 1.

Four hypothetical fitness (W denotes relative fitness) surfaces for the allocation to resistance (R) and tolerance (T). (a, b) Surfaces with two adaptive peaks and a valley of minimum fitness corresponding to intermediate levels of tolerance and resistance. (a) This surface corresponds to a scenario in which tolerance and resistance function as mutually exclusive alternatives, as maximum allocation to tolerance or resistance provides similar fitness benefits, but there are subadditive benefits of having intermediate levels of both strategies. (b) This surface corresponds to the situation in which the extent of the redundancy between tolerance and resistance is reduced because in this hypothetical example maximal allocation to tolerance provides more fitness benefits than maximum allocation to resistance. (c, d) Surfaces with one adaptive peak corresponding to a mixed pattern of allocation to tolerance and resistance (mixed defense strategy). These two panels differ in the position of the fitness peak and illustrate the range of hypothetical combinations of tolerance and resistance that correspond to a complementary defense strategy. When the presence/absence of different species of consumers accounts for qualitative or quantitative changes in the adaptive surface, diffuse selection can be invoked as a mechanism that can maintain the variation in plant defenses (Strauss et al., 2005).

Empirical evidence indicated that individual plants usually allocate resources to both strategies, expressing a range of combined patterns of defense allocation (mixed defense strategy; reviewed in Núñez-Farfán et al., 2007; Fig. 1c,d). More recent theoretical efforts examined the conditions for varied mixed defense strategies within host populations (Fornoni et al., 2004a; Restif & Koella, 2004). When the benefits of expressing both strategies simultaneously are higher than those gained when resistance and tolerance are expressed alone, these strategies represent complementary defense mechanisms (Fornoni et al., 2004a). Hence, if both strategies are complementary rather than redundant, an adaptive landscape with a peak selecting for a mixed allocation pattern of defense will represent an adaptive strategy (Fig. 1c,d). Nevertheless, whether this pattern is truly favored by selection under natural conditions or represents transient states to complete allocation to either tolerance or resistance (Fig. 1a,b) remains unresolved.

The presence of intermediate levels of tolerance and resistance can also result from fluctuating selection pressures related to variations in the amount, types and/or identity of the species imposing damage (reviewed in Núñez-Farfán et al., 2007). In particular, when the pattern of plant defenses selected against an enemy is altered by the presence/absence of an additional consumer, diffuse selection has been invoked as the underlying mechanism (Hougen-Eitzman & Rausher, 1994; Stinchcombe & Rausher, 2002; Strauss et al., 2005). Fluctuating selection acting on the tolerance–resistance adaptive landscape could reflect changes in the extent of redundancy or complementarity between the two strategies and could explain the presence of mixed patterns of defense allocation (Fig. 1). Even when the adaptive surface presents an intermediate peak, its position within the surface may be affected by changes in the selection pattern (Fig. 1c,d). Although diffuse selection has been demonstrated to act independently on tolerance (Stinchcombe & Rausher, 2002) and resistance (Pilson, 1996; Juenger & Bergelson, 1998; Stinchcombe & Rausher, 2001; Lankau, 2007), studies quantifying simultaneous selection on both strategies did not account for this source of variation (Tiffin & Rausher, 1999; Pilson, 2000; Fornoni et al., 2004b). Thus, determining whether tolerance and resistance function as redundant or complementary defenses and whether they depend on the ecological context constitutes a major step in explaining observed mixed patterns of defense allocation. To understand how the effect of several natural enemies conditions the fitness consequences of the simultaneous expression of tolerance and resistance, we experimentally manipulated the presence/absence of different species of natural herbivores to gain a clear picture of the adaptive value of mixed defense strategies.

In a natural habitat, plants are likely to be eaten by diverse species, each with a particular eco-evolutionary dynamic with the host (Johnson & Stinchcombe, 2007). Generalist herbivores are usually more susceptible than specialists to plant secondary metabolites (Ali & Agrawal, 2012), a pattern consistent with the observed adaption of several specialized herbivores to specific toxic compounds (Metcalf, 1986; Shonle & Bergelson, 2000; Agrawal et al., 2012). Whenever the benefits of resistance are reduced by the herbivores’ adaptation, tolerance, as a more general response, could play a major role in the plant's defense strategy, as tolerance can provide fitness benefits in the presence of damage (Agrawal & Fishbein, 2008; Garrido et al., 2012). Hence, when a given host is consumed by several natural enemies with different magnitudes of specialization, tolerance and resistance may function as complementary rather than redundant defense mechanisms. In this study, we performed an ecological genetic field experiment using controlled crosses in the annual herb Datura stramonium and two natural herbivores (Lema daturaphila and Epitrix parvula) to determine (1) the presence of genetic variation in tolerance and resistance against leaf damage; and (2) the pattern and intensity of genotypic selection in tolerance and resistance under combinations (presence/absence) of two natural enemies.

Materials and Methods

Study system

Datura stramonium L. (Solanaceae; jimsonweed) is an annual weed that frequently inhabits disturbed areas throughout North America (Valverde et al., 2001). Within central Mexico, mean herbivore damage can fluctuate between 10% and 50% among populations (Valverde et al., 2001), and 100% of leaf damage to individual plants can occur within populations (Núñez-Farfán & Dirzo, 1994). Plants from this species simultaneously express resistance by producing foliar trichomes and secondary compounds (tropane alkaloids; Shonle & Bergelson, 2000; Valverde et al., 2001) and exhibiting tolerance (Fornoni et al., 2004b). Both defense strategies are heritable, and selection in breeding values has been recorded in two natural populations (Fornoni et al., 2003, 2004b).

In central Mexico, D. stramonium is attacked by two folivore beetles: E. parvula and L. daturaphila (Chrysomelidae) (Fig. 2). E. parvula can consume a broader spectrum of Solanaceous species, while L. daturaphila is mainly a specialized enemy of the genus Datura (Kogan & Goeden, 1970). E. parvula begins to consume the foliage tissue at the seedling stage leading to, in some populations, up to 23% leaf area loss at the end of the plant's life cycle (Núñez-Farfán & Dirzo, 1994; Valverde et al., 2001; Fornoni et al., 2003). By contrast, L. daturaphila tends to oviposit on plants near the onset of reproduction, removing up to 100% of the leaf area of individual plants during the larval stage (Núñez-Farfán & Dirzo, 1994). The study site was located in Teotihuacan, Mexico (19°47′27′′N; 98°51′4′′W, 2294 masl; mean annual precipitation: 559.6 mm; mean annual temperature: 14.8°C; Valverde et al., 2001) within a xerophytic shrub community. At this site, an experimental plot (2048 m2) was established and plowed to remove natural vegetation before the seedlings were transplanted.

Figure 2.

Photographs of two beetles and types of damage imposed on the host plant Datura stramonium (Solanaceae). (a) Lema daturaphila adults. (b) Lema daturaphila third-instar larvae. (c) Leaf damaged by L. daturaphila larvae. (d) Epitrix parvula adult. (e) Leaf damaged by E. parvula adults.

Genetic material

During the summer of 2004, seeds from 115 plants were collected from a natural population of D. stramonium. Thirty seeds from one fruit on each plant were sown in plastic pots (2 l) and germinated in a glasshouse at the Instituto de Ecología of the Universidad Nacional Autónoma de México. One randomly chosen seedling from each family (i.e. seeds from the same fruit) were transplanted to a plastic pot (2 l) and manually self-pollinated to produce full-sibs. In 2005, these selfed full-sibling families were self-pollinated and grown again in the glasshouse to reduce maternal effects. After germination, in the first week of July 2006 a total of 64 full-sib maternal families (1536 plants) were transplanted to the experimental plot.

Experimental design

An ecological genetic field experiment with an incomplete factorial design was performed to manipulate the presence/absence of the most important herbivores of D. stramonium, L. daturaphila and E. parvula. We used a randomized complete block design in which eight siblings per family were assigned to each treatment combination (64 families × 4 blocks × 3 treatments × 2 replicates/family/block). Blocks were used to control for environmental variation. Plants were spaced 1 m apart in the experimental plot.

In the E. parvula treatment (E), beetles were allowed to naturally colonize plants while L. daturaphila were excluded. Lema daturaphila eggs and larvae were removed manually every 5 d, reducing the chances that laid eggs would hatch. Based on previous laboratory surveys, larvae emerge 5–7 d after oviposition (E. Garrido, pers. comm.); thus, eliminating eggs every 5 d ensured that no larvae emerged and ate foliar tissue. The L. daturaphila treatment (L) was established by spraying each plant with SevinXP® (Bayer de México, Ecatepec, México) at a concentration of 25 mg l−1 every week during the plants’ pre-reproductive period, as E. parvula mainly consume the plant during the seedling and juvenile stages (D. Carmona and J. Fornoni, pers. obs.). SevinXP® is a contact carbamyl insecticide reported to have little effect on plants and was used previously in the same study system (Shonle & Bergelson, 2000). An additional set of 50 plants was sprayed with the insecticide and the amount of damage compared with that of the combined herbivore treatment when both herbivores were present (i.e. the natural condition in this population). The mean leaf damage on the sprayed plants was 3.6% (± 0.0012% SE), while the mean leaf damage in the LE treatment was 13 ± 0.004% (Table 1). This indicated that the insecticide was highly effective in deterring herbivores in our experimental plot. We adjusted the frequency of the insecticide applications during the first 3 wk to determine the number of days after which the insecticide was no longer effective and did not affect the preference of L. daturaphila. As after 7 d of insecticide application L. daturaphila adults started laying eggs again, subsequent applications were performed once a week. Results indicated that levels of damage in treatment L were higher (14.6 ± 0.0038%; Table 1) than those observed in the LE treatment, indicating that applying the insecticide did not negatively affect the preference of L. daturaphila. However, we cannot rule out possible side effects that may be associated with our experimental manipulation of the presence of E. parvula. Thus, any inference derived from the selection regime in which plants are eaten only by L. daturaphila should be treated with caution. Plants in the LE treatment were those with natural levels of herbivore damage caused by the folivore species L. daturaphila and E. parvula (LE). Even though the type of leaf damage produced by both herbivores can be distinguished (Fig. 2c,e), damage inflicted by E. parvula is usually masked by the extensive damage produced by L. daturaphila (Fig. 2c). Thus, estimating the amount of damage imposed by each herbivore in the LE treatment is impossible. Plants that were not sprayed with insecticide were sprayed with water as a control. We recorded the following variables for each individual plant: transplant date; number of leaves after 26 d (initial size); percentage of foliar damage; and total fruit production. The percentage of foliar damage was recorded when fruits matured and was estimated using a Color Windows Image Analysis System (WinDIAS-Basic; Delta-T Devices Ltd, Cambridge, UK). Thirty leaves were sampled randomly per plant, unless plants had < 30 leaves, in which case all leaves were collected.

Table 1. Mean (SE) herbivory damage and fruit production under three treatment combinations: presence of Lema daturaphila and Epitrix parvula (LE), presence of L. daturaphila (L), and presence of E. parvula (E)
 TreatmentF (df) P
Damage13.0 (0.005)b14.6 (0.004)a7.4 (0.004)c122.06 (2,1164)< 0.0001
    χ2(df) P
  1. Differences in damage and fruit production among treatments (denoted by different superscript letters) were detected using Tukey's and Wald's test respectively.

Fruit production3.72 (0.528)b4.96 (0.511)a4.76 (0.525)a87.141 (2,1771)0.0375

Data analyses

Variation in damage and fitness among treatments

A one-way ANOVA and subsequent Tukey tests were performed to determine the effects of the amount of damage and plant fitness. Because the total seed number is highly correlated with the number of fruits (Fornoni et al., 2003), it was used as a proxy for maternal fitness. In addition, because D. stramonium is a highly selfing species, maternal fitness probably reflects the majority of seeds sired by a given genotype (Motten & Antonovics, 1992; Núñez-Farfán et al., 1996). Differences among treatments in the number of fruits were determined with the package glm of the program R (R version 2.11.1, 2010) which allows the generalized linear model to be fitted with a quasi-Poisson error distribution. Because we detected overdispersion, we corrected the standard error using a quasi-generalized linear model in which the variance was given by φ × μ, where μ is the mean and φ the estimated dispersion parameter (Zuur et al., 2009).

Operational definition of resistance and tolerance

Resistance for each individual plant was estimated as 1 – the proportion of damaged leaf area (Simms & Rausher, 1987), assuming that all plants were readily available for the herbivore population in the experimental plot. As all plants suffered from herbivore damage within the experimental plot, this assumption is probably validated. Given that we used multiple replicates per family, estimates of resistance at the genetic level are likely to converge to the true resistance value per family.

Tolerance was estimated as the relationship between fitness and the proportion of damaged leaf area (Mauricio et al., 1997; Strauss & Agrawal, 1999; Tiffin & Rausher, 1999; Stinchcombe & Rausher, 2002; Fornoni et al., 2003, 2004b). Before estimating tolerance, we examined whether the relationship between fitness and damage was nonlinear as this can bias the tolerance estimation (Tiffin & Inouye, 2000). Based on preliminary analyses indicating the absence of a general nonlinear effect of damage on fitness (inline image = 0.04; P = 0.8275) and the Akaike information criterion (AIC), we excluded the quadratic term in subsequent analyses. Hence, to obtain family estimates of tolerance, we performed regression analysis on fitness to obtain partial regression coefficients for each family (tolerance estimates; more details in the following section).

Genetic variation for fitness, tolerance, and resistance

As our experimental design was unbalanced and included random effects, we used mixed model methodology to test for significant random effects (Littell et al., 2002). We searched for genetic variations in fitness, tolerance, and resistance to herbivores using mixed models and the restricted maximum-likelihood (REML) iterative algorithm in the statistical module lmer of the program R (R version 2.11.1, 2010). A generalized linear mixed model (GLIMM) was used to detect genetic variation in fitness and tolerance because the fitness estimator (number of fruits) followed a Poisson distribution. In this case, the function glmer from the lme4 package of the program R (R version 2.11.1, 2010) was used with the quasi-Poisson error distribution and a log link function, given that preliminary analyses revealed important levels of overdispersion. The final model for fitness included the fixed effect treatment, and the random effects of block, family, and the interaction term family × treatment. Day of transplant and initial plant size (estimated as the number of leaves) were included as covariables to reduce the error mean square. AIC scores were used to select the best statistical model. Significant effects of family and its interactions with other factors were tested with the likelihood-ratio χ2 test (Zuur et al., 2009). A significant interaction between damage and family indicates the presence of genetic variation in tolerance. This interaction evaluates the presence of differences in the relationship between fruit number and damage among full-sib families. Because fruit number and proportion of damage were regressed using a log link function, our approach for estimating tolerance avoided the kind of bias that arises when the model that estimates tolerance combines additive and multiplicative scales and uses an identity link function (Wise & Carr, 2008). In our model, the interaction damage × family × treatment indicated that the expression of genetic variation in tolerance is conditioned by the herbivore species that cause damage. Following the same mixed model used for fitness, the genetic variation in resistance was analyzed using the GLIMM with a normal error structure.

Natural selection

Natural selection acting on resistance and tolerance was examined at the genotype level (Rausher, 1992) within each treatment following multiple regression analyses. For these analyses, full-sib family estimates for fitness, resistance, and tolerance from previous mixed models (best linear unbiased predictors (BLUP)) were obtained using the ranef function of package lme4 in R. Given that our experimental design was analyzed using mixed model theory because of unbalance, BLUPs are much better estimators of genotypic family values than standard least square means (Littell et al., 2002). Before the analyses were conducted, BLUPs for untransformed fitness values were relativized as (fitness + absolute value of the fitness of the family with lowest fitness)/(mean fitness + absolute value of the fitness of the family with the lowest fitness); thus, the mean relative fitness equals 1 (Pilson, 2000). BLUPs for tolerance and resistance were standardized to a mean of zero and a standard deviation of 1.

Because damage was used to estimate resistance and tolerance, we tested for a genetic correlation between the two defensive traits before conducting the selection analyses. Our results found no evidence of an association between tolerance and resistance, eliminating this source of collinearity in the analyses (see the Results section). Another possible artifact that can also affect the estimation of selection gradients in tolerance occurs when the mean and variance in fitness among genetic families are highly positively correlated. In this case, families with low variance in fitness are likely to express high levels of tolerance because of low statistical power to detect significant shallow slopes resulting in negative directional selection acting on tolerance (Agrawal et al., 2004). Previous studies in which this form of bias has been detected found negative directional selection acting on tolerance and a strong positive association between fitness means and variances (r ≥ 0.81; Agrawal et al., 2004; Baucom & Mauricio, 2008). In the present study, we found a similar association between fitness means and variances (r = 0.79; P < 0.0001) but no evidence of negative directional selection (see the Results section), suggesting that families with high levels of tolerance (shallow slope between fitness and damage) were not necessarily those with lower mean fitness. In addition, a bias affecting selection gradients on tolerance could arise because fitness is used in the predictor (tolerance) and the response variable in subsequent selection analyses. However, in an analysis of covariance on fitness in which the effects of family and family × damage are included, the partial regression coefficients for each family (tolerance estimates) are statistically independent from the family mean fitness values in the absence of multicollinearity. As a consequence of the lack of an algorithm for detecting such nonindependence between random terms, we estimated the variance inflation factor (VIF) from a model in which both effects (family and damage × family) were considered fixed. Results for each treatment indicated that these values were < 1.28, suggesting low chances of multicollinearity between family mean fitness and tolerance (Zuur et al., 2009). Finally, although BLUPs are better estimators of family means than least squares when family replicates are unbalanced, the use of BLUPs in selection analyses has an undesired statistical behavior (Hadfield et al., 2010); thus, any inference about future responses to selection should be made with caution.

To detect directional selection (β) acting on resistance and tolerance, only linear terms were included in the partial regression analyses. A second model was performed that included linear and quadratic effects to test for the presence of stabilizing/disruptive (γii), and correlational (γij) selection (Lande & Arnold, 1983). The estimation of the quadratic selection gradient (γii) was multiplied by 2 to obtain the correct selection strength (Stinchcombe et al., 2008). Differences among treatments in the pattern and the intensity of selection in plant defenses were examined using an ANCOVA on relative fitness. Further analyses between pairs of treatments were performed to test for the presence of differences in selection patterns (LE vs E and LE vs L). Significant differences between treatments E and L indicate that the simulated damage environments imposed different selection regimes.

Visualization of the resistance–tolerance fitness surfaces

Although Lande & Arnold (1983) observed that selection gradients can be used to detect general differences in the pattern of simultaneous selection on two traits, a better picture of selection patterns can be obtained through visualizing fitness surfaces. Detailed portrayals of the resistance–tolerance fitness surfaces in each treatment combination were obtained using a nonparametric spline procedure (Schluter & Nychka, 1994). This graphical analysis represents a useful complement to nonlinear selection gradients as it can detect more complex selection patterns. Selective surfaces were plotted using thin-plate spline fit, a three-dimensional analog of cubic spline (Zuur et al., 2009), using the Tps function from the fields package in R. The smoothing parameters for each spline were chosen based on generalized cross-validation.


Variation in damage and fitness among treatments

Plants growing in different treatments differed in the amount of foliar damage (F2,1164 = 122.06; P < 0.0001; Tukey HSD at 95% confidence; Table 1). The mean estimated damage in the LE treatment (i.e. where L. daturaphila and Eparvula were present) was 13%, and when Eparvula was excluded, the foliar damage imposed by L. daturaphila alone increased to 14.6%. However, when L. daturaphila was excluded, the foliar damage imposed by E. parvula was 7.4%, indicating that the amount of damage caused by the two herbivore species interacted negatively in a nonadditive way. These results showed that the main consumer of foliar tissue in the natural environment was L. daturaphila. Significant differences among treatments in the fruit production were detected (Table 1). Plants in the LE treatment produced a lower number of fruits per plant than in the other two treatments (Table 1). The numbers of fruits produced in the presence of L. daturaphila and E. parvula were similar, but significant differences in the amount of damage were detected (Table 1).

Genetic variation

Significant genetic variation among full-sib families was detected for fitness (fam; inline image = 778.05; P < 0.0001), resistance (inline image = 23.988; P < 0.001), and tolerance (damage × fam; inline image = 22.201; P < 0.0001; Supporting Information Table S1). In addition, the expression of genetic variation in these traits was conditioned by the presence/absence of particular herbivore species (Table S1). Genetic differences among families were observed for all traits in all treatment combinations, except for resistance in treatment E (Table S1). We did not detect the presence of a genetic correlation between resistance and tolerance in any treatment combination (LE: r = 0.123; P = 0.403; L: r = 0.032; P = 0.824; E: r = 0.059; P = 0.686).

Natural selection

Multiple regression analyses detected significant genotype selection acting upon tolerance and resistance. Manipulation of the presence/absence of folivore species altered the selection pattern in resistance and tolerance (Tables 2, S2, S4). In the presence of E. parvula alone (treatment E), analyses detected stabilizing selection favoring intermediate levels of resistance (F value, degrees of freedom, and probability for the full model including linear and quadratic effects: F5,42 = 3.145; = 0.0168; Table 2). Differences in the nonlinear component of selection on resistance were detected between treatments E and LE, indicating the presence of diffuse selection (i.e. differences in the pattern of selection against a focal species when another interacting species is present; Table S4). By contrast, in the presence of the other beetle (treatment L), only marginally directional selection on tolerance was detected (F value, degrees of freedom and probability for the linear model: F2,45 = 2.811; = 0.0707; Table 2). Although directional selection on tolerance was favored only in the presence of L. daturaphila (Table 2), we did not detect differences in the selection pattern in this strategy among treatments (Table S3). Between treatments L and E, significant differences were detected in the nonlinear component of selection on resistance, indicating that each simulated environment exerted a different selective regime on plant defenses in the experimental population (Table S2). In the treatment LE, the analysis detected a significant negative correlational selection gradient acting on the simultaneous expression of tolerance and resistance (F value, degrees of freedom, and probability for the full model including linear and quadratic effects: F5,42 = 3.141; = 0.0169; Table 2).

Table 2. Selection gradients (standard errors within parentheses) (directional (βi), stabilizing/disruptive (γii), and correlational (γij)) on tolerance and resistance obtained from independent multiple regression analyses in three treatments corresponding to the manipulation of the presence/absence of two folivore beetles: (LE) presence of Lema daturaphila and Epitrix parvula, (L) presence of L. daturaphila, and (E) presence of E. parvula
Defensive traitTreatment
  1. Significant effects are indicated in bold, and different letters correspond to statistically significant differences between pairs of treatments (critical α = 0.016 after a Bonferroni correction) in the pattern of selection obtained from an ANCOVA (further details are presented in Supporting Information Tables S2–S4). Each quadratic selection gradient (i.e., Resistance2 and Tolerance2) was corrected following Stinchcombe et al. (2008) (see the Materials and Methods section).

  2. **, < 0.01; , = 0.057.

Directional selection (βi)
 Resistance0.048 (0.054)−0.072 (0.051)0.035 (0.061)
 Tolerance0.063 (0.054)0.097 (0.050)−0.007 (0.065)
Stabilizing/disruptive selection (γii)
 Resistance2−0.128 (0.106)b−0.068 (0.095)b0.264 (0.079)a**
 Tolerance20.065 (0.066)−0.137 (0.073)0.150 (0.090)
Correlational selection (γij)
 Resistance × tolerance0.181 (0.006)a**−0.137(0.073)a−0.027 (0.072)a

Graphical inspection of the selective surfaces in all treatments supported the general result of the multiple regression analyses indicating the presence of one dominant fitness peak in the resistance–tolerance adaptive landscape (Fig. 3). Differences were also evident in the shape and position of the adaptive peaks depending on the presence/absence of the two herbivore species (Fig. 3). When both herbivores were present (treatment LE), the fitness surface indicated the presence of two partially alternative fitness peaks, one represented by high tolerance and intermediate levels of resistance, and the other by high resistance and low tolerance (Fig. 3; LE). Although tolerance and resistance interacted negatively, the adaptive surface indicated that the higher fitness peak corresponded to a mixed pattern of defense allocation (high tolerance and intermediate levels of resistance). This scenario partially matched the hypothetical scenario depicted in Fig. 1(a) as in our experimental population the two fitness peaks should not be considered true equivalent alternatives (the high tolerance–intermediate resistance peak provided higher fitness benefits than the other peak). This pattern contrasts with those observed in the presence of each beetle alone as only one fitness peak is favored in both cases. In the presence of L. daturaphila (treatment L), the fitness peak corresponded to high tolerance and low resistance (Fig. 3L), mimicking the hypothetical scenario depicted in Fig. 1(b) where only one defense mechanism is favored. In the presence of E. parvula (treatment E), the adaptive surface indicated the presence of an intermediate peak for resistance running parallel to the tolerance axis (Fig. 3E). This pattern partially matched the hypothetical surface simulated in Fig. 1(d), as only an intermediate optimum level of one strategy was detected. The observed changes between treatment LE and treatments E and L corresponded to a qualitative difference in the shape of the fitness landscape (i.e. transition between Figs 1a,b and c). In treatment E, the mean observed level of resistance and tolerance (circled cross, Fig. 3) was closer to the adaptive peak than in treatments LE and L (Fig. 3).

Figure 3.

Resistance–tolerance adaptive surfaces (left) and contour plots (right) describing the bivariate fitness function in each treatment. For each contour plot, the circled crosses indicate the mean level of tolerance and resistance observed in the experimental population. Treatments: LE, treatment with both herbivores present, L. daturaphila and E. parvula; L, only L. daturaphila; E, only E. parvula.


Rather than being redundant, tolerance and resistance in D. stramonium functioned as complementary defenses in the presence of only E. parvula or both herbivores, favoring the evolution of a mixed defense strategy. By contrast, L. daturaphila selected for increased tolerance and diminished resistance, suggesting that resistance is useless against this more specialized herbivore (Garrido et al., 2012). Although the idea that resistance and tolerance could function as complementary defenses against natural enemies has been theoretically examined (Tiffin, 2000; Fornoni et al., 2004a; Restif & Koella, 2004), we showed for the first time the validity of this expectation in a natural plant–herbivore system. Our results demonstrated that different experimental methods for altering herbivore damage resulted in different selection modes on resistance and tolerance, affecting the simultaneous evolution of tolerance and resistance and the extent of their complementarity. Although this may account for the presence of intermediate levels of both strategies in our population (circled crosses in Fig. 3), the observed mean level of tolerance and resistance was closer to the adaptive peak predicted when the two herbivores and E. parvula alone were present. Thus, these two scenarios favoring a mixed defense strategy are likely to be the most common selective environments and/or historical selection pressures exerted by E. parvula have been more intense than that imposed by L. daturaphila. Personal observations in the region confirmed that the presence of E. parvula and the two herbivores together are the most frequent combinations in central Mexico (J. Hernández-Cumplido & J. Fornoni, pers. obs.).

An important source of variation that can account for the presence of intermediate levels of tolerance and resistance is diffuse selection when traits are exposed to temporal and/or spatial fluctuation in the selection patterns derived from the presence of different consumer species (i.e. changes in the selective pattern acting on a particular trait involved in an interaction due to the presence/absence of a third species; reviewed in Strauss et al., 2005). Previous studies addressing diffuse selection focused on single traits (Pilson, 1996; Juenger & Bergelson, 1998; Stinchcombe & Rausher, 2002; Lankau, 2007; Sahli & Conner, 2011), although our approach using tolerance and resistance allowed us to test whether diffuse selection acted on the combined expression of the two defense strategies. Analyses detected changes in selection acting on resistance and not on tolerance, although this strategy was selected only in one treatment combination (i.e. treatment L). Further inspection of the adaptive surfaces indicated that, although the presence/absence of L. daturaphila changed the position of a combined fitness peak, the manipulation of E. parvula altered the extent of the complementarity, as when only L. daturaphila was present the combined fitness peak observed in the LE treatment changed to one in which maximum tolerance and minimum resistance were favored. Thus, our results indicated that changes in the amount of damage after our experimental manipulation of two herbivore species can affect the position of the tolerance–resistance fitness peak, and suggest that it may also affect the complementarity between the two defenses.

Selection on tolerance imposed by the beetle L. daturaphila, a common response to specialist herbivores, suggests that this herbivore has a long history of interaction with the plant and is more likely to be locally adapted. A recent cross-infection experiment involving D. stramonium and L. daturaphila in the same study site indicated weak local adaptation by L. daturaphila (Garrido et al., 2012) that may be enough to reduce the benefits of resistance. By contrast, resistance appears to be more effective against E. parvula, a consumer of several genera within the Solanaceae. The closer position of mean resistance and tolerance to the expected fitness peak in the presence of E. parvula and the two beetles together suggests that this herbivore has exerted more frequent and/or intense selection on resistance than L. daturaphila. Although both folivores belong to the Chrysomelidae and to the same functional guild, the narrower diet specialization of L. daturaphila suggests that the adaptive value of tolerance depends to some extent on the coevolutionary state of the interaction (reviewed in Fornoni, 2011).

Our results are consistent with Roy & Kirchner's (2000) theoretical expectation that, whenever tolerance provides a fitness benefit (when L. daturaphila eats the plants), natural selection should fix tolerance at maximum levels, whereas resistance is more likely to be maintained at intermediate levels (in the presence of E. parvula) because of the negative feedback with fitness. Although correlational selection of the form necessary to favor the evolution of a negative genetic correlation between the two strategies was detected, further analyses revealed no evidence that resistance and tolerance are being constrained by a negative genetic correlation between them. Despite genetic variation in tolerance and resistance still being available, we cannot rule out the possibility that further genetic constraints will prevent the evolution of optimum levels of plant defenses (Strauss & Agrawal, 1999; Tiffin & Rausher, 1999; Lankau, 2007; Johnson et al., 2009; Fornoni, 2011; Sahli & Conner, 2011).

Theoretically, a mixed defense strategy would be favored when the nonlinear cost and benefit functions of the two strategies interact to generate an adaptive surface with maximum fitness at intermediate levels (Fornoni et al., 2004a; Restif & Koella, 2004). The shape of the adaptive surface and the mean level of tolerance and resistance provided strong support for the hypothesis that the two strategies functioned as complementary defenses providing more than additive fitness benefits. The absence of redundancy and mutual exclusivity could occur because resistance does not reduce damage completely and tolerance does not fully compensate for the effect of damage (Mauricio, 2000). Our results suggest that functional complementarity could be favored in a scenario in which resistance is partially effective against herbivores and tolerance could provide additional benefits reducing the negative effect of damage on fitness. In this sense, the coevolutionary response of natural enemies could break down the functional redundancy as natural enemies can rapidly adapt to their host resistance (Garrido et al., 2012). This explanation is also consistent with evidence supporting the expectation of mutual exclusivity between tolerance and resistance against two abiotic selective agents (frost and glyphosate damage; Agrawal et al., 2004; Baucom & Mauricio, 2008). Hence, if the presence of a coevolutionary process can break down functional redundancy between tolerance and resistance to natural enemies, future models of plant defense should reconsider the expectation of redundancy.

Concluding remarks

Our results indicated that the average tolerance and resistance values are closer to the adaptive peak predicted in the presence of E. parvula and both herbivores together, suggesting that these scenarios are probably the most common under natural conditions in the studied population. Differences in the selection pressure imposed by two folivore beetles may result in the evolution of markedly different defense strategies. This suggests that, when a plant is eaten by several herbivore species, each with a particular history of interaction with the plant, the probability of finding functional redundancy between resistance and tolerance to the whole community of consumers will decrease. Thus, if different herbivores sharing the same host plant generate selection on resistance and tolerance, or a single enemy generates selection on both, both mechanisms may be maintained, resulting in the expression of a combined defense pattern. Although diffuse selection may actually affect the evolution of defensive traits, the most common ecological setting favors the evolution of a mixed defense strategy. Hence, the simultaneous expression of tolerance and resistance in our experimental population is more consistent with the predicted selective surfaces than with the presence of environmental or genetic constraints. Finally, if conditions for mutual exclusivity against natural enemies are rarely encountered in nature, our results could represent a general pattern.


We thank R. Baucom, S. Benitez-Vieyra, C. Bustos, C. Cordero, C. A. Domínguez, C. Macias, E. Garrido, K. Boege, J. Nuñéz-Farfán, M. T. J. Johnson, J. R. Stinchcombe, N. Turley, and Z. Cano-Santana for valuable comments. We also thank S. Benitez-Vieyra for statistical advice, Ruben Pérez-Ishiwara for technical help, L. Llamas, G. García, E. Campos, Z. Castro, and J. Hernández for valuable field assistance and S. Castro for the photographs of L. daturaphila and E. parvula. We also thank A. Kessler and anonymous reviewers for helpful comments on the manuscript. A. Fuentes provided logistic facilities to perform this study in his farm. The study was financed by grants CONACYT 89872 and PAPIIT 200807 to J.F. This paper constitutes a partial fulfilment of the Graduate Program in Biological Sciences of the National Autonomous University of México (UNAM). D.C. acknowledges the scholarship and financial support provided by the National Council of Science and Technology (CONACyT), and UNAM.