Herbivory selects for tolerance and constitutive defence across stages of community succession

Abstract

Plants defend themselves from herbivory by either reducing damage (resistance) or minimizing its negative fitness effects with compensatory growth (tolerance). Herbivore pressure can fluctuate from year to year in an early secondary successional community, which can create temporal variation in selection for defence traits. We manipulated insect herbivory and successional age of the community as agents of natural selection in replicated common gardens with the perennial herb Solidago altissima. In these genotypic selection experiments, herbivory consistently selected for better defended plants in both successional communities. Herbivore suppression increased plant survival and the probability of flowering only in mid-succession. Despite these substantial differences in the effects of herbivory between early and mid-succession, the selection on defence traits did not change. Succession affected selection only on aboveground biomass, with positive selection in early but not mid-succession, suggesting an important role of competition in the selective environment. These results demonstrate that changes in the community that affect key life-history traits in an individual species can occur over very short timescales in a dynamic secondary successional environment. The resulting community context-driven variation in natural selection may be an important, yet overlooked, contributor to adaptive mosaics across populations.

1. Introduction

Selection by herbivorous insects has shaped the ability of plants to avoid, cope with and overcome the negative impacts of being consumed. In general, plants defend themselves either by minimizing the damage from herbivores (constitutive and induced resistance) or by compensating for the lost tissue and, thus, minimizing the negative fitness effects of the incurred damage (tolerance). Tolerance and induced resistance are phenotypically plastic defence responses that are triggered by herbivore attack, whereas constitutive resistance is the continuously expressed defence level of the plant [1]. These strategies are not mutually exclusive and often plants deploy a mixed defence strategy [2]. However, trade-offs between induced and constitutive resistance are common [3] and there is some less consistent evidence for a similar negative association between resistance and tolerance [4–6]. Such variation in the relative importance of strategies for coping with herbivore pressure, as well as the remarkable variation among plant species, populations and individuals in their respective community contexts, warrants the question of the specific conditions that favour each one of these strategies.

Although herbivory is commonly identified as the main agent of selection on plant defence traits overall [7,8], the strength of selection by herbivores can depend on a whole suite of other factors [2,9]. Variable or low herbivory is in theory predicted to select for inducible over constitutive resistance strategies if resistance is costly, whereas constant strong herbivory should favour high constitutive resistance [1,10]. Because tolerance is inherently linked to plant growth traits [11–13], expression of tolerance can be favoured in conditions unrelated to herbivory through pleiotropy. For example, traits promoting growth may increase fitness in a competitive environment, resulting in concurrent positive selection for competitive ability and tolerance [14], although this is not always the case (e.g. [15]). Such genetically based trait associations can influence the outcome of natural selection more generally if selection on one trait restricts or promotes the expression of another trait [16,17]. Furthermore, the selection on phytochemical defence traits can be altered depending on the surrounding plant community. Such was the case in Brassica nigra, where selection for a defensive trait, leaf sinigrin concentration, was stronger when the surrounding plant community was composed of heterospecifics instead of conspecifics [18]. Taken together, the selective impact of herbivory on plant defence traits can vary depending on genetic trait associations, the surrounding plant community and the context in which plant–herbivore interactions are played out [19].

Ecological communities are formed and reformed in the process of ecological succession. Plant species and populations that persist in the community from early to mid-succession can experience dramatic changes in the associated suite of competitors, herbivores and mutualists over relatively short timescales [20–22]. Selection on plant traits can vary from year to year owing to variation in abiotic factors alone [23], and shifts in community composition as seen during succession can exacerbate temporal variation in selection [24,25]. Temporal variation in fitness optimums has been proposed to contribute to balancing selection and evolutionary stasis [26], to the maintenance of genetic variation [27] and to select for phenotypic plasticity in trait expression [28,29]. Given the ubiquity of agricultural fields which can give rise to secondary successional environments in contemporary landscapes [30], there are an abundance of possibilities for such temporally variable communities to arise. If these pervasive landscape-level shifts in community structure act as sources of variation for selection, they can be highly important for the maintenance of genetic variation in growth and defence traits in plants and that may explain the distribution of adaptive mosaics on a spatial scale.

In this study, we explore whether the strength of selection for plant defence traits by herbivores depends on the successional age of the plant community. We use the tall goldenrod, Solidago altissima, an ecological model species commonly occupying early secondary successional habitats. Our previous work demonstrated that genotypes collected from study plots of varying successional age were expressing higher constitutive defences in a common garden when they originated from older plots [21,25], suggesting strong selection by herbivores. Here, we manipulate the successional stage and arthropod herbivory as agents of natural selection and measure genotypic selection with 36 genotypes in replicate experimental successional fields with a fully crossed herbivory suppression treatment. These successional fields represent second and sixth year of succession since agricultural abandonment and differ dramatically in their plant, herbivore and soil microbial community composition [31–33]. Eliminating herbivory from half of our plots allowed us to examine the effects of succession in other regards, such as changes in aboveground competitive environment and soil microbial community. We measured defence traits (constitutive and induced resistance and specialized metabolites, tolerance) in a separate common garden to obtain genotype mean values for the traits and to explore genetic trait associations among them.

Fundamentally, we addressed the hypothesis that the strength of natural selection on plant defence traits varies with the successional environmental context on very narrow timescales and so can explain the distribution of adaptive mosaics in anti-herbivore defence expression. As we had previously observed a considerable increase in herbivory in the first 6 years of succession in this environment [33], we predicted that herbivores select for constitutive defence traits especially in later years of succession. If constitutive defences are costly and trade-off with inducible defences, the phenotypically plastic defence strategies (tolerance and inducibility) should be selected for in the early succession communities when herbivore pressure is low. If there are positive trait associations between tolerance and growth, successional age may cause concurring selection for both trait complexes. If there are differences in selection for defence traits between communities of different successional age, and these differences are driven by herbivory, herbivore suppression should alter selection for defence traits more in year 6 of succession compared with year 2 of succession.

2. Material and methods

(a) . Study system

Solidago altissima (Asteraceae) is a widespread perennial herb occurring across the United States, Canada and northern Mexico. It occurs in early successional habitats and is commonly found in landscapes disturbed by humans, such as old fields (abandoned, formerly cultivated or grazed lands). It colonizes old field habitats soon after agricultural abandonment, typically becoming the dominant species in the plant community after 3–4 years of succession by forming dense clonal stands with underground rhizomes [31,34]. Over a hundred arthropod herbivore species feed on S. altissima [35], but one of the most abundant herbivores in the study area is the chrysomelid beetle Trirhabda virgata. There is genetic variation in S. altissima in the strength of induction of resistance to T. virgata and a trade-off between the constitutive and induced resistance [33]. Resistance to herbivores is heritable and can evolve over relatively short timescales of less than 12 years [36,37]. Solidago altissima produces phenolic compounds and diterpene acids (specialized metabolites) in its leaves that mediate resistance against insect herbivores [38].

(b) . Plant genotypes

We used 36 genotypes of S. altissima in this study from a collection of genotypes that we knew to differ in constitutive and induced resistance [33]. We originally obtained the genotypes from an old field secondary succession experimental field in Brooktondale, NY (42°23′13″ N, 76°24′00″ W) in the autumn of 2016 by digging up individual plants, separated by at least 2 m. We propagated these plants through at least two rounds of rhizome-to-shoot growth cycles before the experiment to minimize maternal effects. During propagation, plants were kept in the greenhouse with a 16 h photoperiod and fertilized weekly (N : P : K 21 : 5 : 20, 150 ppm at 1 : 100 dilution with Epsom salts at 67.4 g l⁻¹).

(c) . Experimental design: field

To examine the interactive effects of herbivory and succession on selection for defence and tolerance, we conducted a fully factorial experiment in 2018 in Brooktondale, NY. We used the same series of old field secondary succession experimental plots from which we originally collected the S. altissima genotypes. The experimental field consisted of 20 plots of 30 m by 30 m (separated by 10 m) that represented old field succession between year 1 and year 6 after agriculture abandonment (more details in [31,33]). For this study, we used two plots in the second year and two in the sixth year of succession. Although we had only two plots per successional year, we know these plots to be very representative of the community succession in the area in general based on previously published work and our own observations [31–33,35]. The plant, herbivore and soil microbial communities differ considerably between the plots [31–33].

We started the experimental plants from rhizome cuttings in April, and kept them in the greenhouse for three weeks until hardening them in outdoor cold frames for one week prior to planting in the field. We had two replicates per genotype in each successional age (early or mid) and herbivory (ambient or suppressed) combination (altogether 288 plants). We chose to allocate more of the limited numbers of plants to the genotyping effort in the common garden (see below). We established two subplots in each of the four main succession plots on opposite sides from one another, and designated one to ambient control and one to herbivore suppression treatments (electronic supplementary material, figure S1). In mid-May, we planted replicate sets of the 36 genotypes in each of the eight subplots in a six-by-six grid with 50 cm between plants. To enable rhizome harvesting at the end of the season, we kept plants in pots (5.5 l azalea pots with Lambert 111 All Purpose Growing Mix) and embedded them in the soil with minimal disturbance to the surrounding vegetation. Our experimental plants experienced very realistic and strong competition aboveground, which is important in early secondary succession [39], although root competition was limited because of the pots. At planting, we covered the top of the pot with 250 ml of the local soil to inoculate the plants with the local soil microbial community. We treated the herbivory suppression plots biweekly with pyrethroid insecticide esfenvalerate (FenvaStar Plus, LG Life Sciences, Englewood Cliffs, NJ, USA). Following the manufacturer's instructions, we diluted the product by mixing it with water at a ratio of 8.2 ml to 45 l of water and sprayed it evenly across the treated plot. We sprayed ambient herbivory plots with an equal amount of water. The manipulation of the successional stage within a larger replicated succession experiment and the manipulation of herbivory by insecticide treatment within each common garden (see next section) allowed us to simultaneously measure community succession context and herbivory as agents of natural selection on various plant traits with sufficient replication.

We surveyed herbivory in June and measured plant size (height) and survival throughout the growing season every three to four weeks until early August and once more in late September. We counted the number of damaged leaves relative to the total number of leaves and estimated the abundance of the most common insect herbivores (T. virgata, Microrhopala vittata, Dichomeris sp., Corythucha marmorata, Uroleucon caligatum and Philaenus spumarius). At the end of the field experiment in early October we harvested separately the aboveground biomass (stem and leaves), inflorescences and rhizomes. We weighed all harvested material to the nearest 0.01 g after drying at 60°C for 72 h.

(d) . Experimental design: common garden

To obtain measures of constitutive and induced resistance, constitutive and induced specialized metabolites, and tolerance, we established a common garden at Cornell campus in Ithaca, NY. Although the expression of the defence phenotype is likely to slightly differ between the common garden and the successional field experiment, this approach allowed us to accurately estimate phenotypically plastic traits by complete protection of the undamaged control plants and better control over the amount of herbivory on the damage plants. We had 10–12 plants from each genotype (with the exception of one genotype with only six replicate individuals) which we started from rhizomes in April. Half of the plants were assigned to damage and half to control treatments. Plants were potted in the same pots and soil as the field plants and arranged in six blocks that were placed in two 4 m² tents made out of mesh fabric to exclude all additional herbivory (including insects, slugs and mammals). Plants received weekly fertilizer (N : P : K 21 : 5 : 20, 150 ppm at 1 : 100 dilution with Epsom salts at 67.4 g l⁻¹; Jacks Professional, J.R Peters Inc.) and silica (1 : 100; DynaGro Pro-Tekt, ‘The silicon solution’).

In early June, we applied three second instar T. virgata larvae on each plant in the damage treatment and allowed them to feed on the plant for 6 days, until plants had approximately 20% leaf damage. At the end of the damage treatment, we harvested the two youngest fully expanded leaves for bioassay and specialized metabolite analyses. For the phytochemical analysis, we removed the midvein of the leaf, flash froze the remaining tissue in liquid nitrogen and stored it at −80°C until extraction and high-performance liquid chromatography. We quantified 13 phenolic compounds and 4 diterpene acids from the samples. We standardized the signal intensity to the mass of fresh leaf tissue of each sample used for extraction, and used total plant specialized metabolite content (hereafter referred to as PSMs) for all further analyses (more details in the electronic supplementary material).

In the bioassay, the excised leaves were scanned and individually offered to locally collected pre-weighed second instar T. virgata larvae (starved for 10 h). Larvae and leaves were kept in a 9 cm Petri dish with the bottom covered in a thin layer of agar for moisture and at room temperature with 15 : 9 L : D photoperiod. After 87 h, we reweighed the larvae and scanned the leaves to determine the weight gain and leaf consumption of the larvae. From these data, we calculated the relative consumption rate as area of leaf consumed divided by mean weight of the larva between the beginning and end of feeding, to be used as a measure of resistance. We continued to grow the larvae in Petri dishes to identify those that were parasitized. Rate of parasitism by an unidentified nematode was 11% and all parasitized larvae were excluded from further analyses because the infected larvae grew more slowly (F_1,377 = 7.86, p = 0.005) and consumed less leaf material (F_1,377 = 10.70, p = 0.001).

After the bioassay, at the end of June, we caused additional damage on the plants assigned to the damage treatment by cutting each leaf in half perpendicular to the midvein to achieve approximately equivalent damage to that in natural populations of S. altissima in the field. At the beginning of October (ca 3.5 months after damage), we harvested aboveground tissue (stem and leaves) and weighed them to the nearest 0.01 g after drying at 60°C for 72 h. We estimated tolerance by subtracting the genotype mean aboveground dry biomass in the control treatment from that in the damage treatment (D – C; [13]).

(e) . Statistical analyses

We conducted all analyses in R (v.3.5.2; [40]).

(f) . Effects of treatments on herbivory, growth and fitness

We first tested for the effect of herbivore suppression and successional age of the plot on herbivore damage, herbivore abundance, asexual fitness (rhizome biomass) and probability of flowering [41]. We fitted the proportion of damaged leaves in early June, the number of insect herbivores observed on the plants and flowering (yes/no) to generalized linear mixed effects models (lme4: glmer; [42]) with herbivore treatment (ambient/suppression) and successional age (2/6 years) as fixed effects and genotype, plot and subplot as random effects. We included an observational-level random effect to correct for overdispersion for the herbivore damage and abundance models. We used Poisson distribution (log link function) for the herbivore count data and a binomial distribution (logit link function) for proportion of damaged leaves and probability of flowering. We fitted rhizome biomass (square root transformed) to a linear mixed effects model (lmerTest: lmer; [43]) with the same fixed and random effects as in the previous model. To explore whether plant size predicted flowering, and whether sexual and asexual reproduction are associated, we fitted the flowering (yes/no) with a binomial distribution (logit link function) to a mixed effects logistic regression with aboveground biomass, plant height and rhizome biomass (square root transformed) as continuous predictors, and plot, subplot and genotype as random effects (lme4: glmer). Significance of the fixed effects was obtained from Wald's chi-squared test for all generalized linear models and logistic regression (car: Anova) and from F tests for the linear model (lmerTest: anova). Significant differences between levels were obtained by contrasts (emmeans: emmeans; [44]) with Tukey's adjustment.

Of the 288 plants in the field, 106 died by the end of the experiment. More than half (70) of the dead plants had died owing to an attack by rodents (suspected culprit being meadow vole, Microtus pennsylvanicus). To determine whether mortality due to voles differed between treatments we first modelled the deaths by voles with a generalized linear mixed model (lme4: glmer) with a binomial distribution (logit link function). We then tested for differences in mortality that were not due to voles to examine effects of treatments on other mortality. Fixed and random effects were the same as in the previous models and the significance of fixed effects was tested with Wald's chi-squared test. We tested mortality due to voles and other causes separately because voles do not typically feed on S. altissima and this attack may have been motivated by having potted plants.

(g) . Genetic associations among traits

We characterized genetic covariance among all traits by correlating genotype mean values for each pairwise combination of traits using the Pearson correlation coefficients (n = 36 genotypes) [41]. We used genotype means for constitutive and induced resistance, constitutive and induced PSM content and tolerance (aboveground biomass, D – C) obtained from the common garden plants. We calculated resistance as −1 × relative consumption rate of T. virgata on control plants, and induction of resistance as −1 × (D – C relative consumption rate) so that resistance and inducibility increase with increasing trait values. Total aboveground biomass and height (at end of the growing season) were obtained from plants in the field. In addition to the Pearson correlation, for trait pairs where one was used in the calculation of the other (constitutive versus induced resistance, constitutive versus induced PSMs), we tested for the trade-off additionally with bias-corrected Monte Carlo procedure to account for spurious correlations [45].

(h) . Genetic variation and heritability

We calculated broad-sense clonal heritability H² by dividing total genotypic variance by total variance [46] to ensure the assumptions of the selection analyses were met. Ensuring high heritability of traits also enabled the comparison of field and common garden traits. We used linear mixed-effects models and restricted maximum likelihood (REML) with untransformed trait values as recommended by Houle [47] to obtain the estimates of genotypic variance for each trait. For simple traits from the common garden (resistance and PSMs of control and damage plants separately) and all field traits we used a model with each trait as the response variable and genotype and block in the common garden or field plot as random effects. Because genotypes varied in their response to damage treatment (common garden) and herbivore suppression (field) we calculated heritability separately for the damage/herbivore suppression treatments. To test for the genotypic variance in traits that describe a response to damage (induction of resistance, induction of PSMs, tolerance), we used the interaction of genotype with damage treatment [5]. We determined the significance of the random effects with the likelihood ratio test (lmerTest: rand). We divided the p-values from the likelihood ratio test by 2 because tests for significance of clonal genetic variance are one-tailed.

(i) . Selection on resistance and tolerance

We tested whether herbivory and/or succession affected natural selection on plant traits with a multivariate genotypic selection analysis [41,48,49]. We used rhizome mass to calculate relative fitness by dividing the within-treatment genotype mean rhizome mass by the overall mean rhizome mass. We chose to use this measure of clonal reproduction as our fitness measure because we were interested in how natural selection contributes to changes in genotype frequencies in the early successional environment, where the outcome of selection in S. altissima is determined primarily by differential survival and clonal spread in the population. Recruitment from seed occurs only in the first years of succession, after which the shifts in population mean genetic makeup come about through differential success in clonal reproduction, i.e. rhizomal growth [34,50]. We used genotype means from either the field (aboveground biomass, plant height) or common garden (constitutive and induced resistance, constitutive and induced leaf PSM content, tolerance), standardized to a mean of 0 and standard deviation of 1 to facilitate comparison of selection on traits measured on different scales and comparison between field and common garden data. For common garden traits, genotypes had one standardized mean value for each trait of interest, which was then matched to treatment-specific relative fitness measure in each of the four succession by herbivory treatment combinations. We performed a multivariate selection analysis that allowed us to test for the direct selection after correcting for trait covariation with a model that included the standardized traits, succession and herbivory, and all two- and three-way interactions between treatments and traits as fixed effects. Genotype was included as a random effect.

We explored the effects of treatments on selection by examining the significant interactions between traits and treatments. Significant trait × herbivory interaction indicates herbivore-mediated selection on the traits, whereas a significant succession × trait interaction indicates that selection on that trait varies between stages of succession. A significant three-way interaction between succession, herbivory and a trait would suggest that the strength of selection by herbivores differs between successional ages. To quantify treatment-mediated selection, we first obtained selection gradients under each level of the treatments (ambient and suppressed herbivory, years 2 and 6 of succession). Using these selection gradients, we calculated herbivore-mediated selection Δβ_H by subtracting the selection gradient in suppressed herbivory treatment from that in the ambient herbivory treatment. Similarly, we calculated succession-mediated selection Δβ_S by subtracting the selection gradient in year 2 from that in year 6 [51].

3. Results

(a) . Effects of treatments on herbivory, plant growth and fitness

Herbivore suppression decreased herbivore damage in year 6 of succession by 39% but it had no effect on it in year 2 when herbivory was overall lower (figure 1a; succession × herbivory, ${\chi }_1^2=10.72$, p = 0.001). In the ambient herbivory treatment, plants in year 6 plots were 68% more damaged than the plants in year 2 of succession (figure 1a). Herbivore abundance was 106% higher in year 6 compared with year 2 plots overall (figure 1b; succession, ${\chi }_1^2=4.30$, p = 0.038) and herbivore suppression treatment reduced their abundance to near zero (figure 1b; herbivory, ${\chi }_1^2=92.15$, p < 0.001). Only 35 plants out of the 182 plants that survived to the end of the experiment flowered. Probability of flowering increased over 17-fold with herbivore suppression in year 6 of succession (figure 1c; succession × herbivory, ${\chi }_1^2=3.99$, p = 0.046) whereas it was similar across both herbivory treatments in year 2 of succession (figure 1c). Rhizome biomass was unaffected by herbivore suppression or succession (electronic supplementary material, table S1). Probability of flowering increased with increasing aboveground biomass (${\chi }_1^2=6.36$, p = 0.012), and plant height (${\chi }_1^2=4.13$, p = 0.042), but was not affected by rhizome biomass (${\chi }_1^2=0.12$, p = 0.732). Mortality due to causes other than voles was highest in year 6 ambient herbivory treatment and herbivore suppression reduced it by 91%, whereas there was no difference between herbivory treatments in year 2 of succession (figure 1d; succession × herbivory, ${\chi }_1^2=6.72$, p = 0.009). Mortality due to vole attack was higher in the ambient herbivory treatment (0.19 ± 0.15) than in the herbivore suppression treatment (0.05 ± 0.05; herbivory, ${\chi }_1^2=7.43$, p = 0.006). Full results are presented in electronic supplementary material, table S1.

Figure 1.

The effects of herbivore removal (ambient or suppressed) and the successional age of the experimental field (2 or 6 years) on herbivory (a,b), flowering (c) and mortality of S. altissima (d) in a genotypic selection experiment. Mortality does not include that caused by voles. Graphs show estimated marginal means ± s.e. for each treatment. Interaction term succession × herbivory was significant for all except herbivore abundance. Significant differences between levels from Tukey's test with *p < 0.05.

(b) . Effects of herbivory and succession on selection of plant traits

All of the seven traits included in the selection analyses had significant heritable genetic variation among the 36 S. altissima genotypes (electronic supplementary material, table S2). Heritabilities were highest for constitutive and induced total PSM content and tolerance (range 0.33–0.42).

Herbivory altered selection on constitutive PSM content and tolerance. There was a negative directional selection on constitutive total PSM content in the absence of herbivores (β = −0.32, F_1,29 = 4.59, p = 0.041), but under ambient herbivory selection for PSM was not significant (β = 0.09, F_1,30 = 0.37, p = 0.550), resulting in a positive effect of herbivores on selection for PSMs (Δβ_H = 0.41; table 1; figure 2a). The effect of herbivory on selection for tolerance was similarly positive (Δβ_H = 0.33; table 1; figure 2b). The direction of selection changed from positive to negative with herbivore suppression although selection was not significantly different from zero in either herbivory treatment (ambient, β = 0.11, F_1,22 = 1.73, p = 0.202; suppressed, β = −0.21, F_1,27 = 1.64, p = 0.212). There were no significant three-way interactions between succession, herbivory and any of the traits (table 1).

Table 1.

Results from multivariate genotypic selection analyses examining effects of succession and herbivory on selection on growth and defence traits in Solidago altissima. d.d.f., denominator degrees of freedom. p-values < 0.05 are shown in italics.

	β	d.d.f.	F	p
herbivory (H)	−0.007	80	3.39	0.070
succession (S)	−0.983	81	12.21	0.001
aboveground biomass	0.824	91	14.86	<0.001
height	−0.139	79	2.37	0.128
PSM induction	−0.131	28	1.03	0.319
tolerance	0.115	26	0.04	0.844
resistance	0.136	32	0.65	0.424
resistance induction	−0.269	40	0.13	0.722
total PSMs	0.114	31	1.19	0.283
H × S	0.687	80	3.52	0.064
H × aboveground biomass	−0.272	83	0.12	0.728
S × aboveground biomass	−0.897	88	7.20	0.009
H × height	0.567	73	3.23	0.076
S × height	0.292	85	0.00	0.994
H × PSM induction	−0.178	71	1.42	0.238
S × PSM induction	0.225	67	2.85	0.096
H × tolerance	−0.319	65	4.32	0.042
S × tolerance	−0.030	67	0.25	0.621
H × resistance	−0.210	71	0.62	0.433
S × resistance	−0.324	70	2.29	0.135
H × resistance induction	0.303	72	0.36	0.552
S × resistance induction	0.517	74	3.55	0.064
H × total PSMs	−0.432	67	4.60	0.036
S × total PSMs	−0.156	71	0.01	0.910
H × S × aboveground biomass	0.684	77	3.03	0.086
H × S × height	−0.586	70	3.82	0.055
H × S × PSM induction	0.021	67	0.01	0.941
H × S × tolerance	0.173	68	0.58	0.448
H × S × resistance	0.174	72	0.31	0.582
H × S × resistance induction	−0.411	78	1.51	0.223
H × S × total PSMs	0.341	75	1.82	0.181

Figure 2.

The effect of herbivory on selection for (a) constitutive plant specialized metabolites (PSMs) and (b) tolerance in Solidago altissima.

Succession altered directional selection for aboveground biomass from strong positive directional selection in year 2 of succession (β = 0.75, F_1,40 = 24.19, p > 0.001; figure 3) to non-significant selection in year 6 (β = −0.09, F_1,43 = 1.19, p = 0.281; succession × biomass in table 1; figure 3). The effect of succession on selection (Δβ_S) for aboveground biomass was −0.84.

Figure 3.

Effect of successional age of the population (2 or 6 years) on selection on aboveground biomass in Solidago altissima.

(c) . Genetic associations among traits

There were several genetic correlations among traits (electronic supplementary material, table S3). Height and biomass were positively associated (r = 0.429, p = 0.009). Induction of PSMs was negatively associated with tolerance (r = −0.443, p = 0.007) and constitutive resistance (r = −0.445, p = 0.007). There was also a negative association between constitutive and induced resistance and constitutive and induced PSMs, suggestive of trade-offs between these traits. The bias-corrected correlation confirmed a trade-off between constitutive and induced resistance (r = −0.603, p = 0.033). However, after bias correction the trade-off between constitutive and induced PSMs was not significant (r = −0.422, p = 0.148).

4. Discussion

In this study, we explored the strength of selection by insect herbivores on defence strategies in S. altissima in two secondary successional communities that differed in time since agricultural abandonment by only 4 years. We found dramatic differences in the effects of herbivory on the probability of flowering and mortality of the plants between the two successional communities and with that demonstrate substantial temporal shifts in the potential of herbivores to shape plant populations during early succession. Nonetheless, herbivory altered selection for tolerance and constitutive PSMs similarly across both successional communities. Selection on plant aboveground biomass alone was altered by succession. These results highlight how rapidly the community context and selective pressures can shift during early secondary succession, challenging the perennial herbs in these environments to cope with conditions that change from one year to the next.

We found the strongest interactive effects of herbivory and succession on the probability of flowering and mortality. For both traits, herbivore suppression had an effect only in the experimental fields in later (year 6) succession, with herbivore suppression increasing flowering probability and survival. In year 2 of succession, plants were equally likely to flower or die regardless of the herbivory treatment. This pattern is likely driven by the difference in herbivore pressure overall because S. altissima plants received consistently more leaf damage in year 6 of succession than in year 2 in the ambient herbivory treatment (figure 1a; [33]). Although the difference between years 2 and 6 of succession in mortality and flowering probability were not significant in the ambient herbivory treatment per se, the difference in the effects of the herbivore suppression treatment highlights the key role herbivores play in this study system. Our ability to detect this difference may have been undermined by the loss of statistical power owing to unforeseen destruction of some of our study plants by voles. The shift in the importance of insect herbivory between S. altissima populations that were separated by only 4 years of community succession suggests that perennial herbs that persist as individuals in such temporally variable environments must be able to efficiently integrate different types of environmental information (e.g. about competition and herbivory [52]) as well as predict future interactions [53] when inducing phenotypic adjustments to these varying conditions. Strong temporal variation in environmental conditions as seen here can contribute to evolutionary stasis [54] and the maintenance of genetic variation [27].

Unlike flowering probability, asexual clonal reproduction measured as rhizome biomass was not affected by herbivory or succession treatment. The mismatch between effect of herbivory and succession on rhizome biomass and the probability of flowering emphasizes the importance of asexual reproduction for the persistence of S. altissima in the early successional environment. Although trade-offs between sexual and asexual reproduction in plants are common [55], we did not observe such a negative relationship as rhizome mass did not explain flowering probability. Clonality was the primary mode of reproduction in our study plants because 176 (96%) out of the 182 plants that survived had produced new rhizomes by the end of the experiment, whereas only 35 (19%) of them flowered. One of the evolutionary benefits of clonal reproduction is the rapid replication of adaptive genotypes in colonization of new environments [56] and it can thus be an essential determinant of microevolutionary shifts in genotype frequencies within a population [57]. Clonality allows maternal effects to contribute to the phenotype in the following season [58], which can be especially important for increasing defences in plants as a response to growing herbivore populations [59]. While clonal propagation can offer important insights to shifting genotype frequencies and microevolution occurring within one generation in a population [57], sexual reproduction is important for dispersal and colonization of new habitats, especially for plants inhabiting dynamic environments such as those undergoing secondary succession [60]. On the other hand, because investment in clonal propagation enhances sexual reproduction in plants in general [61], our measure of clonal fitness likely predicts the sexual fitness of the genotypes at large.

Herbivory affected directional genotypic selection for tolerance and constitutive PSMs regardless of the successional age of the study plot. Because the effects of herbivory on selection for both constitutive PSMs and tolerance were positive, it appears that herbivores select for a mixed defence strategy with intermediate values for both traits in S. altissima. Models of the evolution of the resistance and tolerance predict that variation in costs of associated traits among populations or environments should favour intermediate values for both traits [62]. This context-dependent variation in costs and benefits of different defence strategies seems likely to occur in early successional species like S. altissima where biotic environmental variation (specifically herbivory and competition) can be very high even within the same population over the course of a plant's lifetime [31,33]. For example, in Datura stramonium, selection on resistance and tolerance differed between two herbivores, but together they selected for mixed defence strategies [63]. In our study, the slope of the selection gradient for both PSMs and tolerance was not significantly different from zero in the ambient herbivory treatment, indicating that the plant population may be well adapted to the current levels and type of herbivory, or the damage level was below a threshold required for observing positive selection on tolerance [64]. However, the negative selection gradient for PSMs in the suppressed herbivory treatment indicates a cost of defence chemistry production when damage is reduced, a cost not apparent for tolerance.

Despite the differences in selection on constitutive PSM content (figure 2a) in ambient and suppressed herbivory treatment, we found no measurable effects of herbivory on selection on constitutive or induced resistance to T. virgata or induction of PSM content. Results of the bioassay are likely explained by species specificity of resistance traits, local adaptation, or dietary habituation of the herbivores used in the bioassay. First, we assessed resistance with only one herbivore species, and although it is typically the most abundant herbivore of S. altissima in the area in general as well as in our experimental plots [33], the plant is a host to over a hundred different arthropod herbivores [35]. Selection by the entire herbivore community might mask the effects of one single, albeit important, herbivore species that would not be detected in our bioassay. Second, the T. virgata beetles used in the experiment were sourced from natural populations and thus may be more or less affected by plant resistance traits, depending on their previous experiences with their host plant [65] or population genetic structure and local adaptation [66]. Heritabilities for resistance measures were relatively low, albeit significant, which could have also contributed to the difficulty in detecting selection. Selection for induction of PSMs would not have been subject to the problem of species specificity or local adaptation as is the case for the bioassay, and yet, likewise we detected no measurable selection for it. There was a negative genetic correlation between tolerance and induction of PSMs, suggesting that perhaps plants employ only one of these plastic strategies and as herbivory affected selection for tolerance, this could have overridden any selection on induction of PSMs.

Ecological species interactions are always played out against the backdrop of the surrounding community. It is well understood in plant–herbivore research that pairwise interactions change through time as, for example, herbivore populations change in size [67] and/or adapt to their host plants [68]. Likewise, the surrounding community can undergo temporal changes in species composition, but the effects that these changes have on the outcome of interactions a given species is involved in (ecological and evolutionary effects of community context) are much less explored. While our study is not optimal for exploring such context dependence in trophic interactions because of the low levels of herbivory in the early stage of succession, the fact that the successional age of the population did not modify selection by herbivores on the defence traits (i.e. there was no significant herbivory × succession interaction) suggests that the effect of herbivores on selection for defence traits is fairly robust across different community contexts in the S. altissima system. Succession alone did not alter selection on any of the defence traits either, which was surprising given the rather dramatic shifts in herbivory we observed in the ambient herbivory treatment, and the fact that competition can also modify selection for defence traits [15,18].

Succession affected selection for aboveground biomass only. We found positive directional selection for higher aboveground biomass in year 2 but not in year 6 of succession. Aboveground biomass in year 6 was overall lower than in year 2, and interestingly even the plants with relatively low biomass were able to reproduce asexually. These results suggest that there may be a shift from shoot competition in year 2 to stronger root/belowground competition in year 3. The very first years of secondary old field succession are typically dominated by fast-growing annuals, which are replaced by perennials (including S. altissima in the study area) after 2 to 3 years [31,69]. In Arabidopsis thaliana, traits that were selected for, as well as the strength of selection, varied depending on the identity of the competing species [70]. Competition in itself can, thus, contribute to the maintenance of genetic variation at small spatial scales depending on the immediate neighbourhood experienced by each plant [71]. Superimposing rapid temporal shifts in the plant community to fine-scale spatial variation in neighbour identity will likely contribute to the maintenance of genetic variation more than either of those processes alone. It may also favour strategies that allow perennial plants to both accurately perceive and construct information about their surroundings [72] and adjust their phenotype accordingly in order to maximize their fitness [73]. The contribution of the perceptive abilities of plants in their adaptation to temporally heterogeneous versus stable environments is an open question. We should note that this result is particularly sensitive to our measure of fitness; had we chosen to use a measure of sexual reproduction as a fitness proxy, the small plants would have zero fitness despite their rhizome production (not a single plant with a standardized biomass of −0.2 flowered). Comparing measures of selection acting via rhizome growth and flowering separately would give interesting insights into the types of selection that act on these two modes of reproduction, and the potential temporal variation over years in the relative contribution of each type of reproduction to individual fitness.

While temporal heterogeneity, or general unpredictability of herbivory, is hypothesized to be one of the main causes of selection for inducible responses to herbivory, studies that explicitly explore such patterns are scarce. Here we report the positive effects of herbivores on selection for tolerance but not for induction of resistance or PSMs. The strong interactive effects of herbivory and succession on flowering probability and mortality underline the role of herbivores in shaping the genotypic compositions of plant populations occupying these dynamic environments. Furthermore, these temporally variable environments are likely highly important for the maintenance of genetic variation and the emergence of adaptive mosaics across spatial scales.

Data accessibility

Data are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.b2rbnzskh [41].

Authors' contributions

A.Ka.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, visualization, writing—original draft; A.Ke.: conceptualization, funding acquisition, investigation, project administration, resources, validation, writing—review and editing.

Both authors gave final approval for publication and agreed to be held accountable for the work performed herein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by the Kone Foundation (A.Ka.) and New Phytologist Trust (A.Ke.).