Ontogenetic changes in the targets of natural selection in three plant defenses
Summary
- The evolution of plant defenses has traditionally been studied at single plant ontogenetic stages, overlooking the fact that natural selection acts continuously on organisms along their development, and that the adaptive value of phenotypes can change along ontogeny.
- We exposed 20 replicated genotypes of Turnera velutina to field conditions to evaluate whether the targets of natural selection on different defenses and their adaptative value change across plant development.
- We found that low chemical defense was favored in seedlings, which seems to be explained by the assimilation efficiency and the ability of the specialist herbivore to sequester cyanogenic glycosides. Whereas trichome density was unfavored in juvenile plants, it increased relative plant fitness in reproductive plants. At this stage we also found a positive correlative gradient between cyanogenic potential and sugar content in extrafloral nectar.
- We visualize this complex multi-trait combination as an ontogenetic defensive strategy. The inclusion of whole-plant ontogeny as a key source of variation in plant defense revealed that the targets and intensity of selection change along the development of plants, indicating that the influence of natural selection cannot be inferred without the assessment of ontogenetic strategies in the expression of multiple defenses.
Introduction
Both plants and animals display considerable ontogenetic variation in the expression of phenotypic traits, allowing them to deal with stage-specific selective pressures and different environmental challenges and ecological contexts as they develop. This variation can be promoted by resource allocation needs to different functions (e.g. growth, reproduction, defense), physiological and ecological costs, and/or restrictions inherent to developmental processes (Herms & Mattson, 1992; West et al., 2001; Hou et al., 2008; Maherali et al., 2009). Furthermore, in many cases, individuals may change their habit or resource use as they develop (i.e. ontogenetic niche shifts; Wilbur, 1980). In this context, morphological, physiological and/or behavioral traits expressed at particular ontogenetic stages can determine survival rates to the following stages, ultimately influencing their reproductive success (Gagliano et al., 2007). Hence, natural selection is expected to adjust the phenotypic expression of multiple traits involved in antagonistic and mutualistic interactions across the lifetime of individuals (Perez & Munch, 2010). For example, life-history studies have revealed that ontogenetic loss in decoy tail coloration in skink lizards can increase survival against predators when individuals are young, but this selective force is relaxed when they are older (Watson et al., 2019). Similarly, changes in the direction and strength of natural selection have been reported for the reef fish Pomacentrus amboinensis, with shifts from favoring small sizes early in their ontogeny to prevent starving, to faster growth in older individuals to reduce predation risk (Gagliano et al., 2007). In the case of plants, despite the accumulated evidence of changes in plant defensive traits (Senner et al., 2015) during the development of leaves (Coley & Barone, 1996; Koricheva & Barton, 2012; Wiggins et al., 2016; Barton et al., 2019) and during the whole ontogeny of individuals (Boege & Marquis, 2005; Barton & Koricheva, 2010), the study of natural selection on such traits has mostly focused on single points across the lifetime of individuals (Barton & Boege, 2017; but see Cope et al., 2019), usually controlling for leaf age (Mauricio & Rausher, 1997; Tiffin & Rausher, 1999; Agrawal et al., 2008a). A few reports suggest, however, that the intensity of natural selection can change across plant ontogeny (Tiffin, 2002; Gómez, 2008), promoting ontogenetic trajectories of ecophysiological (Maherali et al., 2009) and defensive traits (Cope et al., 2019). Whereas the degree to which natural selection can produce evolutionary responses and shape ontogenetic trajectories in plant defense depends on the degree of heritability and the influence of environmental variation, which can vary for different defensive traits. What remains to be explored is how natural selection acts on combinations of multiple defenses across plant development, promoting ontogenetic defensive strategies.
Here, we used a comprehensive approach focusing on the whole-plant ontogeny level, defined as the development of plants through different discrete ontogenetic stages (i.e. from seedling to juvenile/sapling, mature and finally senescent stages). Changes in the expression of defensive traits over the lifetime of plants (hereafter ontogenetic trajectories, sensu Boege & Marquis, 2005) have been reported for many plant species (Boege & Marquis, 2005; Barton & Koricheva, 2010; Quintero et al., 2013), but it remains unclear which are the drivers or mechanisms behind such ontogenetic variation. Because herbivore pressure can be variable along plant development (Hanley et al., 1995; Fenner et al., 1999; Warner & Cushman, 2002; Quintero et al., 2013), it can act as a selective force favoring the expression of greater values of defensive traits at the most vulnerable plant ontogenetic stages (Agrawal et al., 2012; Barton & Boege, 2017). For example, a recent study shows that the adaptive value of salicinoid phenolic glycosides changes during the ontogeny of Populus tremuloides (Cope et al., 2019). The benefit of particular defensive traits is likely to change during plant development as a function of the physiological priorities of different plant stages (Herms & Mattson, 1992), the fitness value of different organs (Boege & Marquis, 2005; Barton & Koricheva, 2010; Villamil, 2017), their efficiency deterring specific herbivores (Van Bael et al., 2003; Boege, 2005; Boege & Marquis, 2006) and/or as a result of ontogenetic niche changes derived from the interaction with different species (Perez & Munch, 2010; Fonseca-Romero et al., 2019; Villamil et al., 2019).
Myrmecophytic plant species represent an especially useful system to examine ontogenetic changes in the adaptive value of defensive traits, because developmental and architectural limitations constrain the expression of rewards for mutualistic ants in young plant stages (Quintero et al., 2013). Hence, their ecological niche changes from being nonmyrmecophytic to the interaction with their ant partners. We hypothesize that myrmecophytic plants should benefit from expressing high values of direct defenses in younger stages, when damage by herbivores and fitness costs are usually high, but they do not yet have the protection by ants (Fig. 1). By contrast, if ant defense is either more effective or less costly for the plants than the production of direct defenses, then natural selection should favor a reduction in the latter and greater production of rewards for ants (Fig. 1), according to the fitness benefits promoted by the presence of their mutualistic partners (Palmer et al., 2010; Stanton & Palmer, 2011; but see Fonseca-Romero et al., 2019).
Several studies have reported direct selection on individual or multiple defense traits such as trichomes (Valverde et al., 2001), secondary metabolites (Mauricio & Rausher, 1997; Shonle & Bergelson, 2000; Agrawal et al., 2008b; Agrawal, 2011) and even biotic defenses (Rudgers, 2004; Rutter & Rausher, 2004; Kessler & Heil, 2011; Quintero et al., 2013). However, plant age, size or ontogenetic stage are usually controlled to assess the direct and indirect impacts of these traits on relative fitness. In this context, the aim of this study was to quantify changes in the intensity of natural selection on different defense traits and assess their adaptive value across the ontogeny of a myrmecophytic plant species. Because early ontogenetic stages do not produce rewards for patrolling ants, we predicted that physical and/or chemical defense should be favored by natural selection at these stages. By contrast, later during plant development, when plants are able to produce rewards for ants, the adaptive value of these indirect defenses should increase as a result of the protection by patrolling ants (Rutter & Rausher, 2004), and because they can be less costly than direct resistance traits (Kessler & Heil, 2011). Accordingly, ontogenetic differences in the amount of plant defenses were expected to influence herbivore efficiency to assimilate plant tissues and hence their biomass accumulation.
In a previous study, we reported that risk of attack and leaf damage by herbivores on Turnera velutina plants used in this study were both greater for seedlings than for older stages (Ochoa-López et al., 2018), which suggests differences in the intensity of natural selection by herbivores across plant ontogeny. In this study, we assessed how the simultaneous phenotypic expression of three main defensive traits (hydrocyanic acid, trichome density (TD) and sugar in extrafloral nectar (SEFN) associated with ant patrolling) at three ontogenetic stages (seedling, juvenile and reproductive) influence the number of seeds produced (i.e. fitness) when plants reached their reproductive stage. We report ontogenetic changes in the direction and strength of natural selection on these defenses and the adaptive value of their joint expression across plant development. Lastly, to further assess the effectiveness of those defenses against the main herbivore, the specialist Euptoieta hegesia, we quantified herbivore assimilation efficiency, sequestration of hydrocyanic acid and performance (biomass gain) when larvae fed on tissues from different ontogenetic stages of T. velutina.
Materials and Methods
Study system
Turnera velutina Presl. (Passifloraceae) is a myrmecophytic shrub endemic to Mexico growing in coastal sand dunes and tropical dry forests (Arbo, 2005; Villamil et al., 2013). Turnera velutina produces axillary flowers, and their fruits are dehiscent capsules with an average of 36 seeds (Sosenski et al., 2017). Flowering occurs mostly from May to June. In this species, significant ontogenetic trajectories have been described for the production of hydrocyanic acid, TD and SEFN (Ochoa-López et al., 2015). Heritable variation has been reported only for the expression of TD (Ochoa-López et al., 2018). In addition, risk of attack by herbivores has been found to be greater in young seedlings than in reproductive plants, as females of the specialist butterfly Euptoieta hegesia Cramer (Lepidoptera: Nymphalidae) prefer to oviposit on leaves of the younger stage (Ochoa-López et al., 2018). Accordingly, the percentage of leaf damage by herbivores has been reported to be 1.5- to two-fold greater for seedlings than for older stages (Ochoa-López et al., 2015).
Experimental population
An experimental population of T. velutina was established in the coast of Veracruz, Mexico (19°36′N, 96°22′W), within the grounds of the Centro de Investigaciones Costeras La Mancha (CICOLMA). The climate in this area is warm sub-humid, with an average annual temperature ranging between 21.1°C in January and 27.3°C in June, and an annual precipitation ranging from 899 to 1829 mm (Travieso-Bello & Campos, 2006), mostly occurring between June and September. In June 2013, we produced an F1 generation (N = 300 plants) of T. velutina, using seeds from 20 maternal plants naturally growing in the shaded parts of the sand dunes. Flowers were self-crossed and excluded from pollinators and fruits were collected when ripe. After removing the elaiosomes, seeds from 20 maternal families were placed in germination trays in a mixture of local soil and vermiculite (1 : 1). Trays were bottom-watered for 3 wk until germination. F1 plants were then transplanted into 2 l pots and watered until their reproductive stage. In June 2014, between nine and 16 F1 plants per family were self-pollinated (using several flowers from each plant) to obtain a total of 2000 (F2) full-sib seeds. We germinated 75 seeds/family as previously described to produce 1200 seedlings. On August 2014, we established 20 plots (1 × 1 m) in the sand dune under the partial shade of vegetation canopy. When plants produced their first true leaf, between three and 10 individuals per family were transplanted to experimental plots using a 10 × 10 cm grid, where they were exposed to all biotic interactions and natural environmental variation.
Ontogenetic trajectories in plant defense
Although we generated plants from 20 genetic families, we only used 16 of them, with at least six surviving plants per ontogenetic stage. Subsets of plants were used to describe the ontogenetic changes in plant defense at three different ontogenetic stages. We quantified hydrocyanic acid (HCN), as a measure of the cyanogenic potential of leaves (Ballhorn et al., 2005), TD, and the amount of sugar produced in extrafloral nectar (SEFN) as a reward for patrolling ants. HCN and TD were quantified once plants reached one of three ontogenetic stages: seedlings (plants with the first two true leaves fully expanded, N = 213), juvenile (plants with the 10th leaf fully expanded, N = 220), and reproductive (plants bearing the first flowers, N = 225). SEFN was quantified only in the latter two stages, as seedlings do not produce extrafloral nectaries. Although we are aware that development is a continuous process (Ellner & Rees, 2007; Rees & Ellner, 2009; Merow et al., 2014), we chose these ontogenetic stages because they represent three key transitions in the functional priorities of plants (Boege & Marquis, 2005). Specific methods for assessing each defense are described in Ochoa-López et al. (2018). Briefly, using a colorimetric test with sodium picrate, we assessed cyanogenic potential (HCN content) on the most apical fully expanded leaf of each ontogenetic stage (i.e. plants were measured only once) (Schappert & Shore, 1995). To estimate HCN content (μg HCN g−1 DW), we used the formula μg HCN = (optical density of sample at 590 nm − optical density of blank)/0.000 1236, obtained from a standard curve (r2 = 0.93, P < 0.0001) using sodium cyanide (Code 7660-1; Caledon Laboratories Ltd, Halton Hills, Canada) as a source of HNC. TD (no. of trichomes mm–2) was measured in the penultimate fully expanded leaf of plants at each ontogenetic stage, using a stereoscopic microscope. Indirect defense was estimated by quantifying SEFN in the three most apical fully expanded leaves (only for juvenile and reproductive plants).
Plant fitness
In plants that reached maturity in the experimental population, we quantified seed number from ripe fruits collected monthly between October 2014 and February 2016.
Herbivore performance
A colony of E. hegesia was established in cages inside the glasshouse at the biological field station of CICOLMA. Caterpillars were collected (N = 84) shortly after hatching and reared in individual plastic containers, where they were randomly assigned to a diet consisting of either fully expanded leaves of reproductive plants or whole seedlings obtained from the same F2 genotypes (but different from the ones used for the natural selection experiment). Dehydration was avoided by covering petioles and seedling roots with a cotton ball. Plant material was offered to larvae ad libitum for 6 d. We estimated larval performance by registering their weight, length and head diameter on a daily basis. During this period, weight of fresh plant tissue was registered when added or subtracted from each container. Assimilation efficiency was estimated as the weight gained by the caterpillar divided by the weight of fresh food eaten during the 6 d. Because E. hegesia has been reported to sequester cyanogenic glycosides in Turnera ulmifolia (Schappert & Shore, 1999), we assessed whether this was the case for the larvae feeding on T. velutina. We assessed HCN contents in the bodies of each larva after feeding on either seedlings (N = 25) or reproductive plants (N = 25). Caterpillars were previously, frozen, lyophilized and dissected. Guts were removed to make sure that the released cyanide was indeed in the caterpillar's tissues. Released cyanide of the larvae tissue was measured using the same protocols as previously described for leaf tissue (Schappert & Shore, 1995; Ochoa-López et al., 2018).
Statistical analyses
Influence of plant defenses across ontogeny on seed production
We first evaluated if the three defensive traits changed along plant development, using a MANOVA and post hoc ANOVAs per defensive trait. Models included ontogenetic stage as a fixed factor, and family as a random explanatory variable. Then, to assess if the phenotypic effects of each plant defense on plant fitness varied as a function of the ontogenetic stage in which they are expressed, we adjusted two generalized mixed models with a Poisson distribution (lme4 package, R; Bates et al., 2015). The first was used to assess linear effects of each defense on seed production, considering family as a random explanatory variable, and ontogenetic stage, cyanogenic potential, TD, SNEF, and their interactions as explanatory fixed variables. The interaction between ontogeny and SEFN was not included in the model, as seedlings and most juvenile plants did do not produce nectar. The second model was used to assess nonlinear effects of the expression of plant defense across ontogeny on seed production. In addition to the previously described fixed and random variables, the quadratic terms of each defense and their interaction with ontogeny were included in the model. Values of each defensive trait were standardized as 
, where 
and σ represent the mean value and standard deviation, respectively, for each defense. Relative fitness values (number of seeds produced by each individual/population average of seed production) were considered as the response variable. A significant interaction between a given defense and ontogeny was considered as evidence that their adaptive value varied as a function of the ontogenetic stage in which they were expressed.
Ontogenetic changes in the intensity of phenotypic natural selection
To assess ontogenetic changes in the direction and/or intensity of natural selection on the three defenses, we performed phenotypic selection analyses, one for each ontogenetic stage, using a multiple linear regression model (Lande & Arnold, 1983), and standardized values of each defensive trait to predict relative plant fitness. After finding a lack of multivariate normal distribution of the three defenses (package mvn; Korkmaz et al., 2014) and exploring their collinearity through pairwise correlations and variance inflation factors, (package car, R software; Fox & Weisberg, 2011), we followed the approximation of Palacio et al. (2019) for multiple phenotypic traits with skewed distributions. We performed two independent models at each ontogenetic stage to obtain unbiased linear and nonlinear selection gradient estimates. Whereas linear models included only the defensive traits as explanatory variables, nonlinear models included the defensive traits, their interactions and quadratic terms. Because the residuals of our models departed from a normal distribution, to assess significance of the selection gradients we used nonparametric bootstrapping to estimate SEs with 95% bias-corrected bootstrap confidence intervals (Efron & Tibshirani, 1994) with the package boot (Canty & Ripley, 2017). Significant selection gradients were defined as those not including zero values within the confidence intervals.
Phenotypic selection analyses are useful to understand the ecological effects of the expression of multiple defenses across development on plant fitness, but do not allow the inference of the potential evolutionary responses of such ontogenetic trajectories. For those traits with significant genetic variation, considering genotypic means rather than phenotypic values ensures that the correlation between fitness and a given trait is genetically determined and not influenced by the environment (Rausher, 1992). Hence, genotypic selection gradients were calculated for TD, for which significant genotypic variation and heritability have been previously reported in the studied population (Ochoa-López et al., 2018). With this purpose, a standard selection analysis was performed for each ontogenetic stage using a linear regression model (Lande & Arnold, 1983) with the standardized values of the family means of TD per stage as explanatory variables, and the average relative fitness for each maternal family.
Defensive ontogenetic strategies
To explore the idea that different genotypes may express particular defensive ontogenetic strategies, which in turn should influence plant fitness, we assessed if particular combinations of the three defenses at each ontogenetic stage had an adaptive value. With this aim, we used the genotypic means of the three defense values per ontogenetic stage and performed a principal component analysis (PCA). Given the lack of genetic correlations among the three defenses across plant ontogeny (Ochoa-López et al., 2018), the mean genotypic value of each defense at a given ontogenetic stage was considered as an independent variable of the same defense in the other ontogenetic two stages. To visualize the ontogenetic strategies of defense for each genotype, we plotted the scores of the first and second principal components (PC1, PC2). This plot allowed the visualization of the functional space of multi-trait defensive strategies across plant ontogeny and revealed how plants expressed singular combinations of the three defenses at each ontogenetic stage. Then, we used the scores of the first principal component (which explained c. 30% of the total variation) to perform a multiple linear regression analysis, considering the relative fitness as the response variable, and the score of each ontogenetic stage as explanatory variables. Significant effects of the PC scores at a given stage would mean that the particular combination of defenses at that stage influenced plant fitness. Interactions among the scores of the three ontogenetic stages were also included to assess the adaptative value of multi-trait defensive strategies across plant ontogeny.
To evaluate herbivore performance as a function of plant ontogenetic stage, we used a generalized linear mixed model with repeated measures (Bates et al., 2015), considering day and ontogenetic stage of their host plant as fixed factors and caterpillar individual number as a random factor. An ANOVA was used to estimate the effect of diet on assimilation efficiency, considering the initial caterpillar weight as a covariate. The amount of HCN sequestration when feeding on seedlings or reproductive plants was assessed using an ANOVA with the ontogenetic stage of their food as fixed factor and considering the total amount of fresh leaf biomass consumed by larvae as a covariate. All statistical analyses were performed in R v.3.6.0 (R Development Core Team, 2019).
Results
Ontogenetic trajectories in plant defense
After 2 years in the field, T. velutina plants from 16 genotypes showed significant ontogenetic trajectories in the expression of three defensive traits (Wilks' λ = 0.2642, F2,639 = 200.73, P < 0.0001; Fig. 2), as previously reported (Ochoa-López et al., 2018). The main pattern was that cyanogenic potential decreased almost twice from the seedling to the reproductive stage (F2,51 = 7.52, P = 0.002; Fig. 2a), TD increased four-fold (F2,51 = 386.9, P = 0.0001; Fig. 2b) and SEFN was produced only in reproductive plants (Fig. 2c). This pattern was consistent for all genotypes (family, Wilks' λ = 0.9243, F19,639 = 0.892, P = 0.702).
Ontogenetic changes in the adaptive value of defensive traits
When considering the simultaneous expression of the three defenses across plant ontogeny, we found that the influence of TD and cyanogenic potential on seed production varied as a function of the ontogenetic stage in which they were expressed (ontogeny × either defensive trait; Table 1), suggesting that the adaptive value of each direct defense changed across plant development. After performing the phenotypic selection analyses, including both defenses at each ontogenetic stage (and SEFN in the case of reproductive plants), we found that, indeed, the targets, direction and intensity of natural selection changed across plant ontogeny. Specifically, negative directional and disruptive selection was found on the cyanogenic potential at the seedling stage (β = −0.3504, γ = 0.3385; Table 2; Fig. 3a). Intermediate values of TD were also favored at this stage (γ = −0.6789; Table 2; Fig. 3b). For juvenile plants, we found negative directional and disruptive selection on TD (β = −0.4621, γ = 0.7871; Table 2; Fig. 3e). Finally, at the reproductive stage, high phenotypic values of TD were favored both at the phenotypic (β = 0.4527; Table 2; Fig. 3h) and genotypic (β = 0.358, P = 0.01) levels. In addition, a positive correlational selection gradient was detected acting on cyanogenic potential and sugar content in extrafloral nectar (γ = 0.3081; Table 2; Fig. 4).
| Effect | df | χ 2 | P-value |
|---|---|---|---|
| Ontogenetic stage | 2 | 1.980 | 3.714 |
| Cyanogenic potential | 1 | 0.791 | 0.3737 |
| Extrafloral nectar | 1 | 2.556 | 0.1098 |
| Trichome density | 1 | 10.744 | 0.0010 |
| Cyanogenic potential × extrafloral nectar | 1 | 12.340 | 0.0004 |
| Cyanogenic potential × trichome density | 1 | 0.839 | 0.3596 |
| Trichome density × extrafloral nectar | 1 | 3.255 | 0.0711 |
| Ontogenic stage × cyanogenic potential | 2 | 12.881 | 0.0015 |
| Ontogenic stage × trichome density | 2 | 73.355 | < 0.0001 |
| Cyanogenic potential2 | 1 | 4.424 | 0.0354 |
| Extrafloral nectar2 | 1 | 0.347 | 0.555 |
| Trichome density 2 | 1 | 33.505 | < 0.0001 |
| Ontogenic stage × cyanogenic potential2 | 2 | 20.853 | < 0.0001 |
| Ontogenic stage × trichome density2 | 2 | 38.242 | < 0.0001 |
- Significant values are shown in bold.
| Ontogenetic stage | Trait | β | CI | γ | CI |
|---|---|---|---|---|---|
| Seedling | HCN TD HCN × TD |
−0.3504 ± 0.17 0.1891 ± 0.17 |
−0.5886, −0.1753 −0.0195, 0.4745 |
0.3385 ± 0.27 −0.6789 ± 0.26 −0.2268 ± 0.22 |
0.0999, 0.6809 −1.2226, −0.3652 −0.5319, 0.0240 |
| Juvenile | HCN TD HCN × TD |
0.012 ± 0.18 −0.4621 ± 0.18 |
−0.2478, 0.3040 −0.9557, −0.1043 |
−0.6189 ± 0.39 0.7871 ± 0.25 0.1305 ± 0.29 |
−2.6893, 0.2113 0.2688, 1.7611 −0.3925, 0.7711 |
| Reproductive | HCN TD SEFN HCN × TD HCN × SEFN SEFN × TD HCN × TD × SEFN |
−0.0614 ± 0.09 0.4527 ± 0.09 0.0131 ± 0.09 |
−0.3137, 0.0986 0.2737, 0.7015 −0.1463, 0.1989 |
0.1880 ± 0.21 0.1437 ± 0.18 0.0726 ± 0.10 0.0753 ± 0.09 0.3081 ± 0.15 −0.1582 ± 0.13 −0.0379 ± 0.12 |
−0.1282, 0.6623 −0.2539, 0.6362 −0.1773, 0.3359 −0.1351, 0.2459 0.0458, 0.6739 −0.4335, 0.0843 −0.2563, 0.1726 |
- Significant selection gradients were determined with 95% bias-corrected bootstrap confidence intervals. Significant values are shown in bold.
After performing a series of bioassays with the specialist herbivore E. hegesia, we found greater amounts of HCN in the tissues of larvae that fed on seedlings than on reproductive plants (F1,48 = 25.53, P < 0.01; Fig. 5a). In addition, larvae consumed more leaf biomass (F1,48 = 24.82, P < 0.01) and grew faster when fed on high cyanogenic seedlings than on leaves from reproductive plants (Fig. 5b). This can be explained by their assimilation efficiency, which was 35% greater when feeding on the younger plant stage (F1,48 = 10.67, P < 0.01; Fig. 5c).
Multi-trait ontogenetic strategies in plant defense
By using a PCA, we visualized the functional space of multi-trait defensive strategies of each genotype across plant ontogeny and revealed how plants expressed singular combinations of defensive traits at each ontogenetic stage (identified with different ellipses and colors in Fig. 6). Greater values of PC1 denote high values of TD and SEFN, and low values of HCN. High scores of PC2 represent high values of SEFN and HCN (Table 3; Fig. 6). When analyzing the covariation between the scores of the first PC on the relative fitness of each genotype, we found a significant effect of particular combinations of defensive traits expressed at the reproductive stage on genotype mean fitness. Interestingly, we found that the three-way interaction term PC1 seedling × PC1 juvenile × PC1 reproductive (F1,8 = 13.49, P = 0.0063; Table 4) had a significant effect on the production of seeds.
| Defensive trait | PC1 | PC2 |
|---|---|---|
| HCN | −0.549 | 0.736 |
| SEFN | 0.561 | 0.675 |
| TD | 0.619 | 0.041 |
| Variance explained | 0.333 | 0.333 |
| Cumulative variance | 0.333 | 0.667 |
- Defensive traits used were hydrocyanic acid (HCN), sugar content in extrafloral nectar (SEFN) and trichome density (TD). PC1/2, first/second principal components.
| Effect | df | F | P-value |
|---|---|---|---|
| PC1 seedling | 1, 8 | 5.264 | 0.0509 |
| PC1 juvenile | 1, 8 | 2.452 | 0.1560 |
| PC1 reproductive | 1, 8 | 13.821 | 0.0059 |
| PC1 seedling × PC1 juvenile | 1, 8 | 0.955 | 0.3571 |
| PC1 seedling × PC1 reproductive | 1, 8 | 0.150 | 0.7086 |
| PC1 juvenile × PC1 reproductive | 1, 8 | 0.053 | 0.8231 |
| PC1 seedling × PC1 juvenile × PC1 reproductive | 1, 8 | 13.486 | 0.0063 |
- Significant values are shown in bold.
Discussion
Ontogenetic changes in the adaptive value of defensive traits
We present experimental evidence showing that the targets, direction and intensity of natural selection on three defense traits changed across the ontogeny of T. velutina plants. Specifically, low cyanogenic potential was favored at the seedling stage, while high TD was unfavored at the juvenile stage but favored at the reproductive stage. Although previous studies have reported variation in the strength and direction of selection as a result of environmental and temporal variation (Grant & Grant, 2002; Siepielski et al., 2009), this is the first study reporting changes in the targets of natural selection on different defenses across plant ontogeny. In addition, we present evidence suggesting that specific combinations of the three defenses at each ontogenetic stage can influence seed production.
Both the risk of attack and foliar damage by the specialist herbivore E. hegesia in the field are greater early during the ontogeny of T. velutina, when plants are not large enough to attract patrolling ants (Ochoa-López et al., 2018). These ontogenetic changes in herbivore damage were similar among genotypes (F19,329 = 0.68, P = 0.84). Given this selective pressure and the lack of ant defense in seedlings, we expected natural selection to favor greater amounts of either HCN or TD as alternative defenses at the youngest stage. Surprisingly, we found directional selection against cyanogenic potential and stabilizing selection for TD at this stage. Although seedlings had the highest concentration of HCN compared with other stages (Fig. 2a; Ochoa-López et al., 2018), individuals with high cyanogenic potential produced fewer seeds when they reached maturity (Fig. 3a). Although the few individuals with the highest cyanogenic potential had a small increment in fitness, this was not comparable with individuals with low cyanogenic potential. The lack of effectivity of HCN as a defense could be a result either of the costs associated with its synthesis (Goodger et al., 2006; Kessler & Heil, 2011; Neilson et al., 2013) or due to the ability of the specialist herbivore E. hegesia to sequester cyanogenic glycosides (Schappert & Shore, 1999; this study). Although cyanogenic glycosides have been related to an effective reduction of leaf damage by generalist herbivores in some plant species (Ballhorn et al., 2005; Thompson & Johnson, 2016), specialist herbivores commonly show an array of strategies to detoxify or sequester these compounds (Engler et al., 2000; Urbanska et al., 2002; Pentzold et al., 2014), rendering them ineffective at reducing herbivore damage (Schappert & Shore, 1999; Gleadow & Woodrow, 2002; Ballhorn et al., 2005, 2007; Shlichta et al., 2014; Hernández-Cumplido et al., 2016). Sequestration of cyanogenic glycosides is a well-known mechanism in specialist caterpillars feeding on cyanogenic plants to reduce risk of predation (Nahrstedt & Davis, 1983; Nahrstedt, 1985; Schappert & Shore, 1999; Nishida, 2002), which seems to be the case for E. hegesia as well.
Indeed, E. hegesia performed better when feeding on younger stages of T. velutina, which could be associated with a greater nutritional value of leaves at this stage. In fact, a previous study reported greater concentration of nitrogen in leaves of juveniles than in leaves from reproductive plants of T. velutina (Damián et al., 2018). Hence, even if cyanogenic glycosides could have a negative effect on caterpillars, it seems to be offset by the benefits of feeding on more nutritious leaf tissue. Although alternative mechanisms of plant defense, such as compensatory responses, have been reported in T. velutina seedlings (Ochoa-López et al., 2015), the replacement of lost tissue at the seedling stage does not seem to be enough to avoid a reduction in seed production when these plants start reproducing. Hence, the observed high contents of HCN in seedlings might be a result of current selection by generalist herbivores and/or of past selection events, before the specialist herbivore was able to sequester cyanogenic compounds. If this is the case, we suggest that E. hegesia has currently overcome the potential arms race with T. velutina on the coevolution of chemical defenses and counter-defenses (Mello & Silva-Filho, 2002).
Trichome density has been previously associated with reduced herbivore damage (Kärkkäinen et al., 2004; Kaplan et al., 2009) and increased plant fitness (Valverde et al., 2001; Bingham & Agrawal, 2010). In fact, previous studies have found positive selection on TD in Asclepias incarnata (Agrawal et al., 2008a) and Arabidopsis spp. (Mauricio & Rausher, 1997; Kärkkäinen et al., 2004; Loe et al., 2007). In the case of T. velutina, we found that the direction and intensity of the selection gradient changed along plant ontogeny (Fig. 3; Table 2). In seedlings, where resources are limited and risk of herbivory is high (Boege & Marquis, 2005), we found that intermediate TDs were favored (Fig. 3b), which could be an alternative to low values of HCN. Whereas in juvenile plants TD was mostly unfavored by selection (Fig. 3e; again, except for a few individuals with high TD and relative fitness), it was significantly favored in reproductive plants (Fig. 3h). Because trichomes play an important role in controlling evapotranspiration, their adaptive value could be related to both a reduction in herbivore damage (Agren & Schemske, 1993; Mauricio & Rausher, 1997; Kärkkäinen et al., 2004; Loe et al., 2007) and water stress avoidance (Ehleringer et al., 1976; Skaltsa et al., 1994; Espigares & Peco, 1995; Sandquist & Ehleringer, 1998, 2003a,b; Agrawal et al., 2008a). Hence, these three ontogenetic stages might have different selective pressures by herbivores, resource availability and water stress. Disentangling the role of these factors, however, warrants further investigation. Because TD had significant genetic variation and heritability in the studied population (Ochoa-López et al., 2018), it is the only trait for which the observed ontogenetic trajectories could potentially evolve through natural selection. By contrast, despite their influence on plant fitness, the evolutionary responses of HCN and SEFN seem to be constrained by environmental heterogeneity. Hence, whereas the ontogenetic trajectories of some defenses could be the result of evolutionary responses to natural selection, others are likely instead to be shaped by plastic responses under variable environments (Whitman & Agrawal, 2009).
For myrmecophytic plants, a reduction in SEFN can decrease ant recruitment, increase herbivore damage and reduce fitness (Kessler & Heil, 2011). Resource-mediated attraction of natural enemies of herbivores generally results in a clear top-down control of herbivore populations, with positive fitness effects for the plant (Miller, 2007). However, direct tests of selection on these indirect defenses are still rare. Previous studies have reported a positive relationship between fitness and the number and size of extrafloral nectaries and the volume of nectar produced (Rudgers, 2004; Rutter & Rausher, 2004). Here, we report a positive correlative selection gradient for sugar content in extrafloral nectar and cyanogenic potential at the reproductive stage of T. velutina, when the interaction between plants and ants is more intense. Although reproductive plants had the lowest cyanogenic potential compared with younger stages, the positive correlative gradient could be the result of more vigorous plants producing more defenses of different kinds. Even when we hypothesized that older plants with active ant colonies should not rely on direct defenses, if there are exploiter ant species that do not provide an effective defense against herbivores, the expression of some amounts of chemical defense should be beneficial. For example, in the myrmecophytic species Volchisia hindsii, the ontogenetic decrease in HCN has been found to vary as a function of the identity of ants colonizing the plants, which can be either true mutualists or opportunistic ant species (Fonseca-Romero et al., 2019). In the case of T. velutina, further investigation is needed to assess whether variation in ant identity and efficiency corresponds to the differential expression of HCN and ant rewards.
A previous study showed that greater amounts of extrafloral nectar are produced in leaves bearing flowers (Villamil, 2017). Therefore, SEFN seems to be related to the attraction of patrolling ants that protect not only leaves, but also flowers, which offer highly nutritious tissues and are highly vulnerable to other consumers (Villamil, 2017). Recent studies also suggest that greater production of SEFN in leaves bearing flowers represent a mutualist–management strategy, by distracting ants from flower nectar to avoid ant–pollinator conflicts (Villamil et al., 2018, 2019).
Ontogenetic strategies in plant defense
Ontogenetic changes in the intensity and targets of natural selection can have important consequences for the evolution of lifetime plant strategies, because relative fitness does not depend on the expression of a single defense at one ontogenetic stage, but rather on what plants do earlier or later during their development. Successful genotypes would be those adjusting their phenotypes to match the best defensive strategy at each ontogenetic stage (Fig. 6). Although we found a significant effect of the expression of different combinations of defenses at each ontogenetic stage (as indicated by the significant three-way interaction among the PCs for each stage), further investigation is needed to interpret the adaptive value of particular ontogenetic strategies in plant defense.
Although other studies have already reported ontogenetic switches in defensive strategies from resistance to tolerance (Boege et al., 2007), from chemical to biotic defense (Ochoa-López et al., 2015; Ochoa-López et al., 2018), or in the expression of different secondary metabolites (Goodger et al., 2013), to our knowledge, this is the first report on how changes of defensive strategies along ontogeny are related to differential reproductive outputs. These ontogenetic strategies can represent adjustments in the expression of each defensive trait when they are more effective or less costly. The optimal defense (Bryant et al., 1992) and the growth-differentiation (Herms & Mattson, 1992) theories predict that selection should favor the production of defensive traits when herbivory risk is higher, in this case the seedling stage. On the other hand, as plants grow, the pool of resources is divided on the different physiological needs of the plant. Later ontogenetic stages have greater availability to acquire resources (and, in the case of myrmecophytic plants, attract mutualistic ants), but also have more physiological needs (e.g. reproduction). We have shown, however, that the adaptive value of defensive traits at different ontogenetic stages also depends on the expression of other defensive traits earlier or later during plant development. Incorporating a demographic approach could further enlighten our understanding of these ontogenetic strategies, as it can reveal how the expression of particular defensive traits can affect plant survival, the transitions among ontogenetic stages and overall plant fitness.
Concluding remarks
Studies on the evolution of plant defense have provided evidence of natural selection on heritable defensive traits (Berenbaum et al., 1986; Simms & Rausher, 1987; Mauricio & Rausher, 1997; Shonle & Bergelson, 2000). Here, we further demonstrated the influence of ontogeny as source of variation determining the adaptive value of three defenses in plants growing in their natural environment over 2 yr. There are three main take-home messages arising from this work: the targets of selection and the adaptive value of different defensive traits can change across plant ontogeny; some ontogenetic trajectories in plant defense are more likely to show evolutionary response than others; and ontogenetic strategies involving the expression of multiple defensive traits influence the overall effects of natural selection. These findings suggest that we need to reconsider how and when selection analyses are performed, to fully understand the evolution of phenotypes.
Acknowledgements
We thank Rubén Perez-Ishiwara for his assistance in fieldwork and laboratory assays. We also thank P. Zedillo, N. Villamil, L. Ochoa, L. López, G. García, C. Peralta, A. Bernal, J. Aguilar, I. Lemus, I. Lemus, M. Castañeda, S. Soria, C. Manriquez, B. Ramírez, M. Maldonado, M. Ramirez, F. Ayhllon, A. López, I. Gongora, S. Salazar, J. Campuzano and B. Esquivel for their invaluable help in the field and shadehouse, and all CICOLMA staff members for all the assistance in the facilities. Special thanks to A. Agrawal and two anonymous reviewers for feedback on previous versions of this manuscript. Funding was provided to KB by PAPIIT-UNAM (IN-211314). SO-L acknowledges CONACyT and the graduate program Posgrado en Ciencias Biológicas at the Universidad Nacional Autónoma de México for academic and financial support.
Author contributions
SO-L, JF, CAD and KB conceived and designed the experimental population and the statistical analyses. SO-L and XD collected data form the experimental population and SO-L performed the statistical analyses. KB and RR designed the experiments of herbivore performance. RR performed the experiments of herbivore performance and analyzed the data. SO-L and KB led the writing of the manuscript, and XD, JF and CAD contributed with further editorial enhancements.
