Macroevolution of plant defenses against herbivores in the evening primroses



  • Plant species vary greatly in defenses against herbivores, but existing theory has struggled to explain this variation. Here, we test how phylogenetic relatedness, tradeoffs, trait syndromes, and sexual reproduction affect the macroevolution of defense.
  • To examine the macroevolution of defenses, we studied 26 Oenothera (Onagraceae) species, combining chemistry, comparative phylogenetics and experimental assays of resistance against generalist and specialist herbivores.
  • We detected dozens of phenolic metabolites within leaves, including ellagitannins (ETs), flavonoids, and caffeic acid derivatives (CAs). The concentration and composition of phenolics exhibited low to moderate phylogenetic signal. There were clear negative correlations between multiple traits, supporting the prediction of allocation tradeoffs. There were also positively covarying suites of traits, but these suites did not strongly predict resistance to herbivores and thus did not act as defensive syndromes. By contrast, specific metabolites did correlate with the performance of generalist and specialist herbivores. Finally, that repeated losses of sex in Oenothera was associated with the evolution of increased flavonoid diversity and altered phenolic composition.
  • These results show that secondary chemistry has evolved rapidly during the diversification of Oenothera. This evolution has been marked by allocation tradeoffs between traits, some of which are related to herbivore performance. The repeated loss of sex appears also to have constrained the evolution of plant secondary chemistry, which may help to explain variation in defense among plants.


Plant–herbivore interactions have driven the adaptive diversification of an arsenal of plant defenses against herbivores (Fraenkel, 1959; Ehrlich & Raven, 1964; Agrawal, 2007; Futuyma & Agrawal, 2009). This diversification is hypothesized to have resulted from the large diversity and abundance of herbivorous arthropods (Hutchinson, 1959; Strong et al., 1984; Novotny et al., 2006; Basset et al., 2012), the negative effects of herbivory on plant fitness (Marquis, 1984; Crawley, 1985; Hawkes & Sullivan, 2001), and over 400 million yr of coevolution (Ehrlich & Raven, 1964; Labandeira, 2007). Despite over a century of research (Hartmann, 2008), we lack a solid understanding of the relative importance of the various plant traits involved in reducing herbivory (i.e. resistance traits) (Coley, 1983; Carmona et al., 2011), tolerating or escaping damage (Strauss & Agrawal, 1999; Núñez-Farfán et al., 2007; Carmona et al., 2011), and the factors that cause variation in these traits among plant species (Berenbaum, 1995; Stamp, 2003; Agrawal, 2007, 2011; Endara & Coley, 2011). Here we seek to shed light on these gaps by studying the macroevolution of defenses in the evening primrose genus Oenothera.

Plants defend themselves against herbivores using a complex array of resistance traits. Early research on the evolution of plant defense focused on the role of plant secondary metabolites (Stahl, 1888; Dethier, 1941; Fraenkel, 1959; Ehrlich & Raven, 1964; Feeny, 1976; Rosenthal & Berenbaum, 1992). Although most biologists still view secondary metabolites as a plant's primary line of resistance (Berenbaum & Zangerl, 2008; Macel et al., 2010), there is clear evidence that the mechanisms and evolution of defense involve a wider diversity of traits, and that not all secondary chemicals function in defense (Coley, 1980; McNaughton et al., 1989; Carmona & Fornoni, 2013). This view was underscored by a recent meta-analysis which failed to detect any consistent effect of standing intraspecific genetic variation in secondary metabolites on resistance to herbivores (Carmona et al., 2011). Instead, physical, morphological, and life-history traits were more consistent predictors of resistance against herbivores. Based on these and other results, Carmona et al. (2011) hypothesized that the defensive function of secondary metabolites might be most apparent over macroevolutionary timescales, where variation in secondary chemistry is greater. If true, then it should be possible to observe the effects of secondary metabolites by experimentally measuring plant traits and herbivore performance from multiple plant species that have recently diversified (e.g. within a single genus or family) and using comparative phylogenetic approaches to expose the role of secondary metabolites in resistance to herbivores (Weber & Agrawal, 2012). Relatively few studies have implemented such an approach (Becerra, 1997; Agrawal & Fishbein, 2008; Kursar et al., 2009; Pearse & Hipp, 2012), and fewer have compared the relative roles of secondary chemistry and nonchemical traits at the macroevolutionary scale of a single clade of confamilial or congeneric species (Agrawal & Fishbein, 2006; Pearse & Hipp, 2009).

Plant species are not all equally defended against herbivores, which is evidenced by the wide variation in the types of resistance traits and degrees of resistance among species. This variation is attributed to a combination of two phenomena. First, plant populations adapt to maximize their fitness to local biotic and abiotic environmental conditions. Secondly, evolutionary tradeoffs prevent populations from maximally defending themselves against all herbivores in every environment. These tradeoffs can involve traits that affect growth, competition, tolerance to stress, mutualisms, and defense against natural enemies (Herms & Mattson, 1992; Fine et al., 2006). Tradeoffs may also occur in allocation to different types of traits (Twigg & Socha, 1996; Rudgers et al., 2004), which seems especially likely given that each plant produces dozens to hundreds of secondary metabolites, plus the additional array of resistance and tolerance traits that do not involve secondary metabolites (Strauss & Agrawal, 1999; Núñez-Farfán et al., 2007; Macel et al., 2010; Walters, 2011). Despite the common belief that tradeoffs are ubiquitous and mediate much of the variation we see in defense, evidence for tradeoffs is mixed, with no consistent overall trend (Koricheva et al., 2004; Leimu & Koricheva, 2006).

In addition to tradeoffs, recent research suggests that selection also creates positively covarying suites of traits that form ‘defensive syndromes’ (Feeny, 1976; Kursar & Coley, 2003; Agrawal, 2007). The defense syndromes hypothesis makes two specific predictions, which together provide support for the existence of defense syndromes (Agrawal, 2011). First, suites of traits should consistently covary with one another, such that groups of species are expected to converge onto different defense syndromes. There is fairly strong support for this prediction (Becerra, 2003; Kursar & Coley, 2003; Agrawal & Fishbein, 2006), but perhaps not at broad phylogenetic scales (Moles et al., 2013). The second prediction is that these syndromes should provide better protection against herbivores than individual traits on their own. There are few specific tests of this second prediction (Agrawal, 2011).

Plant species may also vary in degrees of defense because of inherent constraints on adaptive evolution. For example, suppressed sexual reproduction in plants is expected to decrease the ability of plants to adapt to their parasites and purge deleterious alleles in plant defense genes (Levin, 1975; Bell, 1982; Otto, 2009). We recently tested this hypothesis and found that functionally asexual evening primroses, Oenothera spp., were less resistant to generalist herbivores than sexual species, but more resistant to specialist herbivores (Johnson et al., 2009b). These patterns in resistance were predicted, in part, by the evolution of decreased concentrations of tannins in asexual Oenothera. A major limitation of this previous work was that we measured relatively few traits, and our characterization of plant chemistry was restricted to an assay of protein precipitation capacity (Hagerman, 1987). Such measures of chemistry give an incomplete measure of evolution in plant traits related to resistance against herbivores (Appel, 1993; Salminen & Karonen, 2011).

Here we combine comparative phylogenetics, analytical chemistry, and large-scale experiments to examine the macroevolution of plant resistance in Oenothera. We first test the hypothesis that the composition and concentrations of plant secondary metabolites are evolutionarily conserved. We then ask whether traits negatively covary with one another, thereby showing support for tradeoffs in defensive strategies as predicted by classic plant defense theories. Alternatively, we ask where there are positively covarying suites of traits as predicted by the defensive syndromes hypothesis. Furthermore, we ask, do either individual traits or suites of positively covarying traits predict resistance to herbivores? Finally, we ask, is a loss of sex associated with the evolution of altered concentrations or composition of secondary metabolites? This study builds on our previous work by characterizing the composition of phenolic compounds in the leaves taken from the same plants used to characterize resistance to generalist herbivores in Johnson et al. (2009b). We combine our newly collected phenolic data with the previously collected measures of plant traits and insect resistance to perform novel analyses that answer the hypothesis and these questions.

Materials and Methods

Study system

We used Oenothera (Onagraceae) as a model for studying the macroevolution of defense in plants against herbivores. Oenothera is a New World genus comprising 147 species found from southern South America to boreal Canada (Wagner et al., 2007). Species are largely herbaceous, but they vary in life history from annuals to monocarpic biennials and longer-lived perennials (Evans et al., 2005; Wagner et al., 2007).

Oenothera exhibits a genetic system that renders some species functionally asexual (Cleland, 1972; Darlington, 1980; Ranganath, 2008). Most Oenothera species reproduce via more or less typical sexual reproduction (Cleland, 1972), although some sexual species also experience reduced recombination (Rauwolf et al., 2011). Nearly a third of species (29%) are functionally asexual because of a genetic system called permanent translocation heterozygosity (PTH), which causes a nearly complete suppression of recombination and segregation (Cleland, 1972; Rauwolf et al., 2008). This genetic system occurs in eight plant families (Holsinger & Ellstrand, 1984; Johnson, 2011), and it has independently evolved > 20 times in the Onagraceae (Johnson et al., 2009b, 2011). We refer to PTH as being functionally asexual because the progeny of PTH individuals are genetically identical to their parents when they self-fertilize, which is predominantly the case. Unlike apomixis, PTH individuals are diploid and produce seeds via syngamy, which makes it possible to examine how repeated suppression of recombination and segregation affect plant evolution. Thus, this genetic system avoids the confounding effects of variation in ploidy, which is typically associated with apomixis (Asker & Jerling, 1992; Whitton et al., 2008).

Oenothera species are attacked by a wide diversity of generalist and specialist arthropod species (Dickerson & Weiss, 1920; Johnson, 2011). Oenothera exhibits substantial variation in secondary chemistry, physiology and life-history traits (Zinsmeister & Bartl, 1971; Howard & Mabry, 1972; Dement & Raven, 1973; Zinsmeister et al., 1977; Averett & Raven, 1984; Johnson et al., 2009a), and many of these traits provide resistance against herbivores (Johnson & Agrawal, 2005; Johnson, 2008; Johnson et al., 2009a,b). Herbivores have a large effect on plant fitness (Johnson & Agrawal, 2005) and they can drive rapid evolution in plant traits (Agrawal et al., 2012a).

Our study focused on resistance of Oenothera to both generalist and specialist herbivores. The data on herbivory that we present were obtained in the laboratory and the field, and were originally presented in Johnson et al. (2009b). The generalist herbivores were the beet armyworm (Spodoptera exigua, Noctuidae: Lepidoptera) and the two-spotted spidermite (Tetranychus urticae, Tetranychidae: Acari), which feed on Oenothera as well as species from other plant families. We also assayed resistance to a flea beetle, Altica foliacea (Chrysomelidae, Coleoptera), which primarily feeds on Oenothera spp., but it does occasionally feed on other species in the same order (Myrtales), plus apple (Malus domesticus). For simplicity, we refer to A. foliacea as the ‘specialist beetle’. Finally we assayed resistance of plants to naturally colonizing herbivores in the field, where only generalist feeders were observed. Although the herbivory data were previously published, we present new chemistry data, and all analyses are new.

Experimental methods

This study extends our previous work, which examined the consequences of plant sex for the evolution of resistance against herbivores (Johnson et al., 2009b). Here we characterize the main phenolic metabolites (Fig. 1, Supporting Information, Table S1) and perform comparative phylogenetic analyses to address broader questions about the macroevolution of resistance. To allow for direct comparisons between phenolic chemistry and our previous measures of resistance to arthropods, we used leaf tissue collected from a subset of the same plants studied in Johnson et al. (2009b). In the following we provide a brief description of the experimental setup; full details are available in Johnson et al. (2009b).

Figure 1.

Diagrammatic representation of the biosynthesis of the three classes of phenolic compounds quantified in this study. The diagram shows the main biosynthetic pathways that lead to the production of phenolic compounds (within green boxes) present in the leaves of Oenothera spp. Arrows indicate the flux of metabolites from one pathway to another, where numerous enzymatic steps are involved. Some alternative pathways and biosynthetic loops are not shown for simplicity. Based on Salminen & Karonen (2011) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (Kanehisa & Goto, 2000; Kanehisa et al., 2012).

We conducted a large growth chamber experiment to quantify the performance of generalist herbivores and variation in plant traits for 26 Oenothera species (Fig. 2). The initial study included one additional species (Oenothera filiformis), which we removed because of insufficient material for phenolic analysis. Species were selected to maximize phylogenetic diversity across Oenothera while providing adequate replication of independent transitions between sexual and functional asexual (PTH) reproduction. We did this by selecting species from the two major clades of Oenothera (Wagner et al., 2007) and included 12 sexual and 14 PTH species that represent a minimum of 11 independent transitions between the two genetic systems (Johnson et al., 2009b). In total, we grew 679 plants and assayed resistance of each plant to the generalist caterpillar and mite. These herbivores were placed individually on detached leaves in Petri dishes. For caterpillars, we measured the amount of tissue consumed on each leaf, their fresh weight and their survival after 6 d. We assessed mite survival and the number of eggs laid after 5 d. In the original study, we measured four nonchemical plant traits (leaf toughness, % leaf water content, trichome density and specific leaf area (SLA)), plus protein precipitation capacity (PPC) of leaf extracts (see Johnson et al., 2009b).

Figure 2.

Maximum likelihood phylogeny of the 26 Oenothera species included in the growth chamber and field experiments. We indicate whether species reproduce sexually (SEX) or functionally asexually as a result of permanent translocation heterozygosity (PTH). The x-axis shows the phylogenetic covariance among species, which was used to calculate the variance-covariance matrix in comparative phylogenetic analyses. The numerals indicate which phenotypic cluster of traits within which each species falls according to the dendrogram of phenotypic traits (see Fig. 4). The phylogeny shown is pruned from the RAxML tree inferred in Johnson et al. (2009b). The tree is fully resolved although short branch lengths appear as polytomies.

We conducted a second experiment in the field to measure susceptibility of plants to naturally colonizing herbivores, which were dominated by generalist feeding herbivores. We also used these plants to conduct a Petri dish bioassay of resistance to the specialist beetle. Here we use data from 19 of the original 25 species in the field experiment, for which chemical analyses were performed using tissue collected in the growth chamber experiment. We planted on average 35 replicate plants per species into an old field near Durham (NC, USA). Resistance to the specialist beetle was assayed as the amount of tissue consumed per leaf. Resistance to generalist herbivores in the field was measured as percentage tissue removal from each plant over the entire summer.

Chemical analyses

To characterize phenolic metabolites, we analyzed a subset of samples (= 5 per species) from plants grown in the growth chamber experiment. We focused on characterizing flavonoid glycosides (FGs), CAs and ETs (Fig. 1), because these are the main phenolics present in Oenothera spp. Moreover, these phenolics can confer resistance or susceptibility against herbivores by creating oxidative stress and binding proteins in the guts of herbivores (Appel, 1993; Salminen & Karonen, 2011; Salminen et al., 2011). Some phenolics may also confer susceptibility to herbivores if they serve as feeding stimulants or provide benefits as antioxidants (Kolehmainen et al., 1995; Johnson et al., 2009b).

We excised leaves from plants for phenolic analysis at the same time that leaves were removed for bioassays. A single fully expanded nonsenescing leaf was excised from each of five plants per species using a razor blade. These leaves were stored in a −80°C freezer, freeze-dried for 3 d and homogenized to a fine powder in 1.5 ml microcentrifuge tubes with a 3 mm stainless steel bead using a Geno-Grinder 2000 (OPS Diagnostics, Lebanon, NJ, USA). The metal beads were then removed and samples were stored in a −20°C freezer until phenolic extraction in 2011.

We characterized and quantified phenolic composition of Oenothera leaves using 10 mg of leaf powder from each sample, extracted in 950 μl of acetone: water (7: 3, v/v, containing 0.1% ascorbic acid, w/v) for 3 h with a Vortex-Genie 2T mixer (Scientific Industries, Bohemia, NY, USA). After centrifugation (10 min at 16000 g; Eppendorf centrifuge 5402; Eppendorf AG, Hamburg, Germany), we stored the supernatant in a separate vial and repeated the extraction on the original tissue with fresh solvent. Acetone was removed from the pooled extracts using an Eppendorf concentrator 5301 (Eppendorf AG), and the sample was freeze-dried. Before analysis, we dissolved the sample in 400 μl of water and filtered it through a 0.20 μm polytetrafluoroethylene filter.

We analyzed phenolics from each sample using ultraperformance LC with diode array and electrospray mass spectrometry detectors (UPLC-DAD-MS) (Waters Acquity UPLC and Waters Xevo TQ triple quadrupole mass spectrometer; Waters Corporation, Milford, MA, USA). We used a Waters Acquity UPLC BEH Phenyl (1.7 μm, 2.1 × 100 mm) column with CH3CN (A) and 0.1% HCOOH as eluents. The gradient was as follows: 0–0.5 min, 0.1% A (isocratic); 0.5–5 min, 0.1–30% A in B (linear gradient); the flow rate was 0.5 ml min−1. UV spectra were recorded for each peak between 195 and 500 nm, and full-scan negative mode MS spectra were acquired in the m/z range 100–2000. For quantitative analyses, 5 μl of the Oenothera extract was injected into the UPLC column and ETs were quantified in oenothein B equivalents (280 nm) (Fig. S1), FGs in quercetin equivalents (349 nm) (Fig. S1), and CAs in caffeoyl quinic acid equivalents (315 nm).

Oenothera phenolics separated by UPLC-DAD-MS were first characterized on the basis of their UV spectra. This allowed us to classify peaks as either ETs, flavonoids, or CAs, as each class has their own characteristic UV spectrum (Fig. S1). More specific characterization was achieved by UPLC-ESI-MS using the negative ion mode. Flavonoid glycosides were classified as quercetin glycosides (MS fragment found at m/z 301), kaempferol glycosides (MS fragment found at m/z 285) and myricetin glycosides (MS fragment found at m/z 317), or simply as FGs where the aglycone fragment was not clearly detected and identified. Table S1 provides peak names, retention times, compound assignments and molecular weights for all characterized peaks.

Phylogenetic inference

We used the phylogenetic tree reported in Johnson et al. (2009b) for all comparative analyses. In brief, we obtained DNA sequences for five genetic regions (two plastid, three nuclear) and inferred the single best tree of 113 species in the Onagreae tribe using RAxML (Stamatakis, 2006). The tree was made ultrametric using penalized likelihood in r8s (Sanderson, 2003). We pruned the tree to include the 26 species studied here (Fig. 2) and then converted the tree into a phylogenetic variance-covariance matrix based on phylogenetic distance using the APE package in R (Paradis et al., 2004).

The number of species sampled from a phylogeny can affect the accuracy of parameter estimates using comparative phylogenetic statistics (Blomberg et al., 2003; Munkemuller et al., 2012). In particular, sampling a small number of species decreases the precision of estimates and increases type II error (Blomberg et al., 2003; Munkemuller et al., 2012). Furthermore, parameter estimates can be biased when few individuals are sampled per species and variation within species is ignored (Ives et al., 2007). Our experimental design and statistical approach minimized such biases for the following reasons. First, simulations show that variation in the number of species per se causes little bias in parameter estimates and statistical power is high (> 0.8) when 25 or more species are studied (Blomberg et al., 2003). Secondly, we sampled species from across the two major clades of Oenothera to maximize the probability of capturing the existing range of variation in plant traits within this genus. Thirdly, simulations show that incorporating measurement error (as we do here) greatly reduces biases in parameter estimates (Ives et al., 2007). Finally, studies show that minimum variance parameter estimates can be achieved with as few as four independent replicates per species (Ives et al., 2007), which we exceeded. We are confident that our experimental design and implementation of phylogenetic analyses allows for robust results and conclusions.

Statistical analyses

Phylogenetic signal

We first estimated phylogenetic signal in traits using a version of Blomberg's K statistic, K* (Blomberg et al., 2003). K* was calculated using the Matlab program MEUnivarPHYSIG.m, which incorporates intraspecific variation and measurement error, estimated directly as the standard error from multiple replicate plants of the same species (Ives et al., 2007). Values of K* = 1 imply that covariation in traits among species is equal to that expected under Brownian motion (BM) evolution, whereas K* < 1 implies lower phylogenetic signal. We tested whether there is phylogenetic signal using bootstrapping, where phylogenetic signal was indicated when the observed value of K* lay above the upper 95% confidence interval of bootstrapped values of K* under the assumption of no phylogenetic signal. Whereas K* gives a measure for departure from BM evolution, it does not measure the magnitude of departure from the case of no phylogenetic signal. Therefore, we also employed two related measures: the d statistic from an Ornstein–Uhlenbeck (OU) model of evolution and Pagel's λ. In contrast to K*, which is a nonparametric measure of phylogenetic signal, both d and λ are parametric measures based on a statistical model of the distribution of trait values. Both statistics d and λ are equal to 0 if there is no phylogenetic signal and 1 if there is BM evolution (Blomberg et al., 2003). In addition, d has the attractive property that it is based on an explicit evolutionary model that reflects stabilizing selection (Martins & Hansen, 1997). We estimated d and λ while accounting for measurement error by restricted maximum likelihood in the Matlab programme MERegPHYSIGv2.m, which was newly developed for this project (see Methods S1). We used parametric bootstrapping to determine whether the values of d and λ were significantly > 0. For those variables that did not have associated within-species standard errors (measurement error), such as principle component scores, we estimated the standard Blomberg's K (Blomberg et al., 2003), and d and λ (Lavin et al., 2008) without measurement error.

Covariation among traits

We examined covariation among traits using two methods. First, we estimated pairwise correlation coefficients (r) among all traits while accounting for measurement error around species' mean values using the Matlab program MECorrPHYSIG.m (Ives et al., 2007). In estimating pairwise correlations, we compared the fit of a BM model with that of a nonphylogenetic model that assumed a star phylogeny. We present r-values from the nonphylogenetic analyses because they almost always provided a better fit to the data (average Δlog-likelihood = 11.72; 128/137 nonphylogenetic models had higher LL than BM models), and the results from nonphylogenetic and BM models were highly correlated (r = 0.85, < 0.001). Secondly, we performed hierarchical cluster analysis that allowed us to examine whether traits formed covarying suites that might serve as plant defensive syndromes. This was done using the hclust function in R (R Core Development Team, 2012), with the distance matrix calculated as 1 – (Pearson correlation coefficient) and clusters created using the ‘average’ method. The support for clusters was assessed using the pvclust package in R, which calculated unbiased bootstrap probabilities after 10 000 iterations (Suzuki & Shimodaira, 2006). Thus our approach defines ‘clusters’ based on shared covariation as opposed to discrete clustering of species traits in multivariate space. This approach follows the methods of recent studies on plant defense syndromes (Kursar & Coley, 2003; Agrawal & Fishbein, 2006).

Many traits were correlated, and we simplified this covariation using principal components analysis (PCA). PCA was appropriate because our traits varied continuously and any correlation among variables was generally linear, which we assessed visually by examining all pairwise biplots. We used the prcomp function in R to perform one PCA on just the phenolic chemistry data and then a second PCA on all traits combined. We selected the optimum number of principal component (PC) axes using a scree plot (Legendre & Legendre, 1998), which resulted in just two axes for both PCAs (Tables S2, S3). We did not perform a phylogenetically corrected PCA (Revell, 2009), because of a prevalent lack of phylogenetic signal. Furthermore, the strength of phylogenetic signal varied among traits, making it additionally difficult to adequately correct for phylogeny with existing methods (Revell, 2009).

Phylogenetic regressions

We examined the effects of plant traits on insect performance using phylogenetic regression with the program Regressionv2.m in Matlab (Lavin et al., 2008). We initially conducted analyses that regressed each measure of insect performance against each of the four PC axes. This allowed us to test whether covarying suites of plant traits predicted resistance against herbivores, as predicted by the defense syndromes hypothesis. We then regressed these same measures of herbivory resistance against each individual trait to identify specific plant traits that predicted resistance to herbivores. Because of type I error associated with multiple tests, we also performed multiple regression models and compared the fit of multiple regression models with bivariate linear models using Akaike's information criterion (AIC). However, because the pairwise regressions always resulted in lower AIC than the multiple regression, we only present the former here. We used an OU model of trait evolution for all phylogenetic regressions, because it allows for variation in phylogenetic signal as indicated by the parameter d. We interpreted the effect of a trait on insect performance according to the statistical significance of the slope of the relationship. These analyses did not incorporate measurement error using MERegPHYSIGv2.m, because some of our models exhibited mathematical convergence problems. These problems occur when there are high estimates of measurement error; in these cases, the observed estimates of the residual variation can be lower than those predicted by the independent estimates of measurement error, which leads to an implausible statistical model. To give consistent regression analysis for all traits, we analyzed them all without incorporating measurement error.

To test the effect of plant sex on the evolution of plant secondary chemistry, we used the OU model that incorporates measurement error in MERegPHYSIGv2.m (see Methods S1); these models always converged. The fit of models and statistical significance of the effect of plant sex were assessed by regressing each trait against PTH/sex, which we coded as dummy variables 0 and 1, respectively. We then used parametric bootstrapping with 2000 iterations to determine whether the effect of sex was significantly different from 0, according to whether the 95% confidence interval (CI) of the slope overlapped 0.


Phenolic chemistry

Leaf phenolics were dominated by ETs, CAs and FGs (Table S1), which together comprised, on average, 9.5% (SE = 3.4%) of the total dry leaf mass. We detected between 66 and 97 (mean = 82.3, SE = 0.5) unique phenolic compounds per sample, quantified as the number of unique peaks at 280 nm using UPLC-ESI-DAD (Fig. S1). Total ETs were typically the most abundant metabolites (mean = 8.0 ± 3.5% of leaf mass), but they varied in concentration from < 1% (Oenothera triangulata) to 14% (Oenothera longituba) of the leaf mass. The dimer oenothein B, the trimer oenothein A, and oxidized (ox) oenothein A were the most abundant ETs. There was never more than one unique CA compound found in samples by UV detection, and total concentrations ranged from 0 to 1.9% of the dry leaf mass (mean = 0.5 ± 0.05%). The diversity of FGs (measured at 349 nm) ranged between five and 25 compounds and comprised, on average, 1% (SE = 0.03) of the dry leaf mass (range 0.2–1.9%). Quercetin glycosides and kaempferol glycosides were the most abundant FGs, while myricetin glycosides were detected in two species (Oenothera suffulta and Oenothera speciosa).

Phylogenetic signal in phenolic chemistry

We tested the hypothesis that secondary chemistry exhibits strong phylogenetic signal (Agrawal, 2007). Using the nonmeasurement error version of Blomberg's K, we found moderate to low phylogenetic signal (range 0.34–0.49) in the multivariate composition (i.e. principal components) of secondary metabolites (Table 1). The first principal component (PC1) showed moderate and statistically significant phylogenetic signal by Blomberg's K (= 0.49), and Pagel's λ (λ = 0.88, Table 1), but d of the OU model was not statistically different from zero (= 0.36; Table 1). PC1 summarized variation in the diversity and total concentrations of ETs and FGs, as well as the concentrations of oenothein B and ox-oenothein A (Fig. S2, Table S2). PC2 showed lower phylogenetic signal (Table 1). Similarly, a PCA summarizing variation in all traits (i.e. chemical, physiological, and physical; Fig. S2, Table S3) showed low phylogenetic signal (Table 1).

Table 1. Phylogenetic signal in plant traits and phylogenetic means for sexual and functionally asexual (permanent translocation heterozygosity (PTH)) Oenothera
Trait K* d λ PTHSEX% effect
  1. Percentage effect size was calculated as (SEX/PTH – 1) × 100%. For K*, d, and λ, statistically significant phylogenetic signal (< 0.05) is indicated by bold underlined numbers, and for PTH and SEX statistical significance (< 0.05) is given for the effect of these variables on each trait.

  2. SEX, reproduce sexually; ETs, ellagitannins; FGs, flavonoid glycosides; PC1, PC2, first and second principal components, respectively.

No. of peaks at 280 nm0.350.000.0081.5982.961.7
No. of peaks at 345 nm0.450.320.77 14.44 12.32 −14.7
Oenothein B0.220.000.0043.1734.99−19.0
Oenothein A 0.74 0.49 1.00 4.777.7562.6
ox oenothein A0.390.180.5722.2717.10−23.2
Total ETs 0.53 0.320.7380.2868.36−14.9
Total caffeic acid derivatives0.340.120.546.044.12−31.7
FG4 – quercetin glycoside 0.50 0.46 0.622.091.05−49.6
Total kaempferol 0.53 0.82 0.86 1.672.3641.3
Total quercetin0.340.290.483.252.18−33.0
Total other flavonoids 0.47 0.61 0.86 0.971.6671.6
Total FGs0.370.120.3810.759.38−12.7
Phenolic PC1 0.49 0.36 0.88 -0.15-0.38na
Phenolic PC20.360.370.670.040.52na
All traits PC10.340.240.39−0.320.14na
All traits PC20.440.390.79 0.06 0.73 na

For individual secondary metabolites, the analyses for phylogenetic signal included measurement error. The average K* = 0.44 (95% CI: 0.35–0.52) was only marginally larger than it was for nonchemical traits (K* = 0.39; data not shown, see Johnson et al., 2009a,b). The estimate of phylogenetic signal using d and λ gave similar results to K* (Table 1). Overall, these results show that chemical and nonchemical traits frequently associated with resistance against herbivores show weak to moderate phylogenetic signal in Oenothera.

Covariation among traits

We detected both negative and positive correlations among plant traits (Table S4), providing support for both allocation tradeoffs and possible trait syndromes. The diversity of phenolic compounds at 280 nm was negatively related to total ETs (Fig. 3a) and ox-oenothein A (r = −0.56, 95% CI: −0.84 to −0.17). The diversity of FGs at 349 nm was positively correlated with total FGs (Fig. 3b) and negatively correlated with total CAs (r = −0.43, 95% CI: −0.72 to −0.06). Among major classes of phenolic compounds, total ETs and total FGs showed the strongest negative correlation (Fig. 3c), while total CAs were not significantly correlated with these compounds (|r| < 0.26, Table S4). Allocation to specific types of FGs also exhibited evidence of tradeoffs, where the concentration of kaempferol glycosides was negatively related to the concentration of quercetin glycosides (r = −0.39, 95% CI: −0.68 to −0.01; Table S4). By contrast, specific ETs showed no significant correlations among compounds.

Figure 3.

Pairwise correlations between chemical traits. The raw correlation is shown between: (a) number of peaks at 280 nm and total ellagitannins; (b) number of peaks at 349 nm and total flavonoids; and (c) the negative correlation between total concentrations of flavonoids and ellagitannins. Each point depicted represents the mean for a unique Oenothera species. Range in brackets after r-value refers to the 95% confidence interval.

Phenolic compounds also exhibited correlations with physical, physiological and morphological plant traits. For example, oenothein A exhibited a positive correlation with leaf toughness (r = 0.59, 95% CI: 0.21–0.82) and negative correlations with percentage leaf water content (r = −0.62, 95% CI: −0.84 to −0.25) and SLA (r = −0.63, 95% CI: −0.84 to−0.25). By contrast, total CA was negatively related to leaf toughness (r = −0.38, 95% CI: −0.68 to −0.01) and positively related to SLA (r = 0.5, 95% CI: 0.12–0.77). Trichome density was only significantly correlated with total quercetin glycosides (r = 0.42, 95% CI: 0.04–0.72). Finally, PPC showed strong positive correlations with total ETs, and the specific ET metabolites oenothein A and ox-oenothein A (range in r: 0.53–0.88). PPC was also negatively correlated with total FGs (r = −0.65, 95% CI: −0.84 to −0.34). Thus, the protein binding ability of phenolic compounds in Oenothera is probably solely attributed to ETs.

Hierarchical clustering of trait variation identified traits that might have evolved to form covarying suites of traits underlying syndromes (Figs 2, 4). At the most general level, the dendrogram suggests that species' traits form three general syndromes: species with a low diversity of phenolic compounds, high concentrations of ETs and tough leaves (cluster i); species with high diversity of phenolic compounds and high total FGs (cluster ii); and species with high concentrations of CAs, increased concentrations of quercetin glycosides, higher leaf water content and SLA, and dense trichomes (cluster iii). Total kaempferol did not form a strong cluster with any trait. These three clusters are supported with 90, 96 and 91% approximate unbiased bootstrap support, respectively (Fig. 4).

Figure 4.

Dendrogram showing hierarchical clustering of trait variation measured from 26 Oenothera spp. Clustering was performed using the ‘average’ clustering algorithm in the hclust program of R, which groups traits with similar average ‘distance’. Distance was calculated as 1 – rPearson; small distances correspond to traits that were strongly positively correlated and distance values > 1 correspond to negative correlations among traits. The numbers at each node indicate the approximate unbiased bootstrap support of the node using 10 000 iterations as calculated by the pvclust program in R. The values shown are proportions, where a value of 1 indicates that the grouping of those traits was found 100% of the time. SLA, specific leaf area; FG, flavonoid glycoside; PPC, protein precipitation capacity.

Plants traits that predict resistance to herbivores

Variation in several specific traits was related to the performance of herbivores. Total ETs were associated with reductions in the generalist caterpillar's survival (Fig. 5a), weight gain (t24 = −2.01, = 0.056, R2 = 0.14) and tissue consumption (t23 = −2.03, = 0.066, R2 = 0.15). PPC also had a negative effect on these same measures of performance, but the significance of these effects was weaker (Table S5). Mite survival was positively associated with the diversity of FGs (Fig. 5b), and the number of eggs was positively associated with total kaempferol glycosides (t24 = 2.14, = 0.043, R2 = 0.16). Herbivory by the specialist beetle was also positively related to kaempferol glycosides (Fig. 5c). These beetles consumed more tissue from species with higher trichome density (t17 = 2.63, = 0.018, R2 = 0.29) and lower leaf water content (t17 = −3.39, = 0.004, R2 = 0.40). The amount of herbivory by generalist herbivores in the field was similarly negatively related to % leaf water content (t17 = −2.09, = 0.052, R2 = 0.20). We used the results from Table S5 as a general guide for which traits are likely to be most important in terms of resistance, because overall there were few significant relationships. However, this may be the result of low statistical power, given the modest number of species in our study.

Figure 5.

Regressions showing secondary metabolites – ellagitannins (a), flavonoids (b) and kaempferols (c) – that predict resistance to herbivores. The individual points show the species' raw means while the dotted line shows the regression slope from phylogenetic generalized least-squares analysis, with the corresponding statistics shown above each panel. Additional regression analyses are shown in Table S5.

To test the hypothesis that covarying suites of traits act as defensive syndromes, we regressed herbivore performance against the PC axes summarizing variation in plant traits. Only the relationship between caterpillar survival and PC1chemistry (t24 = −2.17, = 0.04, R2 = 0.16; Table S6) was significant. Although more regressions were significant using nonphylogenetic ordinary least-squares regression, the multivariate descriptors of plant traits provided, at best, a weak predictor of resistance to the diverse herbivores studied (Table S6).

Effects of losing sex on the evolution of plant traits

The diversity and composition of phenolic compounds significantly differed between sexual and asexual Oenothera spp. (Table 1). The diversity of flavonoid compounds was 15% lower in sexual than in asexual lineages. The multivariate composition of all traits combined was also significantly different between sexual and asexual lineages, as evidenced by the significantly higher scores for sexual species along PC2all traits (Table 1). This difference in composition is most likely explained by the multiple phenolic compounds that were strongly associated with PC2 (Table S3), and exhibited large (according to % effect sizes) albeit nonsignificant differences between sexual and asexual Oenothera spp. in univariate analyses (e.g. oenothein B, total CAs, total quercetin, quercetin glycoside-4; Table 1).


In examining the macroevolution of plant secondary chemistry in Oenothera, five results are most important to answering the hypotheses and questions outlined in the Introduction. First, the composition and concentration of chemical and nonchemical traits have been evolutionarily labile during the diversification of Oenothera. Secondly, phenolic traits frequently exhibit tradeoffs with one another and with nonchemical traits (Fig. 3a,c, Table S4). Thirdly, phenolic and nonchemical traits form positively covarying suites of traits or syndromes (Fig. 4, Fig. S2). Fourthly, individual chemical and nonchemical traits predict resistance to generalist and specialist herbivores, while trait syndromes were not strong predictors of resistance. Finally, a loss of sex is associated with altered chemical diversity and multivariate composition of traits. In the following sections, we elaborate on the importance of these results to plant defense evolution.

Phenolic chemistry is evolutionarily labile

The degree to which plant defenses are conserved across a phylogeny provides insight into the tempo of plant defense evolution. The prediction that secondary chemistry is evolutionarily conserved among plant species is based on the fact that many biosynthetic pathways of secondary metabolites are themselves conserved (Malcolm, 1991; Rodman et al., 1998; Wink, 2003; Rausher, 2006; Salminen & Karonen, 2011; Agrawal et al., 2012b). Although the existence of such pathways might be conserved, our results suggest that the composition and concentration of secondary metabolites are evolutionarily labile among congeners. We detected significant phylogenetic signal in the concentration of several individual traits, but overall phylogenetic signal was low for the concentration and composition of phenolic chemistry (Table 1). This finding is consistent with recent studies that show a similar pattern of phenolic chemistry evolution in other plant lineages (Agrawal et al., 2009; Kursar et al., 2009; Pearse & Hipp, 2009). Low to moderate phylogenetic signal has also been found among congeners for nicotine (Adler et al., 2012), cardenolides (Rasmann & Agrawal, 2011), and terpenes (Becerra, 1997).

Two observations suggest that evolutionary lability in the concentration and composition of secondary metabolites should be common. First, the concentration of secondary metabolites exhibits substantial genetic variation within most plant populations. For example, a meta-analysis showed that the heritability of secondary metabolites is on average two to three times higher than other types of traits (Geber & Griffen, 2003). This same pattern is also true in Oenothera (Johnson, 2011). The second observation relates to the molecular mechanisms that control the biosynthesis of secondary chemistry. Biosynthetic pathways are often complex, involving multiple steps and many branching pathways that produce a diversity of metabolites. Although novel biosynthetic pathways arise rarely, flux down the various branches of an existing pathway can be modified by regulatory mutations and structural changes to enzymes that influence their affinity for substrates (Hoekstra & Coyne, 2007; Streisfeld & Rausher, 2011; Bekaert et al., 2012; Rausher, 2013). The combination of these changes can alter both the concentration and the composition of metabolites.

Evolution of tradeoffs

Tradeoffs among ecologically important traits are thought to be inevitable because genetic, selective, and functional constraints make it impossible for populations to maximize fitness in all environments simultaneously (Arnold, 1992). The existence of tradeoffs is the cornerstone of most plant defense theories (Feeny, 1976; Coley et al., 1985; Herms & Mattson, 1992; Stamp, 2003). It is therefore surprising that a meta-analysis of 31 quantitative genetic studies found no overall negative association between resistance traits (Koricheva et al., 2004). However, this apparent challenge to the predicted prevalence of tradeoffs might simply reflect two limitations with the datasets analyzed. First, many of the studies lacked detailed analysis of multiple individual metabolites. Instead they frequently included quantifications of the total concentrations of complex mixtures of secondary metabolites (e.g. FGs, condensed tannins, indolyl glucosinolates). Such data may hide allocation tradeoffs among individual metabolites, which are likely to be common for metabolites produced by a single biosynthetic pathway (Johnson et al., 2009a).

The second limitation relates to how evolutionary constraints are predicted to limit evolution at microevolutionary (i.e. within populations) and macroevolutionary (i.e. across species boundaries) timescales (Agrawal et al., 2010). Within a population, low genetic variance in a trait and strong negative genetic covariance between traits can limit the rate and direction of evolution (Arnold, 1992; Falconer & Mackay, 1996; Agrawal & Stinchcombe, 2009). These types of genetic constraints can be overcome over macroevolutionary timescales, where recurrent mutation and selection can lead to substantial evolution over long periods of time, even in directions that are orthogonal to maximal genetic covariance (Agrawal & Stinchcombe, 2009; Agrawal et al., 2010). In comparative analyses among species, tradeoffs are more likely to reflect a history of selection for optimal investment in traits, and secondarily functional constraints caused by allocation tradeoffs as a result of the branching structure of biosynthetic pathways (Arnold, 1992).

Recent advances in chemical and comparative analyses allow for detailed examination of evolutionary tradeoffs among specific metabolites within and between pathways. The application of these techniques allowed us to detect tradeoffs in allocation to secondary metabolites that would otherwise have been missed. For example, there was a clear tradeoff between allocation to total ETs vs total flavonoids (Fig. 3), which share phosphoenolpyruvate from the glycolysis pathway and dehydroshikimic acid from the shikimate pathway as biosynthetic precursors (Fig. 1) (Salminen & Karonen, 2011). Allocation to specific ETs and CAs also appeared to be associated with decreased diversity of phenolic compounds (Table S4). Given our results (Fig. 3, Table S4), these tradeoffs may reflect adaptive evolution in some lineages to maximize defenses against particular groups of insects while minimizing investment in flavonoids that might confer susceptibility (Fig. 5). These results show the potential benefits of combining advanced chemistry with comparative phylogenetic analyses to understand how tradeoffs among metabolites have shaped the evolution of defense.

Evolution of defensive syndromes

Recent research on the evolution of plant defense traits has focused on the evolution of defensive syndromes (Kursar & Coley, 2003; Agrawal & Fishbein, 2006). The defense syndromes hypothesis posits that natural selection creates positively covarying suites of traits that result in either synergistic negative effects on herbivores or redundancy in defense traits that together provide effective protection against enemies. Our study adds to several other recent studies which found that species frequently converge on particular combinations of traits (Becerra, 1997; Agrawal & Fishbein, 2006; Fine et al., 2006; Travers-Martin & Muller, 2008; Pearse & Hipp, 2012; but see Moles et al., 2013). However, the question remains as to whether the second prediction of the hypothesis is satisfied – covarying suites of traits act as defenses.

Our results suggest that multivariate suites of traits do not in themselves function to increase defense. The principal components of trait variation did not provide a strong predictor of resistance against the herbivores studied (Table S6). By contrast, individual traits did predict resistance to the diversity of herbivores studied (Fig. 5; Table S5). The few studies that have explicitly tested the second prediction of the defense syndromes hypothesis provide very mixed support (Agrawal & Fishbein, 2006; Travers-Martin & Muller, 2008; Pearse, 2011). We believe the reasons for this are twofold. First, if defense syndromes exist, they are most likely to be recognized in the field, where the full complement of herbivores will be present and the importance of such strategies as escaping herbivores will be effective (Kursar & Coley, 2003). Secondly, defense syndromes are unlikely to actually exist in the form originally capitulated because herbivores are just one selective pressure among many in nature. Therefore, defensive syndromes must compete against a broader set of traits under selection. If trait syndromes do exist, they are most likely to reflect adaptation to the vast milieu of abiotic and biotic environmental challenges that plants face (Janzen, 1980; Strauss et al., 2005). Future research on the adaptive function of syndromes could test these ideas through reciprocal transplant experiments coupled with manipulations of the biotic and abiotic environment (Fine et al., 2004). One caveat to these conclusions is that we might have missed some defensive traits that are critical in the formation of defense syndromes yet cannot be detected using our approach. For example, traits related to growth and tolerance can represent important defensive strategies that few researchers have studied in the context of syndromes (Strauss & Agrawal, 1999; Agrawal & Fishbein, 2008; Turley et al., 2013).

The effects of plant sex on evolution of defense

Several recent studies have tested and supported Levin's hypothesis that reduced sexual reproduction leads to decreased plant defenses against parasites (Busch et al., 2004; Johnson et al., 2009b; Hersch-Green et al., 2011; Campbell & Kessler, 2013). In this study, we found evidence that a loss of sex is associated with increased diversity of flavonoid metabolites and altered multivariate composition of phenolic compounds and nonchemical traits (Table 1). A positive correlation between flavonoid diversity and susceptibility to generalist mites further suggests that an increase in flavonoid diversity might confer greater susceptibility to some generalist herbivores. Independent evidence on protein evolution in the flavonoid biosynthetic pathway of Oenothera provides further support for the conclusion that the suppression of recombination and segregation negatively affects flavonoid biosynthesis (E. I. Hersch-Green et al., unpublished). Specifically, across 26 sexual and PTH Oenothera species, three of the four enzymes involved in flavonoid biosynthesis showed evidence consistent with the accumulation of deleterious mutations in asexual lineages. Together, these results are consistent with the interpretation that a suppression of sex in Oenothera prevents optimal control of flavonoid biosynthesis, which manifests itself as the production of more, not fewer, flavonoid metabolites.

Recent evidence also suggests that the effects of plant sex on defense may be more complex than originally envisioned. For example, Campbell & Kessler (2013) found that repeated transitions from self-incompatible (SI) to self-compatible (SC) reproduction in many Solanaceae species were associated with higher induced responses to herbivory in SI than in SC lineages, but no clear difference in constitutive resistance. Patterns that are opposite to the simplest predictions have also been observed. For example, we previously showed that asexual Oenothera species were actually more resistant to specialist insects, probably because of a genetic tradeoff in defense against generalist and specialist herbivores (Johnson et al., 2009b). In tobacco, SI plants exhibit lower concentrations of nicotine than SC plants (Adler et al., 2012). This counterintuitive result is thought to have arisen because the concentrations of nicotine in leaves and nectar are positively correlated, and pollinators that are deterred by nicotine in nectar have consequently driven correlated evolution of lower nicotine in leaves in SI than in SC plant species.

Given that plant species vary almost continuously in the degree of sexual reproduction, and the theoretically predicted effects of sex on adaptive evolution, it seems likely that variation in sex could be among the most important evolutionary constraints acting on the evolution of plant defense. Our results from Oenothera provide some support for this view, but this prediction is in desperate need of additional experimental tests in a wide range of plant systems.


We thank T. Garland and L. Revell for providing important insight into our analyses. A. Agrawal, D. Anstett, C. Fitzpatrick, R. Godfrey, A. Kessler, C. Richter, M. Turcotte, N. Turley and three anonymous reviewers provided helpful comments that improved earlier versions of this paper. M. Rausher provided laboratorh and field space for plant growth. M.T.J.J. was funded by NSERC Canada, the Canadian Foundation for Innovation, the Ontario Government, and a Connaught Award from the University of Toronto. Chemical analyses on the UPLC-DAD-MS system were made possible by a Strategic Research Grant of the University of Turku (Ecological Interactions) to J.P.S. J.P.S. was funded by the Academy of Finland (grant no. 258992) and Anne Koivuniemi helped with the chemical analyses. A.R.I. was funded by NSF DEB-0816613.