Within plant taxa that typically produce latex, there is still substantial variation in the amounts produced following tissue damage. This variation is hierarchical: from phenotypic plasticity within a species, to heritable genetic variation within a species, to across-species variation. For example, following damage by herbivores, some species of Asclepias can more than double latex exudation (Van Zandt & Agrawal, 2004; S Cook, A Erwin & A Agrawal, unpubl). Genetically based variation within A. syriaca accounted for up to four-fold variation in latex production across full-sibling families in a common field environment (Agrawal & Van Zandt, 2003; Agrawal, 2005). Finally, as demonstrated in this study, latex production can vary from nearly zero to upwards of 46 mg exuded upon minor tissue damage.
Latex production showed some evidence of phylogenetic conservatism, yet was also remarkably labile. For example, in the North American subclade of Asclepias series Incarnatae (as emended by M Fishbein, unpubl.; Figure 2), all species produced very low latex amounts (all under 0.5 mg) (Appendix 1). In contrast, in the mostly South American subclade of the same series (Incarnatae) (Figure 2), there was over 20-fold variation in latex production. Similarly, some sister taxa, such as A. tuberosa and A. obovata, have dramatically diverged in latex production, with the former showing an almost complete loss of latex. Given the heritable variation in latex production within a species and some evidence for natural selection acting on this trait (Agrawal, 2005), it is not surprising that closely related taxa can strongly diverge. However, any adaptive role of latex in divergence and speciation has yet to be demonstrated.
Latex as a multivariate defense
Although plant defenses have traditionally been studied as single-trait weapons, a fuller understanding of plant defense ecology and evolution can be gained by simultaneous, integrative studies of suites of traits (Feeny, 1976; Agrawal & Fishbein, 2006; Fine et al., 2006). Clearly, latex functions through multiple modes of action, including physical barriers to consumption and toxicity. Thus, the level of resistance provided should depend, not only on the dose of latex, but also on the concentration of toxins within it. We show an evolutionary trade-off (after accounting for phylogenetic relatedness) between cardenolides and both proteases and latex production, with a concomitant positive correlation between latex production and proteases (Table 1). Positively correlated evolution of latex production and cysteine protease activity is consistent with the finding that the evolution of anti-herbivore defenses in milkweeds is characterized by suites of traits that are expressed in concert (Agrawal & Fishbein, 2006). This result prompts a modification of the proposed ‘high edibility/high physical defense’ syndrome of Agrawal & Fishbein (2006), as it now appears that expression of a chemical defense (cysteine protease) is correlated with two ‘physical’ defenses (latex exudation and trichome density) in Asclepias. Despite long-standing hypotheses of trade-offs in anti-herbivore defenses, such trade-offs are rare (Steward & Keeler, 1988; Agrawal & Fishbein, 2006; Agrawal, 2007). The trade-offs that we have documented between cardenolide quantity (in latex) and both latex production and cysteine protease activity (Table 1) are the first for milkweed defenses, and contributes to the syndromes view of coordinated defense strategies (Agrawal & Fishbein, 2006, Table 1).
The alternative models of trait evolution used to calculate PICs for testing trait correlations produced differing results. The results were somewhat stronger using the two models employing estimated branch lengths, with the result employing a chronogram being the strongest. Of the three models, we favor the gradual model employing a chronogram to account for shared evolutionary history. In this model, branch lengths estimate the temporal duration between speciation events. If rates of defense trait evolution are approximately constant over time, the chronogram most accurately depicts shared opportunities for evolutionary change among related species. We find the gradual model producing a phylogram, where branch lengths are scaled by molecular evolution (i.e., expected rates of nucleotide substitution in non-coding DNA sequences), to be a less intuitive approach. Here, there is no a priori reason to assume that rates of evolution in defense traits should be predicted well by rates of nucleotide substitution. It is reassuring that the results from these differing models are similar, which provides robustness to our preference for using the chronogram. However, the results from employing a speciational model to calculate PICs are quite different. With this model, there are equivalent opportunities for trait evolution at every speciation event, and none of the resulting correlations among defense traits were significant. Gradual models result in the inference of a significant positive correlation (independent of phylogenetic relationship) between the evolution of latex production and cysteine protease activity, along with a significant negative correlation between these two traits and the evolution of cardenolides; the speciational model results in the inference of independent (uncorrelated) evolution between latex production and the other two chemical defenses, along with a marginally significant positive correlation between the activity of cardenolide and protease. As in other studies of the phylogenetic component of character evolution (e.g., Moen, 2006), we do not have a rigorous a priori basis for assessing whether gradual or speciational models of character change are most appropriate for the evolution of latex.
The defense escalation hypothesis
The concept of directional trends was derived from early work on plants (Grant, 1963; Ehrlich & Raven, 1964) and suggests that as lineages diversify, there are directional phenotypic changes in ecologically important traits. Although the existence of directional trends has been a topic of intense scrutiny by paleontologists (McNamara, 1990; Alroy, 2000), and the mechanisms of such trends have been debated (Futuyma, 1989; Grant, 1989) they have less often been tested using data obtained from extant species (e.g., Omland, 1997; Mooers et al., 1999; Baker & Wilkinson, 2001; Hibbett, 2004; Adamowicz & Purvis, 2006; Moen, 2006). One classic example is the work on plant defense, including Ehrlich & Raven's (1964) proposed ‘escalation’ of defense hypothesis. Both Berenbaum (1983) and Farrell & Mitter (1998) made initial attempts to test these ideas and found some evidence for escalation as plant lineages diversified (coumarins in the Apiaceae and cardenolides in Asclepias, respectively). In both of these initial studies, the focus was on biochemical diversification (to more potent forms) within the chemical class being investigated. However, these studies of plant defense escalation did not make explicit use of phylogeny and models of character evolution (Harvey & Pagel, 1991; Mooers et al., 1999; Whittall & Hodges, 2007).
We tested for directional trends associated with the production of latex and two constituent chemical defenses using two general approaches. In the first method, we regressed the phenotypes of species against (i) the number of intervening nodes between each species and the hypothetical ancestor of Asclepias or (ii) the total branch length between ancestor and tips. These results were concordant, indicating that evidence for the decline in latex exudation was robust to assumptions regarding speciational vs. gradual models of evolution.
In the second approach, we reconstructed ancestral states and determined whether phenotypes tended to increase or decrease during speciation events. With this method, we found evidence for increases in all three traits, although the evidence for repeated increases in latex production was the strongest. Several issues complicate interpretation of the conflicting results between the two approaches to testing for directional trends. First, the methods differ in how phenotypic data are used to measure amounts of phenotypic divergence. The regression method uses only observed phenotypic data and implicitly considers simultaneously both the magnitude and direction of phenotypic change, whereas the ancestral state method uses inferred ancestral states and considers only the direction of change at internal nodes. Thus, the most likely cause of the conflicting results for latex production, for example, is that the ancestral states method discovered that more speciation events were associated with increases in latex production, but the regression method discovered that the overall magnitude of decreases in latex production per speciation event exceeded the magnitude of increases (cf. Figure 3). It is thus conceivable that there have been large drops in latex exudation that have occurred a few times during the diversification of Asclepias, but that subsequently these declines have been followed by many small increases in latex values. In this way, there may be an overall decline in latex exudation as the clade diversified, but there may be a preponderance of increases at speciation events.
Because these are correlative analyses, cause and effect are unclear. In fact, Ehrlich & Raven's (1964) hypothesis was based on the notion that the evolution of escalated traits spurred diversification; we have not specifically tested this hypothesis with Asclepias. Here, we have investigated a broader hypothesis, that there has been a dominant trend of directional change in phenotypes during the macroevolution of a lineage. The detection of a directional trend is consistent with a relationship between trait values and diversification, but the causal arrow may point from trait to diversification or vice versa.
Based on our findings, we suggest a hypothesis for evolutionary patterns that include both escalation and decline of particular defenses as plant lineages diversify. Our notion is based on a perspective including an ‘evolving community of herbivores’ and possible directional trends in plant defense that respond to herbivory. Where specialist herbivores dominate the contemporary fauna of a particular plant group, as is the case for milkweeds, there may be relaxed selection for particular defenses, especially those for which the specialist herbivores have evolved a mechanism for circumventing the defense. Latex, cardenolides, and trichomes are possible examples, as milkweed herbivores deactivate latex (Dussourd & Eisner, 1987), have altered ATPase physiologies (Vaughan & Jungreis, 1977; Moore & Scudder, 1986; Holzinger & Wink, 1996; Labeyrie & Dobler, 2004), and shave trichomes on leaves (Malcolm, 1995; Agrawal & Malcolm, 2002; Agrawal, 2007) to overcome these respective barriers to consumption. Nonetheless, as discussed above, these traits are still effective at reducing herbivory by some milkweed specialists. As classic qualitative defenses (Feeny, 1976), these traits are effective barriers against feeding by generalists, even at very low doses. Thus, depending on the costs of particular defenses, their effectiveness against particular groups of herbivores, and the consistency of selection in space and time, we may predict either an escalation or decline in the expression of defense through the plant diversification process.
Furthermore, the conflicting results between the regression and ancestral state approaches to evaluating evolutionary trends suggest that both escalation and decline are important aspects of the history of the evolution of latex defense in Asclepias. Infrequent, large declines in latex investment may be associated with a shift to an alternative defense syndrome (Agrawal & Fishbein, 2006), whereas smaller and more numerous increases in latex may be indicative of escalation within the same syndrome.
Some defensive strategies such as tolerance or re-growth following damage, low nutritive value, or certain toxins that cannot be overcome by specialists, should persist, if not escalate through diversification, as originally proposed by Ehrlich & Raven (1964). Thus, our view is a phylogenetic synthesis of some classic hypotheses, including apparency theory (Feeny, 1976), resistance-tolerance trade-offs (van der Meijden et al., 1988; Simms & Triplett, 1994), and plant responses to selection by generalists vs. specialist (Da Costa & Jones, 1971; Blau et al., 1978; van der Meijden, 1996; Lankau, 2007). We predict that there should be directional phenotypic changes as plant lineages diversify, but that changes in defense investment will depend on the specific defenses and particular guilds of herbivores attacking plants. We advocate the search for patterns of defense investment across taxa and refinements of defense theories in the context of phylogenetic history.