Abstract
As a general rule, plants defend against herbivores with multiple traits. The defense synergy hypothesis posits that some traits are more effective when co-expressed with others compared to their independent efficacy. However, this hypothesis has rarely been tested outside of phytochemical mixtures, and seldom under field conditions. We tested for synergies between multiple defense traits of common milkweed (Asclepias syriaca) by assaying the performance of two specialist chewing herbivores on plants in natural populations. We employed regression and a novel application of Random Forests to identify synergies and antagonisms between defense traits. We found the first direct empirical evidence for two previously hypothesized defense synergies in milkweed (latex by secondary metabolites, latex by trichomes), and identified numerous other potential synergies and antagonisms. Our strongest evidence for a defense synergy was between leaf mass per area and low nitrogen content; given that these “leaf economic” traits typically covary in milkweed, a defense synergy could reinforce their co-expression. We report that each of the plant defense traits showed context-dependent effects on herbivores, and increased trait expression could well be beneficial to herbivores for some ranges of observed expression. The novel methods and findings presented here complement more mechanistic approaches to the study of plant defense diversity, and provide some of the best evidence to date that multiple classes of plant defense synergize in their impact on insects. Plant defense synergies against highly specialized herbivores, as shown here, are consistent with ongoing reciprocal evolution between these antagonists.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
Open Research statement: data and code have been provided for reviewers, and will be made available on Figshare (doi: 0.6084/m9.figshare.20421633) prior to publication.
We have substantially rewritten this manuscript and refined our statistical approaches. We have dropped the 2005 data and simplified our analyses into just "regression" and "random forest" rather than distinguishing between hypothesis testing and exploration. In the process of revising this manuscript, we realized that our regression model structure contained redundant random effects, as plant ID was nested within date. Further, we were correcting for variation in duration (6 or 7 days) in three places: duration was captured by plant ID, but we were also including date, and also using as our growth response the log final weight divided by days (which implicitly accounts for duration under the assumption of exponential growth). We have revised our analysis to only include the single random effect of plant ID, which can capture un-modeled variation between plants in addition to the variation associated with date and the effects of variable duration. We also now simply use log final weight as our response variable for herbivore growth, avoiding an assumed functional form of growth rate in favor of allowing our data to identify this through random effects. This new model structure is cleaner, removing some assumptions and model complexity compared to the original version. We recognized that given the construction of our analysis, traits could in fact switch from being defensive to being beneficial to herbivores if the interaction term was strong relative to the main effects. We have included this as a new part of our analysis, as we found that all of our traits were in fact context-dependent, and we believe this provides some explanation for the inconsistency of defense traits that has been previously reported in the plant defense literature. For consistency and ease of comparisons, we decided our traits should all be scaled such that higher values were expected a priori to confer resistance to herbivores. Accordingly, we have rescaled specific leaf area (SLA) into its inverse, Leaf Mass per Area (LMA), and % Nitrogen to % non-nitrogen. In revising this manuscript, we have also revised our calculations of PRR to range only across observed trait values, and have updated our calculations of %PRV (previously called %PRR) for survival analyses to calculate model fit based on predictions on the logit scale, rather on the probability scale. Finally, we have systematized identification of Random Forest relationships as "synergy", "antagonism", or "unclear" using permutation testing.