Neutrality in plant-herbivore interactions

Abstract

Understanding the distribution of herbivore damage among leaves and individual plants is a central goal of plant-herbivore biology. Commonly observed unequal patterns of herbivore damage have conventionally been attributed to the heterogeneity in plant quality or herbivore behaviour or distribution. Meanwhile, the potential role of stochastic processes in structuring plant-herbivore interactions has been overlooked. Here, we show that based on simple first principle expectations from metabolic theory, random sampling of different sizes of herbivores from a regional pool is sufficient to explain patterns of variation in herbivore damage. This is despite making the neutral assumption that herbivory is caused by randomly feeding herbivores on identical and passive plants. We then compared its predictions against 765 datasets of herbivory on 496 species across 116° of latitude from the Herbivory Variability Network. Using only one free parameter, the estimated attack rate, our neutral model approximates the observed frequency distribution of herbivore damage among plants and especially among leaves very well. Our results suggest that neutral stochastic processes play a large and underappreciated role in natural variation in herbivory and may explain the low predictability of herbivory patterns. We argue that such prominence warrants its consideration as a powerful force in plant-herbivore interactions.

Keywords: R package; functional equivalence; herbivory; intraspecific variation; neutral theory; sub-individual variation.

Figure 1.

Example empirical distributions of among-leaf (a,b) and among-plant (c,d) herbivore damage in the HerbVar dataset. (e) The probability density function of cumulative proportion herbivore damage ϕ_T according to the neutral model with different attack rates λ. The two probability point masses at 0 and 1 are indicated by triangles. To better visualize the mix of discrete and continuous probabilities, we plotted the binned probability of different herbivore damage classes at 5% intervals for different values of attack rates (f–h).

Figure 2.

Observed versus predicted coefficient of variation of among-leaf (a) and among-plant (b) herbivore damage. Slope and intercept are taken from a major axis regression model and plotted as a red line. The 95% confidence intervals are displayed in parentheses. Each point represents a distribution, with the size scaled to the sample size of the distribution. Points above the one-to-one black dashed line indicate overprediction of CV. Similarly, points below the dashed line indicate underprediction of CV. The first constrained and unconstrained axes of RDA of among-leaf (c) and among-plant (d) statistical probes. Each point represents a distribution. High similarity in the shape of the predicted and observed distributions is demonstrated by the high overlap in the predicted and observed data clouds.