Multiple interaction types determine the impact of ant predation of caterpillars in a forest community.

Direct and indirect effects of predators are highly variable in complex communities, and understanding the sources of this variation is a research priority in community ecology. Recent evidence indicates that herbivore community structure is a primary determinant of predation strength and its cascading impacts on plants. In this study, we use variation in herbivore community structure among plant species to experimentally test two hypotheses in a temperate forest food web. First, variation in the strength of predator effects, such as ant predation of caterpillars, is predicted to be density dependent, exhibiting stronger effects when prey abundance is high (density-dependent predation hypothesis). Second, mutualistic interactions between ants and sap-feeding herbivores are expected to increase the abundance of predatory ants, strengthening predation effects on herbivores with cascading effects on host plants (keystone mutualism hypothesis). Using a large-scale predator exclusion experiment across eight dominant tree species, we tracked changes in insect density on 862 plants across two years, recording 2,322 ants, 1,062 sap-feeders, 5,322 caterpillars, and quantifying herbivory on 199, 338 leaves. In this experiment, density-dependent predation did not explain variation in the direct or indirect effects of ants on caterpillars and herbivory. In partial support of the keystone mutualism hypothesis, sap-feeders strengthened top-down effects of ants on caterpillars under some conditions. However, stronger ant predation of caterpillars did not lead to measurable trophic cascades on trees occupied by sap-feeders. Instead, the presence of sap-feeders was associated with increased per capita feeding damage by caterpillars, and this bottom-up effect attenuated the indirect effects of ants on host plants. These findings demonstrate that examining the multi-trophic impacts of mutualisms and predation in the context of the broader community can reveal patterns otherwise masked by compensatory interactions.