In complex food webs, interactions among species in different trophic levels can generate cascading indirect effects that couple top predators with primary producers, thereby affecting ecosystem functioning. Natural selection imposed by top predators on intermediate predators may play a role in shaping the strength of these trophic cascades, but this conjecture remains largely untested. To determine the effects of natural selection on the strength of trophic cascades, we conducted a two‐part experiment in a four‐level aquatic trophic system involving a top predator (fish), an intermediate predator (damselflies), herbivores (zooplankton) and primary producers (algae). We first quantified how predation by fish generated selection on damselfly activity levels after controlling for phenotypic plasticity. We then measured the indirect effects of this selection on primary production (phytoplankton biomass). In both experiments, we varied the density of predators, allowing us to elucidate both trait‐mediated and density‐mediated indirect effects. We found that as fish density increased, damselfly survivorship declined, which generated natural selection favouring less active damselflies. These results are robust after taking into account latent effects of plasticity in response to fish predator cues. The surviving damselflies likely foraged less, freeing herbivores from predation, which in turn reduced primary production. This selection‐driven trait‐mediated indirect effect was only apparent at low damselfly densities, because the consumptive effect of damselflies at high densities overwhelmed the effects of past selection. These results demonstrate how the past action of natural selection imposed by predators affects the phenotypes of prey that consume herbivores, which ultimately influences primary productivity—a selective trophic cascade. Natural selection can therefore act as a mechanism coupling ecological dynamics across trophic levels, which ultimately influences ecosystem functioning. A http://onlinelibrary.wiley.com/doi/10.1111/1365-2435.13102/suppinfo is available for this article.

No abstract