Main Text: Ecologists have traditionally focused on networks of feeding relationships among species (direct trophic interactions) to understand community structure and dynamics . Despite an increasing awareness of the importance of non-trophic and indirect interactions (mediated by a third species or via the environment) , integrating the different types of interactions into a single network has proven extremely difficult and has not yet been done empirically, despite relatively recent efforts and developments . This is because networks quickly become very complex, as the number of indirect interactions increases exponentially with the number of species involved .
Amongst the non-trophic interactions, higher-order interactions (HOIs) or interaction modifications (the modulation of a pairwise interaction by a third species) , play a particularly important role in stabilising species coexistence . For instance, they can attenuate negative interactions, such as when non-prey species increase the persistence of predation-susceptible species by modifying the ability of predators to detect prey . This way, HOIs can promote diversity and the long-term persistence of ecological communities . HOIs are also regulated by feedbacks, meaning that their influence on interactions changes over time with population densities . This context-dependency of interactions makes it difficult to predict the effects of perturbations on ecosystems (e.g., species loss) , with consequences for conservation and management strategies. Thus, investigations of interaction modulation by HOIs are unveiling a new, and previously unaccounted for, higher level of community structuring processes. However, HOIs cannot be observed from studying traditional pairwise networks, unlike other indirect effects (e.g., apparent competition) , because they often involve several other species or environmental factors beyond the interacting pair. The role of HOIs in shaping ecological communities has mostly been explored theoretically or by experiments with particular subsets of communities and tractable laboratory model systems testing specific effects . How pervasive and influential these effects are at a community level, remains largely unknown.
Here we hypothesised that a complex network of HOIs could be constantly modifying pairwise interactions and shaping ecological communities, and that consequently the outcome of pairwise interactions would be a product of many influences from distinct sources. We predicted that the manipulation of any species (or group of species) within the community – or even of one of their by-products (e.g. deadwood, leaf-litter, O2, faeces) – would reverberate throughout the entire community and thus modify apparently unrelated interactions. We performed experimental manipulations of species living on the tropical shrub Baccharis dracunculifolia D.C. (Asteraceae), a highly self-contained system with a diverse arthropod fauna (for details about the study system see the Supplementary Material). In different treatments, we excluded all ant species or live or hatched insect galls of the dominant galler species. Each treatment, including a control with no species exclusion, consisted of 16 replicated plants of B. dracunculifolia . Over two months, every week we quantified the changes in densities of several other species or guilds (hereafter groups; ants, herbivores, predators, and aphids) as well as changes in direct interactions involving two gallers, such as gall induction (herbivory), parasitism by wasps, and inquilinism (sharing occupation of the gall for shelter and feeding) by aphids. Specifically, we combined direct observation and gall dissection data to quantify the effects of groups on each other’s population densities or traits, including direct trophic and non-trophic interactions and the effect of the exclusion of a group on another group (hereafter, density effects; Fig. 1). We also investigated how the direct interactions changed in different contexts defined by the density or exclusion of a third group (for detailed methods see the Supplementary Material). This allowed the construction of a unique “effect network” based on multiple manipulations performed simultaneously on the same system, investigating the same direct interactions under different contexts. Links were categorised into two types: node modulation (node-to-node effects), which are pairwise trophic and non-trophic interactions or density effects; and link modulation (HOIs; node-to-link effects), which are three-way interactions (interaction modification), or four-way interactions (modification of an interaction modification).