Architecture and stability of tripartite ecological networks with two interaction types

Over the past few decades, studies on empirical ecological networks have primarily focused on single antagonistic or mutualistic interactions. However, many species engage in multiple interactions that support distinct ecosystem functions. The architecture of networks integrating these interactions, along with their cascading effects on community dynamics, remains underexplored in ecological research. In this study, we compiled two datasets of empirical plant–herbivore/host–parasitoid (PHP) and pollinator–plant–herbivore (PPH) networks, representing two common types of tripartite networks in terrestrial ecosystems: antagonism–antagonism and mutualism–antagonism. We identified the patterns of subnetwork structures and interconnection properties in these networks and examined their relationships with community stability. Our findings revealed distinct pathway effects of network architecture on persistence and local stability in both PHP and PPH networks, with subnetwork modularity and nestedness showing a few strong direct effects and mediating the indirect effects of subnetwork size and connectance. In PHP networks, subnetwork modularity enhanced persistence and local stability, whereas subnetwork nestedness directly undermined them. However, both subnetwork topologies consistently mediated the destabilizing effects of subnetwork size and connectance on the entire network. In PPH networks, persistence was primarily affected by the plant–herbivore subnetwork, while the size, connectance, and modularity of different subnetworks had opposing effects on local stability. Regarding interconnection properties, the correlation of interaction similarity destabilized PHP networks, whereas the correlation of interaction degree promoted local stability in PPH networks. Further analysis indicated that structure–persistence relationships vary significantly across guilds, and the network-level effects of network architecture can be reversed, negligible, or biased in specific guilds. These findings advance our understanding of how network architecture influences ecosystem stability and underscore the importance of considering multiple interaction types when predicting community dynamics.