Studying how habitat loss affects the tolerance of ecological networks to species extinction (i.e. their robustness) is key for our understanding of the influence of human activities on natural ecosystems. With networks typically occurring as local interaction networks interconnected in space (a meta-network), we may ask how the loss of specific habitat fragments affects the overall robustness of the meta-network. To address this question, for an empirical meta-network of plants, herbivores and natural enemies we simulated the removal of habitat fragments in increasing and decreasing order of area, age and connectivity for plant extinction and the secondary extinction of herbivores, natural enemies and their interactions. Meta-network robustness was characterized as the area under the curve of remnant species or interactions at the end of a fragment removal sequence. To pinpoint the effects of fragment area, age and connectivity, respectively, we compared the observed robustness for each removal scenario against that of a random sequence. The meta-network was more robust to the loss of old (i.e. long-fragmented), large, connected fragments than of young (i.e. recently fragmented), small, isolated fragments. Thus, young, small, isolated fragments may be particularly important to the conservation of species and interactions, while contrary to our expectations larger, more connected fragments contribute little to meta-network robustness. Our findings highlight the importance of young, small, isolated fragments as sources of species and interactions unique to the regional level. These effects may largely result from an unpaid extinction debt, in which case these fragments are likely to lose species over time. Yet, there may also be more long-lasting effects from cultivated lands (e.g. water, fertilizers and restricted cattle grazing) and network complexity in small, isolated fragments. Such fragments may sustain important biological diversity in fragmented landscapes, but maintaining their conservation value may depend on adequate restoration strategies.

To build the meta-network we considered all plant-herbivore-natural enemy interactions (as presence-absence data) recorded in 14 habitat fragments, which involved plant species (herbs and shrubs), endophagous herbivores (gall makers or gall inquilines, and leaf miners), ectophagous herbivores (scale insects and aphids), and their natural enemies (parasitoids, hyperparasitoids and predators). To explore fragment contribution to meta-network robustness, we sorted the 14 habitat fragments according to their area, age and six connectivity metrics combining geographical distances among fragments, neighboring area and similarity in plant resources among fragments. For each fragment removal, we considered secondary extinctions of herbivores and tertiary extinctions of natural enemies as consequence of the local extinction of plant resource, and calculated remnant species richness and trophic interactions. We calculated meta-network robustness as the area under the curve of remnant species and interactions for a specific removal sequence for each trophic guild and interaction type. Finally, we compared observed robustness for each removal rule against calculated robustness for 100 random fragment removals.