Network architecture and vulnerability of macroalgal epiphyte-host interactions in Caribbean coastal ecosystems: implications for marine conservation.

BACKGROUND AND AIMS: Tropical coastal marine ecosystems depend on macroalgal epiphyte-host networks, yet their architecture and vulnerability remain poorly characterised in the Caribbean. This study characterised the topological structure of Cuba's epiphyte-host network, identified determinants of epiphytic richness, and developed an integrated vulnerability index for conservation prioritisation. METHODS: A bipartite network was constructed from 130 host taxa and 266 epiphytic macroalgae. Network topology was analysed via degree distribution, community detection, and centrality. Host habitat distributions were inferred with a hierarchical Bayesian model. Epiphytic richness was modelled using Bayesian negative binomial regression with habitat breadth and betweenness centrality as predictors. Niche overlap was assessed with Jaccard similarity and Monte Carlo tests. Structural robustness was evaluated through sequential host removal, and an Integrated Vulnerability Index combined habitat environmental vulnerability with network irreplaceability. KEY RESULTS: The network exhibits scale-free architecture, where few hosts support most interactions. Betweenness centrality outperformed habitat breadth as a predictor of epiphytic richness. Epiphyte assemblages were highly distinct among hosts (mean similarity < 3%). Simulated loss showed network collapse after removing just 14.6% of central hosts, versus 64.6% under random loss. Rhizophora mangle, Thalassia testudinum, and Digenea simplex emerged as irreplaceable hubs whose loss would trigger secondary extinctions. CONCLUSIONS: The network is efficient yet fragile, depending disproportionately on few engineering hosts. The Integrated Vulnerability Index offers a transferable framework for conservation prioritisation. Prioritising these engineering hosts will safeguard cryptic biodiversity sustained by the interaction network.