Inferring species interactions from co-occurrence networks with environmental DNA metabarcoding data in a coastal marine food-web

Abstract

Improved understanding of biotic interactions is necessary to accurately predict the vulnerability of ecosystems to climate change. Recently, co-occurrence networks built from environmental DNA (eDNA) metabarcoding data have been advocated as a means to explore interspecific interactions in ecological communities exposed to different human and environmental pressures. Co-occurrence networks have been widely used to characterise microbial communities, but it is unclear if they are effective for characterising eukaryotic ecosystems, or whether biotic interactions drive inferred co-occurrences. Here, we assess spatiotemporal variability in the structure and complexity of a North Sea coastal ecosystem inferred from co-occurrence networks and food webs using 60 eDNA samples covering vertebrates and other eukaryotes. We compare topological characteristics and identify potential keystone species, i.e., highly connected species, across spatial and temporal subsets, to evaluate variance in community composition and structure. We find consistent trends in topological characteristics across co-occurrence networks and food webs, despite trophic interactions forming a minority of significant co-occurrences. Known keystone species in food webs were not highly connected in co-occurrence networks. The lack of significant trophic interactions detected in co-occurrence networks may result from ecological complexities such as generalist predators having flexible interactions or behavioural partitioning, as well as methodological limitations such as the inability to distinguish age class with eDNA, or co-occurrences being driven by other interaction types or shared environmental requirements. Deriving biotic interactions with co-occurrence networks constructed from eDNA requires further validation in well-understood ecosystems, and improved reporting of methodological limitations, such as species detection uncertainties, which could influence inferred ecosystem complexity.