Computation of a proxy of the flow of a perturbation across a network: the eigenvector centrality
The more connected a taxon is to the rest of the network, the more a perturbation affecting that taxon is likely to spread across the whole network (Dunne et al. 2002; Delmas et al. 2019). To quantify the degree to which a taxon is connected to other highly connected species, we used the eigenvector centrality metric, which is akin to a simulation of flow across a network (Bonacich 1987; Delmaset al. 2019). Taxa heavily connected to other highly connected taxa hence get higher scores of eigenvector centrality. Finally, to investigate whether there is a pattern of sensitivity and centrality across trophic pathways (pelagic for the lowest values of δ13C and benthic for the highest values) and trophic levels, a linear regression between sensitivity or eigenvector centrality and corrected values of δ13C or δ15N was realised (Fig. S1).