Incorporating interactions into a biogeographical framework may serve to understand how such interactions and the services they provide are distributed in space. We begin by simulating the spatiotemporal dynamics of realistic mutualistic networks inhabiting spatial networks of habitat patches. We proceed by comparing these predictions with the empirical results of a set of pollination networks in isolated hills of the Argentinian Pampas. We first find that one needs to sample up to five times as much area to record interactions as would be needed to sample the same proportion of species. Second, we find that peripheral patches have fewer interactions and harbor less nested networks ---therefore potentially less resilient communities--- compared to central patches. Our results highlight the important role played by the structure of dispersal routes on the spatial distribution of community patterns. This may help to understand the formation of biodiversity hotspots.