Although biological communities are intrinsically dynamic, with both, species and interactions changing over time, interaction networks analyses to date are still largely static. We implemented a temporal multilayer network approach to explore the changes on species roles and on the emergent structure of a seed-dispersal network over five years. Network topology was relatively constant, with four well defined interaction modules spanning across all years. Importantly, species that were present on more years, were also disproportionally important on each year, thus forming a core of temporally reliable species that are critical to the cohesiveness of the multilayer network structure. We propose a new descriptor termed species activity that reflects the number of temporal, spatial or functional layers (e.g., different years, habitats, or functions) that each species integrates, providing a simple and powerful index of species importance for multilayer network cohesion.