Résumé
Darwin was one of the first to hypothesize a connection between niche differentiation and competition and species relatedness, offering an appealing framework to disentangle community assembly processes based on phylogenetic diversity patterns. Community assembly is, however, the result of several processes including potentially confounding factors associated with dispersal limitations and spatial effects, casting doubt about the application of phylogenetic diversity metrics to infer community assembly processes. We implemented a spatially-explicit model involving limited dispersal, drift, trait-based selection and competition to simulate community composition under competing assembly processes in a landscape with contrasted habitat connectivity. The phylogenetic structure of communities globally varied depending on assembly processes and the combination thereof, validating the assumption, made by a large number of studies but seldom tested in a spatially-explicit context, that different assembly processes indeed lead to significantly different patterns of community phylogenetic structure. All the investigated alpha metrics exhibited a poor ability to detect overdispersion under stabilizing processes, and some even unduly recovered a signal of clustering. Some of the most widely used metrics, such as UniFrac, carry a redundant signal with non-phylogenetic metrics, and hence, poorly capture the phylogenetic signal in the data. We identified three metrics, namely Bst or Pst for abundance data and PIst for occurrence data, which best retrieved the correct signal of phylogenetic structure under different assembly processes. Spatial effects may blur the phylogenetic structure of communities and decrease our ability to infer underlying processes. However, meaningful results may be obtained when the appropriate comparisons are made. In particular, phylogenetic clustering under equalizing processes must be tested on inter-habitat comparisons because it is the differential filtering of species between habitats that reveals the impact of equalizing processes. Our simulations further suggest that a significant phylogenetic structure of communities can be retrieved even in species-poor communities, except when the communities being compared are dominated by a single, most abundant species. We therefore conclude with best practices to adequately infer assembly processes with useful phylogenetic diversity metrics.