Expanding genetic graphs' potential to analyse ecological connectivity: assessment of graphs construction methods

International audienceDispersal movements are often constrained in human-shaped landscapes, thereby threatening species survival. Landscape genetics approaches are commonly used to analyse ecological connectivity because genetic data well reflect dispersal capacities. When species occupy discrete habitat patches, graph-theoretic methods are a particularly relevant approach to study dispersal-driven gene flow. The links of a genetic graph can be weighted using different genetic distances between populations (nodes). Similarly, graph pruning (link set selection) can rely on different criteria. However, despite growing interest in genetic graphs, the influence of these parameters remains mostly unknown. Here, we assessed the relative influence of genetic distance and graph pruning method on inference. To that purpose, we simulated gene flow between 50 populations on a landscape resistance surface. We constructed genetic graphs using different genetic distances and pruning methods (dispersal distance threshold, topological constraints, conditional independence). We then compared metrics derived from these graphs to analogous metrics describing the topology and connectivity of the dispersal network driving gene flow during the simulation.Genetic graphs consistently reflected the dispersal pattern. However, the total number of links exhibited strong variations between graphs, notably affecting results based on node-level metrics. Hence, precise definition of study objectives is key to relevant genetic graph construction and metrics analysis. Besides, we showed that the pruning method based on conditional independence is compatible with different genetic distances. This study lays down a new framework to use genetic graphs and opens avenues for landscape genetics research on ecological connectivity