A river runs through it: Causal graphs capture rivers' complex control on the genetic structure of populations

Earth's physiographic features shape the genetic evolution of organisms, but understanding how such features act as barriers to gene flow requires quantifying characteristics of both the barrier and the organism. Many barrier characteristics, however, are interdependent and not fully captured by traditional multivariate statistics. Here, we evaluate the use of directed acyclic (causal) graphs and structural equation modeling (SEM) to test the Riverine Barrier Hypothesis using 27 river-spanning population genomic datasets of terrestrial plants and animals associated with 24 rivers across the contiguous United States. These data were paired with seasonality, river width, and river discharge data. SEM analysis revealed patterns not captured by standard approaches. River width had a strong effect on population differentiation, with distinct direct and indirect effects for high and low dispersers. Results suggest a negative width-Fst relationship for low dispersers, which we interpret to be due to topographic context of higher elevation or bedrock-incised rivers. In contrast, high dispersers had a positive relationship, indicating wider rivers present a greater barrier to dispersal. The total effect of river discharge was negligible because its direct effects on population differentiation were canceled out by indirect effects on other river features. Overall, the best-fitting SEM explained 52% of population differentiation for low dispersers and 13% for high dispersers, consistent with the idea that high-dispersing species are less impacted from river barrier effects. This proof of concept shows the utility of causal graphs and SEM at modeling complex relationships between Earth's physiographic features and the organisms that evolve with them.