Widespread Non-Random Species Associations in Freshwater Fishes

The question of how species communities assemble is one of the most fundamental yet unresolved topics in ecology. For stream fishes, findings from studies performed in single catchments were mixed: in many cases assemblages appeared random whereas interspecific competition or environmental heterogeneity shaped species membership in others. A generalizable and robust test of community assembly requires a broad-scale investigation consisting of multiple independent sets of interacting assemblages across vast environmental gradient. Here, we compiled a comprehensive dataset comprising ~580 fish species in >9500 sampling sites spread across the conterminous US to investigate how freshwater fish communities are structured in lotic systems. Using an Empirical Bayes approach that emphasized pairwise instead of conventional (but problematic) matrix-wide species co-occurrence metrics, fish communities were structured in a non-random fashion. Most sub-basins had more co-occurring species pairs as well as more negatively associated species pairs than expected from randomly assembled assemblages. The degree of structure in fish meta-communities across sub-basins can be explained by environmental heterogeneity, strength of interspecific competition, and frequency of predator-prey interactions. Our results establish that structured fish communities are the norm rather than the exception and that multiple ecological processes combine to drive structure.