Understanding drivers of antagonistic interactions across temporal and spatial scales is important for predicting community structure. In particular, studies examining spatial variation in ecological networks are critical for anticipating community responses to anthropogenic change. Most studies examining spatial interaction turnover focus on bipartite networks begging the question of whether the results are also reflected in unipartite, multi‐trophic networks. To examine the spatial turnover in food web interactions, the environmental and ecological drivers of this, and the influence of interaction turnover on the preservation of individual species’ roles, we used a spatially expansive multi‐trophic antagonistic ecological network data set of 129 lakes spanning over 1000 kms. We used β‐diversity metrics to quantify spatial turnover in interactions and calculated the relative contributions of interaction rewiring and turnover in top, intermediate, and basal species to network turnover. We then investigated the relative and combined role of multiple ecological drivers (e.g., abundance, thermal tolerance) and environmental drivers (e.g., latitude, total phosphorus) on internal network structure. Finally, we used a motif analysis to measure the effect of spatial interaction turnover on the variation in individual species’ roles. We observed high interaction turnover across lakes, driven primarily by turnover in basal species but also the rewiring of interactions among shared species, driven, in part by underlying environmental gradients (e.g., species richness). Contrary to previous food web models applied to single sites, none of the ecological drivers we considered were effective predictors of lake‐specific interactions perhaps indicating an important distinction between network model accuracy at regional and local extents. Finally, despite high spatial turnover in interactions, species’ roles were highly conserved across the study lakes demonstrating the potential of species’ roles for predicting community structure. These findings demonstrate how integrating species’ fundamental roles into trait‐based approaches may improve our predictions of ecological networks at local scales.