One of the central challenges of evolutionary biology is to understand how coevolution organizes biodiversity over complex geographic landscapes. Most species are collections of genetically differentiated populations, and these populations have the potential to become adapted to their local environments in different ways. The geographic mosaic theory of coevolution incorporates this idea by proposing that spatial variation in natural selection and gene flow across a landscape can shape local coevolutionary dynamics. These effects may be particularly strong when populations differ across productivity gradients, where gene flow will often be asymmetric among populations. Conclusive empirical tests of this theory have been particularly difficult to perform because they require knowledge of patterns of gene flow, historical population relationships and local selection pressures. We have tested these predictions empirically using a model community of bacteria and bacteriophage (viral parasitoids of bacteria). We show that gene flow across a spatially structured landscape alters coevolution of parasitoids and their hosts and that the resulting patterns of adaptation can fluctuate in both space and time.