Network analysis has been harnessed to elucidate microbial interactions and co-occurrence patterns in complex datasets across a variety of habitats. Ecological interpretation of these analyses has often been limited due to a lack of testable hypotheses and ecological theory that can be applied to correlation based analyses. This has been partially addressed by a series of metrics that determine which taxa are “keystones” in the network, defined as taxa that are central in determining microbiome structure and who’s removal results in a fracturing of network topology. These taxa are potentially indicators of particular ecosystem states, targets for microbiome engineering and ideally suited to monitoring and incorporation into predictive models. Here we find keystone taxa across spatiotemporal gradients in three habitats: continental scale soil and marine surveys from the Australian Microbiome Initiative and clinical biofilms from limb wound infections. These taxa are keystones regardless of their abundance category, and are dynamic over gradients using balance-tree distribution modelling. Combined, they represent an inventory of taxa that can be targeted for experimental study and forecasting and who’s population genomes will shed light on microbial functional potential in these habitats.