Legumes can preferentially select beneficial rhizobial symbionts and sanction ineffective strains that fail to fix nitrogen. Yet paradoxically, rhizobial populations vary from highly beneficial to ineffective in natural and agricultural soils. Classic models of symbiosis focus on the single dimension of symbiont cost-benefit to sympatric hosts, but fail to explain the widespread persistence of ineffective rhizobia. Here, we test a novel framework predicting that spatio-temporal and community dynamics can maintain ineffective strains in rhizobial populations. We used clonal and multistrain inoculations and quantitative culturing to investigate the relative fitness of four focal Bradyrhizobium strains varying from effective to ineffective on Acmispon strigosus. We found that an ineffective Bradyrhizobium strain can be sanctioned by its native A. strigosus host across the host's range, forming fewer and smaller nodules compared to beneficial strains. But the same ineffective Bradyrhizobium strain exhibits a nearly opposite pattern on the broadly sympatric host Acmispon wrangelianus, forming large nodules in both clonal and multistrain inoculations. These data suggest that community-level effects could favour the persistence of ineffective rhizobia and contribute to variation in symbiotic nitrogen fixation.

The patterns and drivers of bacterial strain dominance remain poorly understood in natural populations. Here, we cultured 1292 Bradyrhizobium isolates from symbiotic root nodules and the soil root interface of the host plant Acmispon strigosus across a >840-km transect in California. To investigate epidemiology and the potential role of accessory loci as epidemic drivers, isolates were genotyped at two chromosomal loci and were assayed for presence or absence of accessory "symbiosis island" loci that encode capacity to form nodules on hosts. We found that Bradyrhizobium populations were very diverse but dominated by few haplotypes-with a single "epidemic" haplotype constituting nearly 30 % of collected isolates and spreading nearly statewide. In many Bradyrhizobium lineages, we inferred presence and absence of the symbiosis island suggesting recurrent evolutionary gain and or loss of symbiotic capacity. We did not find statistical phylogenetic evidence that the symbiosis island acquisition promotes strain dominance and both symbiotic and non-symbiotic strains exhibited population dominance and spatial spread. Our dataset reveals that a strikingly few Bradyrhizobium genotypes can rapidly spread to dominate a landscape and suggests that these epidemics are not driven by the acquisition of accessory loci as occurs in key human pathogens.