The emergence of high-throughput DNA sequencing methods provides unprecedented opportunities to further unravel bacterial biodiversity and its worldwide role from human health to ecosystem functioning. However, despite the abundance of sequencing studies, combining data from multiple individual studies to address macroecological questions of bacterial diversity remains methodically challenging and plagued with biases. Here, using a machine-learning approach that accounts for differences among studies and complex interactions among taxa, we merge 30 independent bacterial data sets comprising 1,998 soil samples from 21 countries. Whereas previous meta-analysis efforts have focused on bacterial diversity measures or abundances of major taxa, we show that disparate amplicon sequence data can be combined at the taxonomy-based level to assess bacterial community structure. We find that rarer taxa are more important for structuring soil communities than abundant taxa, and that these rarer taxa are better predictors of community structure than environmental factors, which are often confounded across studies. We conclude that combining data from independent studies can be used to explore bacterial community dynamics, identify potential 'indicator' taxa with an important role in structuring communities, and propose hypotheses on the factors that shape bacterial biogeography that have been overlooked in the past.

The rhizosphere microbiome is considered to play a key role in determining crop health. However, current understanding of the factors which shape assembly and composition of the microbiome is heavily biased toward bacterial communities, and the relevance for other microbial groups is unclear. Furthermore, community assembly is determined by a variety of factors, including host genotype, environment and agricultural management practices, and their relative importance and interactions remain to be elucidated. We investigated the impact of nitrogen fertilization on rhizosphere bacterial, fungal, nematode and protist communities of 10 contrasting oilseed rape genotypes in a field experiment. We found significant differences in the composition of bacteria, fungi, protist and nematode communities between the rhizosphere and bulk soil. Nitrogen application had a significant but weak effect on fungal, bacterial, and protist community composition, and this was associated with increased relative abundance of a complex of fungal pathogens in the rhizosphere and soil, including Mycosphaerella sp. and Leptosphaeria sp. Network analysis showed that nitrogen application had different effects on microbial community connectivity in the soil and rhizosphere. Crop genotype significantly affected fungal community composition, with evidence for a degree of genotype specificity for a number of pathogens, including L. maculans, Alternaria sp., Pyrenopeziza brassicae, Olpidium brassicae, and L. biglobosa, and also potentially beneficial Heliotales root endophytes. Crop genotype had no significant effect on assembly of bacteria, protist or nematode communities. There was no relationship between genetic distance of crop genotypes and the extent of dissimilarity of rhizosphere microbial communities. Field disease assessment confirmed infection of crops by Leptosphaeria sp., P. brassicae, and Alternaria sp., indicating that rhizosphere microbiome sequencing was an effective indicator of plant health. We conclude that under field conditions soil and rhizosphere nutrient stoichiometry and crop genotype are key factors determining crop health by influencing the infection of roots by pathogenic and mutualistic fungal communities, and the connectivity and stability of rhizosphere microbiome interaction networks.