Fig. 4: Different in vitro and in planta patterns of inhibition of pathogens by commensals.

a, NMDS based on Bray–Curtis distances between samples infected with the three synthetic communities across two experiments in August 2018 (Aug) and October 2018 (Oct). The abundance of all 14 barcoded isolates was measured in all communities, including PathoCom and CommenCom, which contained only seven of the 14 isolates, to account for potential cross-contamination and to avoid technical bias. For PathoCom, n = 170, n = 151 for CommenCom and n = 182 for MixedCom. b, Abundance change of the 14 barcoded isolates in MixedCom when compared with their exclusive community in infected plants (that is, PathoCom for ATUE5 and CommenCom for non-ATUE5). Abundance mean difference was estimated with the model [log₁₀(isolate load) ~ treatment × experiment + treatment + experiment + error] for each individual strain. Thus, the treatment coefficient was estimated per isolate. Dots indicate the median estimates, and vertical lines represent 95% Bayesian credible intervals of the fitted parameter. c, Taxonomic representation of the 14 barcoded isolates tested in vitro for directional interactions. Ring colours indicate the bacterial isolate classification, ATUE5 or non-ATUE5. Directional inhibitory interactions are indicated from yellow to black. The experiments were repeated three times with two technical replicates. Only inhibitions observed in at least two independent experiments and in both technical replicates were considered. d, Correlation network of relative abundances of all 14 barcoded isolates in MixedCom-infected plants. Strengths of negative and positive correlations are indicated from yellow to purple. Boldness of lines also indicates the strength of correlation, with all correlations >|± 0.2| shown. Node colours indicate the bacterial isolate classification, ATUE5 or non-ATUE5. e, In planta abundance change of the seven ATUE5 isolates in non-ATUE5 inclusive treatments in comparison with PathoCom. Abundance mean difference was estimated with the model [log₁₀(isolate load) ~ treatment × experiment + treatment + experiment + error] for each individual strain. Thus, the treatment coefficient was estimated per isolate. Dots indicate the median estimates, and vertical lines represent 95% Bayesian credible intervals of the fitted parameter. ‘Combi’ indicates combination of the isolates C3,C4,C5 and C7, and n = 23.