Fig. 2

Microbial species abundance changes with differences in trained and stimulus-specific immune responses. A Ex vivo cytokine expression upon stimulation with M. tuberculosis (M. tb) or S. aureus at baseline (b) and at 2 weeks (2w) and 3 months (3m) after BCG vaccination. P-values were determined by paired Kruskal-Wallis tests. Significant correlations between changes in cytokine expression at 2 weeks (B) and 3 months (C), measured as fold change against baseline levels (Spearman correlation, P < 0.05). Diagonal cells show significance of changes for each individual cytokine versus baseline (T-test, *P < 0.05, ***P < 0.001). D Significant MSPs associated with trained immune responses subject to FDR < 0.2 and prevalence > 20%. The effect of each MSP was estimated by a linear model of the fold change in TNF-α, IL-1β, and IL-6 responses versus the baseline, adjusted for age and sex (“Methods”). E Significant MSPs associated with specific response (IFN-γ stimulated by M. tuberculosis) and subject to FDR < 0.2 and prevalence > 20%, using a model of the same form as in D. F Distribution of unadjusted P-values for each MSP in the trained and specific response models. Skew from uniform distribution suggests more pronounced effects of MSPs on cytokine expression fold change. G Individual effects on cytokine fold changes by each MSP and time point, quantified as explained variance multiplied by effect sign. Number of associations per MSP and confidence in species-level annotation are reported on the left. H MSP 112 (Roseburia) abundance decreases two of the three trained immunity phenotypes in both subsequent time points. Relative abundance of the MSP was converted to ranks, and ranks were sorted into bins by equal-frequency binning. P-values in titles refer to the trained immunity linear model. Coefficients in panels are Spearman correlations (*P < 0.1, ***P < 0.001). I MSP 091 (Ruminococcus) and MSP 181 (Eggerthella lenta) ranked abundance versus specific response phenotype. Relative abundance of the MSP was converted to ranks, and ranks were sorted into bins by equal-frequency binning. P-values in titles refer to the specific response linear model. Coefficients in panels are Spearman correlations (*P < 0.1, **P < 0.01)