Fig. 5

Early-life disruption of juvenile skin and gut microbiota predicts later-life resistance to infection. Paths included in the best structural equation model (SEM) based on Akaike information criterion are depicted with solid lines; paths not included in the best model are depicted with dotted lines. The best model shows that the experimental water treatment reduced bacterial phylogenetic diversity in juveniles, and in turn, bacterial diversity in juveniles negatively predicted worm establishment in the guts of adults. The best model combines gut and skin bacterial diversity of juveniles into a latent variable (because they are positively correlated; GLM, χ ² = 31.95, df = 1, P < 0.0001) and therefore we use a factor score to present these data; numbers below the “skin” and “gut” boxes are factor loadings. In contrast, treatment did not affect bacterial diversity of adults, bacterial diversity of juveniles was not related significantly to bacterial diversity of adults, and bacterial diversity of adults was not related significantly to adult resistance to infection. The SEM only included tanks for which there was a complete set of samples for skin and gut bacteria of juveniles, gut bacteria of adults, and infection data (n = 15; solid points); clear points represent the remaining data. Control pond water treatment is represented in green and experimental water treatments (sterile pond water, sterile pond water plus antibiotics) are represented in orange. Individual regressions based on the entire available data set provide similar results as the SEM on the subset of complete data and thus we conservatively provide the results from only the subset. Error bars indicate the s.e.m. P-values and standardized coefficients from the best model are shown next to each path. R ²-values indicate total variance explained by predictor variables in the top model. The χ ² of the best model was 0.38 (df = 2, P = 0.83), indicating that the model was a good fit to the data. This model accounted for 92% of the model weight