FIG 2.

Diversity of healthy or diseased skin microbiota among the three cities. (a) PCoA of the healthy controls from the three cities. Within a healthy population, geographic (city) variation in skin microbiota is remarkable (F = 22.09, P = 0.001). (b) PCoA of the lesional samples from the three cities. The between-city difference was reduced after AD infection (from F = 22.09 to F = 14.11), indicating that AD drives the convergence of skin microbiota from distinct cities. (c and d) Shared and specific bacterial genera among the cities, for healthy (c) and lesional (d) samples. (e) Heatmap of bacterial genera in both healthy and lesional skin (only those with relative abundance >1% were shown; *, significantly changed). (f) Skin microbiota predict city origin. Healthy samples were used as training set, with healthy (left), nonlesional (middle), and lesional (right) samples, respectively, as the testing set. In the ternary plot, the closer one sample is to an apex, the more likely it is predicted to be from that city. The likelihood to correctly predict Denver samples is lower than the other cities, and moreover, the Denver samples are more difficult to separate from Beijing than from Qingdao (f, middle diagram). The reason for this observation, however, is not clear.