Facial Skin Microbiota-Mediated Host Response to Pollution Stress Revealed by Microbiome Networks of Individual

Abstract

Urban living has been reported to cause various skin disorders. As an integral part of the skin barrier, the skin microbiome is among the key factors associated with urbanization-related skin alterations. The role of skin microbiome in mediating the effect of urban stressors (e.g., air pollutants) on skin physiology is not well understood. We generated 16S sequencing data and constructed a microbiome network of individual (MNI) to analyze the effect of pollution stressors on the microbiome network and its downstream mediation effect on skin physiology in a personalized manner. In particular, we found that the connectivity and fragility of MNIs significantly mediated the adverse effects of air pollution on skin health, and a smoking lifestyle deepened the negative effects of pollution stress on facial skin microbiota. This is the first study that describes the mediation effect of the microbiome network on the skin's physiological response toward environmental factors as revealed by our newly developed MNI approach and conditional process analysis. IMPORTANCE The association between the skin microbiome and skin health has been widely reported. However, the role of the skin microbiome in mediating skin physiology remains a challenging and yet priority subject in the field. Through developing a novel MNI method followed by mediation analysis, we characterized the network signature of the skin microbiome at an individual level and revealed the role of the skin microbiome in mediating the skin's responses toward environmental stressors. Our findings may shed new light on microbiome functions in skin health and lay the foundation for the design of a microbiome-based intervention strategy in the future.

Keywords: direct association; microbiome network of individual; microbiome network of population; pollution; single-sample network; skin microbiome.

Figures

FIG 1

Framework of network analyses by microbiome network of individual (MNI). (A) In order to construct an MNI for each subject/sample, we first built a population network of n samples and then eliminated the effect of all but the sample of interest. Next, we repeated this process to obtain an MNI for each sample (detailed information is in Materials and Methods). (B) Samples were collected from 58 subjects, including rural smokers (n = 15 from Chongming smokers [CS]), urban smokers (n = 14 from Shanghai smokers [SS]), rural nonsmokers (n = 14 from Chongming nonsmokers [CN]), and urban nonsmokers (n = 15 for Shanghai nonsmokers [SN]). We constructed an MNI for each subject, totaling 58 individual MNIs. (C) The mediation role of the skin microbiome was revealed by MNI analysis for each subject. In particular, the moderated mediation effect of network properties of MNIs was statistically tested using conditional process analysis.

FIG 2

Analysis of the microbiome network of populations (MNP) for each group of subjects under different pollution conditions. (A) MNPs of each group. Each node represents a microbial taxon (OTU). Associations were detected using the SPIEC-EASI method. Clearly, there were significant differences in terms of network between the four groups. Three dominating genera (accounting for over 50% of the total abundance of all species), Cutibacterium (green), Corynebacterium (blue), and Staphylococcus (red), were colored. (B) Connectivity properties (node number, edge number, mean degree, and average path) of the MNPs of each group. (C) Fragility of the MNP of each group. Fragility was measured by the percentage of remaining nodes in the giant (largest) component, which was indicated on the y axis. The percentage of nodes/vertices removed was indicated on the x axis. The fragility of each network characterizes the robustness of each network. Both a larger area under the curve and a high robustness value indicate the corresponding network was more robust/less fragile.

FIG 3

Analysis of the microbiome network of individual (MNI) for each subject under different pollution conditions. There were 58 MNIs corresponding to 58 subjects, respectively. (A) The MNI of each subject. Fifteen MNIs of the 15 subjects in the lowest-pollution group (CN) had the densest connections between the four groups. Three dominating genera (accounting for over 50% of the total abundance of all species), Cutibacterium (green), Corynebacterium (blue), and Staphylococcus (red), were colored. (B) Violin chart demonstrating the connectivity property of MNIs for each group of subjects, including node number, edge number, mean degree, and average path (Wilcox test P value: **, <0.05; ***, <0.001; ****, <0.0001). Clearly, the network connectivity decreased (CN > CS > SN > SS) with decreased pollution stress. (C) Line chart with error bars (median) demonstrated the fragility change as nodes were removed in each MNI network. The robustness of MNIs decreased (CN > CS > SN > SS) with decreased pollution stress and was consistent with the result of MNP. Both a larger area under the curve and a higher robustness value indicated the corresponding network was more robust.

FIG 4

Mediation analysis reveals the mediation role of the skin microbiome. Only models that had consistent mediation effects are contained in the table. Each model contains three variables (X, exposure; M, mediator; and Y, outcome). Each number in the table represents the coefficient of each path (a, X to M; b, M to Y; c, X to Y; total, total effect a*b+c; mediation, mediation effect a*b). Red numbers indicate significant coefficients with a negative value, black numbers indicate significant coefficients with a positive value, and gray numbers indicate nonsignificant coefficients.

FIG 5

Moderation analysis with conditional process analysis linking pollution and microbiome network properties to skin attributes. (A to I) In each model, regional air pollution was a mediator, and smoking was a moderator. MNIs can mediate the damage of pollution on skin indices, while smoking can moderate the path between pollution and MNI properties. As a result, we obtained multiple models that were biologically plausible and were also statistically significant, tested by conditional process analysis (moderated mediation). Here, the number near the path was path coefficient, which was labeled with P value significance levels (*, <0.05; **, <0.001; ***, 0.0001).