The Skin Microbiome of Cohabiting Couples

Abstract

Distinct microbial communities inhabit individuals as part of the human skin microbiome and are continually shed to the surrounding environment. Microbial communities from 17 skin sites of 10 sexually active cohabiting couples (20 individuals) were sampled to test whether cohabitation impacts an individual's skin microbiome, leading to shared skin microbiota among partner pairs. Amplified 16S rRNA genes of bacteria and archaea from a total of 340 skin swabs were analyzed by high-throughput sequencing, and the results demonstrated that cohabitation was significantly associated with microbial community composition, although this association was greatly exceeded by characteristics of body location and individuality. Random forest modeling demonstrated that the partners could be predicted 86% of the time (P < 0.001) based on their skin microbiome profiles, which was always greater than combinations of incorrectly matched partners. Cohabiting couples had the most similar overall microbial skin communities on their feet, according to Bray-Curtis distances. In contrast, thigh microbial communities were strongly associated with biological sex rather than cohabiting partner. Additional factors that were associated with the skin microbiome of specific body locations included the use of skin care products, pet ownership, allergies, and alcohol consumption. These baseline data identified links between the skin microbiome and daily interactions among cohabiting individuals, adding to known factors that shape the human microbiome and, by extension, its relation to human health. IMPORTANCE Our work characterizes the influence of cohabitation as a factor influencing the composition of the skin microbiome. Although the body site and sampled individual were stronger influences than other factors collected as metadata in this study, we show that modeling of detected microbial taxa can help with correct identifications of cohabiting partners based on skin microbiome profiles using machine learning approaches. These results show that a cohabiting partner can significantly influence our microbiota. Follow-up studies will be important for investigating the implications of shared microbiota on dermatological health and the contributions of cohabiting parents to the microbiome profiles of their infants.

Keywords: cohabitation; high-throughput sequencing; human skin; microbiome; random forest modeling.

FIG 1

Microbial diversity of the 10 body locations sampled. (A) Pie charts illustrating the relative abundance of microbial families present at >1% and the phyla to which they belong, organized by each of the 10 body locations sampled. (B) PCoA plot calculated using the Bray-Curtis dissimilarity metric. The 330 samples from all body locations are included and are denoted by body location.

FIG 2

Ordinations (PCoA) generated by using the Bray-Curtis dissimilarity metric for each of the 10 body locations sampled. Lines connect samples from a participant. Female samples are denoted by circular points, whereas male partners are represented by triangles. Where a single sample per person was collected for specific body locations (i.e., back, navel, torso), no lines connect the participant samples. Samples from different couples are indicated by the different colors.

FIG 3

Samples were matched with another sample in the data set that possessed the most similar microbial community. Matches were analyzed to determine the percentage of samples belonging to self, partner, or another participant. (A) Proportion of samples that had the lowest Bray-Curtis distance with either another sample from within an individual, from within a cohabiting couple, or to any of the other participants. (B) Proportion of samples that had the lowest Bray-Curtis distance with nonself samples. (C) Proportion of samples that had the lowest Bray-Curtis distance with nonself, opposite-sex samples. The dotted line represents the threshold that would be expected by random chance from the 20 participants.

FIG 4

Heatmap summarizing the significant (P < 0.05) metadata factors that were collected from a participant survey. Categories with higher F values by PERMANOVA have higher variation in community dissimilarity within 10 body locations. White regions of the heatmap represent nonsignificant results. Body locations and metadata categories were arranged into dendrograms using the Bray-Curtis dissimilarity metric. Back, navel, and torso sites were associated with single samples, which likely contributed to fewer significant F values compared to other locations that were sampled on each side of the body.

FIG 5

Bar plots of the data set with the correct couple composition compared to randomly assorted incorrect pairings. Distribution of the estimated supervised learning error rates (A) and PERMANOVA F values of 1,000 unique artificially shuffled partner pairings (B). The dotted lines represent the position of the result from the correctly matched couples’ data set (P < 0.001 and P = 0.006).