Temporal variability is a personalized feature of the human microbiome

Abstract

Background: It is now apparent that the complex microbial communities found on and in the human body vary across individuals. What has largely been missing from previous studies is an understanding of how these communities vary over time within individuals. To the extent to which it has been considered, it is often assumed that temporal variability is negligible for healthy adults. Here we address this gap in understanding by profiling the forehead, gut (fecal), palm, and tongue microbial communities in 85 adults, weekly over 3 months.

Results: We found that skin (forehead and palm) varied most in the number of taxa present, whereas gut and tongue communities varied more in the relative abundances of taxa. Within each body habitat, there was a wide range of temporal variability across the study population, with some individuals harboring more variable communities than others. The best predictor of these differences in variability across individuals was microbial diversity; individuals with more diverse gut or tongue communities were more stable in composition than individuals with less diverse communities.

Conclusions: Longitudinal sampling of a relatively large number of individuals allowed us to observe high levels of temporal variability in both diversity and community structure in all body habitats studied. These findings suggest that temporal dynamics may need to be considered when attempting to link changes in microbiome structure to changes in health status. Furthermore, our findings show that, not only is the composition of an individual's microbiome highly personalized, but their degree of temporal variability is also a personalized feature.

Figure 1

Body habitats exhibited different levels of temporal variability both in diversity (A) and membership (B). In (A), each point represents the temporal variability of a single individual colored by gender (red = female, blue = male) with black bars representing the median for a given body habitat and metric. Statistical differences were observed for each metric across body habitats (Kruskal-Wallis, P ≤0.01) and comparisons based on pairwise Mann-Whitney U-test are denoted by asterisks (* = corrected P ≤0.05, ** = corrected P ≤0.01). In (B), the smaller, lighter shaded bars in each plot are for all phylotypes except singletons and the larger, darker bars are only for the 100 most abundant phylotypes for each individual. Error bars in (B) are ±1 SEM.

Figure 2

Boxplots of unweighted (A) and weighted (B) intra-individual UniFrac distances for each body habitat. A broad range of temporal variability in microbial community membership (A) and structure (B) was observed across body habitats and within body habitats across individuals. Individuals are sorted by median in each plot. Green bars depict individuals who did not report antibiotic use during the study period while blue bars indicate individuals who took antibiotics. The median values for each body habitat are shown with vertical red lines. Dotted horizontal lines in each plot divide the study population into first and fourth quartiles and depict ‘stable’ and ‘variable’ individuals, respectively. Non-parametric Mann-Whitney U-tests were used to determine the affect of antibiotic use on temporal variability within each body habitat. P values are shown in each panel. Note that statistical differences were observed for each metric across body habitats (Kruskal-Wallis, P ≤0.01).

Figure 3

Relationship between diversity and variability of microbial communities associated with each body habitat. Diversity was measured as the median Shannon Diversity Index for each individual over the 3-month sampling period. Variability was measured as intra-individual median weighted (white boxes) and unweighted (gray circles) UniFrac distance. Each point represents values of the time-series data for one individual. Spearman rank correlation coefficients are presented for statistically significant relationships (P ≤0.01). Note that similar patterns were observed with other alpha diversity metrics (Additional file 11).

Figure 4

Average taxonomic composition was different among stability classes across individuals. Individuals were assigned to stability classes based on quartiles (first = stable (blue), second and third = average (red), fourth = variable (green)) of median weighted UniFrac distances for each body habitat. Significant differences were observed across forehead (A) and gut (B) communities but not in palm (C) or tongue (D) communities as determined by rank transforming the most abundant bacterial families (>1% in any group) for each body habitat and testing for differences between stability classes using the nonparametric Kruskal-Wallis analysis of variance. Significance is denoted with asterisks (* = corrected P ≤0.05, ** = corrected P ≤0.01).