Figure 1: Experimental design
(a) Configuration of sampling site in Flagstaff office 1. This configuration was similar to those set up in all offices. Signs on the wall adjacent to wall sampling plate describe the project, as request that the materials not be touched. (b) Diagram of single sampling plate illustrating nine sampling swatches (circles) of three different materials, one row for tracking equilibrium relative humidity of the materials (Row #1), one row for infrequent sampling (Row #2), and one row for frequent sampling (Row #3). (c) Samples were collected from rows 2 and 3 of all sampling plates from three offices in each of our three cities in four intensive sampling periods over the course of one year. Coloring of sampling swatches in this figure illustrates the change in bacterial Phylogenetic Diversity over the year.
Figure 2: Bacterial community dissimilarity
(a) Weighted UniFrac, a quantitative measure, and (b) unweighted UniFrac, a qualitative measure for the Flagstaff samples (similar patterns are observed across all cities). Darker colors indicate that groups of samples are more dissimilar and lighter colors indicate that groups of samples are more similar. The labels indicate the location in the office followed by the material type. For example, Ceiling/Carpet indicates the samples of carpet that are installed on the ceiling sampling plate, and investigation of the first column of (a) shows that the carpet samples on the ceiling are more similar to the carpet samples on the wall than they are to the carpet samples on the floor.
Figure 3: Disturbance due to repeat sampling, though detectable, is small compared to other variables
(a) Phylogenetic diversity (PD) of frequently sampled swatches (“Row 2”, see Figure 1b) was consistently lower than PD of the frequently sampled swatches, suggesting that sampling decreases the PD of the sites. (b) Weighted UniFrac shows that the samples with the same sampling frequency are more similar to each other than samples with different sampling frequencies, suggesting an effect of repeat sampling. However comparison of this difference to effects of biological interest show that sampling frequency has a larger effect on the bacterial communities than the material does (which we show in Figure 2 is effectively null), but a small effect than the location of the plate in the office or the sequencing run, which we show in Figures 2 and 5 to impact our observed bacterial communities. (c) Similar results are observed with unweighted UniFrac.
Figure 4: Confusion matrices illustrating the performance of city and office SVM classifiers
(a) True (actual) city and predicted city for our SVM classifier when trained and tested on office microbiomes by city. Dark colors on the diagonal indicate the predicted label is very frequently correct. (b) True and predicted office for our SVM classifier when trained and tested on office microbiomes from all cities labeled by individual office. Note that the diagonal is not as dark as in (a), illustrating that the predicted office labels are not correct as often as the predicted city labels. When an incorrect office is predicted, it is often from the correct city, as indicated by the darker colors surrounding the diagonal. (c-e) True and predicted office for our SVM classifiers when trained and tested on office microbiomes from cities individually.
Figure 5: Investigation of sequencing run effect on observed bacterial community composition
(a) Weighted and unweighted UniFrac principal coordinates analysis (PCoA) and Phylogenetic Diversity (PD) by sequencing run for sequencing runs 1, 2 and 3. (b) Weighted and unweighted UniFrac PCoA and PD by season. (c) Weighted and unweighted UniFrac PCoA and PD for samples of the same sites from summer and fall when sequenced in a single sequencing run. (d) Weighted and unweighted UniFrac PCoA and PD of technical replicate samples sequenced on sequencing runs 1, 3, and 4.
Figure 6: Community richness as a function of equilibrium relative humidity of sites for bacterial (16S) and fungal (ITS) communities
We observe weak correlations between community richness and equilibrium relative humidity (ERH) of sites. While these correlations are statistically significant, we do not find these relationships to be convincing for reasons discussed in Results and Supplementary Results.
Figure 7: Human-based source tracking of office microbiome samples
(a) Percent contribution of microbiomes of different sites of the human body to the office bacterial communities. Unknown indicates contribution from a source other than the body sites tested here. (b) Percent contribution of microbiomes of different human subjects to the office bacterial communities. Unknown indicates contribution from a source other than the individuals tested here.
Figure 8: Longitudinal analysis of bacterial Phylogenetic Diversity (PD) over one year
(a) Flagstaff, (b) San Diego, and (c) Toronto. Each “violin” represents a two week period from the beginning, middle, and end of our four sampling periods.[b]
Supplementary Table 1
Descriptions of offices included in this study.
Supplementary Dataset 1
Differentially abundant OTUs across office locations, as determined by ANCOM.
Supplementary Dataset 2
Differentially abundant OTUs across frequently and infrequently sampled sites, as determined by ANCOM.
Supplementary Dataset 3
Differentially abundant OTUs across cities, as determined by ANCOM.
Supplementary Figure 1: Experimental design
Analog of Figure 1, illustrating fungal community results.
Supplementary Figure 2: Fungal community dissimilarity
Analog of Figure 2, illustrating fungal community results.
Supplementary Figure 3: Disturbance due to repeat sampling, though detectable, is small compared to other variables
Analog of Figure 3, illustrating fungal community results.
Supplementary Figure 4: Confusion matrices illustrating the performance of city and office SVM classifiers
Analog of Figure 4, illustrating fungal community results.