Humans differ in their personal microbial cloud

Dispersal of microbes between humans and the built environment can occur through direct contact with surfaces or through airborne release; the latter mechanism remains poorly understood. Humans emit upwards of 106 biological particles per hour, and have long been known to transmit pathogens to other individuals and to indoor surfaces. However it has not previously been demonstrated that humans emit a detectible microbial cloud into surrounding indoor air, nor whether such clouds are sufficiently differentiated to allow the identification of individual occupants. We used high-throughput sequencing of 16S rRNA genes to characterize the airborne bacterial contribution of a single person sitting in a sanitized custom experimental climate chamber. We compared that to air sampled in an adjacent, identical, unoccupied chamber, as well as to supply and exhaust air sources. Additionally, we assessed microbial communities in settled particles surrounding each occupant, to investigate the potential long-term fate of airborne microbial emissions. Most occupants could be clearly detected by their airborne bacterial emissions, as well as their contribution to settled particles, within 1.5–4 h. Bacterial clouds from the occupants were statistically distinct, allowing the identification of some individual occupants. Our results confirm that an occupied space is microbially distinct from an unoccupied one, and demonstrate for the first time that individuals release their own personalized microbial cloud.


INTRODUCTION 1
Humans harbor diverse microbial assemblages in and on our bodies (HMP Consortium, 2012), and these 2 distinctly human-associated bacteria can be readily detected inside of buildings on surfaces, in dust, and  Kembel et al., 2012Kembel et al., , 2014. Human-associated bacteria disperse into and throughout 5 humans within the built environment than those found on surfaces or in resuspended dust, since recently 43 emitted microbes are more likely to be physiologically active and have not been subjected to prolonged 44 desiccation or UV exposure before colonization can successfully occur. 45 In order to understand the human contribution to bioaerosols within built environments, and the 46 extent to which this emitted bioaerosol pool contributes to the residual human-microbial signal detected 47 in indoor dust and on surrounding indoor surfaces, we characterized the airborne bacterial cloud of a 48 person sitting in a sanitized experimental climate chamber (Fig. S1). Background bacterial biomass in the 49 chamber was reduced by a combination of surface disinfection and ventilation control. Over the course of 50 two separate experiments, we used high-throughput DNA sequencing methods to characterize airborne 51 bacterial community composition emitted by 11 different human occupants. During the first experiment, 52 we compared airborne bacterial assemblages to those detected simultaneously in an identical, adjacent 53 unoccupied side of the chamber. This was repeated for three different people, each for 4-and 2-hour 54 sampling periods. To assess the potential for these airborne particles to result in a detectable human 55 microbiome signal as settled dust on surrounding surfaces, we sequenced DNA from settling dishes in 56 each sampling period, and compared those to airborne assemblages. Given that occupants could each be 57 clearly detected and differentiated from one another, we designed a second experiment to further explore 58 the distinguishability of the personal microbial cloud. For this we sampled 8 different people for 90 59 minutes each, and with air flowing at 1 air change per hour (ACH) and 3 ACH. Each occupant's personal 60 microbial emissions were compared among occupants, and to filtered supply and exhaust air from the 61 occupied chamber to assess personal detectability within a building's ventilation system.  Occupants self-reported their comfort, and any necessary temperature adjustments were made without 121 tempering air, but rather by adjusting radiant floor temperature. Air temperature and relative humidity 122 were monitored using data loggers (#U12-012, Onset Computer Corporation, Bourne, MA). Temperatures 123 inside the climate chamber throughout both experiments ranged from 22-26 • C, and relative humidity 124 ranged from 25-45%.

125
During the first experiment, each day consisted of a single occupant in the chamber for one 240-and 126 one 120-minute sampling period, with a break between sampling periods. During the second experiment, 127 each occupant was in the chamber for two separate 90-minute periods, once at 3 air changes per hour 128 (ACH) ventilation rate, and again for 1 ACH.

129
Particle count data was collected at a rate of 2.83 L minute −1 in 1 minute intervals, and size fractionated 130 with the AeroTrack 9306-V2 (TSI Inc., Shoreview, MN, USA). Three different particle size classes (2.5-5 131 µm, 5-10 µm and 10+ µm) were considered for this study. All particle counts were averaged over 10 132 minute intervals (5 for the second experiment) and converted to L minute −1 ratio above simultaneous 133 unoccupied values for the first experiment. Since no unoccupied chamber was used in the second 134 experiment, we calculated a particle deposition loss coefficient (Tracy et al., 2002) by comparing particle 135 counts in the occupied chamber to particle counts in the supply duct system. Filters and settling dishes 136 were immediately packaged, transported on dry ice, and stored at -80 • C until further processing. 138 To avoid confounding effects introduced during library preparation, all samples were randomized across 139 extraction batch, amplification batch, and processing order. Air filters and settling dishes from both

5/28
PeerJ reviewing PDF | (2015:06:5401:1:1:NEW 28 Aug 2015) Reviewing Manuscript experiments were all processed using methods specifically for low-biomass samples adapted from Kwan et al. (Kwan et al., 2011), and amplicon libraries were constructed following methods from Caporaso et    the second experiment were processed using the QIIME v. 1.8 pipeline, except that OTUs were clustered 197 using USEARCH v. 7 (Edgar, 2013). We retained and demultiplexed 7.5 x 10 7 sequences with expected 198 error rates less than 0.5. Taxonomy was assigned to OTUs using the RDP classifier and Greengenes

First experiment
All samples in the first experiment were rarefied to 1000 sequences per sample to achieve approximately 208 equal sampling depth. β -diversity was calculated using the Canberra taxonomic metric, and ordinations 209 were constructed using iterative non-metric multi-dimensional scaling (NMDS). Community differences 210 were assessed using permutational multivariate analysis of variance tests (PERMANOVA). Since commu-211 nity differences were tested with permutational tests, we report p-values down to, but not below, 0.001.

212
Clustering was conducted with an average linkage method based on Canberra distances. Indicator species  Table S1. Representative sequences from each OTU were 216 BLAST'ed against the NCBI 16S isolate database, resulting in putative species assignments and NCBI 217 accession numbers.

218
Second experiment 219 The goal of the second analysis was different from the first. We were primarily interested in the Dietziaceae, Aerococcaceae, and Tissierellaceae. Thus we didn't rarefy the second dataset, but rather 225 created a subset of relative abundances for analysis. Additionally, since several of these human-associated 226 groups were the most abundant and distinguishing among occupants, we used the Bray-Curtis dissimilarity 227 metric for multivariate analyses, and the Jaccard distance (as a percent of shared OTUs) to show the 228 average shared relationships in Fig. S4. Since community differences were tested with permutational tests, 229 we report p-values down to, but not below, 0.001. Clustering was conducted with Ward's linkage method

236
Sequencing of bacterial 16S rRNA genes across the two experiments resulted in more than 14 x 10 6 237 quality-filtered sequences. Since the objective of the first experiment was to determine the detectability of Reviewing Manuscript a single occupant in a cleaned room, we first focused on differentiating occupied air from unoccupied. 240 In the first experiment, bacterial assemblages in samples from occupied and unoccupied air were signifi-241 cantly different, regardless of occupant, collection method or trial duration (p = 0.001; from PERMANOVA 242 tests on Canberra distances; Table 1). When considering individual sampling periods (Table 2 and Fig. 1), 243 all three individuals could be clearly detected above background airborne communities after 4-and 2-hours 244 from the airborne bacteria collected on air filters. Only Subjects 1 and 3 were consistently detectable from 245 particles in settling dishes at both time intervals; Subject 2 was significantly detectable during the 4-hour 246 sampling period, but not during the 2-hour sampling period (p = 0.34).

247
These community differences were evident in a few specific human-associated bacterial taxa. Indicator  Table 3  (e) Occupant microbial clouds were more similar to other samples from the same person than to other occupants, regardless of sampling method. This difference was significantly more pronounced than that of unoccupied samples taken simultaneously during sampling periods (Fig. S1). Error bars represent ± 1 standard error on pairwise Canberra similarities.  Reviewing Manuscript

Occupants differ in their personal microbial cloud
In addition to our finding that occupants were detectable from their microbial contributions of bioaerosols 258 and/or settled particles, bacterial assemblages were also unique to each of the three occupants, meaning 259 that samples from each individual were statistically distinct and identifiable to that occupant (p = 0.001; 260 from PERMANOVA on 4-hour air filters from each occupant; Fig. 1  first experiment. All OTUs discussed above were also significant indicator taxa (Table 3).  (Table 3); 2) the relative abundances of these families were always elevated in occupied into and out of the occupied chamber, respectively (Fig S1b). We then analyzed the targeted subset of 300 human-associated bacterial OTUs, described above, to determine if and how many occupants could be 301 statistically differentiated just by the air around them. 302 We found that each of the eight occupants emitted their own characteristic concentration of airborne 303 particles. These particle concentrations were correlated with the proportion of human-associated bacteria 304 in the surrounding air, and subsequently with our ability to identify each unique occupant from their 305 microbial cloud (Fig. 2a-c). As before, some occupants' microbial clouds were more detectable than 306 others, and for each person this was predicted by the proportion of targeted human-associated OTUs in 307 an occupant's respective dataset. Samples where the targeted subset of OTUs composed more than 20% 308 of the total generally clustered correctly by occupant, while those with less were generally unable to 309 be classified as being from a specific occupant (Fig. 2a, d & e). The same apparent 20% threshold also 310 applied to the human cloud signal detected in the exhaust air leaving the chamber. We were only able  Reviewing Manuscript among all occupants, some individual OTUs within these genera were indicative of individual occupants 322 (Fig. 4), indicating that species-or strain-level variation in airborne bacteria can inform future microbial 323 cloud and identifiability studies.  Figure 3. Three example cases of detectability in the occupied chamber and the exhaust ventilation system: (a) Subject 11 was an example of ideal detection in the ventilation system -we were able to find sufficient human-associated OTU concentrations to correctly classify the air leaving the occupied chamber. (b) Most occupants, however, did not emit sufficient bacterial concentrations to be detected in the ventilation system, even when they were readily detected within the occupied chamber. (c) Two subjects emitted nearly undetectable concentrations of particles (Fig. 2c) and human-associated bacterial OTUs (Fig. 2b), and were thus impossible to detect or identify in either the occupied chamber or the exhaust ventilation system. combination of resuspended dust, emission from clothing, and active particle emission from occupants.

335
In our study, we made all attempts to eliminate the potential for resuspended dust by heavily cleaning 336 the interior of a controlled climate chamber and eliminating most movement within the chamber. We own detectable microbial cloud (Fig. 2c).

381
Airborne particles, regardless of their biological nature, were also optically measured throughout the 382 experiment in addition to microbial communities, to better understand personalized particle emissions 383 from different people (Qian et al., 2014a). Particle emissions from the eleven occupants in this study 384 varied substantially but were consistent for each person (Table S1 and Fig. 2c). We might expect that the 385 occupant emitting the most particles would also be the most easily discernible from their microbial cloud.

386
This was generally the case, since airborne particle concentrations tended to correlate with the proportion   Figure S1: Schematic of the experimental chamber during both experiments. (a) The objective of the first experiment was to distinguish occupied from unoccupied airborne bacterial communities. Thus the test chamber was split into two identical portions, and air was collected on both sides simultaneously. Supply air velocity (entering through the ceiling plenum) was determined to replace the volume of air removed by vacuum sampling, as well as create slight positive pressure within the occupied chamber. (b) The second experiment was designed to distinguish among occupants, so the test chamber was not divided, but rather vacuum samples were taken in the supply ventilation system, surrounding the occupant in the chamber, and also in the exhaust ventilation system. During the second experiment, supply air velocity resulted in 1 & 3 ACH.

26/28
PeerJ reviewing PDF | (2015:06  Figure S3. Significant differences among occupant personal microbial clouds are not explained by temporal changes in background airborne microbial assemblages. We detected marginal differences in background bacteria (i.e., day-to-day temporal changes). These differences, however, were negligible when compared to differences among the different occupants. (left) Community similarities in the left plot (occupied samples from the first experiment; same data as shown in Fig. 1e) show that occupants were more similar to other samples from the same person than to other occupants, regardless of sampling method. This difference was significantly more pronounced than that of unoccupied samples (right) taken simultaneously during sampling periods. Error bars represent ± 1 standard error on pairwise Canberra similarities.  Figure S4. At 1 air change per hour (left), occupants were, on average, detectable inside the chamber, but less so in exhaust air. When air exchange rates were tripled (right), these signals disappeared, and not a single occupant was consistently detectable, even in occupied indoor air. Bars show average Jaccard Similarity values ± 1 standard error.