10th International Aerosol Conference September 2 - September 7, 2018 America's Center Convention Complex St. Louis, Missouri, USA
Abstract View
Spatial and Longitudinal Influences on Accurately Predicting a Microbiome “Biofingerprint”
ANDREW HOISINGTON, Christopher Stamper, Katherine Bates, Christopher Lowry, Air Force Institute of Technology
Abstract Number: 1557 Working Group: Bioaerosols
Abstract Indoor bacterial communities have been linked to the human occupants. More specifically, the human skin bacterial microbial communities have been observed to influence the bacteria identified on surfaces (Hewitt et al., 2012; Dunn et al., 2013) , likely from the contact of the surfaces with hands. Humans share in common only approximately 10% of their skin microbes (Gonzales et al., 2014), which might provide a unique marker for each individual who has touched a surface as well as a way to model the passage of bacteria from person to person via a fomite. Lax et al. (2014) found in a longitudinal study of residential microbes that humans sharing the same home had the most similar microbial skin communities, and that the skin microbial communities of individuals could be used to accurately predict the individual’s home. This study was conducted in two separate locations; work and home environments. Samples were collected once a week for the three consecutive weeks at both locations from the human and built environmental microbiomes. Bacterial DNA was extracted from the samples, amplified, and sequenced on an Illumina MiSeq platform. Bacterial communities were characterized and supervised learning models were used to determine if samples are associated with the participants. Further supervised learning models were used to determine the predictive power for associating possessions and spaces to the owner/ occupant. Finally, supervised learning models and LEfSe analysis were used to determine specific taxa that are biomarkers of the gender associated with a sample. We determined that supervised learning models can associate a participant with their possessions and spaces with a high degree of certainty (90% and 94% accurate for work and home environments, respectively). We also determined that there are varying levels of predictive power associated with the different sample types in determining the identity of the owner/ occupant. In general surfaces commonly in contact with the hand (i.e. computer mouse and computer keyboard) were very good at training supervised learning models to predict the owner/ occupant (~73% accurate). While samples that rarely come in direct contact with the hand (i.e., bedroom floor) were not very good at training supervised learning models to predict the owner/ occupant (~13% accurate). We were able to show that, over the three-week sampling period, sample types had relatively stable bacterial communities, and that some sample types are more longitudinally stable than others. For instance, the computer mouse was highly stable longitudinally, while the hand is more variable. Overall, this study showed that bacterial communities can be used to accurately associate items and spaces to a specific owner/ occupant with a high degree of certainty. Effectively adding to the growing body of evidence that individuals leave a microbial “biofingerprint” on items they touch and spaces they occupy.