The experimental design for this study was the topic of a Sloan Foundation-sponsored workshop convened on June 7th, 2012 at the University of Chicago. Among the participants were experts from diverse scientific fields including building engineers, medical practitioners, and microbial ecologists. The final project design was strongly influenced by the group’s suggestions over the course of the presentation, and the final version of the project design, along with the PowerPoint presentation from the June preliminary meeting, is available below.

This is a two-part study. Firstly, a 1 year-long study of microbial diversity, building science and medical microbiology will be performed in the University of Chicago’s new hospital pavilion, involving 10 patient rooms, 2 nursing stations, staff, water and air sampling, both daily and weekly. The second will involve an in-depth look at a single patient room at 16 time points to determine the impact of wounded patients entering the room in multiple rounds in a US Army hospital in Germany. These activities are very different, and while they are examining similar aspects, they are not designed to be immediately comparable, but rather to explore different dynamics in very different hospital systems.

In general hospitalized patients at the University of Chicago are represented within 3 categories of care. 1. A remedial problem in a healthy patient requiring complex medical or surgical intervention. These patients enter the hospital with a relatively normal microbiome, stay for 1-3 days and leave. 2. Patients with a potentially remedial problem requiring complex medical or surgical intervention who have severe clinical illness requiring prolonged confinement. These patients (liver transplant patients, pt with leukemia, dialysis patients, patient with acute life-threatening surgical problems- bleeding, aneurysms, heart failure/infarction, dissections, artery occlusions, etc.) may be colonized by highly pathogenic bacteria as a result of their illness and prior exposure to hospital environments and antibiotics) and will require prolonged confinement and likely will become colonized by even more pathogens that they will harbor, virulize and disseminate. 3. Patients with chronic subacute illnesses. These patients may be the elderly with pneumonia, malnutrition, etc that come from nursing homes with multiple pathogens and require prolonged confinement, antibiotics, and medical and surgical intervention. Tracking the spread of microbial species throughout the Chicago hospital will be accomplished via complementary air, water, surface, and human bio-monitoring techniques applied weekly over the course of a year, beginning one month prior to the opening of the hospital. We have chosen 5 patient rooms and the accompanying nursing station on the 10th floor (oncology) and 5 patient rooms and the accompanying nursing station on the 9th floor (surgery). These rooms will be on the northern aspect of the building, and those on the 10th floor will be exactly above those on the 9th floor sharing the same water and air handlers, the same square footage, layout, and building plan. In these 10 patient rooms, we will use duplicate sterile swabs to collect bacteria from 5 surfaces of interest: the floor, bedrail, outgoing air filter, cold water supply/faucet, and glove box (these areas were chosen based on the APSF funded workshop meeting in June 2012). Duplicate nasal, hand, and inguinal fold samples will also be acquired from each patient (Table 1). Airborne microorganisms in these rooms will be collected onto filters using active air sampling. Additionally, microbial samples (30-50) to test for effect of surface moisture, temperature and air-grill return on microbial community structure. Two of these rooms will be chosen to perform this sampling regimen daily over 365 days (during the days the room is empty we will collect human samples from the sampler themselves), producing 5,840 samples ( duplicate samples will be taken – so 11,680 actual swabs). The other 8 rooms will be sampled weekly for 52 weeks using the same strategy, generating 3,328 samples (in duplicate – therefore 6,654 actual swabs). Eight nursing staff or equivalent from each floor will be sampled at six sites: interior nose, hand, uniform cuff, pager, shoe, and cell phone; this will be performed every week for 52 weeks generating 2,496 samples (in duplicate – therefore 4,992 actual swabs). Finally, two nursing station (1 on each floor) will be examined at seven sites: countertop, computer mouse, phone handle, chair, and main corridor floor, hot water, and cold water; this will be performed every week for 52 weeks, generating 728 samples (in duplicate – therefore 1,456 swabs). In total we will sample 12,392 samples between January 2013 and January 2014. This is about 60% of what we processed for the current EMP (, and is well within our capacity.

We will also collect larger surface area samples in each location weekly for viral sample analysis. These samples will survey a 12’’x12’’ surface using Rayon swabs, subsequently stored at -80C. We will perform this in collaboration with Prof. Scott Kelley, with whom we will pursue a separate funding opportunity to explore the viral community development alongside the microbial community structure.

Existing laboratory infrastructure will be utilized for the processing of samples, which will consist of both culture and culture-independent methods. Amplicon sequencing of 16S rRNA, 18S rRNA, and fungal ITS will be performed on the Illumina HiSeq platform to identify community structure. The occurrence of biologically relevant species such as methicillin-resistant Staphylococcus aureus and vancomycin-resistant enterococci will be assessed using previously established agar plate culturing techniques [1]. Quantitative PCR will be used to determine the absolute abundance of bacteria for a sub-selection of samples based on the 16S and 18S rRNA community analysis.

The emergence and spread of antimicrobial resistance determinants and multidrug resistant bacteria present a serious challenge to modern day public health. To address the challenge of identifying the potential for drug resistance in a rapid and high-throughput manner, we will collaborate with researchers from the Naval Research Laboratory’s Center for Bio/Molecular Science & Engineering – Dr Chris Taitt, Dr Gary Vora, and CPT Benjamin Kirkup – who have developed the Antimicrobial Resistance Determinant Microarray (ARDM) which provides DNA-based analysis for over 250 resistance genes covering 12 classes of antibiotics [2]. Examples of ARDM content for resistance determinants in nosocomial pathogens include mecA (methicillin resistance, i.e. MRSA), carbapenemases (blaKPC, blaVIM, blaGES, blaIMP), extended-spectrum β-lactamases (including blaTEM, blaSHV, and blaCTX-M families), tetX (hydrolysis of all known tetracyclines), antimicrobial modifying enzymes and efflux pumps, rRNA protecting proteins (tetracycline resistance) or modifying enzymes (macrolide resistance, armA for aminoglycoside resistance), as well as Acinetobacter-specific genes (blaOXAs, aminoglycoside modifying enzymes, qacE) and resistance determinants of emerging concern (qepA, qnrS for quinolone resistance, e.g., blaNDM encoding New Delhi metallolactamase), among others. In a collaborative effort with scientists from Walter Reed Army Institute of Research, the Naval Medical Research Unit-3 (NAMRU-3) Cairo, and Mercy Hospital, Sierra Leone, these researchers have recently completed ARDM analysis of over 120 drug-resistant bacteria and have demonstrated significant differences in genetic resistance patterns between geographic areas where the bacteria were isolated. DNA extracted from each sample (same DNA used for amplicon and qPCR analysis) will be applied to an ARDM to identify the presence of multi-drug resistant isolates. This information will be combined with culture assays and 16S rRNA analysis to determine the development of pathogenic reservoirs in the analyzed areas of the hospital.

For the US Army military hospital in Germany, a selected ICU room (likely LRMC 318) will be sampled three times after cleaning and prior to occupation by the first patient. Prior to sampling, the room will be studied for ventilation rates and flow patterns. At the time of occupation, samples of discarded clothing and bandaging will be recovered, as well as daily air samples (through a button aerosol sampler, SKC; NIST Technical Note 1737, 2012) and surface samples (high contact surfaces, horizontal surfaces, vertical surfaces, HVAC, sink, etc); attempting to minimize impact on the operations of the staff. It is hypothesized that clothing samples, in particular, will trap shed skin and the bacteria borne by that skin, while bandaging represents samples from the wound itself. Unpublished data based on pillowcases supports these assumptions (G. Flores, R. Knight, R. Dunn and N. Fierer, unpublished). After the patient has left the room, the room will be sampled once prior to cleaning and three times after cleaning. Samples of linens and discarded materials (ventilator tubing, for example) will also be taken. Sampling of this same room will be repeated with a second patient and after the second patient has departed. This totals 16 time points; with approximately 100 samples processed per time point (Figure 1).

Figure 1 – Military Treatment Facility sampling strategy. There will be 16 sampling events with approximately 100 samples per event. Red arrows represent cleanings. Samples will be taken once per day at a set period during the day while patients are resident. This schedule will adjust depending on the requirements of the facility (shortage of rooms, perhaps) and the lengths of stays for individual patients (usually 2-3 days).

Statistical analysis of this data will be performed using non-parametric multivariate techniques for community composition and univariate analysis of variance tests for diversity measures via MoBEDAC (which will be used for data release). Using these methods, we will identify the most influential environmental parameters (e.g. temperature, building material, adjacent microbiome, etc) that shape microbial community at each sampled location. The benefit of a multivariate crossed analysis is that we will be able to determine if a particular combination of the environmental parameters (an interaction) has a synergistic effect on community composition, causing an even greater change in the presence of certain HAPs. We will also examine whether there is a core microbiome commonly shared across a significant subset of samples. Univariate tests of diversity indices will use higher-way ANOVA, and calculated with distribution-free, permutation-based (PERMANOVA) routines [3]. Additionally, following taxonomic characterization of the communities, using the QIIME pipeline [4], and production of an abundance matrix of operational taxonomic units against experimental condition, community similarity between samples will be represented by calculating a Bray-Curtis similarity matrix and UniFrac distances [5]. Non-metric multidimensional scaling will be used to visualize the relationship between the experimental factors and formally tested using a combination of permutation-based PERMANOVA and fully non-parametric ANOSIM tests [6]. Temporal patterns will be explored using temporal autocorrelation techniques to determine the linear regression of community structure metrics against time; this will be used to explore succession between the different temporal scales and different time-frames of different patient stays. State-of-the-art artificial neural network software developed by our group [7] will be employed to generate models for predicting the development of microbial communities based on the bacterial abundances observed in this study. Source tracking algorithms [8] will be used to trace movement of communities between sampling sites. Taken together, these analyses will provide the first report of the driving factors behind the development of microbial communities in a hospital, and pave the way for designing buildings that are resistant to colonization by pathogens.

Table 1: Sampling sites in Chicago hospital

  Samples/Replicate Replicates Days Total Samples
Patient Rooms Nose, Hand, Inguinal Fold, Floor, Bedrail, Air Filter, Cold water tap, Glove Box 8 52 3,328
2 365 5,840
Staff Nose, Hand, Phone, Pager, Uniform Cuff, Shoe 8 52 2,496
Nurse Station Countertop, Computer Mouse, Phone, Chair, Corridor Floor, Hot Tap Water, Cold Tap Water 2 52 728

While the sample selection was designed based on a best-case scenario, taking into consideration financial constraints we recognized that the sensitivity of this method to identify taxonomic resolution and temporal/spatial patterns. The limitations of this sampling design are that it is not possible to track or observe every surface (both building and human) at every time point, and as such despite the size of this study we will have to use statistical analysis (outlined) below to interpolate patterns between locations and time points. However, the design has taken this into consideration, and as such has multiple time scales, and spatial scaling to enable regression based exploration of patterns of organism dispersal and community succession. This being understood, the proposed methodology is the most appropriate balance between breadth and depth of analysis of community profiles and cost of study, and will represent the most significant effort to date to explore such dynamics.

Central to the research effort will be the analysis of both the human demographics (age, race, sex, reason for being hospital (visitor, patient, staff) that interact with the environment, and the building properties that influence the microbial community structure. To make this possible we will employ the following techniques:

i. Human Demographics and Movement

The nGage radio frequency identification (RFID) system developed by Proventix will be deployed in all of the patient rooms in this study. The system is used by many hospitals (mostly in Alabama) to monitor hand hygiene. Similar systems are used in additional hospitals. The technology is not, by any means, widely used for hand hygiene monitoring, but almost all hospitals use the technology for tracking supplies, nurses, and/or equipment. We currently have permission from the appropriate administration group in the hospital, and are including tracking healthcare provider movements in the IRB. This system is comprised of two types of devices: communication units mounted near the hand-washing station in each room, and uniquely identifiable RFID tags worn by hospital staff. The wall-mounted sensor is designed to detect the RFID tags present in the room, as well as when each tag is within close proximity to the hand-washing station. Intended as a method to monitor whether hospital staff wash their hands upon entering and exiting a patient room, this infrastructure will additionally be leveraged to track movements of staff among patient rooms and continuously monitor the number of occupants within each room (by distributing RFID-equipped badges to visitors). This data will enable correlations of microbial succession to human traffic patterns. For instance, if the degree of room-to-room contamination is influenced by the order in which nurses perform rounds. In addition, all patient movement will be recorded, e.g. when the patient is moved to a different location and returned to the room for a procedure.

ii. Building Science

The building science measures are designed to support the microbial measurements by providing meaningful data on several building parameters that may influence the microbial community. Most of the measurements focus on using HVAC filters as samplers as we have done in earlier and ongoing measurements [9,10]. This will provide an integrated measurement of particle-bound contaminants, particularly bioaerosols, in each patient room. The overall approach will be to utilize the existing near-floor return grills as sampling locations. Based on the site visit in June 2012 and conversations with hospital administrators earlier this month, we will use a very thin piece of highly-efficient filter media on the exterior of the filter grill and attached with a large magnet (or clips if the grill is not magnetic). This has several advantages over using a whole filter, including:

  1. It is much thinner and therefore has a lower pressure drop and will not affect airflows in the system
  2. It will be much easier to extract the DNA than it would from an ordinary filter.
  3. It will be easy and fast for the field team to collect samples every week.
  4. The aesthetic impacts will be small.

The filter media will be removed and replaced every week. In order to yield average airborne microbial concentrations, we will also use the filter media as a flow measurement device by measuring the static pressure across the filter. These measurements, in combination with the particle size-resolved efficiency of the filter media (provided by the manufacturer), will also provide size resolution of the captured particles. The pressure sensor will be connected to a datalogger that measures temperature and relative humidity for the space.

To support the airborne concentration measurements, it is essential to know the amount of outdoor air in the room. To assess this, we will measure the carbon dioxide concentrations in the air handling unit(s) that serve the patient rooms. In particular we will measure the outdoor air concentration, the return air concentration entering into the unit(s), and the mixed air concentration leaving the units. These measurements will provide the fraction of outdoor air for each room and in combination with the airflow measurements discussed above and below will provide the amount of outdoor air in each room.

The patient rooms are also negatively or positively pressurized with respect to the rest of the hospital. To assess the amount and sign of pressurization, we will also install a datalogger and pressure sensor on the supply air grill in each room. Similar to the filter media example above, we will use the grill as a flow measurement device: by measuring the static pressure drop across the grill we will know the supply air flow into the room. The difference between the inflow and the outflow (measured above at the filter media) will provide an estimate of the pressurization as well as provide information about the air exchange rate of the room. The datalogger used for this sensor will also measure temperature and relative humidity of the supply air, as well as the light levels in the room (this is an piece of building factor metadata that is typically included for microbial samples.)

The last building factor that is important to measure is the occupancy of the room. This will be done with the RFID tags as discussed above and will be supplemented with a carbon dioxide measurement in the return air in each room. A mass balance on carbon dioxide in the room will provide a check on the room occupancy, as well as provide more nuanced ventilation information because it will allow for an assessment of the infiltration of air from the other parts of the hospital.

In combination, these building measurements will provide a robust set of metadata that will provide valuable context to the microbial community samples.


  1. Wiener-Well Y, Galuty M, Rudensky B, Schlesinger Y, Attias D, et al. (2011) Nursing and physician attire as possible source of nosocomial infections. American Journal of Infection Control 39: 555–559. doi:10.1016/j.ajic.2010.12.016.
  2. Taitt CR, Leski T, Stenger D, Vora GJ, House B, et al. (2012) Antimicrobial resistance determinant microarray for analysis of multi-drug resistant isolates. In: Southern SO, Prabhakar S, Montgomery KN, Taylor CW, editors. SPIE. Vol. 8300. p. 83710X. doi:10.1117/12.924569.
  3. Anderson MJ (2001) A new method for non-parametric multivariate analysis of variance. Austral Ecology 26: 32–46. doi:10.1111/j.1442-9993.2001.01070.pp.x.
  4. Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K, Bushman FD, et al. (2010) QIIME allows analysis of high-throughput community sequencing data. Nature Methods 7: 335–336. doi:10.1038/nmeth.f.303.
  5. Lozupone C, Knight R (2005) UniFrac: a New Phylogenetic Method for Comparing Microbial Communities. Applied and Environmental Microbiology 71: 8228–8235. doi:10.1128/AEM.71.12.8228-8235.2005.
  6. Clarke KR (1993) Non-parametric multivariate analyses of changes in community structure. Austral Ecology 18: 117–143. doi:10.1111/j.1442-9993.1993.tb00438.x.
  7. Larsen PE, Field D, Gilbert JA (2012) Predicting bacterial community assemblages using an artificial neural network approach. Nature Methods. doi:10.1038/nmeth.1975.
  8. Knights D, Kuczynski J, Charlson ES, Zaneveld J, Mozer MC, et al. (2011) Bayesian community-wide culture-independent microbial source tracking. Nature Methods 8: 761–763. doi:10.1038/nmeth.1650.
  9. Noris F, Siegel JA, Kinney KA (2011) Evaluation of HVAC filters as a sampling mechanism for indoor microbial communities. Atmospheric Environment 45: 338–346. doi:10.1016/j.atmosenv.2010.10.017.
  10. Noris F, Siegel JA, Kinney KA (2009) Biological and Chemical Contaminants in HVAC Filter Dust. ASHRAE Transactions 115: 484–491.