The culture-independent high-throughput sequencing methods employed in the NICU facilities revealed far more microbial diversity than previously revealed by culture-dependent or targeted molecular PCR analyses of NICU surfaces. Every surface we sampled was inhabited by tens to hundreds of bacterial genera, averaging approximately 100 bacterial genera per surface. These included genera containing many known opportunistic pathogens (), as well as abundant groups whose pathogenic potential and ability to resist antibiotic treatment are poorly understood (Table S1
). While we detected substantially more diversity with the 16S rRNA methods than typically found with culture-based methods, many of the genera in our study are commonly found in culture-based studies of hospital environments and specifically associated with Hospital-Acquired Infections (HAIs) in neonatal patients. Species of Enterobacter
, and Staphylococcus
were found abundantly in both NICUs (). Members of these genera are known to cause nosocomial infections in infants 
. We also found evidence of other opportunistic pathogens that routinely cause nosocomial infections, including Acinetobacter
, and Stenotrophomonas
. We also observed a substantial number of organisms that are not readily cultured (Table S1
Our Principal Coordinates Analysis (PCoA) of the pairwise weighted UniFrac distances between samples in the NICUs found that nine of the NICU1 surfaces were easily separable from the rest of the surface samples from both NICU1 and NICU2 (top left of ). Most of the NICU samples clustered with other indoor samples from office, healthcare centers and restrooms. However, PCoA showed that those nine NICU1 samples were clearly divergent from the rest of the samples (, pink points). Many of the restroom surface samples (green points) were also quite different from indoor office surfaces (yellow) and air (indoor hospital air (red), outdoor hospital air (purple).
A closer inspection of the microbial diversity in the divergent NICU1 samples indicated that an excess of Enterobacteriaceae sequences was responsible for the divergence of these samples (). Members of the Enterobacteriaceae (e.g., E. Coli
) commonly inhabit the digestive tract and can be found abundantly in feces. E. coli
in particular are very well known hospital ICU pathogens that appear to spread and proliferate quite easily in hospitals 
, and many of whom appear to be developing multi-drug resistance 
. Another very consistent finding was the high proportion of bacterial genera associated with human skin, particularly Propionibacterium
, which was one of the most common in both NICUs. We also found considerable proportions of Corynebacterium
, Lactobacillus, Staphylococcus
, and Streptococcus
, all of which are very common on hand surfaces 
According to Flores et al. (2011), some restroom samples were either dominated by gut- (fecal), vagina- and soil-associated bacteria, while others were dominated by skin-associated bacteria. The majority of the office surfaces contained both skin and soil-associated bacteria 
and clustered with the frequently hand-touched restroom surfaces (e.g., door and handle surfaces; ). SourceTracker shows the source of NICU surface microbes is often human skin to the exclusion of the other sources that were investigated here (). Interestingly, we noticed that NICU1 surfaces seem to resemble human skin far more than NICU2 surfaces. We suspect that this might result from more recent cleaning of NICU2, but unfortunately we do not know when each room was last cleaned. These findings lend considerable weight to the notion that human hands are important vectors for transmitting bacteria in NICU facilities.
Our data provide evidence that NICUs harbor pools of diverse bacteria, that NICU diversity is similar to other indoor surface environments, and that human skin is a primary contributor indicating that hand transfer (touch) can move organisms through such settings. Future work in hospitals should attempt to integrate these molecular methods with long-term assessment of surface bacterial diversity and infection rates over time and record the cleaning schedule to investigate the rate at which skin microbes colonize IHEs.
One clear limitation to our study is that we cannot determine which of the microbes we identified were viable. Non-viable organisms cannot directly cause infection, although they may still contribute antibiotic resistance genes to the wider bacterial community. Numerous previous studies have shown the potential ability of many of the microbes detected by molecular methods to be viable, and it is generally easy to grow microbes from any given indoor surface. Setting aside the futility of trying to cultivate dozens or hundreds of different microbes from even a single sample, future work in this area would benefit from the combination of molecular and cultivation assays, increasingly rapid sequencing technologies, and perhaps the addition of molecular assays that simultaneously determine diversity and viability.
A drawback of this approach is the lack of taxonomic resolution at the strain level, which can be problematic for differentiating pathogens from their non-pathogenic close relatives. Short reads of the 16S rRNA obtained (as obtained with recent sequencing technologies) are effective at providing broad, genus-level characterization of microbial communities. Longer reads or different marker genes that provide strain-level resolution for taxonomic groups of interest (i.e., genes with higher rates of accumulated mutations and therefore more divergence between species and strains) will likely be necessary to accurately detect the presence of specific pathogenic bacterial species and strains.
As these high-throughput methods become cheaper and easier, and as the associated bioinformatics becomes more accessible, techniques such as those described here could be routinely applied in detecting or monitoring the spread of bacteria in NICUs. By detecting departures from “typical” NICU bacterial diversity, an early warning system for infectious agents could be developed. To achieve these goals, more data (including NICU surface time-series data) will need to be gathered to understand what normal bacterial diversity and temporal variability looks like on NICU surfaces. This information is essential to accurately identifying deviations from normality.