It is now well established that the human microbiome contributes fundamentally to human health and disease. Critically ill patients are certain to experience disturbances of the microbiota either due to underlying diagnosis at admission and/or unintended consequences of medical treatment but the specific disturbances encountered by intensive care unit (ICU) patients have not yet been studied at high resolution (1). Older culture-dependent studies have clearly established three principles on this subject: critically ill patients are rapidly colonized by nosocomial pathogens (2–4); colonization by a specific pathogen sharply increases the risk of an infection by that organism (5, 6); manipulation of the microbiota (e.g., selective gut decontamination) can improve clinical outcomes (7, 8).
To date, the small number of culture-independent studies of the ICU microbiome has focused largely on individual body sites (e.g., gut or trachea) without considering concomitant colonization patterns at other sites (9–11). The goal of this study was to characterize spatial and temporal variations in the bacterial populations observed at five body sites (mouth, skin, gut, urinary tract, and trachea) in critically ill adult surgical patients. We compared these data to publicly available data from two large studies of the human microbiome. First, we compared results to skin, oral, and fecal samples in the American Gut Project (AG), a crowd-sourced initiative in which citizen scientists submit self-collected samples (12). Second, we compared results to analogous samples from the Human Microbiome Project (HMP) sponsored by the National Institutes of Health (13). Taken together, our analysis provides a granular view of dysbiosis in critically ill patients and provides new details about how the microbiota of critically ill patients differs from that of healthy volunteers.
PATIENTS AND METHODS
Thirty-two patients admitted to the intensive care unit under the trauma/acute care surgery services at the University of Pittsburgh Medical Center Presbyterian Hospital from July to October 2015 were enrolled in accordance with the UPMC Quality Improvement Committee.
Sample collection occurred within 48 h of ICU admission and every 3 to 4 days thereafter. GI samples were either stool samples or rectal swabs. Oral swabs were taken from the tongue dorsum. Skin samples were taken from the antecubital fossa of one arm and chest combined. Tracheal aspirates were collected from patients requiring mechanical ventilation. Urine samples were taken from catheters. Specific protocol for each body site collection is included in Supplemental Methods, http://links.lww.com/SHK/A428.
Microbial DNA was extracted from samples using the UltraClean DNA Isolation kit (urine samples) or the PowerSoil DNA Isolation kit (all other body sites) (MO BIO Laboratories, Inc, Carlsbad, Calif). Sample processing details are provided in Supplemental Methods, http://links.lww.com/SHK/A428.
16S amplicon PCR and sequencing
Bacterial 16S rRNA gene sequences were amplified and sequenced with the Illumina MiSeq. 16S amplicons were produced utilizing fusion primers adapted for the Illumina MiSeq. Amplicons targeting the V4 region were generated using 515F and 806R primers. Samples were sequenced with blank extraction controls and no-template added polymerase chain reaction (PCR) controls at the University of Illinois Roy J. Carver Biotechnology Center. Further details are provided in Supplemental Methods, http://links.lww.com/SHK/A428.
Forward and reverse reads were merged into a single amplicon sequence using PEAR. QIIME, UPARSE, and UCLUST were used for demultiplexing, quality filtering, and taxonomic assignments (14–16). Taxonomic assignments were made at a 97% threshold. Specific details are provided in Supplemental Methods, http://links.lww.com/SHK/A428.
HMP and AG dataset
We compared 16S rRNA sequencing results to publicly available datasets from non-critically ill adults, HMP and AG. The HMP included 227 healthy adult volunteers who were screened and determined to be free of major disease, but detailed comparisons to HMP sequencing data are limited because the sequencing was performed with different primer sets and a different sequencing platform. Direct comparisons to AG 16S rRNA sequencing data are possible because the AG uses the same primer sets and sequencing platform used in this study. The AG dataset was filtered to include only samples from healthy adult volunteers meeting specific health criteria (Table S1, http://links.lww.com/SHK/A425). After filtering, 977 AG samples were included in our analysis.
Sequence and statistical analysis
Sequence data was analyzed using Quantitative Insights into Microbial Ecology (QIIME) with default parameters and normalized numbers of sequencing reads. Alpha and beta diversity were calculated using QIIME. Significant differences were assessed using ANOVA and post-hoc Tukey HSD test where appropriate to account for multiple hypothesis testing. Variations in beta diversity were assessed with the PERMANOVA and PERMDISP algorithms in QIIME. LEfSe was used to determine differentially abundant taxa in experimental conditions (17). Only taxa with an average relative abundance >1% in at least 1 group of samples were considered for this analysis. Proportion of samples with dominant organisms and loss of site specificity was compared across groups using a Z test. A P value of 0.05 was used to determine significance in all statistical tests.
Microbial diversity among critically ill patients is decreased compared to published microbiome datasets
A total of 382 samples from five body sites were collected from 32 critically ill patients. Demographic data for the study subjects are summarized in Table S2 (http://links.lww.com/SHK/A426). After removing low-quality samples, 79 GI samples, 82 oral cavity samples, 79 skin samples, 54 tracheal aspirates, and 50 urine samples remained for analysis.
Alpha diversity is an ecological measure of how many taxonomic groups are within each sample and the evenness in their distribution. There are several indices that can be used to quantify alpha diversity; they all assign a numerical value to the degree of diversity of a single microbial community or sample. Relative to AG participants, we found that alpha diversity in ICU GI and skin samples was stable but reduced in all three time periods considered (0–5 ICU days, 6–10 ICU days, >10 ICU days) (Fig. 1). In the oral cavity samples, alpha diversity initially was similar to healthy controls but diminished over time (P <0.05). Alpha diversity of tracheal aspirates and urine samples was low and did not differ significantly with time. There was no publicly available dataset with which to compare tracheal and urine samples. The impact of gender and whether the patient was an acute care surgery or trauma patient on alpha diversity was evaluated and found to have no significant impact (Figure S1, http://links.lww.com/SHK/A419, and Figure S2, http://links.lww.com/SHK/A420).
Microbial composition of ICU patients across multiple body sites differs from unhospitalized adults
Beta diversity is a measure of dissimilarity between multiple ecological communities. Several indices such as the abundance Jaccard distance metric exist to quantify beta diversity; they all assign a numerical value to the degree of dissimilarity between the composition of two microbial populations. Using such metrics, a distance matrix can be constructed and used to compare all samples. Applying the abundance Jaccard distance metric, we found that the microbial composition within ICU skin, colon, and oral samples differed significantly from samples taken from the same body sites in healthy controls from the AG (Fig. 2). We then divided patients based on gender and whether they were an acute care surgery or trauma patient to determine if these variables impacted microbial composition. We found that there was no significant effect for either of these variables (Figure S3, http://links.lww.com/SHK/A421 and Figure S4, http://links.lww.com/SHK/A422).
LEfSe was used to determine specifically which taxa distinguished ICU and healthy individuals in the gut, skin, and oral cavity (17). Results are shown in Figure 3. At all three body sites, we observed a depletion of important commensal organisms in the ICU with a corresponding proliferation of pathogenic taxa. Relative to healthy controls, GI samples in the ICU were enriched with Enterococcus and depleted of Faecalibacterium, Blautia, and Ruminococcus. Oral samples were enriched with Mycoplasma and Staphylococcus but depleted of the commensals Veillonella and Neisseria(18–20). Skin swabs were unusually enriched with Mycoplasma, Stenotrophomonas, and Prevotella but were depleted of Propionibacterium and Corynebacterium.
Dominant species are commonly pathogenic at multiple body sites in critically Ill patients
Recent studies have established that dominance of a pathogen predisposes patients to developing a clinical infection caused by that same organism (21). We analyzed ICU GI, oral, and skin samples to identify the presence of dominant pathogens using a definition of dominance as 50% relative abundance within a mixed bacterial community. As shown in Figure 4A, the proportion of ICU samples containing a dominant organism in the gut is 17.7% versus 19.4% in the AG. In nearly all cases, the dominant taxon seen in the AG fecal samples was Bacteroides, a ubiquitous commensal. By contrast, the dominant taxa in ICU samples included pathogens such as Enterococcus and Campylobacter.
The proportion of skin samples with dominant organisms was 39.5% and 26.8% among ICU and AG subjects, respectively. Nearly all dominant taxa seen in the AG skin samples are typical colonizers from the genera Corynebacterium and Staphylococcus. Among ICU skin samples, Staphylococcus was found to be dominant more frequently than in AG samples (36.7% vs. 13.3%) and these ICU samples also harbored other dominant pathogens such as Stenotrophomonas and Mycoplasma. The proportion of ICU and AG oral samples containing a dominant taxon was 28.0% and 17.9%, respectively. A wide range of dominant oral taxa were observed in ICU patients, including Acinetobacter and Mycoplasma. Proportion of tracheal and urine samples from ICU patients with dominant taxa in 61% and 76% of samples, respectively (Figure S5, http://links.lww.com/SHK/A423). Dominant pathogens within tracheal aspirates were similar to those seen in oral samples, and dominant pathogens within urine samples included many taxa seen in both GI samples (e.g., Enterococcus) and skin samples (e.g., Corynebacterium).
Loss of site specificity among microbial communities in the ICU
Prior studies in healthy volunteers have demonstrated that microbial communities generally differ significantly across disparate body sites (22–26). We observed significant overlap among microbial communities within ICU skin, tongue, and tracheal samples (PERMDISP, P >0.05) but not GI or urine samples (Figure S6, http://links.lww.com/SHK/A424). To further explore this “loss of site specificity,” we examined how frequently a single taxon was observed simultaneously within the skin, tongue, and GI samples from ICU and HMP subjects. We used HMP data for this comparison because samples from multiple body sites were available from individual subjects, in contrast to the AG. As shown in Figure 4B, over 20% of ICU sample sets of skin, GI, and tongue at a single time point were marked by the presence of a genus found in all three sample types at a minimum relative abundance of 2%. Even with a stringent cutoff of 10% relative abundance, 5.2% of sample sets harbored an organism at all three body sites compared with 1.3% in the HMP. Moreover, the responsible organisms were typically organisms with known pathogenic potential including Acinetobacter, Staphylococcus, and Streptococcus compared with the HMP where the only genus to span three body sites was Prevotella. These data suggest that critically ill patients are more likely than healthy volunteers to display loss of site specificity marked by the abundance of potentially pathogenic taxa at three body sites simultaneously.
Tracheal microbiome data and occasionally precedes culture-proven infections
All clinically relevant infections that developed during the study period are detailed in Table S3 (http://links.lww.com/SHK/A427). The most frequently observed nosocomial infection was culture-proven pneumonia diagnosed by bronchoalveolar lavage (BAL) (observed in 8 of 32 subjects). BAL results for these patients are shown in Figure 5 along with corresponding microbiota profiles obtained from tracheal aspirates. As shown, 9 of 14 positive BAL cultures were preceded by tracheal aspirates containing the offending pathogen at a minimum relative abundance of 10%.
In this observational analysis, we examined how microbial communities vary across time and space in critically ill patients, and compared our results to data from control samples. Our results indicate that ICU study subjects demonstrate decreased microbial diversity within GI, skin, and oral samples relative to participants in AG. Loss of diversity is an important finding that likely predisposes critically ill patients to nosocomial infection and colonization by pathogens (11, 21). The observed decreased diversity undoubtedly reflects a combination of selection pressures including systemic antibiotic therapy, medical therapy, and topical antiseptic agents. Future efforts to monitor and modify the ICU microbiome may benefit from efforts to increase diversity.
The membership of bacterial communities within GI, skin, and oral samples from critically ill patients differed sharply from the membership seen in AG participants. At each of these body sites, we observed loss of common commensal taxa and gain of taxa with pathogenic potential. The “missing” taxa in the ICU included Faecalibacterium and Ruminococcus in the gut, Neisseria, Veilonella, and Rothia on tongue swabs, and Corynebacterium on skin swabs (18, 19, 27). The significance of the relative absence of these taxa is yet to be determined. In their place, we observed numerous abundant pathogens in ICU samples, including Enterococcus in the gut, Mycoplasma on the tongue, and Acinetobacter on skin swabs. Similar profiles were observed in tracheal aspirates and urine samples.
Loss of site specificity, which has not previously been described, also appears to be a key feature of the ICU microbiome. We observed the same phenomenon in a survey of critically ill children (unpublished data). The data reported here indicate that critically ill individuals commonly harbor taxa that are present at significant abundance at multiple body sites simultaneously. Under normal circumstances, it is thought that specific physiologic features (e.g., pH or presence of mucin along the gut epithelium) select for specific microbial communities (28). Therefore, we speculate that the altered physiology of ICU patients no longer favors site-specific communities. The lack of site specificity likely also reflects the presence of pathogens that possess the ability to colonize and reproduce in diverse habitats (e.g., facultative anaerobes that thrive in anaerobic and aerobic environments).
A second key feature of the ICU microbiome is the increased prevalence of dominant pathogens. To the best of our knowledge, no prior studies have systematically assessed the presence of dominant taxa in reference datasets. We found that samples from healthy volunteers in the AG commonly contained dominant taxa, but invariably these were ubiquitous commensals such as Bacteroides in the gut. By contrast, ICU samples were far more likely to contain a dominant pathogen such as Enterococcus or Staphylococcus. These patterns likely are clinically relevant as it has been shown that the presence of dominant pathogen is a risk factor for nosocomial infection (11, 21, 29).
The findings presented here provide a high-resolution analysis of dysbiosis in the ICU but they are not unexpected. Future studies of the microbiome in the ICU will need to address the potential translational value of such knowledge, and one distinct possibility is that real-time monitoring of the microbiota can offer opportunities to learn about and reduce the incidence of nosocomial infection. For this reason, we were intrigued to find that seven of eight patients with culture-proven pneumonia demonstrated some level of correlation between culture results and the dominant or significantly abundant genera seen in the microbiome surveys. Prospective trials are required to define whether specific patterns of dysbiosis represent actionable findings and whether rational therapeutic interventions (e.g., administration of probiotics or fecal microbiota transplantation) based upon these patterns can be used to improve clinical outcomes.
This study was limited by its sample size, the heterogeneity of the study population, and the exposure of study subjects to disparate antibiotic regimens. Another limitation was the comparison of microbiome sequencing data from ICU patients to publicly available data from two large microbiome initiatives, AG and HMP. These initiatives both feature large populations of unhospitalized adult study subjects, and therefore we elected not to enroll a smaller, in-house control group of non-critically ill adults. However, the use of microbiome reference sets has its limitations (16, 26). The AG data were generated with the same PCR primers and sequencing platform as our data thereby limiting technical bias. However, the quality of the self-reported health information of the volunteers cannot be verified. By contrast, HMP study subjects were carefully assessed to verify health status but gene sequencing was performed with a different set of PCR primers and sequencing platform. A related consideration is that sample collection varies with each study. For example, oral samples in this study and the HMP study were tongue swabs, but AG oral samples were saliva samples. Similarly, the site of skin swab collection varied somewhat with each study (30).
Although it has long been recognized that ICU patients are colonized with pathogenic organisms, few studies have used high-throughput sequencing to characterize colonization patterns in detail and to compare them to patterns seen in healthy volunteers. Here, we have studied critically ill adult surgical patients and identified several key features of the ICU microbiome. In addition to the risk of infection, there are many other aspects of the disturbed physiology of critically ill patients (nutrition and metabolism) that likely are related in some way to dysbiosis. Recently, a nascent body of translational research in other fields of medicine has proven that the status of the microbiome can sharply impact clinical outcomes at the level of the individual patient (31, 32). Future studies will be required to determine if real-time monitoring and modification of the microbiota can also be leveraged to improve short and long-term outcomes in critically ill patients.
The authors acknowledge the staff and patients of the University of Pittsburgh Medical Center for their support of this research.
1. Krezalek MA, DeFazio J, Zaborina O, Zaborin A, Alverdy JC. The shift of an intestinal “microbiome
” to a “pathobiome” governs the course and outcome of sepsis following surgical injury. Shock
2016; 45 5:475–482.
2. Berthelot P, Grattard F, Mahul P, Pain P, Jospé R, Venet C, Carricajo A, Aubert G, Ros A, Dumont A, et al. Prospective study of nosocomial colonization and infection
due to Pseudomonas aeruginosa in mechanically ventilated patients. Intensive Care Med
2001; 27 3:503–512.
3. Ziakas PD, Anagnostou T, Mylonakis E. The prevalence and significance of methicillin-resistant Staphylococcus aureus colonization at admission in the general ICU setting: a meta-analysis of published studies. Crit Care Med
2014; 42 2:433–444.
4. McKinley L, Becerra B, Moriarty H, Short TH, Hagle M, Reymann A, Valentine S, Duster M, Warrack S, Safdar N. Vancomycin-resistant Enterococcus co-colonization rates with methicillin-resistant Staphylococcus aureus and Clostridium difficile in critically ill veterans. Am J Infect Control
2016; 44 9:1047–1049.
5. Shimizu K, Ogura H, Goto M, Asahara T, Nomoto K, Morotomi M, Yoshiya K, Matsushima A, Sumi Y, Kuwagata Y, et al. Altered gut flora and environment in patients with severe SIRS. J Trauma Inj Infect Crit Care
2006; 60 1:126–133.
6. Shimizu K, Ogura H, Hamasaki T, Goto M, Tasaki O, Asahara T, Nomoto K, Morotomi M, Matsushima A, Kuwagata Y, et al. Altered gut flora are associated with septic complications and death in critically ill patients with systemic inflammatory response syndrome. Dig Dis Sci
2011; 56 4:1171–1177.
7. Silvestri L, van Saene HKF. Selective decontamination of the digestive tract: an update of the evidence. HSR Proc Intensive Care Cardiovasc Anesth
2012; 4 1:21–29.
8. Hurley JC. Is selective decontamination of the digestive tract safe? Clin Infect Dis
9. Johanson WG, Pierce AK, Sanford JP. Changing pharyngeal bacterial flora of hospitalized patients. Emergence of gram-negative bacilli. N Engl J Med
1969; 281 21:1137–1140.
10. Zaborin A, Smith D, Garfield K, Quensen J, Shakhsheer B, Kade M, Tirrell M, Tiedje J, Gilbert JA, Zaborina O, et al. Membership and behavior of ultra-low-diversity pathogen communities present in the gut of humans during prolonged critical illness. mBio
2014; 5 5: e01361–14–e01361–14.
11. Kelly BJ, Imai I, Bittinger K, Laughlin A, Fuchs BD, Bushman FD, Collman RG. Composition and dynamics of the respiratory tract microbiome
in intubated patients. Microbiome
2016; 4 1:7.
12. Debelius JW, Vázquez-Baeza Y, McDonald D, Xu Z, Wolfe E, Knight R. Turning participatory microbiome
research into usable data: lessons from the American Gut Project. J Microbiol Biol Educ
2016; 17 1:46–50.
13. Turnbaugh PJ, Ley RE, Hamady M, Fraser-Liggett CM, Knight R, Gordon JI. The human microbiome
2007; 449 7164:804–810.
14. Zhang J, Kobert K, Flouri T, Stamatakis A. PEAR: a fast and accurate Illumina Paired-End reAd mergeR. Bioinforma Oxf Engl
2014; 30 5:614–620.
15. Edgar RC. UPARSE: highly accurate OTU sequences from microbial amplicon reads. Nat Methods
2013; 10 10:996–998.
16. McDonald D, Birmingham A, Knight R. Context and the human microbiome
2015; 3 1:52.
17. Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, Huttenhower C. Metagenomic biomarker discovery and explanation. Genome Biol
2011; 12 6:R60.
18. Zaura E, Keijser BJ, Huse SM, Crielaard W. Defining the healthy “core microbiome
” of oral microbial communities. BMC Microbiol
2009; 9 1:259.
19. Zaura E, Mira A. The Oral Microbiome
in an Ecological Perspective. Lausanne, Switzerland: Frontiers Media SA; 2015.
20. Colombo APV, Boches SK, Cotton SL, Goodson JM, Kent R, Haffajee AD, Socransky SS, Hasturk H, Van Dyke TE, Dewhirst F, et al. Comparisons of subgingival microbial profiles of refractory periodontitis, severe periodontitis, and periodontal health using the human oral microbe identification microarray. J Periodontol
2009; 80 9:1421–1432.
21. Taur Y, Xavier JB, Lipuma L, Ubeda C, Goldberg J, Gobourne A, Lee YJ, Dubin KA, Socci ND, Viale A, et al. Intestinal domination and the risk of bacteremia in patients undergoing allogeneic hematopoietic stem cell transplantation. Clin Infect Dis
2012; 55 7:905–914.
22. Costello EK, Lauber CL, Hamady M, Fierer N, Gordon JI, Knight R. Bacterial community variation in human body habitats across space and time. Science
2009; 326 5960:1694–1697.
23. Caporaso JG, Lauber CL, Costello EK, Berg-Lyons D, Gonzalez A, Stombaugh J, Knights D, Gajer P, Ravel J, Fierer N, et al. Moving pictures of the human microbiome
. Genome Biol
2011; 12 5:R50.
24. Huttenhower C, Gevers D, Knight R, Abubucker S, Badger JH, Chinwalla AT, Creasy HH, Earl AM, FitzGerald MG, Fulton RS, et al. Structure, function and diversity of the healthy human microbiome
2012; 486 7402:207–214.
25. Faust K, Sathirapongsasuti JF, Izard J, Segata N, Gevers D, Raes J, Huttenhower C. Microbial co-occurrence relationships in the human microbiome
. PLoS Comput Biol
2012; 8 7:e1002606.
26. Lozupone CA, Stombaugh J, Gonzalez A, Ackermann G, Wendel D, Vazquez-Baeza Y, Jansson JK, Gordon JI, Knight R. Meta-analyses of studies of the human microbiota
. Genome Res
2013; 23 10:1704–1714.
27. Aas JA, Paster BJ, Stokes LN, Olsen I, Dewhirst FE. Defining the normal bacterial flora of the oral cavity. J Clin Microbiol
2005; 43 11:5721–5732.
28. Arumugam M, Raes J, Pelletier E, Le Paslier D, Yamada T, Mende DR, Fernandes GR, Tap J, Bruls T, Batto J-M, et al. Enterotypes of the human gut microbiome
2011; 473 7346:174–180.
29. Caporaso JG, Lauber CL, Walters WA, Berg-Lyons D, Huntley J, Fierer N, Owens SM, Betley J, Fraser L, Bauer M, et al. Ultra-high-throughput microbial community analysis on the Illumina HiSeq and MiSeq platforms. ISME J
2012; 6 8:1621–1624.
30. Grice EA, Segre JA. The skin microbiome
. Nat Rev Microbiol
2011; 9 4:244–253.
31. Vétizou M, Pitt JM, Daillère R, Lepage P, Waldschmitt N, Flament C, Rusakiewicz S, Routy B, Roberti MP, Duong CPM, et al. Anticancer immunotherapy by CTLA-4 blockade relies on the gut microbiota
2015; 350 6264:1079–1084.
32. Taur Y, Jenq RR, Perales M-A, Littmann ER, Morjaria S, Ling L, No D, Gobourne A, Viale A, Dahi PB, et al. The effects of intestinal tract bacterial diversity on mortality following allogeneic hematopoietic stem cell transplantation. Blood
2014; 124 7:1174–1182.