Gut colonization with vancomycin-resistant Enterococcus shapes the gut microbiome in the intensive care unit

Abstract

Introduction

Gut pathogen colonization with Vancomycin-resistant Enterococcus (VRE) is common in the intensive care unit (ICU) and is associated with worse clinical outcomes, yet the timing of VRE colonization and its collateral effects on the gut microbiome are incompletely understood.

Methods

Medical ICU patients admitted with sepsis and receiving broad-spectrum antibiotics were sampled via deep rectal swabs at ICU admission and on ICU Day 3, 7, 14, and 30. Rectal swabs were cultured for VRE on selective media and analyzed via 16S rRNA gene sequencing.

Results

Ninety patients were sampled (340 longitudinal swabs). VRE positivity rose from 20% at ICU admission to a peak of 33% by ICU Day 14 and then modestly declined to 31% by ICU Day 30. Paralleling this, alpha diversity fell while Enterococcus relative abundance rose through ICU Day 14 with both returning to baseline by ICU Day 30. The median relative abundance of Enterococcus was 38% (IQR 7.4 to 75%) for VRE positive samples compared to 0.01% (IQR 0 to 19%) for VRE negative samples (rank-sum p<0.01); 38 samples had ≥90% Enterococcus and 8 samples were 100% Enterococcus by sequencing. VRE was associated with lower alpha diversity (median Shannon index of 1.90 (IQR 0.89 to 2.66) if VRE positive versus 2.64 (IQR 1.58 to 3.22) if VRE negative, p<0.01).

Conclusion

VRE gut colonization peaked at ICU Day 14 followed by a modest decline and was associated with low alpha diversity. Improved understanding of dynamic changes in the gut microbiome may facilitate successful future ICU interventions.

INTRODUCTION

Gut pathogen colonization is common in the medical intensive care unit (ICU) and is associated with worse outcomes.^1–3 In an observational cohort of 301 medical ICU patients, the presence of enteric pathogens such as E. coli, Pseudomonal spp., Klebsiella spp., or Clostridioides difficile at the time of ICU admission was associated with subsequent culture-proven infections with the same organism during the following 30 days.¹ In this same cohort, gut colonization with the enteric pathogen vancomycin-resistant Enterococcus (VRE) had unique associations. Not only was culture for VRE associated with increased risk for subsequent VRE infections, but culture for VRE or 16S sequencing showing Enterococcus domination (≥30% 16S reads) was associated with increased risk for all-cause death or infection.

Among bacterial pathogens, VRE has a distinct combination of characteristics which facilitates domination of the gut microbiome during critical illness. VRE is intrinsically hardy and survives well on inhospitable ICU surfaces, rapidly acquires antibiotic resistance, produces bacteriocins to suppress non-VRE strains, and scavenges carbohydrates to gain an advantage when nutrients are limited

Studies in non-ICU patient cohorts provide further evidence of a causal link between gut colonization and subsequent systemic infection.^4–8 Focusing on VRE, Gouliouris et. al found that six of nine patients who developed VRE bacteremia were previously gut colonized with the infection-causing VRE clone.⁹ Environmental swabs often showed corresponding VRE clones, but patient gut colonization consistently preceded environmental detection of VRE.

These and other studies imply that VRE may be uniquely important as a marker of microbiome colonization resistance in the ICU or in similar settings, and raise hope that treatment of VRE colonization might reduce VRE or non-VRE infections.¹⁰ The recent success of microbiome-based therapies for recurrent C. difficile infection (CDI)^{11, 12} provides some proof-of-concept that microbiome-based strategies could be applied toward gut pathogen colonization to improve patient outcomes. However, if such strategies are to be successful, the dynamics of VRE colonization—and the overall dynamics of the gut microbiome in the medical ICU—will need to be better understood.

This study sought to describe how the gut microbiome changes during the 30 days following medical ICU admission, and how VRE gut colonization status may impact these changes. It explores the relationship between VRE and the rest of the gut microbiota to provide insight into the role of VRE and its clinical implications in the ICU. The study re-utilized samples and data from a previously described randomized controlled trial which tested prebiotic inulin for the purpose of preventing gut pathogen colonization and infection.

METHODS

Population

The study cohort was drawn from a single-center, randomized, double-blind, placebo-controlled trial (NCT03865706).¹³ The trial randomized patients 1:1:1 to placebo, inulin 16 g/day, or inulin 32 g/day to prevent gut pathogen colonization and infection. The primary outcome of the trial was null, and the trial data was re-utilized for the current study. In brief, adults were included in the trial if they were admitted to the ICU with sepsis²⁰ as the primary diagnosis and could be enrolled within 24 hours of ICU admission. All subjects were required to be receiving a broad-spectrum antibiotic, usually piperacillin-tazobactam or ceftriaxone. Exclusion criteria included testing positive for COVID-19 or limited treatment goals such as “do not resuscitate” or “no escalation of care” orders.

Samples

VRE culture and 16S sequencing were performed from protocolized rectal swabs. Rectal swabs were gathered with the participant in the left lateral decubitus position, with fecal soilage used to ensure adequate sampling. Two deep rectal swabs were collected by a study nurse on all participants at the time of ICU admission and prior to receipt of the first dose of the study intervention. Subsequent rectal swabs were performed on ICU Day 3, 7, 14, and 30 unless patients died or were discharged from the hospital. At each timepoint, one of the two rectal swabs was flash frozen at −80°C for 16S rRNA gene sequencing and the second swab was added to soy broth and cultured for VRE.

Vancomycin-resistant Enterococcus culture

Patient stool samples were inoculated on CHROMagar TME VRE and incubated overnight at 37°C. Positive pink to mauve colonies were sub-cultured onto blood agar plates to confirm growth. VRE colonies were further tested using colony PCR and for antibiotic susceptibility via the Kirby-Bauer disk diffusion method. Susceptibility was assessed using a 0.5 McFarland suspension and gram-positive-specific antibiotic panels following the CLSI (Clinical and Laboratory Standards Institute)¹⁴ disc diffusion guidelines. Only isolates that exhibited resistance to vancomycin according to CLSI guidelines were categorized as VRE-positive. These susceptibility results were cross-referenced with sequencing data to confirm microbial alignment across both culture and sequencing-based methods.

16S rRNA gene sequencing

Rectal swabs were thawed and DNA was extracted on a 96-well plate handling station (Qiagen), including positive and negative controls. From extracted DNA, the V3V4 region of the 16S rRNA gene was amplified using primers with Illumina adapters and the resulting libraries were barcoded and sequenced on an Illumina MiSeq platform at 600 cycles. Sequencing data was processed using DADA2 and R with all analyzed samples meeting a minimum requirement of 7,500 reads after quality filtering.¹⁵ Phylogenetic trees were constructed using the MAFFT and FastTree modules in QIIME2. Taxonomic classification was conducted with a native naïve RDP Bayesian classifier against the Silva version 138 database and to produce diversity indices and the relative abundance for Enterococcus and other taxa.¹⁶ SCRuB was used downstream to account for potential well-to-well contamination.¹⁷ Sequencing data are publicly available through the NCBI Sequencing Read Archive (SRA) (accession number PRJNA1128691).

Random forest model and feature selection

Random forest models were constructed to examine the association between microbial data and patient outcomes.¹⁸ Two models were developed: one including Enterococcus-related features and another excluding them. Taxonomic features with read counts exceeding 2,000 across all samples were incorporated into the analysis. Feature selection was based on the Mean Decrease Gini index, retaining variables with scores above 0.5 to enhance model interpretability. Data were divided into training (75%) and test (25%) sets using the createDataPartition function. The model was built using the training set, while the testing set, which consisted of unseen data, provided an unbiased evaluation of its performance. Models were constructed with an ensemble of 100 decision trees, and cross-validation was applied to the training set to fine-tune model parameters. Performance metrics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic (auROC) were calculated for both the training and testing sets.

Statistical Analysis

Demographic data were collected and summarized using descriptive statistics. Chi-squared tests were used to compare categorical measures and rank-sum tests were used to compare continuous measures. Longitudinal analyses were performed based on data from the patients who remained hospitalized and could be sampled (i.e., without imputing results for patients who died or were discharged). For sequencing data, alpha diversity indices including Shannon and Chao1 were calculated using the phyloseq package (version 1.36.0).¹⁹ Principal Coordinates Analysis (PCoA) was performed using the unweighted UniFrac distance matrix to visualize beta diversity, with statistical differences assessed via PERMANOVA. Differential abundance testing was conducted using ANCOM2,²⁰ which accounts for the compositional nature of microbiome data and includes adjustment for longitudinal samples. P-values were adjusted using the Benjamini-Hochberg method to control for false discovery rates. Data were analyzed in R version 4.1.0.

RESULTS

Population and initial VRE gut colonization

Ninety medical ICU patients with sepsis provided a total of 340 longitudinal samples for the analysis. Eighteen of 90 (20%) patients were gut colonized with VRE at ICU admission and 72/90 (80%) were not. Among all samples, 93/340 (27%) were positive for VRE. The most common ICU-acquired infections were caused by P. aeruginosa (N=13), K. pneumoniae (N=10), and E. coli (N=5). There were 3 patients with infections due to VRE during the study including 2 urinary infections and one VRE pneumonia. Of these 3 patients with VRE infections, 2 patients were colonized with VRE at all timepoints and the third patient was never VRE colonized.

Clinical characteristics at ICU admission and VRE gut colonization

Patients who were VRE positive at the time of ICU admission were older and were more likely to be Hispanic compared to patients who were VRE negative at the time of ICU admission (Table 1). VRE positive patients had increased comorbidities and were more likely to come from nursing homes, although these differences were not statistically significant. There was no difference in acute severity of illness at ICU admission comparing VRE positive versus negative patients.

Table 1.

Demographic and clinical characteristics at admission to the ICU, stratified by gut colonization with VRE.

Characteristic	VRE Colonized (N=18)	Not VRE Colonized (N=72)	p-valuea
Female sex (N, %)	5 (28%)	35 (49%)	0.11
Age in years (median, IQR)	70 (62–76)	60 (44–70)	0.04
Race (N, %)			0.37
White	9 (50%)	26 (36%)
Black	3 (17%)	16 (22%)
Hispanic	2 (11%)	21 (29%)
Other	4 (22%)	9 (13%)
Ethnicity (N, %)			0.03
Hispanic	3 (17%)	32 (44%)
Non-Hispanic	15 (83%)	40 (56%)
Home environment (N, %)			0.12
Independent	6 (33%)	34 (47%)
Home with help	4 (22%)	23 (32%)
Assisted living facility	8 (44%)	15 (21%)
Charlson comorbidity score (N, %)	5 (4–5)	3 (2–6)	0.11
ICU admission characteristics
Sepsis, based on Sepsis-3 criteria	30 (100%)	30 (100%)	---
Broad-spectrum antibiotics administered within +/− 24 hours of ICU admissionb	30 (100%)	30 (100%)	---
End-organ failure at ICU admission
Use of mechanical ventilationc	12 (67%)	60 (83%)	0.11
Use of vasopressorsd	13 (72%)	51 (72%)	0.97
Urine output < 500 mL / 24 hours	9 (50%)	29 (40%)	0.46
SOFA score (median, IQR)	9 (7–10)	8 (6.5–10)	0.46

^aChi-squared test for categorical measures and rank-sum for continuous measures.

^bReceipt of broad-spectrum antibiotics was an inclusion requirement and was operationalized as receipt of one or more of any of the following antibiotic classes within +/− 24 hours of ICU admission: β-lactam/β-lactamase inhibitor combination antibiotics, cephalosporins (gen 2 or greater), fluoroquinolones, lincosamides (clindamycin), metronidazole, and monobactams (e.g., meropenem).

^cMechanical ventilation: classified as use of mechanical ventilation including CPAP or BiPAP at ICU admission.

^dUse of vasopressors: classified as receipt of dopamine, dobutamine, epinephrine, norepinephrine, or vasopressin (any dose) at ICU admission.

Abbreviations: IQR: interquartile range; ICU: intensive care unit; SOFA: sequential organ failure assessment; CPAP: continuous positive airway pressure.

Acquisition of VRE

Overall, VRE positivity rose from admission, peaking on Day 14, and then modestly declining to Day 30. The rates of VRE gut colonization among those who remained hospitalized were 20%, 28%, 29%, 33%, and 32% on ICU days 0, 3, 7, 14 and 30 respectively. There was substantial loss of patients due to death/discharge after ICU Day 14 so VRE status was also imputed using a carry-forward approach; using this approach, the rates of VRE colonization were similar at 20%, 28%, 30%, 33%, and 33% respectively.

Changes in VRE status

We sought to understand potential noise in the VRE culture data by looking at patients who switched VRE status. In general, VRE status was stable over time; most patients were either VRE positive at ICU admission or acquired VRE soon after admission, and few patients acquired VRE late during the ICU stay (Fig. 1). Quantifying this, among the 18 patients who were VRE positive at ICU admission, 11 patients (61%) remained VRE positive based on their last available sample whereas only 19/72 patients (26%) who were VRE negative at ICU admission converted to VRE positive by the time of their last available sample (p<0.01).

Figure 1.

VRE gut colonization through the study, shown by ICU Day. Each row represents a pattern of VRE colonization during the 5 timepoints when patients were sampled. Outlined circles are VRE negative samples, filled circles are VRE positive samples, and gray circles without an outliner are timepoints when no sampled was gathered because the patient died or was discharged. Blue bars to the right indicate how many patients were present with each pattern. For example, in the first row there were 6 patients who were negative for VRE at all 5 study timepoints; in the second row there were 21 patients who were negative for VRE on ICU Days 0 to 14 and who did not provide a Day 30 sample.

Association between VRE in culture and Enterococcus in 16S sequencing

Among all 340 samples, the median relative abundance of Enterococcus was 4.7% (IQR 0 to 41%). There were 38 (11%) samples which had ≥90% relative abundance of Enterococcus and 8 (2.4%) samples from 5 unique patients which were 100% Enterococcus by sequencing. The median relative abundance of Enterococcus was 38% (IQR 7.4 to 75%) among samples that were VRE positive compared to 0.01% (IQR 0 to 19%) among samples that were VRE negative. When Enterococcus domination was defined as a relative abundance of ≥30% as per Taur et al., the rates of Enterococcus domination were 56% among samples that were VRE positive versus 19% among samples that were VRE negative (p<0.01).^{21, 22} Longitudinally, VRE positivity was associated with increased relative abundance of Enterococcus through ICU Day 30 (Fig. 2; rank-sum p<0.05 at all timepoints). Regardless of VRE status, Enterococcus relative abundance peaked at ICU Day 7 to 14.

Figure 2.

Relative abundance of Enterococcus at each study timepoint, organized by VRE culture status. The unadjusted rank-sum p-value is shown, comparing VRE positive samples versus negative samples at each timepoint.

Impact of VRE gut colonization on the microbiota

We examined the relationship between VRE status based on culture and fecal microbiota composition based on 16S sequencing. Among all samples, VRE positivity was associated with lower fecal microbial diversity with a median Shannon index of 1.90 (IQR 0.89 to 2.66) among samples that were VRE positive compared to a median Shannon index of 2.64 (IQR 1.58 to 3.22) among samples that were VRE negative (rank-sum p<0.01, Table 2 and see Fig. S1). On PCoA, the samples separated based on VRE status (PERMANOVA p<0.01, Fig. S2). Longitudinally, patients who were VRE positive at ICU admission had lower alpha diversity throughout ICU hospitalization, although both VRE colonized and VRE non-colonized patients showed some recovery of diversity by ICU Day 30 (Fig. 3A & 3B and Fig. S3). Additionally, VRE gut colonization was associated with decreased putatively beneficial short chain fatty acid-producing bacteria (median relative abundance 6.9% for VRE positive vs. 10.8% for VRE negative samples, p=0.04).

Table 2.

Microbiome effects through 30 days, stratified by VRE status. Diversity (Shannon index) is first shown based on the corresponding sample VRE status and then shown based on the patient’s ICU admission sample VRE status.

Measure	VRE Positive Sample (N=93)	VRE Negative Sample (N=247)	p-value*
Shannon index
ICU admission (N=90)	1.93 (1.36 to 2.92)	2.96 (2.02 to 3.48)	0.03
ICU Day 3 (N=90)	2.37 (1.71 to 3.00)	2.65 (1.88 to 3.23)	0.39
ICU Day 7 (N=78)	1.41 (0.90 to 2.03)	2.40 (1.76 to 3.07)	<0.01
ICU Day 14 (N=60)	1.47 (0.71 to 2.15)	2.18 (1.63 to 3.33)	<0.01
ICU Day 30 (N=22)	2.03 (0.94 to 2.85)	2.76 (2.35 to 3.42)	0.10
	Patient VRE Positive at ICU Admission (N=18)	Patient VRE Negative at ICU Admission (N=72)	p-value*
Shannon index
ICU admission (N=90)	1.93 (1.36 to 2.92)	2.96 (2.02 to 3.48)	0.03
ICU Day 3 (N=90)	2.21 (0.90 to 2.65)	2.68 (1.90 to 3.15)	0.10
ICU Day 7 (N=78)	1.73 (0.83 to 2.23)	2.28 (1.28 to 2.94)	0.04
ICU Day 14 (N=60)	1.25 (0.56 to 2.65)	2.07 (1.47 to 2.99)	0.14
ICU Day 30 (N=22)	2.74 (1.44 to 3.13)	2.71 (1.93 to 3.32)	0.80
Within-individual change in Shannon index from admission to last available sample	+0.29 (−0.34 to +0.83)	+0.40 (−0.23 to +0.99)	0.66

^*Rank-sum p-value.

Figure 3.

Fecal microbial diversity through the study. (A) Diversity for all 90 patients, stratified based on VRE status. Unadjusted rank-sum p-values are shown, comparing Shannon index for VRE negative versus positive samples at each timepoint. (B) Diversity (right-hand axis) and Enterococcus relative abundance (left-hand axis) for one individual, showing when that individual received vancomycin and beta-lactam/beta-lactamase inhibitors.

Additive impact of vancomycin on the gut microbiota

We hypothesized that there would be an additive interaction between VRE status and receipt of intravenous vancomycin—i.e., that Enterococcus domination and loss of fecal microbial diversity would be more prominent among those who were VRE positive and received vancomycin compared to those who were VRE positive but did not receive vancomycin. All samples were classified by VRE status and based on whether the patient was receiving vancomycin at the time when the sample was collected. Receipt of vancomycin significantly modified the effects of VRE status. Specifically, samples that were positive for VRE and collected while the patient was receiving vancomycin were the highest in Enterococcus relative abundance (Fig. 4A) and the lowest in alpha diversity (Fig. 4B); the next highest in Enterococcus and next lowest in diversity were samples positive for VRE that were not collected while the patient was receiving vancomycin. Comparatively, beta-lactams with beta-lactamase inhibitors (BL-BLI), which were the most common broad-spectrum antibiotic received, had less of an additive effect with VRE status (Fig. S4).

Figure 4.

The impact of VRE status and receipt of vancomycin on the gut microbiota. (A) Enterococcus relative abundance, stratified based on VRE status and receipt of vancomycin. (B) Shannon diversity, stratified based on VRE status and receipt of vancomycin. Unadjusted rank-sum p-values are shown.

Differentially abundant taxa based on VRE gut colonization

Next, a heatmap was generated to compare microbiota features from 16S sequencing with VRE gut colonization status based on culture at each timepoint (Fig. 5). Across multiple timepoints, Enterococcus was associated with VRE positivity, as expected. Additionally, Bacterioides thetaiotamicron and Bacteroides vulgatus were associated with VRE positivity at multiple timepoints whereas Pseudomonal spp. were associated with VRE negativity at multiple timepoints.

Figure 5.

Differentially abundant taxa based on VRE gut colonization, organized by timepoint and depicted as a heatmap. Red shows associations with VRE positivity and blue shows associations with VRE negativity. * indicates false discovery-adjusted p<0.01 and ** indicates false discovery-adjusted p<0.001.

Random forest model for VRE status

Last, to better understand which specific taxa were associated with VRE status, a random forest model was then constructed to identify the taxa which were most associated with VRE positivity, after controlling for study timepoint and for the study intervention. In this model, 16S sequencing results were predictive of VRE status with an auROC of 80% (Fig. S5A) and a model accuracy of 81% (Fig. S5B). Aside from Enterococcus, the microbiota features most associated with VRE positivity were Bacteroides caccae, Lactobacillus gasseri, and Clostridium incertae sedis (Fig. S5C). To test for novel features associated with VRE positivity and to add to the robustness of the results, the model was then re-run after excluding Enterococcus. In this random forest model excluding Enterococcus, the model auROC was 69% (Fig. S6A) with a model accuracy of 71% (Fig. S6B). The features most associated with VRE status were Erysipelatoclostridium ramosum, Bacteroides vulgatus, Lactobacillus gasseri, Bacteroides caccae, and Bacteroides betaiotathetamicron (Fig. S6C).

DISCUSSION

This reanalysis of longitudinal trial data²³ from 90 medical ICU patients (340 samples) describes how colonization with VRE shapes the dynamics of the gut microbiome in the ICU. Compared to historical cohorts comprised of healthy volunteers, the initial gut microbiome on the day of ICU admission was extremely low in alpha diversity and enriched in Enterococcus.^{24, 25} This was true even when VRE was absent, but was more dramatically so when VRE was present. Among the 93 samples positive for VRE, the median relative abundance of Enterococcus was 38% or over 3,000-fold enriched in Enterococcus compared to VRE negative samples. Taur et al. used the concept of “domination” to describe gut microbiomes which are subsumed by enteric pathogens, operationalizing domination as ≥30% of 16S sequences assigned to a single taxon.^{21, 22} Using this framework, 56% of the VRE colonized samples in this study were dominated by Enterococcus including 38 samples (11%) which were ≥90% Enterococcus. To a remarkable extent, the results of 16S sequencing are largely determined when a VRE rectal swab from a medical ICU patient is positive.

This longitudinal data yields new understanding of the dynamic changes within the gut microbiome among septic patients in the medical ICU.^26–28 Through ICU Day 14 there was a slow but steady decline in alpha diversity with a consequent rise in Enterococcus relative abundance. After ICU Day 14, there was recovery of the gut microbiome with rising diversity and less Enterococcus such that by ICU Day 30 both VRE positive and VRE negative patients exceeded their ICU baseline in terms of alpha diversity. The VRE positive patients made greater gains in diversity compared to the VRE negative patients, although they never quite caught up to their VRE negative counterparts.

These essentially descriptive results have important implications for future ICU-based interventions such as pre- or probiotics which seek to restore the gut microbiome to prevent secondary infections. One interpretation of this data is that it will be difficult for interventions to slow the decline in diversity during initial ICU hospitalization. Certainly, this was true for the trial intervention (inulin), which was delivered from ICU Days 0 to 7 and did not impact diversity.²³ Instead, future interventions may be more successful if they are delayed until later during ICU hospitalization—for example, if they seek to augment the gut microbiome when it is on the upswing starting on ICU Day 14. This timing may correlate with stoppage of broad-spectrum antibiotics for most patients, and it may be difficult or impossible to improve gut microbiome before broad-spectrum antibiotics are stopped.²⁹ This contrasts with current prebiotic strategies for recurrent CDI, where prebiotics are delivered once or for the first few days after cessation of anti-C. difficile antibiotics.^{11, 12}

Prior studies have highlighted the importance of VRE in the medical ICU where it predicts death or infection over and above standard clinical scores.^{1, 30} This study reinforces the conceptualization of VRE as a non-causal marker for a gut microbiome with low alpha diversity and low pathogen colonization resistance. Future studies should seek to determine which kinds of adverse clinical outcomes are best predicted by VRE status, and whether VRE should be analyzed by culture alone or paired with sequencing data. VRE positivity, with or without other biomarkers, may be useful as an inclusion criterion for future ICU microbiome trials.

A surprising ancillary finding in this data was that taxa which are typically regarded as protective—such as Blautia, Ruminococcus, or Bacteroides caccae—were associated with VRE positivity at multiple study timepoints while taxa which include common pathogens—e.g., Pseudomonal spp.—were associated with VRE negativity.^{31, 32} Ironically, the absence of VRE may open up an environmental niche for other pathogens including Pseudomonas. The ICU microbiota may frequently fall into “enterotypes” of patients who have low fecal microbial diversity and VRE domination, low fecal microbial diversity and domination with other common ICU pathogens like Pseudomonas, or relatively high fecal microbial diversity. This speaks to the complexity of the gut microbiome in the medical ICU.

There are strengths to this study. The data was derived from a gold standard, randomized clinical trial with rigorous sample handling and strictly protocolized sample collection at up to 5 timepoints. Prior studies have performed serial rectal swabs to determine the timing of VRE gut colonization in the ICU^{33, 34} but few have performed sufficiently dense samples to allow curve-fitting for gut pathogen colonization in the ICU and none have characterized VRE by both culture and sequencing.³⁵ The study also has limitations. Patients dropped out of the study due to death or discharge, with substantial loss to follow-up after 14 days. It was single-center and our rates of VRE (and our high Enterococcus relative abundance) exceed rates at institutions or in settings with less overall antibiotic use and a lower prevalence of multidrug-resistant organisms.³⁶ Trial criteria required that all patients be exposed to broad-spectrum antibiotics, which may inflate the abundance of Enterococcus and VRE. Sixty of 90 patients in the study received an experimental intervention, although this intervention did not appear to impact the gut microbiome. Last, while all patients were admitted for sepsis, while all patients were admitted for sepsis, there was considerable heterogeneity between them in terms of their origins (e.g., nursing home residency) and medical history, factors which may impact the microbiome. This heterogeneity may move some results towards the null, although it may also improve generalizability to other medical ICUs.³⁷

In sum, in this longitudinal cohort of 90 medical ICU patients, we found dense VRE colonization and a startling degree of Enterococcus relative abundance. VRE status partitioned microbiome diversity and was associated with specific taxa apart from just Enterococceae. As far as the dynamics of the microbiome, there were declines in diversity through ICU Day 14 and then a gradual rise back to baseline by ICU Day 30. These results may help to guide the timing of future microbiome-based interventions in the medical ICU.

Data availability

Sequencing data are available through the NCBI Sequencing Read Archive (SRA) (accession number PRJNA1128691). Additional metadata can be supplied on request.