Introduction

The human microbiota harbors a vast diversity of microorganisms (mainly bacteria), with the highest concentrations being located in the oropharyngeal and gut microbiota [1, 2]. A balanced composition of the microbiota is assumed to be associated with health, while an altered composition (referred to as dysbiosis) is associated with a broad range of intestinal and extra intestinal conditions [3]. The gut microbiota composition is dominated by strictly anaerobic bacteria, while opportunistic pathogens such as Enterobacterales and enterococci are subdominant [4]. Nonetheless, antibiotic exposure can alter the composition of the microbiota in promoting resistant, opportunistic pathogens which may be subsequently involved in infections [5,6,7]. More recently, the presence of Enterobacterales in rectal samples or the dominance of Enterococcus in patients at admission to the intensive care unit has been associated with a higher risk of death [8, 9]. Recently, the composition of the microbiota was associated with mortality at D28 in patients hospitalized in intensive care unit (ICU), with a lower diversity being associated with death [10]. Beyond colonization, we showed that the oropharyngeal and rectal concentrations of extended-spectrum beta-lactamase (ESBL)—producing Enterobacterales—could be a powerful tool to estimate the risk of subsequent ventilator-associated pneumonia (VAP) caused by ESBL-producing Enterobacterales in mechanically ventilated patients [11]. Altogether, the oropharyngeal microbiota and the gut microbiota are potential sources of knowledge from which the prognosis of critical care patients can be assessed.

More specifically, ICU patients suffering from severe presentations of COVID-19 frequently experience a high rate of nosocomial infections such as bloodstream infection (BSI) and VAP, which extend the course of ICU stay and worsen the prognosis [12, 13]. While several host risk factors for severe COVID-19 have been identified [14], factors related to the microbiota have been little studied. Nonetheless, the SARS-CoV-2 can affect the gut microbiota and induce dysbiosis. Early shotgun metagenomic sequencing analyses of fecal samples from a dozen of COVID patients showed enrichment of opportunistic pathogens and depletion of beneficial commensals [15]. Also, the composition of the intestinal microbiota has been associated with the severity of COVID [16, 17].

Our objective was to investigate the association between oropharyngeal and intestinal concentrations of opportunistic pathogens, the overall composition of the gut microbiota, and the prognosis of patients admitted to intensive care with severe forms of SARS-CoV-2 infection. Our hypothesis is that the presence of these opportunistic pathogens is a possible marker of altered gut microbiota, and that it is associated with the outcome of patients hospitalized with severe forms of COVID-19.

Methods

Ethics

Our ICU is part of the French COVID consortium (registered in clinicaltrials.gov NCT04262921). The French COVID scientific committee approved the NOSOCOVID ancillary study and the additional oropharyngeal and rectal samples. The study was conducted with the understanding and consent of each participant or surrogate. The French COVID received ethical clearance on February 5, 2020 by the CPP-Ile-de-France VI (ID RCB: 2020-A00256-33). The OutcomeRea™ database provided anonymous extractions of daily data collected during the ICU stays in accordance with the French law (“Commission Nationale de l’Informatique et des Libertés” #999,262).

Population

From March to September 2020 (first wave of COVID-19), we conducted a single-center observational prospective cohort study in critically ill patients which included consecutive patients ≥ 18 years hospitalized in ICU of Bichat university hospital (Paris, France) for a proven severe SARS-CoV-2 infection. All patients admitted to the ICU with COVID-19 were included. SARS-CoV-2 infection was proven by PCR test performed in the virology laboratory of our hospital. There were no exclusion criteria, except for the patient without proven COVID-19 infection. Rectal or oropharyngeal samples were sampled twice a week during the ICU stay. Indeed, we collected oropharyngeal and rectal swabs (E-Swab, Copan, Brescia, Italy and Deltaswab, Greiner, Courtaboeuf, France) during nursing cares for each patient at admission and twice a week, on Monday and Thursday for the first month, and on Tuesday and Friday for the following months (change for logistic matters). When a patient was admitted on Monday, the next sample was on Thursday. The patient continued to be swabbed until discharge or death. The oropharyngeal swab was rubbed in the throat until it was saturated with throat secretions. As we experienced occasional shortages of E-swabs during the inclusion period, we used alternative swabs such as Regular Swab with Amies Agar Gel (Copan), Regular Swab and Liquid Amies Medium (Copan), Human DNAsa, RNAsa and DNA-free certified swabs, steriles (Deltalabs, Barcelona, Spain) in addition to brain heart infusion (BHI) broth (bioMérieux, Marcy-l’Etoile, France). The samples were then sent on a daily pace to the bacteriology laboratory.

This study is an ancillary study of the French-Cohort and is also a part of the OutcomeRea™ database allowing to extract the data anonymously. We collected general characteristics: age, sex, history, and first day of COVID symptoms. We calculated body mass index (BMI) from admission height and weight and collected the treatments (e.g., antibiotics, immunomodulatory treatments, steroids) initiated before ICU admission if the patient had been transferred from another unit. Digestive decontamination using topical antibiotics was not used. Routine biological tests at admission included complete blood count, plasma proteins (C-reactive protein, procalcitonin, interleukin-6, ferritin), d-dimers, and lactate-dehydrogenase. Regarding severity assessment, we calculated the sequential organ failure assessment (SOFA) score and the simplified acute physiology score 2 (SAPS 2) at admission, and the SOFA score on each day of hospitalization in the ICU. For death at D90, we took the last available data in our information system. Time-dependent variables were initialized to 0 and remained fixed after discharge. There were 42 (44%) deaths and 16 (17%) patients were lost to follow-up before D90.

Culture methods

We measured the absolute concentration of pathogens in the oropharyngeal and rectal samples using the following protocol depicted in the Additional file 1: Fig. S1. For each rectal and oropharyngeal swab, 10 µL were plated onto seven agar media: (1) Columbia CNA + 5% sheep blood (bioMérieux, Marcy-l’Etoile, France) for the selection of Gram-positive bacteria, (2) Drigalski (bioMérieux) for the selection of Enterobacterales and non-fermenting Gram-negative bacilli, (3) Cetrimide (bioMérieux) for the selection of Pseudomonas aeruginosa, (4) ChromID® S. aureus Elite SAIDE (bioMérieux) for the selection of Staphylococcus aureus, (5) ChromID® ESBL (bioMérieux) for the selection of ESBL-producing Enterobacterales, (6) ChromID® CARBA (bioMérieux) for the selection of carbapenemase-producing Enterobacterales and (7) BBL™ CHROMagar™ Candida (Becton-Dicksinson, Rungis, France) for the selection of yeasts. Plates were then incubated at 37 °C for 24 to 48 h. The rest of the fluid from the rectal and throat e-swabs was stored at − 20 °C after initial viral inactivation (30 min at 37 °C) [18]. The identification of bacteria was obtained by mass spectrometry (Maldi Biotyper, Bruker Daltonics, Bremen, Germany). The quantification (expressed in colony-forming unit [CFU] per mL) was performed according to the protocol depicted in the Additional file 1: Fig. S1. As for antibiotic resistance, the following tests were performed: Alere™ PLP2A test (Abott, Rungis, France) for the detection of methicillin-resistant S. aureus, β LACTA™ test (Bio-Rad, Marne-La-Coquette, France) for the detection of C3G-resistant Enterobacterales and Xpert® CARBA (Cepheid, Sunnyvale, CA) for the detection of carbapenemase-producing Enterobacterales. A complete antibiogram (disk diffusion) was performed for P. aeruginosa and Acinetobacter sp.

16S rDNA sequencing

The composition of the intestinal microbiota was assessed by 16S profiling. DNA extraction of the rectal swabs was performed with the QIAamp PowerFecal Pro DNA kit (Qiagen, Courtaboeuf, France). PCR amplification of the v3-v4 segments of the 16S rDNA gene was performed using the primers proposed by Illumina (San Diego, CA). The amplicons were sequenced in paired end using a MiSeq (Illumina) device, with the target of > 10,000 read pairs per sample. The software SHAMAN was used for the analysis of the data [19]. We defined for each sample the genus richness (number of unique genera per sample) and the Shannon and the inverse Simpson indices for assessing diversity (i.e., how the different taxa are distributed). The higher the values, the more balanced the taxa are distributed.

Statistical methods

Patients’ characteristics were expressed as number (percentage) for categorical variables and median (interquartile range [IQR]) for continuous variables. Comparisons were made using Fisher exact tests for categorical variables and Wilcoxon tests for continuous variables. All the oropharyngeal and rectal culture and microbiota variables were tested in univariate Cox models as time-dependent covariates. We also tested the impact of daily SOFA and antimicrobial use using the same method. Then, we performed a multivariable Cox model introducing the microbiota variables selected in the univariate Cox models and adjusting for prognostic covariates (age, chronic illness, daily SOFA score and daily antibiotic use). The proportionality of hazard risks for time-fixed covariates was assessed using martingale residuals. For all tests, a two-sided α of 0.05 was considered as significant. All statistical analyses were performed with SAS software, Version 9.4 (SAS Institute, Cary, NC) and R (version 3.6.3). The primary outcome was death from all causes at day 90. The secondary outcomes were death in ICU, in-hospital death, all-cause death on days 42.

Results

Population

Ninety-five consecutive patients were admitted in the Medical and Infectious Diseases ICU, in Bichat University hospital, (Paris, France) for a SARS-CoV-2 infection from March 19 to September 25, 2020. Patients’ characteristics are detailed in Table 1. Of note, they were mostly men (78.9%), overweight (median BMI 28.4 kg/m2 [24.7; 32.4]). Forty-six (48.4%) had a coexisting condition. The median time from COVID symptom onset to ICU admission was 11 days [8; 15]. During the first 48 h of ICU hospitalization, 37 patients (38.9%) received invasive ventilation, and of these, 10 (10.5%) received extracorporeal membrane oxygenation (ECMO), 15 patients (15.8%) were dialyzed, and 41 (43.2%) had vasopressors. All-cause mortality at D90 was 44.2% (42/95). The median length of stay in the ICU was 11 [6; 20] days. One third of patients (34 patients) were treated for ICU-acquired pneumonia occurring after a median of 8 days after ICU admission (IQR [6; 20]) and one quarter of patients (24 patients, 51 episodes) were treated for bloodstream infection (BSI) diagnosed 12 days in median after ICU admission (IQR [8; 15]). Last, Enterococcus spp., S. aureus and Candida spp. were, respectively, identified in 2, 9 and 5 episodes of BSI. Other bacteria recovered from blood cultures were (coagulase negative Staphylococci (n = 21), Streptococcus sp. (n = 5), Enterobacterales (n = 5), P aeruginosa (n = 4) and anaerobes (n = 2).

Table 1 Initial patients’ characteristics and outcomes
Full size table

Culture results

We collected 765 samples (386 oropharyngeal and 379 rectal). The average time from admission to the first oropharyngeal and rectal swabs was 3 days (IQR [2; 4]). Details on per pathogen culture results are available on Additional file 1: Table S2. The crude colonization rate with Enterococcus spp. increased along with the days of hospitalization: 6.4% of patients were colonized by Enterococcus spp. in the oropharynx at admission, while 35.6% patients still hospitalized after 25 days were colonized (Fig. 1). This was less marked for the gut, with 58.9% patients colonized at admission vs. 77.1% after 25 days. In colonized patients, the median concentrations of Enterococcus spp. at admission were 5.5 and 6 log of colony-forming units (CFU) per mL in the throat and the gut, respectively, which remained in the same range (5–6 log CFU/mL) over time (Fig. 1). Among enterococci, E. faecalis was the most frequent species (Additional file 1: Fig. 2).

Fig. 1
figure 1

Boxplot representation of the oropharyngeal (blue) and rectal (red) concentrations of Enterococcus spp. according to the sampling time range. Bars indicate the percentage of positive sample and number in red indicates median (IQR) values of positive samples

Full size image
Fig. 2
figure 2

Boxplot representation of the oropharyngeal (blue) and rectal (red) concentrations of S. aureus according to the sampling time range. Bars indicate the percentage of positive sample and number in red indicates median (IQR) values of positive samples

Full size image

The prevalences of oropharyngeal and rectal colonization by S. aureus, were 14.9% and 6.7%, respectively (Fig. 2). In both oropharynx and gut, the carriage prevalence dropped to 4.1% and 1.4% for patients hospitalized for 25 days or more. The median oropharyngeal and rectal concentrations at admission were both 4 log CFU per mL. Yet, for patients still colonized in the oropharynx, the concentration ranged from 5 to 6 UFC/mL, while it decreased in the gut (Fig. 2).

Last for Candida spp., the prevalences of oropharyngeal and rectal colonization were 36.2% and 37.8%, respectively (Fig. 3). The carriage prevalence tended to peak between days 8 and 14 in the oropharynx and days 4 and 14 in the gut, before returning at baseline prevalence. The median oropharyngeal and rectal concentrations at admission were both 4. C. albicans was the most frequent species (Additional file 1: Fig. S2).

Fig. 3
figure 3

Boxplot representation of the oropharyngeal (blue) and rectal (red) concentrations of Candida sp. according to the sampling time range. Bars indicate the percentage of positive sample and number in red indicates median (IQR) values of positive samples

Full size image

16S rDNA sequencing

16S rDNA sequencing could be achieved on 339 out of 379 rectal swabs: Two were missing and 38 were excluded due to low quality (less than 10.000 reads obtained). At admission, average richness (genus level), Shannon and inverse Simpson indices were 30, 2.5 and 8.0, respectively (Additional file 1: Table S1). In patients colonized with Enterococcus spp. at admission, we found an association between richness and diversity (inverse Simpson indices) and the gut concentrations of Enterococcus spp. (Additional file 1: Fig. S3). However, in patients colonized with Candida spp., such association was not found (Additional file 1: Fig. S3).

Association with mortality

In the univariate analysis (Table 2), the serial absolute abundance of Enterococcus spp., S aureus, and Candida spp., either in the oropharynx or in the rectum, was strongly associated with mortality at D90. Of note, such association was not found with other opportunistic pathogens (Enterobacterales, P. aeruginosa, ESBL-producing Enterobacterales). In multivariable analysis adjusted for age, chronic diseases, daily SOFA score, and daily use of anti-infective agents, a high oropharyngeal or rectal colonization with Enterococcus spp., S. aureus and Candida spp. was again found to be independently associated with mortality at D90. Of note, neither intestinal richness nor diversity as determined by 16S rDNA sequencing was associated with mortality (Table 2).

Table 2 Univariate analysis of day 90 mortality*
Full size table

We found that certain anti-infective agents were also strongly related to the primary outcome. Indeed, in univariate analysis and confirmed in the multivariable model, receiving antifungal treatment active against Candida, or antibiotic treatment active against anaerobic bacteria, Enterococcus spp. and methicillin-resistant S. aureus (MRSA) was strongly associated with mortality (Tables 2 and 3, and Additional file 1: Tables S3, S4).

Table 3 Adjusted impact of Enterococcus spp., S. aureus and Candida spp. oropharyngeal and rectal abundances on day 90 mortality (combined analyses)
Full size table

Discussion

We showed a direct relationship between the serial concentrations of Enterococcus spp., S. aureus and Candida spp. in the gut and oropharynx of severe COVID-19 patients and mortality at D90. This relationship persisted after adjustment for mortality risk factors, daily severity of organ failure, and daily use of anti-infective therapy during the ICU stay. Considering the daily use of antibacterial agents or the richness and diversity of the gut microbiota did not add significant prognostic information.

In a previous study, Freedberg et al. showed that a dominance of Enterococcus spp. at ICU admission was associated with mortality [9]. Another study from Agudelo-Ochoa et al. similarly reported that in ICU patients, the abundance of intestinal Enterococcus spp. was higher in sepsis patients who died compared to sepsis patients who survived [20]. In COVID-19 patients, high intestinal concentrations of Enterococcus spp. have been associated with severe presentations [16, 17]. Recently, our group observed a strong association between the diversity of the intestinal microbiota of ICU patients and the relative abundance of Enterococcus spp., in that a low diversity was associated with high relative abundances of Enterococcus spp. [21]. Altogether, these observations support that the quantification of Enterococcus spp. could be a potential biomarker reflecting dysbiosis. Furthermore, in patients who received allogeneic hematopoietic cell transplantation, Stein-Thoeringer et al. showed that the type of diet (especially lactose intake) could influence the abundance of Enterococcus spp., which increase was associated with a higher risk of graft-versus-host disease (GvHD) [22]. In all, data supporting that Enterococcus spp. could be a biomarker of interest for ICU patients is accumulating, and simple measurement methods are expected. In this perspective, we used a simple, cheap culture method to measure the oropharyngeal and intestinal concentrations of Enterococus spp.. Like Enterococcus spp., Candida spp. has also been observed to be prevalent in ICU patients [23], but alongside with S. aureus, had not been associated with a poor outcome.

Our analysis also revealed that the association between the concentrations of Enterococcus spp., S. aureus and Candida spp. and D90 mortality persisted even after considering the exposure to antifungal and antibacterial agents active against these species. One possible hypothesis is that when the gut microbiota is altered, S. aureus, Enterococcus spp. and Candida spp. may not be involved in infectious processes per se which could respond to antibiotics, but rather promote inflammatory reactions and multi-organ failure. Indeed, in mouse models, Candida infection acts synergistically with S. aureus to promote the pro-inflammatory response [24, 25]. Using a mouse model of properly resuscitated peritonitis, Panpetch et al. found that Candida ingestion decreased animal survival and increased pro-inflammatory cytokine production [26, 27]. We also found that the administration of treatment active against anaerobic bacteria, Enterococcus/MRSA and antifungal drugs active against Candida spp. was linked to a poor prognosis. One hypothesis is that such administration fuels dysbiosis and thereby promotes the emergence of opportunistic pathogens. Another hypothesis is that the use of such drugs may reflect the suspicion or occurrence of infections themselves associated with a poor prognosis.

Our study has limitations. First, it is a single-center study and our results may not apply in other settings. Also, our population was homogeneous in that we only included severe COVID-19 patients. Our results should be replicated in other ICU populations and in non-COVD-19 severe patients. While stool remains the reference for the analysis of the gut microbiota, several studies have shown that 16S rDNA sequencing from rectal swabs is a suitable alternative. Swabs reliably reproduce the bacterial composition of the stool microbiota at alpha and beta diversity levels [28]. Obtaining stool in intensive care units is challenging, and in routine practice, rectal swabs are commonly used. In addition, despite the fact that we used several different types of swabs during the study, we believe that the impact on the culture or 16S rDNA sequencing results should be minimal as storage conditions were optimal, kept at room temperature for less than 24 h then stored at − 20 °C [29]. The samples were taken twice a week for practical reasons. Though, time from admission to sample was not perfectly identical between patients and may have introduced at random inter-individual variability. Last, we found that the oropharyngeal and gut concentrations of Enterococcus spp., S. aureus and Candida spp. were significantly associated with to the risk of death, the underlying mechanisms remain hypothetical.

In conclusion, we observed that in severe critically ill SARS-CoV2 patients, the oropharyngeal and intestinal concentrations of Enterococcus spp., S. aureus and Candida spp. as assessed by simple quantitative cultures were associated with the mortality, even when taking into account risk factors of mortality and evolution of the main organ failures. Our results open perspectives for simple, cheap microbiota-based markers for the prognosis of critically ill patients.