key: cord-1011435-gymtl9as
authors: de Castilhos, Juliana; Zamir, Eli; Hippchen, Theresa; Rohrbach, Roman; Schmidt, Sabine; Hengler, Silvana; Schumacher, Hanna; Neubauer, Melanie; Kunz, Sabrina; Müller-Esch, Tonia; Hiergeist, Andreas; Gessner, André; Khalid, Dina; Gaiser, Rogier; Cullin, Nyssa; Papagiannarou, Stamatia M; Beuthien-Baumann, Bettina; Krämer, Alwin; Bartenschlager, Ralf; Jäger, Dirk; Müller, Michael; Herth, Felix; Duerschmied, Daniel; Schneider, Jochen; Schmid, Roland M; Eberhardt, Johann F; Khodamoradi, Yascha; Vehreschild, Maria J G T; Teufel, Andreas; Ebert, Matthias P; Hau, Peter; Salzberger, Bernd; Schnitzler, Paul; Poeck, Hendrik; Elinav, Eran; Merle, Uta; Stein-Thoeringer, Christoph K
title: Severe dysbiosis and specific Haemophilus and Neisseria signatures as hallmarks of the oropharyngeal microbiome in critically ill COVID-19 patients
date: 2021-10-25
journal: Clin Infect Dis
DOI: 10.1093/cid/ciab902
sha: 1788688500f38234e091fb37728e80a812311bc4
doc_id: 1011435
cord_uid: gymtl9as

BACKGROUND: At the entry site of respiratory virus infections, the oropharyngeal microbiome has been proposed as a major hub integrating viral and host immune signals. Early studies suggested that infections with Coronavirus 2 (SARS-CoV-2) are associated with changes of the upper and lower airway microbiome, and that specific microbial signatures may predict COVID-19 illness. However, the results are not conclusive, as critical illness can drastically alter a patient’s microbiome through multiple confounders. METHODS: To study oropharyngeal microbiome profiles in SARS-CoV-2 infection, clinical confounders, and prediction models in COVID-19, we performed a multi-center, cross-sectional clinical study analyzing oropharyngeal microbial metagenomes in healthy adults, patients with non-SARS-CoV-2 infections, or with mild, moderate and severe COVID-19 (n=322 participants). RESULTS: In contrast to mild infections, patients admitted to a hospital with moderate or severe COVID-19 showed dysbiotic microbial configurations, which were significantly pronounced in patients treated with broad-spectrum antibiotics, receiving invasive mechanical ventilation, or when sampling was performed during prolonged hospitalization. In contrast, specimens collected early after admission allowed us to segregate microbiome features predictive of hospital COVID-19 mortality utilizing machine learning models. Taxonomic signatures were found to perform better than models utilizing clinical variables with Neisseria and Haemophilus species abundances as most important features. CONCLUSION: In addition to the infection per se, several factors shape the oropharyngeal microbiome of severely affected COVID-19 patients and deserve consideration in the interpretation of the role of the microbiome in severe COVID-19. Nevertheless, we were able to extract microbial features that can help to predict clinical outcomes.

The microbiome constitutes a signaling hub regulating host immunity, mucosal homeostasis and defense against pathogens [1] . The oropharyngeal and lung microbiome has been previously studied during respiratory syncytial virus (RSV) and influenza respiratory tract infection [2, 3] . At these body sites, the microbiome was found to modulate viral infections through adaptions of mucosal immunity or inhibition of mucosal virus adherence [4] , which eventually affect the clinical course of these infections [5] . Viruses can also facilitate pathogen colonization, eventually leading to bacterial superinfections and acute respiratory distress syndrome [6, 7] . A few studies have already explored associations of the nasal, oropharyngeal or lung microbiome with severity of COVID-19 or its role as a prognostic biomarker with so far inconclusive results [8] [9] [10] [11] [12] [13] . In particular, most of these studies investigated microbiome profiles of patients with severe COVID-19 that were admitted to a hospital or even an ICU for emergency medical treatment and not mildly or moderately affected cases. Upper vs. lower respiratory tract sampling, collection timepoint after admission, or sampling while on antibiotics or during invasive mechanical ventilation are major factors that can determine diversity and composition of the microbiome at these sites [14, 15] . Considering that treatment-and patientrelated factors can contribute to extreme dysbiosis and expansion of pathobionts in patients with critical illness, confounder analyses are mandatory in these studies to elucidate associations beyond noise and to identify potential biomarkers for disease outcomes.

Recruitment. Seven German medical centers participated in the recruitment of inpatient subjects (both sexes, 21-93 years of age) with laboratory confirmed COVID- 19 Participant metadata. Epidemiological, clinical, laboratory, treatment and outcome data of inpatient study participants were extracted from medical records using a standardized version of the WHO case record form for severe acute respiratory infections [16] . Baseline epidemiologic characteristics, symptoms of infection and antibiotic use were captured from healthy volunteers and outpatient participants.

Assessment of COVID-19 severity according to the WHO guidance document for clinical management of COVID-19 (May 27 th , 2020). Symptomatic patients who were tested positive for SARS-CoV-2 infection without evidence of pneumonia or hypoxia were categorized as mild COVID-19. These were largely patients that were screened for the infection in the outpatient test center. Patients admitted to the hospital with signs of pneumonia but no signs of severe pneumonia (including SpO 2 ≥ 93% on room air) and tested positive for SARS-CoV-2 were considered to have a moderate disease.

Severe COVID-19 was assigned to cases with severe pneumonia (respiratory rate > 30 breaths/min; or SpO 2 < 93% on room air), acute respiratory distress syndrome, sepsis or septic shock. For six patients in the severe COVID-19 group that received mechanical ventilation, we do not have gender, age or clinical outcome information. Clinical outcomes were monitored until December 22 nd , 2020.

Oropharyngeal samples for microbiome analyses were collected cross-sectionally primarily around admission or in a few cases at later time points (see Results section). Laboratory confirmation of SARS-CoV-2 was performed by RT-PCR in certified diagnostic departments of the participating university hospitals or in our laboratory in accordance with the protocol established by the WHO.

Shotgun metagenome sequencing was carried out on DNA extracted from oropharyngeal biospecimen following standard Illumina workflows of our laboratory. Metagenomic data processing was performed after demultiplexing using either Kraken 2 / Bracken for microbial taxonomy profiling.

A detailed description of the methods is available online as supplemental material.

A c c e p t e d M a n u s c r i p t 6

To study the role of the human oropharyngeal microbiome in SARS-CoV-2 infection and illness, we collected oropharyngeal biospecimens from a multi-center patient and healthy volunteer cohort in Germany Between March and November 2020 (n = 322). We recruited healthy adult individuals, patients with mild, moderate or severe COVID-19, and SARS-CoV-2 negative patients with mild upper respiratory infections or with critical pneumonia. Demographic and clinical characteristics are reported in Table 1 and the Methods section. Mild COVID-19 was assigned to patients presenting with symptoms of upper respiratory tract infections that were not hospitalized (Table S1 ). Patients with moderate or severe disease were hospitalized with lower respiratory tract symptoms. In our cohort, 65 cases with severe COVID-19 required intensive care unit (ICU) therapy, and 42 out of them deteriorated over the course of hospitalization and required invasive mechanical ventilation or extracorporeal membrane oxygenation (ECMO); 22 patients (21.6%) died by the conclusion of our study. Table S2 presents pre-existing conditions and medical treatments of the hospitalized patients of our study cohort.

To characterize the oropharyngeal microbiome profiles in our cohort, we used shotgun metagenome sequencing on oropharyngeal samples from all study subjects that were collected either during SARS-CoV-2 screening of suspected cases or during/after admission of patients with confirmed or suspected COVID-19. The number of metagenomic reads obtained from the oropharyngeal biospecimens and the fractions of human, bacterial and viral reads are listed in Table   S3 ( Table S4 for complete list of individual bacterial reads). Analyzing alpha diversity using the Shannon index, hospitalized COVID-19 patients showed a lower bacterial diversity ( Figure 1A ), but the differences were not significant after FDR adjustment of p-values (mild vs. moderate: p=0.73; moderate vs. severe: p=0.08). The Simpson's alpha diversity index, which also weighs taxonomic abundances, did not differ significantly between groups ( Figure S1 ). Richness and Chao index, a richness estimator, were found reduced in the microbiome of hospitalized patients ( Figure S1A ).

Differences in microbiome compositions between each group were most apparent in a principal coordinate analysis (PCoA) with significant separations according to COVID-19 severity (PERMANOVA: F=10.6, p<0.001; Figure 1B ). We further classified samples with taxonomic compositions highly unlike those of healthy controls as "dysbiotic" 15 as measured by Bray-Curtis beta-diversity distances 16 . We observed that patients with moderate and severe COVID-19 showed a significantly higher dysbiosis index of the oropharyngeal microbiome compared to healthy controls, URT patients and mild COVID-19 patients, respectively ( Figure 1C ). We further asked whether other patient anthropometrics and clinical variables contribute to the variation of the oropharyngeal microbiome than COVID-19 severity. As shown in Figure 1D high microbial dysbiosis were observed in our cohort ( Figure S1B ).

Hospitalized COVID-19 patients, notably severe cases, are frequently treated with antibiotics [17, 18] . In our inpatient cohort, 79.4% of all patients received broad-spectrum antibiotic treatments (Table S2) , mainly piperacillin/tazobactam (72.8%), carbapenems (21.0%), ampicillin/sulbactam and amoxicillin/clavulanic acid (9.9%) or macrolides (8.6% of all antibiotics administered).

Administration of these antibiotics during oropharyngeal sample collection significantly affected the compositional diversity of the microbiome (Figures 2A and S2A ; PERMANOVA: F=25.5, p<0.001).

This was reflected by a significantly higher dysbiosis index of samples collected during antibiotic exposure in patients with severe COVID-19 ( Figure 2B ), with similar trends in patients with moderate disease (p=0.22). The alpha diversity was not affected by antibiotic administration in hospitalized patients ( Figure 2B ). The time spent in an ICU has also been reported to affect the structure of the oral and upper respiratory tract microbiome with longer ICU stays being associated with lower diversity [19] . In our study, most of the oropharyngeal samples were collected at or early after hospital admission, and only a few swabs were taken at later timepoints ( Figure S2B ). Stratifying samples collected early (day 0-3, n=51), intermediate (day 4-14, n=34) or late after hospital admission (n=17) had a strong effect on the microbial community structure as observed in a PCoA ( Figure 2C ; PERMANOVA: F=3.6, p<0.001). A loss of diversity and an increased dysbiosis that is associated with the length of hospital stay was only observed in severe COVID-19 cases, i.e., those admitted to an ICU, and only in samples collected at a late timepoint ( Figure 2D ). As this community disruption was seen only in severe COVID-19, we hypothesized that intensive medical care with intubation and mechanical ventilation affects the structure of the oropharyngeal microbiome. As presented in Figure   2E , mechanical ventilation at the time of specimen collection has a significant effect on the compositional diversity of the microbiome in hospitalized COVID-19 patients (PERMANOVA: F=4.9, p<0.001). Although trending towards lower levels, especially in samples from late collection timepoints, the alpha diversity was not significantly reduced in the oropharyngeal microbiome collected during invasive mechanical ventilation ( Figure 2F ). In contrast, samples taken at late timepoints and during invasive ventilation had a significantly increased dysbiosis score ( Figure 2F ). So far, we found that antibiotic exposure, sampling timepoint, intubation and invasive ventilation determine structure and composition of the oropharyngeal microbiome. This is reflected by mono-domination patterns of individual bacterial species observed in patients with moderate and severe COVID-19, whereas a rather diverse ecology of microbes was observed in healthy controls, patients with non-SARS-CoV-2 URT, or mild COVID-19 ( Figure S3A ). In order to circumvent these A c c e p t e d M a n u s c r i p t 8 cases of extreme dysbiosis induced by late sampling that may confound clinical association studies, we focused on specimens collected early after hospital admission (day 0-3) for further species vs. outcome analyses (n = 51 hospitalized patients; Figure S3B : overview of the taxonomic composition).

We aimed to determine whether specific taxa are able to discriminate mortality as the major outcome of hospitalized COVID-19 patients. To address this question, we applied a machine learning (ML) analysis which provides a powerful approach for unsupervised biomarker discovery in microbiome research [20] . Here, we utilized random forest (RF)-based ML models that are little prone to overfitting, can handle effectively high-dimensional data and heterogenous types of input features (e.g., integrating taxonomic and clinical variables as combined input) and provide interpretable quantification of the importance of each input feature for the performance of the model [21] . We aggregated the microbiome and clinical data of the subset of hospitalized patients and initially assessed how well the mortality in these patients can be predicted based on their species-level oropharyngeal microbiome. After partitions of the data into training and validation sets of patients, RF models were trained using the training sets and tested for their power to predict mortality in the validation sets (see Methods section in the supplementary materials for details). Microbiome features enabled the RF models to discriminate mortality, even with a better performance than RF models solely based on the clinical variables age, gender, obesity and antibiotic treatment (Figures 3A and   3B ). Given the power of the obtained RF models to discriminate outcomes based on the oropharyngeal microbiome, we next analyzed the importance of the bacterial species in these models and found that Haemophilus influenzeae, parainfluenzae, pittmaniae and Neisseria subflava were among the most important species that segregated patients succumbed to COVID-19 versus survivors in our cohort ( Figures 3C and 3D) . We further assessed the contribution of high vs. low abundances of these bacteria on survival as determined by splitting along the median of their relative abundances. A low abundance of all four species identified in the RF models significantly increased the risk for mortality using Cox proportional-hazards models, and Neisseria subflava showed the strongest effect (HR 16, 2.0-121.6 [95% CI]) ( Figures 3E and S4A ). The risk for mortality in patients with low Neisseria subflava abundance remained significantly elevated even after multivariate adjustment of the model for the covariates age, gender and center (i.e., hospital in which patients were treated) ( Figure S4B ). Administration of antibiotics, or sampling during invasive ventilation did not affect high vs. low Neisseria subflava abundance ( Figure S4C ). The SARS-CoV-2 viral load was also not affected by high vs. low Neisseria subflava abundance ( Figure S4D ). We further analyzed the general architecture of the oropharyngeal microbiome in samples with high vs. low Neisseria abundance, but we did not observe a significant effect on Shannon diversity or on the dysbiosis index ( Figure 3F) ; although there were a few positive and negative correlations with other species in the oropharyngeal microbiome as found in a network analysis ( Figure S4E) . (Table S4 ). These species, however, were not found to dominate the oropharyngeal microbiome, and this discrepancy can be explained by site-specificity of potentially pathogenic microbes and methodological differences.

COVID-19 severity has been linked to several immunological changes such as lymphocytopenia [22] , disinhibited release of proinflammatory cytokines [16] , hyperactivation of monocytes and macrophages [23] , dampened type-I interferon (IFN) signaling [24] , or a hypercoagulable state [25] . Given that secondary bacterial infections are associated with increased risk for mortality in viral respiratory infections, the upper and lower respiratory tract microbiome has been explored in COVID-19 patients, first in case series and later in more systematic, but monocentric clinical studies with overall varying and inconsistent results [11, [26] [27] [28] .

In our study, we asked whether a SARS-CoV2 infections and its severity induced changes in diversity measures of the microbiome at the entry site of infection. We observed that mild illness or infections in outpatients was not associated with significant changes in diversity or in taxonomic composition compared to healthy controls or patients with other viral upper respiratory tract infections. Similar observations were previously published in an Italian cohort of mildly affected COVID-19 cases [28] . However, patients admitted to the hospital, and notably to an ICU, presented with reduced alpha diversity and increased dysbiosis of the oropharyngeal microbiome. In a recent study of patients admitted to an ICU for various medical conditions, McDonald et al. longitudinally explored the oral and gut microbiome and found patterns of microbial dysbiosis at both sites. Of note, the disruption of the microbiome was related to the length of ICU stay and, secondly, attributed to ventilation and antibiotic treatment [19] . In our cohort of severely affected COVID-19 patients, specimen collection at prolonged timepoints during ICU stay, administration of broad-spectrum antibiotics, and invasive mechanical ventilation were identified to destabilize the oropharyngeal microbiome with a loss of diversity and severe dysbiosis, which goes along previous results on disruption of the lung microbiome in critically ill COVID-19 patients [12] .

A c c e p t e d M a n u s c r i p t 10 In addition to a general characterization of the oropharyngeal microbiome and covariates affecting diversity and composition, we asked whether configurations of the microbiome can be utilized to segregate mortality as major clinical outcome in hospitalized COVID-19 patients applying machine learning models which we have used in the past for the prediction of metabolic responses to diet [29] . We restricted this analysis to samples collected early after admission -at timepoints where we and others found little effects on community disruption by hospitalization and other external factors [19] . In our cohort, taxonomic features were more suitable to discriminate this clinically critical endpoint than selected clinical variables such as age, gender or obesity that were previously reported as major risk factors for lethal COVID-19 [30, 31] . The abundances of Neisseria subflava and Haemophilus spp. were found as most important features associated with mortality, all of which are abundant in the local oropharyngeal microbiome [32] . This finding confirms a previous study by Merenstein et al. (2021) who also reported a significantly lower abundance of Neisseria and Haemophilus in the oropharyngeal microbiome of patients with severe disease. We found no association of high vs. low abundance of Neisseria with a general diversity disruption or any association with specific genes or pathways encoded in the microbiome suggesting the Neisseria or Haemophilus spp. are not a surrogate marker for general community changes, but may be relevant for pathophysiological processes in SARS-CoV-2 infections. For instance, commensal Neisseria spp.

have been reported to regulate innate immune responses and cytokine production in experimental infection models [33] .

Limitations. Due to its cross-sectional design with a one-time oropharyngeal microbiome collection either in the outpatient setting or during hospital admission or later during the hospital stay, and a lack of follow-up data on outpatients, we cannot infer any specific microbiome configuration at the very beginning of the infection or even in an asymptomatic pre-infectious state that would help to predict the full course of infection including its severity. Therefore, the predictive power of our observations is limited to hospitalized COVID-19 patients. With these limitations notwithstanding, utilization of the readily accessible oral microbiome signature at admission may enable data-driven patient stratification, leading to improved follow-up, tailored management from earlier disease stages and optimization of allocation of critical care supplies and staff prior to clinical deterioration. 

), honoraria for lectures, presentations, speakers bureaus, manuscript writing or educational events from SKK Kliniken Heilbronn GmbH (Georg Thieme Verlag), Terrapinn (Touch Medical Media), and BMS GmbH & Co KGaA (MSD); payment for expert testimony from Wilhelm-Sander Stifung, Else-Kroner-Fresenius Stiftung, Schering Stiftung, and Nordforsk; travel support from Amgen Inc

Interaction between microbiota and immunity in health and disease

Nasopharyngeal Microbiota, Host Transcriptome, and Disease Severity in Children with Respiratory Syncytial Virus Infection

Association between the respiratory microbiome and susceptibility to influenza virus infection

Direct interactions with influenza promote bacterial adherence during respiratory infections

Transient TLR activation restores inflammatory response and ability to control pulmonary bacterial infection in germfree mice

The co-pathogenesis of influenza viruses with bacteria in the lung

Molecular analysis of oral and respiratory bacterial species associated with ventilator-associated pneumonia

Metagenomic analysis reveals oropharyngeal microbiota alterations in patients with COVID-19

The lower respiratory tract microbiome of critically ill patients with COVID-19

SARS-CoV-2 does not have a strong effect on the nasopharyngeal microbial composition

Signatures of COVID-19 severity and immune response in the respiratory tract microbiome

Impaired antibacterial immune signaling and changes in the lung microbiome precede secondary bacterial pneumonia in COVID-19

Alterations in the human oral and gut microbiomes and lipidomics in COVID-19

Composition and dynamics of the respiratory tract microbiome in intubated patients

Reconsidering ventilator-associated pneumonia from a new dimension of the lung microbiome

Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

Antibiotic use in patients with COVID-19: a 'snapshot' Infectious Diseases International Research Initiative (ID-IRI) survey

6-month consequences of COVID-19 in patients discharged from hospital: a cohort study

Extreme Dysbiosis of the Microbiome in Critical Illness

A Review and Tutorial of Machine Learning Methods for Microbiome Host Trait Prediction

A Framework for Effective Application of Machine Learning to Microbiome-Based Classification Problems

Haematological characteristics and risk factors in the classification and prognosis evaluation of COVID-19: a retrospective cohort study

Pathological inflammation in patients with COVID-19: a key role for monocytes and macrophages

Host-Viral Infection Maps Reveal Signatures of Severe COVID-19 Patients

COVID-19-associated coagulopathy and antithrombotic agents-lessons after 1 year

Genomic Diversity of Severe Acute Respiratory Syndrome-Coronavirus 2 in Patients With Coronavirus Disease

The lung tissue microbiota features of 20 deceased patients with COVID-19

Nasopharyngeal Microbiota Profiling of SARS-CoV-2 Infected Patients

Personalized Nutrition by Prediction of Glycemic Responses

Factors associated with COVID-19-related death using OpenSAFELY

Case characteristics, resource use, and outcomes of 10 021 patients with COVID-19 admitted to 920 German hospitals: an observational study

Non-pathogenic Neisseria: members of an abundant, multi-habitat, diverse genus

The Commensal Neisseria musculi Modulates Host Innate Immunity To Promote Oral Colonization

A c c e p t e d M a n u s c r i p t 13 A c c e p t e d M a n u s c r i p t