Oral Microbiome Alterations and SARS-CoV-2 Saliva Viral Load in Patients with COVID-19

ABSTRACT Bacterial-viral interactions in saliva have been associated with morbidity and mortality for respiratory viruses such as influenza and SARS-CoV. However, such transkingdom relationships during SARS-CoV-2 infection are currently unknown. Here, we aimed to elucidate the relationship between saliva microbiota and SARS-CoV-2 in a cohort of newly hospitalized COVID-19 patients and controls. We used 16S rRNA sequencing to compare microbiome diversity and taxonomic composition between COVID-19 patients (n = 53) and controls (n = 59) and based on saliva SARS-CoV-2 viral load as measured using reverse transcription PCR (RT-PCR). The saliva microbiome did not differ markedly between COVID-19 patients and controls. However, we identified significant differential abundance of numerous taxa based on saliva SARS-CoV-2 viral load, including multiple species within Streptococcus and Prevotella. IMPORTANCE Alterations to the saliva microbiome based on SARS-CoV-2 viral load indicate potential biologically relevant bacterial-viral relationships which may affect clinical outcomes in COVID-19 disease.

B iological factors that influence SARS-CoV-2 acquisition and subsequent disease severity are not well understood, particularly with respect to potential interactions between the virus and the human microbiome. Given the extensively reported detection of SARS-CoV-2 in saliva and the possibility of transmission via saliva (1,2), the oral microbiome may represent one such key constituent and viral reservoir. Oral dysbiosis has been linked to many local and systemic diseases, including periodontitis, and could potentially influence COVID-19 disease severity (3)(4)(5). Defining the oral microbiome in COVID-19 disease is a necessary step in determining if factors such as oral hygiene could be a modifiable risk factor for severe disease (6).
It has been suggested that oral and respiratory tract microbiota could similarly express enzymes such as the transmembrane serine protease 2 (TMPRSS2), which may enhance viral entry into host cells and further viral infection (7). Results of a recent study of bronchoalveolar lavage fluid from patients with and without COVID-19 suggest that components of the host microbiome can modify heparan sulfate, a key cofactor to SARS-CoV-2 infectivity (8). Direct interaction between influenza and components of the oral microbiome, such as neuraminidase-producing streptococci, is thought to result in increased viral load (9). These studies support the hypothesis that the host microbiome is an important mediator of disease severity in respiratory viral infections. However, the relationship between oral microbiota and SARS-CoV-2 is currently poorly defined.
Here, we characterize the saliva microbiome and saliva SARS-CoV-2 viral load in COVID-19 and control patients hospitalized during the peak of the COVID-19 outbreak in New York City. Bacterial composition did not differ markedly between COVID-19 patients and controls. However, we identified numerous differentially abundant bacterial taxa associated with SARS-CoV-2 saliva viral load, providing evidence for bacterialviral interactions in the saliva and suggesting that the host microbiome may represent a cofactor affecting disease course.

RESULTS
Clinical characteristics of patient cohort. We collected saliva within 24 h of hospitalization from 53 COVID-19 patients who tested positive for SARS-CoV-2 based on nasopharyngeal swabs and who did not require intensive care at admission. During the study period, SARS-CoV-2 testing was performed on all patients admitted to the hospital, and we also enrolled 59 control patients who were newly hospitalized and confirmed to be SARS-CoV-2 negative (Table S2). Demographics of case and control cohorts were largely comparable, with no significant differences in age, sex, race, or ethnicity (Table 1). However, SARS-CoV-2 positive patients did have significantly higher median body mass index (BMI; 29.5 versus 26.4, P = 0.04). Notably, in our cohort, control patients had a significantly higher median Charlson comorbidity index (3 versus 2, P = 0.02), driven by higher rates of coronary artery disease (CAD; 23.7% versus 5.7%, P = 0.02) and chronic kidney disease (CKD; 27.1% versus 11.3%, P = 0.06). Compared to controls, COVID-19 patients had similar rates of receipt of antibiotics within 48 h of admission (P = 0.62) but were more likely to receive supplemental oxygen (P = 0.03), primarily via nasal cannula (30.2% of COVID-19 patients). One patient received remdesivir prior to saliva sample collection. With regard to hospital course and outcomes, relatively few COVID-19 patients had clinical decompensation (n = 6, 11.3%) or died in the hospital (n = 4, 7.5%). Saliva microbiome in COVID-19 patients and controls. Of the enrolled patients, we were able to perform saliva microbiome analyses by 16S rRNA gene sequencing on 46 COVID-19 patients (86.8%) and 54 controls (91.5%). There were no significant differences in a-diversity comparing COVID-19 patients and controls (Shannon P = 0.10; Chao P = 0.21), and there was no evidence of clustering by COVID-19 status in b-diversity analyses (permutational multivariate analysis of variance [PERMANOVA] P = 0.11) ( Fig. 1; Fig.  S1). However, differential abundance analyses did show significant alterations in three amplicon sequence variants (ASVs), including enrichment of Prevotella pallens in COVID-19 patients and enrichment of Rothia mucilaginosa and a Streptococcus spp. in control patients (DESeq2, adjusted P value [P adj ] of ,0.05).
Saliva SARS-CoV-2 viral load in COVID-19 patients. Cycle threshold (C T ) values from SARS-CoV-2 reverse transcription (RT-PCR) testing can be used as a surrogate for viral load and have been associated with patient outcomes (10). We performed RT-PCR on saliva from both COVID-19 and control patients. Among COVID-19 patients, 28 (62%) had detectable SARS-CoV-2 in the saliva, and one patient who tested negative for SARS-CoV-2 on nasopharyngeal swab had detectable viral load in their saliva. Thus, RT-PCR testing of saliva in this cohort had 65.1% (95% confidence interval [CI] 49.1% to 79.0%) sensitivity and 98.1% (95% CI 90.1% to 99.9%) specificity to identify COVID-19 patients (with nasopharyngeal swab results as the gold standard). Detection of SARS-CoV-2 in saliva was not significantly associated with clinical characteristics among COVID-19 patients.
We then categorized COVID-19 patients based on saliva viral load as follows: negative, low viral load (C T of $30.4 [the median value among COVID-19 patients with detectable SARS-CoV-2 in saliva] and ,40), and high viral load (C T , 30.4). Across these groups, we did not observe any significant differences in demographic or clinical characteristics (Table S3). There was no association between saliva viral load and symptom duration prior to hospital admission. Saliva viral load was not associated with clinical decompensation or death, although these outcomes occurred in relatively few patients.
Differences in saliva microbiome composition based on saliva SARS-CoV-2 viral load in COVID-19 patients. We next tested whether SARS-CoV-2 viral load in saliva was associated with saliva microbiome diversity or composition. There were no differences in Shannon (P = 0.34) or Chao (P = 0.66) a-diversity based on saliva viral load, and there was no evidence of clustering by viral load on b-diversity analyses (PERMANOVA P = 0.97) ( Fig. 2; Fig. S1). However, we identified multiple taxa that were differentially abundant between saliva-negative COVID-19 patients and those with detectable saliva viral load, with alterations in numerous ASVs in genera Streptococcus, Prevotella, and Actinomyces, among others ( Fig. 3; Table S4). COVID-19 patients with detectable SARS-CoV-2 in saliva had increased P. pallens, Streptococcus infantis, Streptococcus parasanguinis clade 411, Streptococcus sanguinis, Actinomyces sp. HMT180, and Treponema spp. and decreased Prevotella denticola, Prevotella oris, Saccharibacteria strain HMT356, Streptococcus peroris, and an additional Streptococcus spp. Comparisons of COVID-19 patients with low viral load versus undetectable virus and high viral load versus undetectable virus produced highly similar results ( Fig. 3; Table S4).
While there was no difference in the proportion of COVID-19 patients requiring supplemental oxygen based on saliva viral load (Table S3), we speculated that receipt of supplemental oxygen could affect the relationship between saliva viral load and microbiome composition. However, after adjusting the DESeq2 generalized linear model for receipt of supplemental oxygen, the large majority of originally identified taxa that were altered based on saliva viral load remained significantly differentially abundant ( Fig. 3; Table S4).

DISCUSSION
In our cohort of hospitalized patients, we did not find marked differences in the saliva microbiome between those with and without COVID-19 disease. However, among COVID-19 patients, we did find significant differences in saliva microbiome communities based on saliva SARS-CoV-2 viral load. Further, differential abundance of these taxa was also independent of administration of supplemental oxygen, which suggests that the addition of oxygen into the local oral environment by either nasal cannula or face mask does not markedly affect the relationship between the microbiome and viral load.
Prior studies of other respiratory viruses, such as influenza, have established that the saliva microbiome can be a mediator of disease severity. Bacterial-viral interactions between neuraminidase-producing streptococci and influenza may directly affect viral load (9). The nasal and posterior pharyngeal microbiome is highly distinct in household contacts who do or do not develop influenza after close contact exposure to an index patient (11), which suggests that host microbiome influences susceptibility to influenza infection after exposure. Recently, evidence has been emerging that similar transkingdom relationships may exist with SARS-CoV-2. Analyses of bronchoalveolar lavage fluid from COVID-19 patients and controls identified alterations in bacterial communities capable of modifying heparan sulfate, which is required for SARS-CoV-2 binding to angiotensin converting enzyme 2 (ACE2) (8). Bacterial components of both the respiratory and gut microbiomes influence expression of and cofactors needed for viral binding to ACE2, the host-derived receptor for SARS- CoV-2 (8, 12, 13). Our findings are also consistent with bacterial-viral associations which may have potential clinical significance and warrant further study. Additionally, the role of inflammation in the oral microbiome should also be considered. One hallmark of COVID-19 disease is a robust inflammatory response in patients, which can lead to a hyperinflammatory state and death (8,(14)(15)(16). Specific components of the oral microbiome, such as Prevotella nigrescens and Streptococcus mutans, are known to be associated with inflammation and oral dysbiosis (3). It is also possible that COVID-19-induced inflammation could directly alter oral bacterial composition. Further investigation into the effect of inflammation on the oral microbiome during SARS-CoV-2 infection is necessary.
Interestingly, we observed only minimal differences in the saliva microbiome of COVID-19 patients compared to that of controls. This suggests that the presence of SARS-CoV-2 may not in and of itself markedly alter the oral microbiome. This is consistent with previous work demonstrating that influenza infection does not appreciably affect oropharyngeal microbiome composition (17). However, possible mediation by the microbiome of susceptibility to SARS-CoV-2 infection cannot be excluded, as we did not assess SARS-CoV-2 exposure in controls. However, two other studies did recently observe an oral microbiome diversity in COVID-19 patients significantly lower than that in healthy controls (18,19), along with a decrease in butyrate-producing bacteria (18). A metagenomic analysis identified enrichment in opportunistic oral pathogens Veillonella and Megasphaera in COVID-19 patients, but, similarly to this study, did not find significant changes in alpha-diversity when comparing noncritically ill COVID-19 patients to healthy controls (20). These findings taken along with the findings in this study underscore the need for larger studies of the oral microbiome in COVID-19 patients.
Assessing the diagnostic performance of saliva testing for COVID-19 was not a focus of the current study, and the study was underpowered to identify clinical features associated with SARS-CoV-2 viral load in saliva. However, the results do add to growing data on saliva testing to identify COVID-19 patients. In this study, RT-PCR for SARS-CoV-2 in the saliva had lower sensitivity compared to that seen in prior reports on saliva-based testing in patients hospitalized with COVID-19 disease (1,21,22). These studies included patients presenting with more severe disease, and saliva testing may have decreased sensitivity in less symptomatic and asymptomatic patients. Another consideration is the quality of saliva sample the patients produce. COVID-19 patients can have difficulty producing saliva (23,24), and hyposalivation in COVID-19 patients can be exacerbated by delivery of supplementary oxygen, especially via facemask. Both hyposalivation and supplemental oxygen delivery could affect the production of high-quality saliva samples and adversely affect the performance of RT-PCR SARS-CoV-2 testing on these samples in real-life clinical situations.
This study represents one of the first comprehensive analyses of the saliva microbiome in patients with active, symptomatic SARS-CoV-2 infection. Strengths include the rapid saliva collection in newly admitted patients, the availability of relevant clinical and laboratory data for both cases and controls, and the ability to integrate RT-PCR and 16S rRNA sequencing data on the saliva samples collected. The study included a large proportion of black and Hispanic patients, populations disproportionately affected by COVID-19 yet historically underrepresented in clinical studies. Limitations include a relatively small sample size, which may have affected our ability to detect biologically meaningful differences in microbiome composition between COVID-19 patients and controls. Given the extraordinary circumstances the medical center faced at the height of the COVID-19 surge in New York City, recruiting and consenting patients for research studies was challenging. Clinical C T values for corresponding nasopharyngeal specimens from this cohort were also largely unavailable due to these constraints. Assessment of the corresponding nasopharyngeal microbiome may have been informative but was beyond the scope of this study. Additionally, data on oral health such as periodontal disease were not available, so any relationship between oral health and the oral microbiome could not be evaluated in this study. Future studies are needed to address whether oral health and preexisting periodontal disease affect the saliva microbiome during SARS-CoV-2 infection.
As the COVID-19 pandemic continues to pose a global public health threat, it is essential to understand how the virus interacts with the host and how these interactions affect severity of disease. The saliva microbiome represents an array of commensal and potentially pathogenic bacteria which can act as cofactors in the disease process. With high rates of hospitalized patients with COVID-19, common complications, such as bacterial pneumonia, underscore the importance of understanding viral-bacterial community dynamics. Further studies are needed to better elucidate the mechanisms of interactions between the saliva microbiome and SARS-CoV-2 and what effect these interactions have on the course of COVID-19 disease. Additionally, an improved understanding of saliva microbiome dynamics in the setting of SARS-CoV-2 infection may allow for the development of strategies to decrease transmission and complications.

MATERIALS AND METHODS
Sample processing and nucleic acid extraction. Consented patients provided approximately 1 to 2 ml saliva via self-collection in Zymo DNA/RNA Shield saliva collection kits containing 2 ml Zymo DNA/ RNA Shield for viral inactivation and stabilization. Inactivated samples were separated into two 1-ml aliquots and stored at 280°C. DNA and RNA were extracted in parallel from 300 ml sample using the ZymoBIOMICS DNA/RNA miniprep kit (Zymo).
SARS-CoV-2 RT-PCR. We used the 2019-nCoV research use only (RUO) primers and probes kit to amplify the SARS-CoV-2 N1 gene and human RNAse P (RP) gene (IDT). Reverse transcription PCR (RT-PCR) was performed using the TaqMan Fast Virus 1-step master mix (ThermoFisher), with a total reaction volume of 20 ml, including 5 ml of RNA template. Single RT-PCRs were performed for each sample on 96well plates, with each plate including a no template control (NTC) with 5 ml of molecular-grade water. Cycle threshold (C T ) values were obtained from RT-PCR amplification curves, and samples with C T of ,40 were considered SARS-CoV-2 positive.
16S rRNA sequencing and data processing. We amplified the 16S rRNA V1-V2 region using Illumina adapter-ligated primers (25), with 2.5 ml (5 ng) DNA template in a total reaction volume of 25 ml (12.5 ml KAPA HiFi HotStart ReadyMix, 5 ml each of forward and reverse primers) with the following cycling protocol: 95°C for 3 min, 25 cycles of 95°C for 30s, 55°C for 30s, and 72°C for 30s, and 72°C for 5 min. The Illumina Nextera XT v2 index sets A to D were used to barcode sequencing libraries. Libraries were sequenced on an Illumina MiSeq using the v3 reagent kit (600 cycles) and a loading concentration of 12 pM with 10% phiX spike-in.
Raw sequencing reads were adapter-trimmed and demultiplexed after FASTQ conversion in BaseSpace (Illumina). Quantitative Insights Into Microbial Ecology (QIIME2) and DADA2 were used to trim, dereplicate, and filter chimeric sequences before generating amplicon sequence variant (ASV) tables (26). Based on the quality score profiles of sequencing reads, forward reads were truncated at 280 bp and reverse reads were truncated at 260 bp prior to merging, ambiguities in the overlap region were not allowed, and default parameters were otherwise applied in the R dada2 package filterAndTrim () function [truncLen=c(280,260); trimLeft=c (5,5), maxN = 0, maxEE=c(2,2), truncQ = 2]. After dereplication and merging of reads, chimeric reads were identified by consensus across samples using the dada2 function removeBimeraDenovo(). All samples passed the imposed minimum of 10,000 reads after quality filtering for inclusion in this analysis. The MAFFT and FastTree modules in QIIME2 were used to generate a phylogenetic tree of all ASV sequences. ASV taxonomy was assigned using a naive Bayesian classifier trained on a custom database consisting of Greengenes 99% clustered operational taxonomic units (OTUs) and the eHOMD 16S rRNA RefSeq version 15.2 database for increased resolution of oral taxa assignments (27,28). Unique sequences across the Greengenes and eHOMD databases were clustered at 99% identity and used as the training data set for the QIIME2 naive Bayesian taxonomic classifier. Read counts before and after quality filtering and ASV assignment are provided in Table S1. The R phyloseq package was used to calculate microbiome aand b-diversity metrics (Shannon, Chao indices, and UniFrac distance matrix) (29).
Quantification and statistical analysis. Categorical variables were compared using chi-square or Fisher's exact test. Continuous variables were assessed for normal or normal-like distribution using visualization of QQ-plots. Variables with normal or near-normal distribution were compared using two-sided t-tests, and nonnormally distributed variables were compared using Kruskal-Wallis or Mann-Whitney-U nonparametric tests. UniFrac b-diversity distance matrices were compared across groups using permutation ANOVA (PERMANOVA). For all comparisons, significance was defined as a P value of ,0.05. Differential abundance analyses were performed using DESeq2, with significance defined as Benjamini-Hochberg adjusted P adj of ,0.05. The statistical tests used are clearly indicated for each comparison throughout the text and in table and figure legends. All statistical analysis and table and figure generation were performed using R version 3.6.1, and R Markdown files detailing all statistical analyses and visualizations are available in a public GitHub repository (https://github.com/mka2136/covid19_saliva_microbiome).
Experimental model and subject details. This was a prospective cohort study of patients newly admitted to the medical service at Columbia University Irving Medical Center between 7 April 2020 and 9 May 2020. The study was approved by the Columbia University Institutional Review Board. The electronic medical record was screened to identify patients admitted within the previous 24 h to an inpatient, non-intensive care medical service. Inclusion criteria included age of $18 years, ability to produce a saliva sample, and ability to provide informed consent. Patients were excluded if they were intubated, otherwise medically unstable or unconscious, or admitted to an intensive care setting. Hospitalized patients with a nasopharyngeal swab negative for SARS-CoV-2 but with history of prior COVID-19 disease were excluded. Patients included in the control group had a negative nasopharyngeal swab for SARS-CoV-2 on admission, as per our institution's clinical practice at the time of the study, and otherwise met the same inclusion and exclusion criteria as those in the case group.
Per routine clinical practice during the study period, all patients were tested in the emergency room for SARS-CoV-2 by nasopharyngeal swab PCR. Written informed consent was obtained from SARS-CoV-2-negative patients. Verbal consent was obtained from patients with a positive or pending SARS-CoV-2 test to minimize patient contact.
The following patient-level data were extracted from the electronic medical record: age, sex, race, ethnicity, body mass index (BMI), comorbidities, and receipt of antibiotics prior to or at admission. Additional hospitalization data collected included oxygen requirement at the time of admission, COVID-19 treatment (including methylprednisolone and remdesivir) received during the hospitalization, and relevant laboratory results (white blood cells [WBC], C-reactive protein [CRP], erythrocyte sedimentation rate [ESR], neutrophil-to-lymphocyte ratio [NLR], interleukin-6 [IL-6], ferritin). Clinical outcomes were recorded, including hospital length of stay, clinical decompensation (defined as need for intubation and/or transfer to an intensive care setting), discharge to hospice, discharge to home, and death. All details regarding sample size and demographic characteristics of the full cohort are included in Table 1.
The study was approved by the Columbia University Institutional Review Board (protocol number IRB-AAAS9837).
Data availability. Raw sequencing reads generated in this study have been deposited to the NCBI Short Read Archive (SRA) under NCBI BioProject PRJNA669421. All R code used for statistical analyses and table and figure generation are available in a public GitHub repository (https://github.com/mka2136/ saliva_covid19_microbiome). Further information and requests for resources or reagents should be directed to the lead contact, Julian A. Abrams. This study did not generate new unique reagents.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 0.2 MB. We declare no conflicts of interest.