The Upper Respiratory Tract Microbiome Network Impacted by SARS-CoV-2

Abstract

The microbiome of upper respiratory tract (URT) acts as a gatekeeper to respiratory health of the host. However, little is still known about the impacts of SARS-CoV-2 infection on the microbial species composition and co-occurrence correlations of the URT microbiome, especially the relationships between SARS-CoV-2 and other microbes. Here, we characterized the URT microbiome based on RNA metagenomic-sequencing datasets from 1737 nasopharyngeal samples collected from COVID-19 patients. The URT-microbiome network consisting of bacteria, archaea, and RNA viruses was built and analyzed from aspects of core/periphery species, cluster composition, and balance between positive and negative interactions. It is discovered that the URT microbiome in the COVID-19 patients is enriched with Enterobacteriaceae, a gut associated family containing many pathogens. These pathogens formed a dense cooperative guild that seemed to suppress beneficial microbes collectively. Besides bacteria and archaea, 72 eukaryotic RNA viruses were identified in the URT microbiome of COVID-19 patients. Only five of these viruses were present in more than 10% of all samples, including SARS-CoV-2 and a bat coronavirus (i.e., BatCoV BM48-31) not detected in humans by routine means. SARS-CoV-2 was inhibited by a cooperative alliance of 89 species, but seems to cooperate with BatCoV BM48-31 given their statistically significant, positive correlations. The presence of cooperative bat-coronavirus partner of SARS-CoV-2 (BatCoV BM48-31), which was previously discovered in bat but not in humans to the best of our knowledge, is puzzling and deserves further investigation given their obvious implications. Possible microbial translocation mechanism from gut to URT also deserves future studies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00248-022-02148-9.

Introduction

Since 2019, the coronavirus disease 2019 (COVID-19) pandemic caused by severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2) has triggered a massive global health crisis, claiming more than six millions of lives (e.g., [30, 31]. Numerous studies have shown that there is a wide variation in clinical severity in SARS-CoV-2 infection, which is related to many factors, including age, sex, body mass index, hormone secretion levels, and genetic factors [13, 38]. The microbial community (microbiome) plays an important role in the pathogenesis and progression of many respiratory diseases [5, 7, 15, 16]. Exploring the affect of COVID-19 on the respiratory microbiome may help research into disease progression and the appearance of certain symptoms.

Some studies have explored the relationships between SARS-CoV-2 infection and respiratory microbiome. In terms of microbial diversity, COVID-19 may cause a decrease in the upper respiratory tract (including the oral, nose, throat, oropharynx, nasopharynx) and an increase in the lung [14, 52–54]. The diversity of respiratory microbiome may be negatively correlated with the severity of COVID-19 [37, 39, 54]. The impact of COVID-19 on microbial composition is mainly reflected in the increased species abundance of pathogens. The pathogens enriched in the respiratory tract of COVID-19 patients included Klebsiella, Acinetobacter, Serratia, Pseudomonas, Veillonella, Megasphaera, Prevotella, Peptoniphilus, and Anaerococcus [14, 26, 39, 41, 42, 52, 53]. However, it is also possible that these changes in respiratory microbiome of COVID-19 patients may be more closely related to ICU stay, oxygen support type, antibiotic use, and other factors than SARS-CoV-2 infection [23]. Most studies focus on bacterial communities, but ignore viral communities. Currently, research on viral community in the upper respiratory tract (URT) of COVID-19 patients is still limited [14, 37].

Here, we investigate the characteristics of URT microbiome impacted by COVID-19 by re-analyzing RNA metagenomic-sequencing datasets of 1737 nasopharyngeal samples collected from COVID-19 patients. We applied complex network analysis approach to investigate core/periphery nodes and strong clusters of URT microbiome network, aiming to shed lights on the possible interactions between SARS-CoV-2 and other microbes including bacteria, archaea, and RNA viruses. The comprehensive network analysis revealed the enrichment of gut-associated bacteria, many of which are opportunistic pathogens for URT, in the URT microbiome, possibly translocated from gut triggered by COVID infections. Furthermore, we found a “cooperative alliance” consisting of 89 microbes against SARS-CoV-2, which might play an important role in the disease progression. A somewhat puzzling discovery is the presence of a bat coronavirus (i.e., Bat coronavirus BM48-31/BGR/2008, BatCoV BM48-31), which was previously discovered in bat [10], but, to the best of our knowledge, not yet in humans, seems to be the only “ally” of SARS-CoV-2 in the URT microbiome network of COVID-19 patients.

Materials and Methods

Dataset of Upper Respiratory Tract Microbiome

URT microbiome dataset used for this study were from 1903 nasopharyngeal (NP) swab samples tested positive for SARS-CoV-2 by RT-qPCR, whose total RNA was extracted and processed with RNA sequencing-based metagenomics [19]. Metagenomic sequencing data and genome assemblies are publicly available at NCBI’s GenBank and SRA databases under BioProject PRJNA622837. Metagenomic data were classified to the species-level using Kraken 2 v2.1.2 against the reference sequence databases [50, 51]. The reference sequence databases were built based on the complete bacterial, archaeal, and viral genomes, the GRCh38 human genome, and UniVec Core database accessed from NCBI RefSeq on June 3, 2021. After taxonomic classification, the species abundance was estimated using Bracken v2.6 [24, 25]. Samples not annotated with severe acute respiratory syndrome-related coronavirus (SARSr-CoV) were removed, and the microbiomes of the remaining 1737 samples were used for subsequent analyses.

Co-occurrence Network Analysis

To ensure the reliability of correlation calculation, species present in less than 1% of the samples were excluded for correlation calculation. FastSpar, an efficient and parallelizable implementation of the SparCC algorithm, was used to construct network based on the true abundance of taxon (Friedman et al. 2012, Watts et al. 2019). Correlation relationships with a p value ≤ 0.01 after false discovery rate (FDR) adjustment and absolute coefficient > 0.1 were used to construct the co-occurrence network. The properties of network were calculated using the iGraph R-package. The P/N (positive to negative links) ratio estimated in this study is a network property that was introduced by Ma [27]. Cytoscape v2.8.3 was used for the visualization of the network. The MCODE (molecular complex detection) algorithm was used to detect network clusters.

The core/periphery network (CPN) analysis was used to detect the core/periphery structure in the network, which consists of two types of nodes: dense cohesive core nodes and sparse connected periphery nodes. An ideal CPN consists of core nodes that are fully linked to each other and periphery nodes that are fully linked only to the core [6]. Formally, let G = (V, E) be an undirected, unweighted graph with n nodes and m links, and let A = (a_ij) be the adjacency matrix of G. If node i and node j are linked, then a_ij = 1, otherwise a_ij = 0. Let δ be a vector of length n consisting of 1 and 0, where 1 represents the core node and 0 represents the periphery node. Let P = (p_ij) be the adjacency matrix of the ideal CPN of n nodes and m links. The detection of the core-periphery structure is an optimization problem to find vector δ, such that the objective function (ρ) achieves its maximum based on the following expression:

$$\rho =\sum _{i,j}{A}_{\mathit{ij}}{P}_{\mathit{ij}}$$ 1

With vector δ, it is then easy to classify nodes as either core or periphery. We implemented the CPN analysis in Python using code provided by Ma and Ellison [28].

Results

Composition of the URT Microbiomes in COVID-19 Patients

We identified 6049 species, including 5754 bacteria, 147 archaea, 72 eukaryotic RNA viruses, and 76 prokaryotic RNA viruses, from the 1737 URT metagenomic samples of COVID-19 patients. In addition to severe acute respiratory syndrome-related coronavirus (SARSr-CoV, a species-level taxon including subspecies of SARS-CoV-2), these eukaryotic RNA viruses identified also included some other respiratory pathogens, such as Human coronavirus NL63, Human coronavirus HKU, Human metapneumovirus, Enterovirus, and Rotavirus, but most of them were present in less than 10 samples (Supplementary Tables S1). It suggested that some COVID-19 patients have developed co-infections with other viruses. We also identified two bat viruses in the URT of COVID-19 patients, BatCoV BM48-31 and Bat sapelovirus, of which BatCoV BM48-31 was present in 178 samples and Bat sapelovirus in one sample (Supplementary Tables S1). Possible explanations for the presence of bat viruses in humans have been discussed in later section.

The most abundant phylum across all samples was the Proteobacteria (68% of total abundance), followed by the Pisuviricota (24%), Firmicutes (4.6%), Actinobacteria (1.4%), and Uroviricota (1.4%). The 1737 microbial communities can be roughly divided into two categories, one dominated by bacteria of the family Enterobacteriaceae and the other dominated by virus of the family Coronaviridae. The relative abundance of Enterobacteriaceae decreased with the increase of Coronaviridae (Fig. 1a). The most widely species was severe acute respiratory syndrome-related coronavirus (a species-level taxon including subspecies of SARS-CoV-2), which was present in all samples, followed by Staphylococcus aureus (99.8%), Escherichia coli (99.8%), Klebsiella pneumoniae (99.4%), and Salmonella enterica (99%). These five species accounted for 88% of the relative abundance of the total microbiomes (Fig. 1b).

Fig. 1.

Composition of upper respiratory tract microbiome: a at family level and b at species level of COVID-19 patients: X-axis is 1737 URT samples and y-axis is relative abundance

Co-occurrence Network Analysis for COVID-19-URT Microbiomes

The co-occurrence network of COVID-19-URT microbiomes was constructed based on the species that were present in 1% or more of the samples. Out of the 6104 species identified, 32.6% were present in 1% or more of the samples, including 1956 bacteria, 23 archaea, 5 eukaryotic viruses, and 9 prokaryotic viruses. After removing the links (correlations) with insignificant p values (p > 0.01) or weak correlation coefficients (|R|≤ 0.1), 1840 nodes (species) and 135,105 links remained in the network (Fig. 2). The basic network properties are listed in Supplementary Table S2.

Fig. 2.

Network of upper respiratory tract microbiome in COVID-19 patients: links in cyan are positive correlation; links in orange are negative correlations; circles represent core species; triangles represent periphery species; nodes are colored differently in terms of their phylum identity. See supplementary Table S10 for the details of links

We divided the species in the network into core and periphery groups by using CPN analysis, of which 938 species belonged to core and 902 to periphery (Supplementary Tables S3 & S4). It is worth noting that all 5 eukaryotic viruses in the network belonged to the core: SARSr-CoV, BatCoV BM48-31, Murine leukemia virus, Adeno-associated dependoparvovirus A (AAV), and Megavirus chiliensis.

The Strongest Cluster (Guild) in the URT Microbiome Network of COVID-19 Patients

There were 28 clusters in the COVID-19-URT microbiome network (supplementary Table S5), of which the strongest cluster consisted of 277 nodes (15% of the total nodes) and 28,657 links (21% of the total links). The strongest cluster contains two parts, one was a cooperative sub-cluster dominated by Proteobacteria (129/185), in which one-third (50/185) of the species were from Enterobacteriaceae; another was composed of species that were antagonistic or competitive this sub-cluster (Fig. 3). The interactive patterns within each of these two parts were positive, but almost all correlations between them were negative. In the following, we refer to this Proteobacteria-dominated sub-cluster as PDSC for short. Supplementary Table S6 lists the species composition of the top three strongest clusters in the network.

Fig. 3.

The strongest cluster in the URT microbiome network of COVID-19 patients: links in cyan are positive correlation; links in orange are negative correlations; circles represent core species; triangles represent periphery species; nodes are colored differently in terms of their phylum identity. Specific species information is listed in supplementary Table S6

Ratio of Positive to Negative Interactions in the COVID-19-URT Microbiome

Previous studies by our team found that most human microbiome networks were predominantly mutualistic, and diseases might even lead to a reduction in antagonistic interactions [20, 21, 22, 27, 29]. However, the ratio of positive to negative links (P/N ratio) in the network is 0.677, that is, the negative links are 1.477 times of the positive links (Supplementary Table S7, Fig. 4), indicating that the COVID-19-URT microbiome was not a mutualistic community. Bacteria, especially Proteobacteria species, held most of the negative links (Supplementary Table S7).

Fig. 4.

Ratios of positive to negative links (P/N ratios) in the URT microbiome network of COVID-19 patients: a the P/N ratio in the whole networks, and the ratios within and between kingdoms. b The P/N ratios within and between main bacteria phyla. F, Firmicutes; P, Proteobacteria; B, Bacteroidetes; A, Actinobacteria

The interactions between archaea, between viruses, and between archaea and viruses were mainly positive [log₁₀(P/N ratio) > 0, Fig. 4a]. There were more negative interactions than positive between bacteria, between bacteria and archaea, and between bacteria and viruses [log₁₀(P/N ratio) < 0, Fig. 4a]. The interactions between viruses are very few and mostly positive (Supplementary Table S7). Prokaryotic viruses hold 88% of the interactions between viruses and bacteria. Both eukaryotic and prokaryotic viruses have more negative interactions with bacteria than positive ones. For bacteria, the P/N ratios of intra-phyla were all greater than 1 [log₁₀(P/N ratio) > 0], and the P/N ratios of inter-phyla were all less than 1 [log₁₀(P/N ratio) < 0, Fig. 4b].

Niche overlap is one of the important reasons for competition or antagonism (negative interaction) among species. These results suggest that niche overlap within URT microbiome of COVID-19 patients occurs mainly within bacteria as well as between bacteria and archaea and viruses. More specifically, niche competition within bacterial community mainly occurred intra-Proteobacteria and between Proteobacteria and other phyla.

Sub-networks Associated with SARSr-CoV and Other Eukaryotic Viruses

In the COVID-19-URT microbiome network, SARSr-CoV was suppressed by a cooperative alliance of 89 species and was in a cooperative relationship with only one species, i.e., BatCoV BM48-31 (Fig. 5). The reason why it is called cooperative alliance is that the relationships between these 89 species were positive or cooperative. All species but four bacteria in the anti-SARSr-CoV cooperative alliance were the members of PDSC in the strongest cluster, including 1 prokaryotic virus (Enterobacteria phage fd), 1 archaea (Sulfolobus acidocaldarius), and 83 bacteria (including 58 Proteobacteria, 12 Firmicutes, 8 Actinobacteria). Specific species information is listed in supplementary Table S7. Much like SARSr-CoV, BatCoV BM48-31 was also jointly inhibited by a group of species and was positively linked only to SARSr-CoV (supplementary Fig. S1, Table S8). All 129 species negatively linked to BatCoV BM48-31 were from PDSC, and more than 4/5 of them also inhibited SARSr-CoV (supplementary Table S8).

Fig. 5.

Sub-network associated with SARSr-CoV in the URT microbiome network of COVID-19 patients: links in cyan are positive correlation; links in orange are negative correlations; circles represent core species; triangles represent periphery species; nodes are colored differently in terms of their phylum identity

The network also contains three other eukaryotic viruses, Adeno-associated dependoparvovirus A (AAV), Murine leukemia virus, and Megavirus chiliensis. AAV was positively correlated with 42 species, and the correlations within these 42 species were positive (supplementary Fig. S2A). Murine leukemia virus held negative links with three bacteria, and Megavirus chiliensis held positive link with a bacteria (supplementary Figs. S2B & S2C).

Discussion and Conclusion

The URT is the initial site of entry and replication of SARS-CoV-2, and its colonization of microbial community is closely related to respiratory health. However, a limited number of studies have examined the signatures of the URT microbiome infected by SARS-CoV-2. The purpose of this study was to understand the impact of COVID-19 on URT microbiome including bacteria, archaea, and viruses from two aspects of species composition and interspecies interactions.

In terms of species composition, we found that COVID-19 leads to the predominance of gut-associated bacteria (i.e., Enterobacteriaceae) in the URT microbiomes of patients. The normal URT microbiomes are dominant by Corynebacterium, Staphylococcus, Streptococcus, Dolosigranulum, and Moraxella, and the species abundance of Enterobacteriaceae is low [1, 18, 32]. However, the URT microbiomes of many patients were enriched with Enterobacteriaceae, especially pathogens, E. coli, K. pneumoniae, and S. enterica.

One explanation for this phenomenon is the uncontrolled overgrowth of these gut-associated bacteria, which were present in low abundance in the URT before infection. The infection of SARS-CoV-2 may first destroy the cooperative relationships between the normal URT species and the inhibitions of the normal bacteria to opportunistic pathogens. This gives pathogens the opportunity to proliferate and establish their own guilds to compete for resources. The strongest cluster (guild) found in the COVID-19-URT microbiome network may confirm this progress. Network module analysis can be used to study the guild in ecology, which refers to a group of species exploiting the same kinds of resources in similar ways [20, 21, 46, 52, 53]. The strongest guild in the COVID-19-URT microbiome consists of two parts: PDSC, a highly dense cooperative sub-guild dominated by Enterobacteriaceae, and the species that antagonize PDSC. PDSC had a high network density, in which interactions were all cooperative. However, interactions between the PDSC-antagonistic species were sparse, with even 25 species negatively correlated only with PDSC and not interacting with the rest of species in this group. The interactions between PDSC and PDSC-antagonistic species were almost entirely competitive or antagonistic. It illustrates a common phenomenon in ecological guilds, where the two groups are competing for the same resources. The 13 out of 92 species exploited by the PDSC were belonged to Corynebacterium. Corynebacterium spp. are closely related to respiratory health and have been considered as potential keystone species in the URT microbiome, as they play an important role in the exclusion of potential pathogens [3, 18, 32, 48]. Therefore, we speculated that although Enterobacteriaceae pathogens exist in URT of healthy people, they are inhibited by probiotics such as Corynebacterium and maintain a very low abundance. Infection with SARS-COV-2 results in strong interference of URT microbiome by internal and external factors, thus destroying this inhibition effect. Another possible explanation for the dominance of gut-associated bacteria in COVID-19 URT is that SARS-CoV-2 infection, like HIV infection, triggers translocation of the gut microbiome.

URT microbiome, as gatekeeper of the respiratory system, is one of the most important sources of lower respiratory tract and lung microbiome [2, 32, 35, 45]. The gut-associated bacteria in URT may enter the lung and alter the composition of the lung microbiome. Enrichment of the lung microbiome with gut-associated bacteria may propel alveolar inflammation and injury, which may lead to the aggravation or even deterioration of the disease [8, 9, 40]. In addition, a study has shown that the decrease of odor identification might be related to the increase of Enterobacteriaceae in URT [17]. One possible explanation is that Enterobacteriaceae are capable of producing butyrate, whose unpleasant smell might interfere with olfactory performance [18]. During the COVID-19 outbreak, an increasing number of patients have reported a disorder of the sense of smell [47, 49, 55]. Olfactory dysfunction has been one of the symptoms and sequelae of COVID-19. The increase of gut-associated bacteria in URT may be one of the causes of olfactory dysfunction, and controlling these bacteria may improve this symptom.

We also found through co-occurrence network analysis that SARSr-CoV and BatCoV BM48-31 were inhibited by a cooperative alliance in which most of species were members of the PDSC. Moreover, SARSr-CoV and BatCoV BM48-3 were the each other’s only cooperators in the microbiome. This result suggested that SARSr-CoV and its “friend” are common enemies of both normal and dysbiosis URT microbiome. Although PDSC, the high-density cooperative guild dominated by gut-associated pathogens, seized resources and suppresses beneficial microbes collectively, it also played the most important role in warding off these two viruses. What roles that the cooperative alliance consisting of species against SARS-CoV-2 play in clinical progression, severity, and recovery of COVID-19 will need to be addressed in future studies.

Another point of note is the cooperative interaction between SARSr-CoV and BatCoV BM48-31. BatCoV BM48-31 was identified from Rhinolophus blasii bat collected in Bulgaria in 2008 and belongs to lineage B betacoronaviruses, together with SARS-CoV-2 and SARS-CoV from SARSr-CoV [10]. Sequences of this virus are similar to those of SARS-CoV-2 and SARS-CoV, but there is still some phylogenetic distance between them [58], Boni et al. 2020, [43]. Although it has been shown that some bat SARS-like coronavirus viruses have the ability to infect human cells, they do not include BatCoV BM48-31 [11, 33, 34, 43]. Therefore, we should emphasize that the presence of BatCoV BM48-31 in the URT microbiome is somewhat puzzling. More cautiously, we cannot exclude the small probabilities of errors, from arguably exhaustive three sources: possible mis-annotation from using Kraken 2 software pipeline, possible contamination of the original samples, and/or data handling/storage errors in the databases from which we obtained the raw datasets for reanalysis. A mis-annotation could mean that the annotated BatCoV BM48-31 and SARS-CoV-2 could just be the same virus or a close variant, as suggested by one of the anonymous expert reviewers. In terms of species level annotation, compared with other methods, Kraken 2 has a relatively low false positive rate, and the accuracy is relatively high [44, 36]. Kraken 2 has also been used in many studies to identify viruses from metagenomic sequencing data (e.g., [4, 56, 57]. Although BatCoV BM48-31 belongs to the same lineage B betacoronaviruses with SARS-CoV-2, there is still a considerable amount of phylogenetic distance between them. In other words, there is still a large difference between the sequences of BatCoV BM48-31 and SARS-CoV-2. The results of Kraken 2 should be robust as long as the difference between the two virus sequences is greater than Kraken 2’s error rate, which seems to be the case from existing applications of Kraken 2 [12].

Finally, if the previous three small probability events are unnecessarily cautioned, i.e., the presence of BatCoV BM48-31 in the URT microbiome is truly positive, then further investigations on the infection mechanisms of BatCoV BM48-31, in particular, its possible cooperative infection action with SARS-CoV-2, should have important implications for understanding the origin of SARS-CoV-2. Given that BatCoV BM48-31 was discovered more than a decade ago, its co-presence SARS-CoV-2 in humans a decade later during the COVID-19 pandemic is puzzling and certainly worthy of further investigations.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contribution

WL performed the study and wrote the manuscript. ZSM designed and performed interpretations. All authors read and approved the final manuscript.

Data Availability

The dataset used in this study was downloaded from NCBI Sequence Read Archive (SRA) database (accession no. PRJNA622837).

Declarations

Ethics Approval

Ethics approval was not required for this study.

Conflict of Interest

The authors declare no competing interests.