Nine COVID-19 children between 7-139 months old were enrolled in this study together with 14 age-matched healthy control children. A total of 103 specimens including 27 sets of paired specimens (at least two of throat swab, nasal swab and feces) were collected from children with COVID-19 (Supplementary Fig. 1). The children were being followed between 25-58 days after symptom onset. All samples were subjected to high-throughput sequencing of the V4-region of bacterial 16S rRNA gene (Methods).
Respiratory and gut microbiome dynamics in COVID-19
We analyzed the 16S-rRNA gene sequences of all specimens from three body sites, and obtained 2,187 sub-OTUs (sOTUs) that represent 15 known phyla including 200 known genera (Supplementary Table 1). Using the DMM method (Supplementary method), we identified 8 community types (Fig.1a). The specimens from healthy children clustered into two community types, one bears the signature of stool samples (H-GUT), and another represents the collection of all three kinds of samples (throat swab, nasal swab and feces) (H-MIX). H-GUT is significantly separated from H-MIX (Fig. 1b), with significantly lower richness and evenness (Fig.1c). H-GUT was featured by Moraxella, a commensal in nasal passages of infants, implying that it might not represent a normal gut microbiome status. Because the development of infant microbiome is influenced by maternal materials from multiple sites (stool, vaginal, oral and skin) among which bacteria from maternal gut being the most important contributor34, infants and children may share the same or very similar microbial community structures of the nasal cavity, throat and gut28 35. Therefore, H-MIX represents the gut and respiratory tract microbiomes of healthy children. In fact, the nasal cavity, throat and gut still maintain similar microbial structures in adulthood.
Bacteria from stool specimens of COVID-19 children fell into three distinct community types (COVID-GUT I-III), and those from nasal and throat swabs formed another three distinct types (COVID-TN I-III) (Fig. 1a). All COVID-19-related types are significantly separated from the type H-MIX except COVID-TN-I that overlaps with H-MIX (Fig. 1b). In particular, three respiratory tract-related types and three GUT-related types of COVID-19 children are also significantly separated from each other, and distinctly different from healthy children. These results indicated that SARS-CoV-2 infection significantly changed the gut and respiratory tract microbiota of children, and the separation of bacterial community structures between the gut and respiratory tracts suggested that the normal development of the microbiota may be impaired.
All COVID-19-related types showed lower richness and evenness than H-MIX, except for COVID-GUT-I that has the most similarity to H-MIX and relatively normal microbiome structure. There was a gradual decrease from community type I to III for both gut and respiratory tract (Fig. 1c), indicating a progressive deterioration (dysbiosis) of the microbiome. Overall, the dysbiosis appeared to be more severe in the respiratory tract than in the gut.
Indicator genera of eight DMM clusters
To characterize eight microbial community types, we identified 35 indicator genera (Fig. 2a). The H-MIX type was characterized by 11 genera, and the predominant commensal bacteria contained Prevotella, Streptococcus, unclassified Pasteurellaceae, and Actinomyces (Fig. 2b). Some of the indicator bacteria in H-MIX were shared by the community types COVID-GUT-I (e.g. Prevotella, Porphyromonas, Finegoldia, Anaerococcus, etc.) and COVID-TN-I (e.g. Prevotella, Neisseria, Fusobacterium, unclassified Pasteurellaceae, Leptotrichia etc.). As a dysbiosis status, community type COVID-GUT-III was dominated by Bacteroides, as well as Parasutterella that is associated with irritable bowel syndrome and other intestinal chronic inflammation36. Community type COVID-TN-III was dominated by highly abundant Pseudomonas and Herbaspirillum, and it had higher levels of genera of Corynebacterium, Comamonadaceae, Burkholderia, Achromobacter, Brevundimonas, Ralstonia, Phyllobacterium, and Burkholderiales than other community types (Fig. 2b). Genus Pseudomonas is a notorious human pathogen associated with various diseases (e.g. pneumonia), and the samples were overwhelmed by the dominant species Pseudomonas veronii (100% sequence identity)37. Apart from COVID-TN-III, genus Pseudomonas also dominated community type COVID-TN-II with Streptococcus, and COVID-GUT-II with Bacteroides. Furthermore, Achromobacter and Burkholderia are associated with cystic fibrosis38 39, and most other genera are environmental bacteria. The predominance of Pseudomonas together with the colonization of various environmental bacteria in type COVID-TN-III imply an extreme dysbiosis in upper respiratory tract.
The dynamic change of children during COVID-19
Recently, we observed synchronous restoration of the microbiomes of both respiratory tract and the gut towards normal structure in COVID-19 adults within a short time (6-17 days) after symptom onset (Xu et al., unpublished observation). Distinct from adults, the microbiome community compositions were extremely dynamic in children during COVID-19, and the changes of the community types in the respiratory tract and gut were divergent (Fig. 3a). The respiratory (especially nasopharyngeal) microbiome of 7/8 children (except CV05) appeared to evolve from early healthy (H-MIX) or high-diversity community structures (COVID-TN-I) to late low-diversity dysbiosis structure (COVID-TN-III), indicating a steady deterioration in composition and function of the respiratory microbiome despite a fast clinical recovery (Fig. 3a). Surprisingly, the respiratory dysbiosis was sustained at least 19-24 days after discharge (i.e., 42-58 days after symptom onset) in three children (CV01, CV02 and CV09).
In contrast, the gut microbiome alternation varied greatly among these COVID-19 children. Improvement or restoration was observed in three children (CV01, CV02 and CV05), but a worsening trend of unstable bacterial genera occurred in another three children (CV03, CV04 and CV09) (Fig. 3a). For example, the community type of CV09 improved from COVID-GUT-II to COVID-GUT-I on day 7 after symptom onset, but deteriorated to COVID-GUT-III on day 37. For CV03, whose microbiome got worse from a gut community type COVID-GUT-II on day 19 to a respiratory community type COVID-TN-III on day 27, and returned to COVID-GUT-II on day 43. The shift from a slightly dysbiosis gut community type to a severely dysbiosis respiratory community type implies microbial translocations from respiratory tract to gut. The restoration or worsening of the gut microbiome showed no association with clinical recovery (discharge from the hospital) or the presence or absence of SARS-CoV-2 RNA in the gut (Fig.3a).
Our data clearly demonstrate a progressively worsening of microbiome in both the respiratory tract and gut of children during the course of COVID-19. The worsening was predominantly driven by Pseudomonas species P.veronii (Fig. 3b and Supplementary Fig. 2), which was the most prevalent Pseudomonas species identified, and had a relative abundance of over 20% in most COVID-19 children. Genus Streptococcus (mainly S.mitis) also contributed to the worsening of the microbiome40. On the other hand, the presence of probiotic Bifidobacterium and the most important butyrate-producing bacteria Faecalibacterium were inversely correlated with the existence of Pseudomonas (Fig. 3b and Supplementary Fig. 2), despite these beneficial bacteria presented at a very low relative abundance and often decreased in late disease stage.
Bacteria–bacteria co-occurrence networks
The co-occurrence network analysis revealed significant microbial cross-talk among different body sites of children with COVID-19 (Fig. 4). There were three main co-occurrence networks identified. Positive co-occurrence relationships were observed within and between bacteria from the respiratory tract and the gut (FDR-adjusted P<0.001, Pearson correlation r > 0.4), indicating the presence of frequent bacterial cross-talk between different body sites. Similar to our observation in adults, the co-occurrence networks were relatively separated by different diversity-levels of community types, but not by body sites. For example, bacteria from the community type COVID-TN-III are closely associated with those from COVID-TN-II and COVID-GUT-II (Fig. 4). In this network, Pseudomonas was positively correlated with some environmental bacteria. In another two networks, probiotic (e.g. Bifidobacterium and Faecalibacterium) were mainly correlated with commensals and community types H-MIX, COVID-GUT-I, COVID-TN-I and COVID-TN-II (Fig. 4). Therefore, the community types are more representative of microbiome status.