Temporal association between human upper respiratory and gut bacterial microbiomes during the course of COVID-19 in adults

SARS-CoV-2 is the cause of COVID-19. It infects multiple organs including the respiratory tract and gut. Dynamic changes of regional microbiomes in infected adults are largely unknown. Here, we performed longitudinal analyses of throat and anal swabs from 35 COVID-19 and 19 healthy adult controls, as well as 10 non-COVID-19 patients with other diseases, by 16 S rRNA gene sequencing. The results showed a partitioning of the patients into 3-4 categories based on microbial community types (I-IV) in both sites. The bacterial diversity was lower in COVID-19 patients than healthy controls and decreased gradually from community type I to III/IV. Although the dynamic change of microbiome was complex during COVID-19, a synchronous restoration of both the upper respiratory and gut microbiomes from early dysbiosis towards late more diverse status was observed in 6/8 mild COVID-19 adult patients. These findings reveal previously unknown interactions between upper respiratory and gut microbiomes during COVID-19.

More importantly -Supplementary figure 1 indicates that some of the COVID-19 patients (e.g. p05, p11) never tested positive for COVID-19, yet they seem to be considered COVID-19 patients in the analyses (Figure 1). Please explain. Patients 05 and 11 are the only ones where community type IV was observed in early stages of disease, so this obviously needs clarification. Supplementary methods are mentioned in the text (line 73) but were not provided.

Specific comments:
The paper states that all patients (except p09) had mild symptoms. Is it a normal procedure to hospitalize patients with mild symptoms? L 82-83: "The vast majority of the specimens of COVID-19 patients were divided into four community types, called I-IV, and 6 specimens were included in the NP type". I did not understand this analysis. Was this a cross-validation? How did you proceed with the 6 specimens that were classified as "non-COVID-19 patients"?
Paragraph starting in L 108: I think this is an overinterpretation of the result. For example: "Prominent microbiome community type shifts from early lower-diversity community types (NP, IV or II) towards later higher-diversity types (II or I) were observed in 9/24 COVID-19 adults who had specimens at two or more time points." In other words -these transitions were observed in only 37% of the data. The graphs show that the results are not that obvious, and community shifts are far from "prominent".
Section starting on L 128: Were anal swabs also collected for healthy and non-covid-19 patients?
L 164 -166: How were those bacterial genera selected? L 197 -200: I think this is an overstatement. It can be observed that some bacterial genera cooccur, but there too few samples and time points to make inferences about community succession. I suggest to also correct these networks for sample size. The authors could build networks of groups II and I with just 4 samples for example, to confirm that the increased complexity in these groups is not an artifact of sample size. This is an interesting paper in an emerging field, and seeks to answer an important question concerning the gut-lung axis. The primary claims of the paper are that the throat and gut microbiota of COVID-19 infected adults can be characterised into different community types, beginning with a low diversity population soon after infection and a restoration back to diverse microbiota that in synchronous in the respiratory tract and the gut. Overall, the researchers have done very well to draw conclusions from a limited dataset to provide timely information to assist with the current pandemic. Their attention to the precise details of the data and their longitudinal sampling is commendable. This work is very interesting and could provide valuable insight into the field and the long-term consequences of this novel disease, but additional considerations may help to improve the confidence of these conclusions: Major Comments: -Patient characteristics of healthy adults and non-COVID-19 patients should be included as supplementary information.
-Care should be taken describing the results of throat swabs. The authors describe the sampling as coming from the "posterior oropharynx" (Line 323) which is only one part of the upper respiratory tract. Referring to results more generally as "respiratory tract" may give the impression that results are associated with changes in the lower respiratory tract, particularly given references to the airways and lung throughout the discussion (e.g. lines 288-291) -Line 83: It was noted that 6 COVID samples clustered with the NP type samples. Was there anything to distinguish these samples from other COVID samples? -Authors discuss a return from low diversity to high diversity community types over time in throat samples, yet this was only observed in 9/24 patients (Line 112) while a reverse pattern was observed in four patients (Line 118-119). To my mind, this does not appear to demonstrate a clear trend that these patterns are consistent features of COVID-19 and additional sampling may be required to validate these findings. Alternatively, authors may be able to place additional emphasis on the association between the community types and time since appearance of symptoms in samples which were not a part of the longitudinal series but nevertheless support the hypothesis (e.g. p34, p18, p21, p28 in Fig 1e) -Line 124-125: "These results indicate that the change of the respiratory microbiome might be closely associated with disease progression in COVID-19". The strength of this statement could be substantially improved by matching community types to some measure of clinical severity (e.g. symptom score, viral load, oxygen saturation, etc.) if this data is available.
-Line 161-163: Could the lack of correlation between community divergence and other parameters be due to the relatively small sample size available? -Line 182-184: Authors discuss the hypothesis that co-occurrence networks between bacteria reflect crosstalk between the gut and lungs as a result of bacterial translocation, which is later described as a result of damage to the respiratory and gastrointestinal mucosa ; Lines 261-262). Other processes may also explain these co-occurrences including (1) bacteria migrating from the oropharangeal site to the gut via swallowing and passage through the gastrointestinal tract (especially as several oral taxa are implicated) or (2) induction of immune responses at both sites which apply similar selective pressures to the microbiota. Could the authors please address these hypotheses. -Similarly, is there any evidence available for a breakdown of mucosal barriers which may support the hypothesis of the authors (e.g. serum LPS or citrulline, etc.)? This may not be possible given the status of the patients but could provide valuable support for the hypothesis.
-Line 202-204: Increasedd Bifidobacterium and Faecalibacterium in throat samples were noted as evidence of restoration in microbiome composition, yet to the best of my knowledge these genera are not considered to be a normal part of the oropharangeal microbiome. Can the authors please provide some citations to demonstrate that these genera are representative of a healthy oropharangeal microbiome. -Increases in Bifidobacterium and Faecalibacterium are noted as signs of improvement and restoration (e.g. Lines 164-170; 202-204) but the authors note several other commensal genera such as Bacteroides, Roseburia, Blautia, and Coprococcus, while other probiotic strains such as Lactobacillus are also prominent members of the microbiota. Can the authors explain why they chose to focus on Bifidobacterium and Faecalibacterium specifically, and not other probiotic genera? If other genera do not display similar patterns, does this perhaps indicate that the restoration of the microbiota is incomplete and may have long-term consequences? -Line 228-230: The authors indicate that inconsistencies in changes observed during longitudinal sampling indicate that diversity characteristics of the throat microbiome were affected by COVID-19. Can the authors please explain why they determined this direction of causality (i.e. is it possible that diversity characteristics may have rather influenced the progression of COVID-19 and not vice versa).
-Line 268-272: It is stated that the gut microbiome appeared to have a faster restoration to increased bacterial diversity, but this phenomenon did not appear to be described in the results section. Indeed, Figure 3a seems to suggest that the progression to more diverse community types was occurring at the same time in the two body sites. Could the authors please provide clarity for the justification of this statement.
-Similarly, in Lines 270-272 the authors describe bacterial crosstalk which promotes restoration of the respiratory microbiota. How do the authors propose that bacteria are translocated from the gut to the oropharynx, particularly if mucosal integrity is beginning to recover concurrent with restoration of the microbiome?
Minor comments: -Title should be revised to make it clear that this work specifically investigated bacterial communities, and not the broader microbiome (including viruses, archaea and fungi) -Line 30: It may not be accurate to say that this paper "addresses the question" of whether microbiomes "affect disease progression". Although the authors show longitudinal changes, they do not demonstrate causality or a functional role and should be cautious of over-emphasising their results -Line 36: replace "they had a" with "their -descriptions of "bacterial interactions" (e.g. line 36, line 193, etc) may be interpreted to mean that the authors have evidence of bacterial cells interacting, whereas the analysis performed only demonstrated correlations and co-occurrence. Revised word should be considered to avoid any confusion -Lines 49-55: Sentence is long and difficult to follow. Consider breaking it up into smaller sentences -Line 61: delete "but" -Please avoid using "significantly" where statistical analysis has not been applied (e.g. line 122 refers to a change in a single sample). Consider perhaps using "substantially" or a similar word to emphasis the magnitude of changes.
-Line 166: The authors state that that relative abundances of commensals "appeared to be" correlated. This term is ambiguous -if a correlation was statistically significant then the result shouls be stated more conclusively. If it was merely identified as an association or trend then the term "correlated" should be avoided as this is a specific type of statistical analysis. -Line 210: reference to "near-normal microbiota" may be difficult to define, especially in the gut samples where no samples were available from healthy controls. Perhaps rephrasing to "more diverse" or "greater abundance of commensals" may be a more accurate description -Line 217-219: authors describe that the respiratory microbiome is more easily affected by the infection, but also admit that this has not been examined yet. The first half of this sentence should be rephrased to demonstrate this is a hypothesis and not a know phenomenon -please describe how "mild" and "severe" cases of COVID-19 were distinguished in patients -Line 233: please change "baseline" to "early infection" or similar, to avoid any ambiguity about whether the samples were collected prior to infection. -Line 265: it seems contradictory to state that the gut microbiome is both "more stable" and "more plastic: than the respiratory microbiota. Please clarify -Line 306: please change "will be particularly useful" to "may be useful" as SCFA production is influence by factors other than microbiome composition (e.g. diet, cross-feeding, etc.) -Line 303 and 310: Please capitalise Pseudomonas

Reviewer #1 (Remarks to the Author):
The manuscript from Xu and colleagues describes temporal changes in microbiome composition during the course of SARS-CoV-2 infections. The study finds that the microbiome of COVID-19 patients can be classified into 3 or 4 categories (for throat and gut samples respectively), reflecting the transition from dysbiosis (Types III and IV) to a microbiota that is more similar to the one found in healthy individuals (Type I). A similar pattern was observed in the gut and in the respiratory tract. Overall, the manuscript addresses an important knowledge gap and brings a coherent story. Although the results are reasonable, I think the small number of samples prevents appropriate statistical tests to support the main conclusions of the paper. I understand that clinical data can be difficult to collect, but at very least the manuscript needs to be upfront about these limitations and should take them into consideration when interpreting the results. I have several major concerns that would need to be addressed before I can recommend this manuscript for publication. Authors: Thank you for your crucial comments and suggestions. We agree that the small number of patients is a major limitation of our paper, and appreciate your understanding in the difficulty of sample collection. We discussed this carefully in the Discussion section of the revised MS.
The main conclusions of this study are based on the diversity and composition of community types and their putative link with a dysbiosis-to-health transition. In particular, community types III and IV have a very low diversity and are dominated by a few seemly pathogenic species, which leads the authors to conclude that these samples are more severe cases of dysbiosis. However, those community types are also the ones with the lowest number of samples. In the respiratory tract dataset for example: community type IV has only 4 samples, while type II has 44, therefore it is logical that community type IV has a lower diversity. I suggest the diversity indices to be corrected for sample size before they are used as a dysbiosis marker. Authors: As suggested, we employed the Margalef's index (Clifford HT, Stephenson W. 1975. An introduction to numerical classification. London: Academic Press; Perspectives in marine Biology: 323-349) for comparing the a-diversity difference among community types. The Margalef's index is created to compensate for the effects of sample size by dividing the number of species in a sample by the natural log of the number of organisms collected. The sample size-controlled diversities reflected by Margalef's index are similar to those observed in our previous analyses (Fig. R1), indicating that sample size has minor influence on the microbiome diversity. We added the results of sample size-controlled diversities in both supplementary Fig. S3 and revised MS (Lines 89-90). The other argument was that community types III and IV are dominated by pathogenic bacteria, especially Pseudomonas and Rothia. The paper specifically argues that "Pseudomonas is a well-known pathogenic bacterium, and rarely found in healthy individuals." (Line 281). This is not correct, Pseudomonas can be pathogenic but it is also commonly found in healthy individuals. It is actually one of the most common bacteria in the saliva (Ruan, et al. 2020 "Healthy Human Gastrointestinal Microbiome: Composition and Function After a Decade of Exploration." Digestive Diseases and Sciences). Rothia, the most abundant bacteria in community type IV, is also a common member of a healthy oral microbiome (Uranga et al. 2020 "Commensal Oral Rothia mucilaginosa Produces Enterobactin, a Metal-Chelating Siderophore." MSystems). Therefore, I cannot see any evidence that the community types I-IV reflect a transition from health to dysbiosis. Authors: We thank the reviewer for correctly indicated that Pseudomonas and Rothia are normally non-pathogenic and can be detected in healthy individuals. We carefully checked the raw data and performed a blasting analysis against rRNA_typestrains/16S_ribosomal_RNA database from NCBI. The result showed that the identified Pseudomans species had highest sequence similarity (100%) to the nonpathogenic species Pseudomonas lactis (Fig. R3 in page 12). Pseudomonas lactis was initially isolated from bovine raw milk (Fig. R4 in page 13).
With respect to Rothia spp., they are Gram-positive coccobacilli that cause a wide In this study, we found that healthy controls and most COVID-19 patients carried very low abundance of Pseudomonas and Rothia, and therefore a high abundance of Pseudomonas lactis (often in bovine raw milk) and Rothia in community type III and IV are consistent with a probable disruption of homeostasis, as the result of inflammation that is known to exist in most COVID-19 patients. In view of the concerns and the knowledge in literature, we modified our description on the significance of the variation of these bacterial species in the revised manuscript. . There is also no information about the actual health status of the patients to support the link with health and dysbiosis. We can assume that the patients were admitted to the hospital in poor health and progressively got better, but I would expect that the improvement is not linear (i.e. did any of the patients got worse during their stay at the hospital, before getting better?) Authors: The reviewer is quite right to make these assumptions. The number of samples from dysbiotic-like community types are very small. In the respiratory tract for example, community type IV was only observed in 3 patients, including the one with severe symptoms. Overall, these numbers preclude making meaningful statistical tests to confirm if community types are associated with COVID-19 transition stages. I suggest considering using a continuous variable (rather than categories) to associate community diversity with COVID-19 progression, which might give more statistical power to address the study's questions. Authors: We agree that the small sample (patient) numbers (especially those with community type IV) is a major limitation of this study, and it is hard to draw a strong association between community types and COVID-19 transition stages. I found some inconsistencies in the description of the sampling that needs clarification: The paper reads "A total of 63 subjects, including 35 laboratory-confirmed COVID-19 patients, 10 SARS-CoV-2 negative patients with various diseases (non-COVID-19) and 15 healthy adults were enrolled in this study". 35 + 15 + 10 = 60, so what are the 3 extra subjects? Authors: We thank the reviewer for pointing out errors in counting. We have checked the numbers to make sure their consistency in the revised manuscript.
More importantly -Supplementary figure 1 indicates that some of the COVID-19 patients (e.g. p05, p11) never tested positive for COVID-19, yet they seem to be considered COVID-19 patients in the analyses (Figure 1). Please explain. Patients 05 and 11 are the only ones where community type IV was observed in early stages of disease, so this obviously needs clarification. Authors: First, all recruited patients were confirmed to have COVID-19 infection by the local CDC. The criterion for diagnosis of COVID-19 infection was positive for two or more different SARS-CoV-2 genes by RT-qPCR assay.
During the pandemic, China national and local CDC are responsible for prevention, control and management of COVID-19 infection. Some of the recruited patients (e.g. P05, P11, P13, P25, P27 etc.) were initially found/confirmed by local CDC in Nantong city, and then admitted, as required, to Nantong Third Hospital Affiliated to Nantong University for treatment and quarantine. At the time of sampling, some patients have recovered and become test-negative. Alternatively, the particular samples tested might have been false-negative.
We added more information on these patients in supplementary Fig. S1. In addition, we mentioned that community type IV was observed in early stages of disease in only two patients P05 and P11 although SARS-CoV-2 RNA has become negative in their throat swabs.
Supplementary methods are mentioned in the text (line 73) but were not provided. Authors: More detailed methodology is provided in the "Methods" section of the main text, rather than in a Supplementary file. We deleted "See Supplementary Methods" from this sentence.

Specific comments:
The paper states that all patients (except p09) had mild symptoms. Is it a normal procedure to hospitalize patients with mild symptoms? Authors: Yes, in China, all confirmed patients (including asymptomatic, mild or severe cases) are required to be admitted to designated hospitals for treatment and/or quarantine. L 82-83: "The vast majority of the specimens of COVID-19 patients were divided into four community types, called I-IV, and 6 specimens were included in the NP type". I did not understand this analysis. Was this a cross-validation? How did you proceed with the 6 specimens that were classified as "non-COVID-19 patients"? Authors: In our study, we included 10 SARS-CoV-2 RNA negative patients who were hospitalized because of other diseases, and designated them as non-COVID-19 patients (NP). In the DMM cluster analysis, samples of the NP group clustered together to form an independent cluster, and showed a community feature distinct from the groups of healthy control or COVID-19 patients.
To characterize each cluster, group distribution was analyzed and compared (new supplementary Fig. S2). The H cluster that only contains healthy controls represents normal throat microbiome. In the NP cluster, there was a significantly higher proportion of NP patients (55.6%) than COVID-19 patients (33.3%) and healthy controls (11.1%) (P<0.01). Thereby, the NP cluster is more likely representative of a diverse microbiome status associated with various other diseases (new supplementary Fig. S2). Whereas the COVID-19 patients are more homogeneous at a given disease stage, and their microbiome was mainly predominated by community type II regardless of in throat swabs (56.6%) and in anal swabs (53.3%).
Paragraph starting in L 108: I think this is an over-interpretation of the result. For example: "Prominent microbiome community type shifts from early lower-diversity community types (NP, IV or II) towards later higher-diversity types (II or I) were observed in 9/24 COVID-19 adults who had specimens at two or more time points." In other words -these transitions were observed in only 37% of the data. The graphs show that the results are not that obvious, and community shifts are far from "prominent". Authors: We appreciate this critical crucial comment, and toned down our conclusion accordingly in the revised manuscript.
Section starting on L 128: Were anal swabs also collected for healthy and non-covid-19 patients? Authors: The anal swabs were not collected from healthy and non-COVID-19 patients. L 164 -166: How were those bacterial genera selected? Authors: The top indicator bacteria with at least 0.5 indicator values were selected from each cluster ( Fig. 1d and 2d), and subjected to analyses with several core functional bacteria (e.g. Faecalibacterium, Lactobacillus and Bifidobacterium) in gut.
L 197 -200: I think this is an overstatement. It can be observed that some bacterial genera co-occur, but there too few samples and time points to make inferences about community succession. I suggest to also correct these networks for sample size. The authors could build networks of groups II and I with just 4 samples for example, to confirm that the increased complexity in these groups is not an artifact of sample size. Authors: This is a reasonable comment. According to the suggestion, we reanalyzed the 16S datasets from all 74 samples of 13 COVID-19 patients (corresponding to old Supplementary Fig. S3 or new Supplementary Fig. S4) by log transform the raw data before performing spearman correlation test. A total of 153 co-occurred pairs with |r| > 0.7 under FDR-adjusted P < 0.05 were identified and visualized by Cytoscape version 3.8.0. We presented the new co-occurrence networks in the revised manuscript. In the new cooccurrence networks, we found that 1) there were cross-talks of microbial compositions between and within host niches; 2) there was a competitive relationship between Gut-type-II and Gut-type-I mediated by significantly negative correlation between gut bacterial genera Neisseria and Bacteroides; 3) Bacteroides may modulate the cross-talk between Throat-type H and Gut-type-I (shown in page 15). L 313: the manuscript states "Lastly, fecal microbiota transplantation may be considered as another treatment choice." This is a dangerous statement to make. Considering the current circumstances, I can imagine people may try to perform fecal transplants at home to treat COVID-19 after reading this. There is no data in this manuscript to support that fecal transplants can be used to treat COVID-19. Authors: We appreciate the reviewer's critical comment and therefore deleted this sentence from the revised manuscript.

Reviewer #2 (Remarks to the Author):
In the present study, the authors used throat and anal samples from 35 COVID-19 adults and 15 controls to profile changes of the microbiome composition associated with COVID-19 infection by 16S rRNA gene sequencing. The authors claim to show alterations of respiratory and intestinal communities associated with the viral infection, which may potential impact on the outcome. Moreover, they claim having uncovered a new airway-gut microbial axis. The authors address interesting and important questions. However, I have serious concerns regarding lack of statistical power, choice of statistical methodology, lack of external validation and often times highly speculative or erroneous conclusions from the present results. Specific comments are listed below. Authors: We thank the reviewer for these critiques. We have carefully revised our manuscript according to all comments and suggestions as detailed below.
I have serious doubts that throat swabs represent "the respiratory microbiota", wouldn't specimens collected form deeper respiratory tract localizations better represent the airway microbiota (although I acknowledge that these are more difficult to collect). Authors: We revised the text to be more precise. Specifically, "respiratory microbiota" was more clearly written as "upper respiratory microbiota" in the revised manuscript.
Wot 35 cases and only 15(!) controls the study is severely limited by the sample sizes and the representability of the given results is highly questionable. Authors: We agree that sample size is a limitation of this study. Despite the small sample size, we discovered 13 of 19 healthy controls (throat swabs) can yield microbiomes clustered clearly different from that of other groups in the DMM modelling analysis ( Fig. 1a and 1b). However, we discussed this issue and cautioned the readers that our results may not be representative of all patient groups.
The community typing using DMM modelling is highly problematic as the models were trained on a very small cohort -and the validity of the inferred community types needs to be validated in an (sufficiently large) independent cohort before further downstream results are inferred based on assumptions derived from this model. Authors: Even with small sample size, this type of analyses have yielded interesting information previously. In the original paper that describes DMM modelling method (Holmes I, Harris K, Quince C. 2012. Dirichlet Multinomial Mixtures: Generative Models for Microbial Metagenomics. PLoS ONE 7: e30126.), the authors used two datasets to test the DMM availability. Based on the small dataset (78 samples) of inflammatory bowel disease (IBD) phenotypes, the authors found ileal Crohn's disease (ICD) was associated with a more variable community. In another study (Vandeputte D, et al. 2016. Stool consistency is strongly associated with gut microbiota richness and composition, enterotypes and bacterial growth rates. Gut 65: 57-62.), the authors used 16S DNA datasets from faecal samples of 53 healthy women for performing DMM modeling analysis and found strong associations between stool consistency and enterotype distribution. In our study, 112 throat swabs and 45 anal swabs were used to investigate the upper respiratory and gut microbial community type distribution in COVID-19 patients, respectively. We believed that the obtained results are reliable. Of course, we would like in the future to validate our current results with an increased sample size.
The separation in the ordination plots between the "community types" is not surprising given that the very inferred or separated beforehand on the same cohort with a different method (DMM), this only shows that the methods comply. The clusters could show up simply by chance, the reliability of the clusters/community types has to proofed using an independent cohort. Authors: We thank the reviewer for concurring the methodology is sound. In DMM modeling ( In fact, all our samples were classified as different clusters by using the minimum Laplace approximation value and posterior probabilities equal to 1. Of course, it would be better if another independent cohort is used to test the generalizability and reliability of our findings in future studies. Line 90, why were "the top 30 genera" chosen to represent the community types? This seems arbitrary to me, why not e.g. 50? Why isn't the number of genera rather chosen based on abundance or persistence, e.g. genera with at least 0.1% (or 1%) mean relative abundance and presence in at least 50% of the samples, this would be a more convincing definition of a kind of core microbiota. Authors: We selected the top genera number based on what is needed to form a separate community. We agree the selection is a bit arbitrary. To address, we also performed analyses with varying genera numbers.
In throat microbiota, the top 30 genera contributed to 66% of cumulative differences that are similar to total variations of all bacterial genera applied to DMM modeling. In anal microbiota, the top 30 genera reached up to 68% of cumulative differences. Those descriptions can be seen from "Indicator analysis in throat and gut community types" section in Methods. We also added related descriptions in the text of revised MS (Lines 100-114).
We also compared the difference between top30 (contributed to 66% of cumulative differences) and top40 genera (contributed to 72% of cumulative differences) for the identification of indicators of each throat microbial cluster. We found that the results were similar (Fig. R3) because five new identified indicators from the top40 genera had the average or median abundances higher than 1% only in 3-13% of samples but not in at least 50% of samples. So, those top30 genera representing at least 66% of cumulative differences is reliable to identify the indicator of throat & gut microbial clusters. The inference of any function or medical implication of identified bacteria here is highly problematic as the authors work with 16S data. E.g., line 96 ff. Pseudomans is not necessarily pathogenic -as by far not species or strains belonging to this genus exhibit pathogenic properties. Authors: We agree with the limitation when only 16S data were used. After checking the raw data by blasting analysis against rRNA_typestrains/16S_ribosomal_RNA database from NCBI, we found that the identified Pseudomans species had highest sequence similarity (100%) to the non-pathogenic species Pseudomonas lactis (  Line 164, the choice of "representative bacteria" is highly problematic, as indeed the functional relevance of throat bacteria in general, let alone in the context of COVID-19, is uncertain. Also, the term "probiotics" for Bifidobacterium and Faecalibacterium is inadequate, only certain strains of certain Bifidobacterium species are entailed in some commercially available probiotics, a (next-generation) probiotic potential of F. prausnitzii remains to be proven convincingly yet. Authors: We agree with the reviewer. First, in the revised MS, we selected the top indicator bacteria with > 0.5 indicator values from each cluster ( Fig. 1d and 2d) and several major core functional bacteria (e.g. Faecalibacterium, Lactobacillus and Bifidobacterium) in gut as the representative bacteria, and re-performed the analysis.
Second, the majority of the species of Bifidobacterium and Faecalibacterium are beneficial to host health. Anti-Inflammatory roles of gut bacteria At least, these studies suggested benefical roles of gut bacteria Faecalibacterium prausnitzii to host health, although some of these bacteria may not be called as probiotics. Accordingly, we revised the term "probiotics" as "potential beneficial bacteria". Generally, the throughout the text there is an uncritical use of the attribution of being beneficial or pathogenic, to date this ascription is possible for only very few bacteria based on genus-level taxonomic resolution alone. Authors: We agree with the reviewer. "Pathogenic" may be a relative concept. Some bacteria in healthy individuals might be pathogenic in immunocompromised individuals or under inflammatory status. For caution, we used relatively more neutral terms (e.g. potential pathogenic bacteria) in the revised MS.
The whole part on bacterial cooccurrence networks is highly dubious. As microbiome data generated by 16S sequencing is of compositional nature, appropriate correlation methods have to used here, such as SparCC https://bitbucket.org/yonatanf/sparcc/src/default/ Authors: As suggested, we re-performed the co-occurrence network analysis using the 16S data normalized by the centered log ratio transformation as described below.
At least before spearman correlation the 16S data should be transformed according to the centered log ratio transformation. https://www.frontiersin.org/articles/10.3389/fmicb.2017.02224/full?report=reader Authors: As suggested, we transferred the 16S datasets from all 74 samples of 13 COVID-19 patients according to the centered log ratio transformation before performing spearman correlation. A total of 153 co-occurred pairs with |r| > 0.7 under FDR-adjusted P < 0.05 were identified and visualized by Cytoscape version 3.8.0. In new co-occurrence network, we found 1) obvious cross-talks of microbial compositions between and with niches; 2) a competitive relationship between Gut-type-B and Guttype-A mediated by significantly negative interaction between gut bacterial genera Neisseria and Bacteroides, which might determine the microbiome shift from Gut-type-B to Gut-type-A during the progress of COVID-19; 3) Bacteroides may modulate the cross-talk between Throat-type H and Gut-type-A, which might be beneficial for the restoration of throat and gut microbiota associated with COVID-19. The  Also, the conclusion that any statistically significant correlations indicate "bacterial translocation" (line 194) is highly speculative. This hypothesis needs to be addressed with more sophisticated methods. Authors: We added some new results (e.g. serum LPS levels) and explanations in the revised MS, and toned down some conclusions including those on bacterial translocation.
From the given study design and based on the statistical methods applied, it cannot be concluded that microbiota is altered by COVID-19 and restored upon resolution of infection. First, because criteria for alteration in this context are very difficult to define, and second because in order to show that a longitudinal study is required to show changes in microbial composition during and after infection. Authors: Alteration in gut microbiota of COVID-19 patients have been previously reported (Zuo et al. 2020;Gu et al. 2020). In this study, by comparing with the healthy controls, we clearly found that the upper respiratory tract microbiotas of COVID-19 patients are altered in multiple levels of α-diversity (Fig. 1c), β-diversity (Fig. 1b), bacterial community types (Fig. 1a), and indicator bacteria (Fig. 1d), which are consistent with and support previous observations. In particular, we observed the dynamic changes of the microbiota profiles at both upper respiratory tract and gut during acute infection and recovery phase of COVID-19. It is clear that the bacterial community types (along with the α-diversity and indicator bacteria) have changed over time since COVID-19 infection. In particular, the most prevalent microbial community type associated with COVID-19 patients is Throat-Ⅱ, significantly distinct from that in healthy controls. Similarly, Gut-Ⅱ significantly associated with COVID-19 patients is significantly different from known healthy gut microbial community structures

Reviewer #3 (Remarks to the Author):
This is an interesting paper in an emerging field, and seeks to answer an important question concerning the gut-lung axis. The primary claims of the paper are that the throat and gut microbiota of COVID-19 infected adults can be characterised into different community types, beginning with a low diversity population soon after infection and a restoration back to diverse microbiota that in synchronous in the respiratory tract and the gut.
Overall, the researchers have done very well to draw conclusions from a limited dataset to provide timely information to assist with the current pandemic. Their attention to the precise details of the data and their longitudinal sampling is commendable. This work is very interesting and could provide valuable insight into the field and the long-term consequences of this novel disease, but additional considerations may help to improve the confidence of these conclusions: Authors: We thank the reviewer for these positive comments.

Major Comments:
-Patient characteristics of healthy adults and non-COVID-19 patients should be included as supplementary information. Authors: We provided the clinical information of the COVID-19 patients whenever possible, with the caveat that some demographic and clinical characteristics of healthy adults and non-COVID-19 patients are not available due to the emergency nature of the pandemic, and lock-down measure in some places.
-Care should be taken describing the results of throat swabs. The authors describe the sampling as coming from the "posterior oropharynx" (Line 323) which is only one part of the upper respiratory tract. Referring to results more generally as "respiratory tract" may give the impression that results are associated with changes in the lower respiratory tract, particularly given references to the airways and lung throughout the discussion (e.g. lines 288-291) Authors: We thank the reviewer for pointing out this point. We revised "respiratory tract" and "airway" into "upper respiratory tract" in the revised MS.
-Line 83: It was noted that 6 COVID samples clustered with the NP type samples. Was there anything to distinguish these samples from other COVID samples? Authors: In our study, 10 SARS-CoV-2 RNA negative patients who were hospitalized because of other diseases were included, and defined as non-COVID-19 patients (NP) group. In the DMM cluster analysis, all samples of the NP group clustered with those of 6 COVID-19 patients and two healthy controls, and formed a cluster, distinct from community features of majority of COVID-19 patients as well as healthy control. Because a significantly higher proportion (55.6%) of NP patients were included in this cluster than COVID-19 (33.3%) and healthy controls (11.1%) (P<0.01), this cluster represents more likely a diverse microbiome status associated with various other diseases (please see new supplementary Fig. S2 in page 8).
The 6 COVID-19 samples in NP cluster exhibited different bacterial characteristics from other COVID-19 samples, and can be distinguished from other COVID-19 samples by corresponding indictor bacteria (e.g. Enterobacteriaceae in NP type; other bacteria such as Neisseria, Pseudomonas etc. in community types I-IV) (Fig. 1d). The inclusion of 6 COVID-19 samples in the NP cluster might indicate that they shared similar microbiota profile to the NP group, and COVID-19 has diverse effects on microbiome.
-Authors discuss a return from low diversity to high diversity community types over time in throat samples, yet this was only observed in 9/24 patients (Line 112) while a reverse pattern was observed in four patients (Line 118-119). To my mind, this does not appear to demonstrate a clear trend that these patterns are consistent features of COVID-19 and additional sampling may be required to validate these findings. Alternatively, authors may be able to place additional emphasis on the association between the community types and time since appearance of symptoms in samples which were not a part of the longitudinal series but nevertheless support the hypothesis (e.g. p34, p18, p21, p28 in Fig 1e) Authors: We thank the reviewer's comment and suggestion. First, we toned down our conclusion on the association between a microbiota change and COVID-19 disease stage. Second, we performed the correlation relationship analysis between the community types and sampling time. There was no significantly association to be identified. However, when we analyzed the community type characteristics of all samples, we found that the majority of COVID-19 samples belonged to microbiome community type II regardless of in upper respiratory tracts or the gut (New supplementary Fig. S2, see page 8). In other words, COVID-19-associated microbiome was characterized by community type II. Furthermore, several lines of evidence also support the microbiome shift from community type II to type I during COVID-19 (Figs.  1e, 2e, 3a, and 4). We revised related sections in the revised MS. In spite of this, we believe that the alternations and dynamics changes of microbiome caused by COVID-19 are divergent among different individuals (please see New supplementary Fig. S6 in page 6) -Line 124-125: "These results indicate that the change of the respiratory microbiome might be closely associated with disease progression in COVID-19". The strength of this statement could be substantially improved by matching community types to some measure of clinical severity (e.g. symptom score, viral load, oxygen saturation, etc.) if this data is available. Authors: Thank you for this suggestion. We performed this analysis using available clinical parameters (e.g. total T lymphocyte, B lymphocyte, NK cells, CD4/CD8 ratios, etc.) (new supplementary Fig. S10: below). These results showed that there was no significant association between the microbiome diversity and clinical parameters except that upper respiratory microbiome richness appeared to negatively correlate with NK cell counts. Fig. S10. Correlation of microbiome diversity with clinical parameters.

New supplementary
-Line 161-163: Could the lack of correlation between community divergence and other parameters be due to the relatively small sample size available? Authors: Relatively small sample size might be a reason for weak or no correlation between community diversity and clinical parameters (new supplementary Fig. S10: above).
-Line 182-184: Authors discuss the hypothesis that co-occurrence networks between bacteria reflect crosstalk between the gut and lungs as a result of bacterial translocation, which is later described as a result of damage to the respiratory and gastrointestinal mucosa (Lines 194-195; Lines 261-262). Other processes may also explain these co-occurrences including (1) bacteria migrating from the oropharangeal site to the gut via swallowing and passage through the gastrointestinal tract (especially as several oral taxa are implicated) or (2) induction of immune responses at both sites which apply similar selective pressures to the microbiota. Could the authors please address these hypotheses. Authors: Thank the reviewer for these suggestions. We discussed these potential mechanisms.
-Similarly, is there any evidence available for a breakdown of mucosal barriers which may support the hypothesis of the authors (e.g. serum LPS or citrulline, etc.)? This may not be possible given the status of the patients but could provide valuable support for the hypothesis. Authors: Thank you for this suggestion. We measured the serum LPS levels of the COVID-19 patients. High levels of serum LPS were found in some patients such as P05, P09, P10, P13, etc, implying disruption of mucosal barriers and potential microbial translocation. In spite of high LPS levels in some patients, our data are more consistent with the interpretation that a cross-talk between the respiratory and gut microbiomes occurred more likely through respiratory and gastrointestinal tracts. Furthermore, we performed the correlation analysis, and found that the microbial richness index was negatively correlated with the serum LPS levels (New supplementary figure S11). Figure S11. Correlation between microbiome diversity and LPS levels.

New supplementary
-Line 202-204: Increased Bifidobacterium and Faecalibacterium in throat samples were noted as evidence of restoration in microbiome composition, yet to the best of my knowledge these genera are not considered to be a normal part of the oropharangeal microbiome. Can the authors please provide some citations to demonstrate that these genera are representative of a healthy oropharangeal microbiome.  (Fig. 1d and 2d), and were subjected to the analyses together with the several core functional bacteria (e.g. Faecalibacterium, Lactobacillus and Bifidobacterium) in gut.
Although Bifidobacterium and Faecalibacterium are not the normal part of the oropharangeal microbiome, the majority of the species of Bifidobacterium and Faecalibacterium are documented to be beneficial bacteria and can be used as probiotics. In fact, the relative abundance of Bifidobacterium and Faecalibacterium was indeed substantially lower in the upper respiratory tract than the gut (Fig. 3b), supporting that Bifidobacterium and Faecalibacterium are not the most common bacteria in oropharangeal microbiome. Although the increase in Bifidobacterium and Faecalibacterium in throat samples might not be noted as evidence of restoration of microbiome composition in upper respiratory tract, their increase in gut at least reflects the improvement and restoration of gut microbiome.
Several parameters, including α-diversity, bacterial community types, and representative bacteria, were used to reflect the restoration of the microbiota. According to our current results, the microbiotas of COVID-19 patients have not been restored to healthy type (Fig. 1c). Therefore, we presume that COVID-19 has a long-term consequence on health and deserve further follow-up investigation.
-Line 228-230: The authors indicate that inconsistencies in changes observed during longitudinal sampling indicate that diversity characteristics of the throat microbiome were affected by COVID-19. Can the authors please explain why they determined this direction of causality (i.e. is it possible that diversity characteristics may have rather influenced the progression of COVID-19 and not vice versa). Authors: Although it is difficult to draw a solid conclusion on the causality between SARS-CoV-2 infection and the altered microbiota, according to our results (comparison with the microbiota of healthy controls in diversity, community types and indicator bacteria) and previous studies (Zuo T, et  , it is more likely that SARS-CoV-2 infection first alters the microbiota in respiratory tract and the gut, and then the changed microbiota leads to a long-term influence on the health of these patients. In fact, the relationship between respiratory virus infection (e.g. influenza, RSV, rhinovirus, etc.) and microbiota has been investigated (Dubourg G, Edouard S, Raoult D. Relationship between nasopharyngeal microbiota and patient's susceptibility to viral infection. Expert Rev Anti Infect Ther. 2019 Jun;17(6):437-447). In this study, we provided further evidences to support the interaction between respiratory virus infection and microbiota. In particular, the dynamic changes of microbiota in both upper respiratory tract and gut over time since SARS-CoV-2 infection provide new insight into the understanding of mechanism of COVID-19 and the causality of virus infection and microbiota.
-Line 268-272: It is stated that the gut microbiome appeared to have a faster restoration to increased bacterial diversity, but this phenomenon did not appear to be described in the results section. Indeed, Figure 3a seems to suggest that the progression to more diverse community types was occurring at the same time in the two body sites. Could the authors please provide clarity for the justification of this statement.
-Similarly, in Lines 270-272 the authors describe bacterial crosstalk which promotes restoration of the respiratory microbiota. How do the authors propose that bacteria are translocated from the gut to the oropharynx, particularly if mucosal integrity is beginning to recover concurrent with restoration of the microbiome? Authors (to both questions above): We agree that the progression to more diverse community types was occurring at the same time in the upper respiratory tract and gut. Therefore, we corrected previous description into "synchronous occurrence" in the revised MS.
The co-occurrence networks analysis suggests the cross-talk between the respiratory and gut microbiomes (new Fig. 5). In fact, the detection of Bifidobacterium and Faecalibacterium (the core functional bacteria in gut) in the upper respiratory tract in spite of lower relative abundance might support the cross-talk. Although we also detected high level of serum LPS in some COVID-19 patients (supplementary Table  S2), we presume that the cross-talk between upper respiratory and gut microbiomes might be mediated through oropharyngeal-gastrointestinal tracts or diet.
Minor comments: -Title should be revised to make it clear that this work specifically investigated bacterial communities, and not the broader microbiome (including viruses, archaea and fungi) Authors: We emphasized "bacterial" microbiomes in title and some places of the main text.
-Line 30: It may not be accurate to say that this paper "addresses the question" of whether microbiomes "affect disease progression". Although the authors show longitudinal changes, they do not demonstrate causality or a functional role and should be cautious of over-emphasising their results Authors: We agree with reviewer. We edited related sentences and accordingly toned down some conclusions in the revised MS.
-Line 36: replace "they had a" with "their -Line 61: delete "but" -Lines 49-55: Sentence is long and difficult to follow. Consider breaking it up into smaller sentences -Please avoid using "significantly" where statistical analysis has not been applied (e.g. line 122 refers to a change in a single sample). Consider perhaps using "substantially" or a similar word to emphasis the magnitude of changes. -Line 210: reference to "near-normal microbiota" may be difficult to define, especially in the gut samples where no samples were available from healthy controls. Perhaps rephrasing to "more diverse" or "greater abundance of commensals" may be a more accurate description -Line 233: please change "baseline" to "early infection" or similar, to avoid any ambiguity about whether the samples were collected prior to infection. -Line 306: please change "will be particularly useful" to "may be useful" as SCFA production is influence by factors other than microbiome composition (e.g. diet, crossfeeding, etc.) -Line 303 and 310: Please capitalise Pseudomonas Authors (to eight questions above): Thank you very much for these editing suggestions. We did all these changes in the revised MS.
-descriptions of "bacterial interactions" (e.g. line 36, line 193, etc) may be interpreted to mean that the authors have evidence of bacterial cells interacting, whereas the analysis performed only demonstrated correlations and co-occurrence. Revised word should be considered to avoid any confusion Authors: We changed "interactions" into "co-occurrence".
-Line 166: The authors state that that relative abundances of commensals "appeared to be" correlated. This term is ambiguous -if a correlation was statistically significant then the result should be stated more conclusively. If it was merely identified as an association or trend then the term "correlated" should be avoided as this is a specific type of statistical analysis. Authors: We changed "correlated" into "associated".
-Line 217-219: authors describe that the respiratory microbiome is more easily affected by the infection, but also admit that this has not been examined yet. The first half of this sentence should be rephrased to demonstrate this is a hypothesis and not a known phenomenon Authors: The altered respiratory microbiome was previous observed during other respiratory virus infections (e.g. RSV, influenza etc.). So, we changed the sentence as "As an open system with direct contact with environment and the primary site for respiratory infections, the respiratory tract microbiota is more easily affected." However, the effect of SARS-CoV-2 infection has not been examined".
-please describe how "mild" and "severe" cases of COVID-19 were distinguished in patients Authors: We added the definitions and criteria of mild" and "severe" cases of COVID-19 into the Methods section.
-Line 265: it seems contradictory to state that the gut microbiome is both "more stable" and "more plastic: than the respiratory microbiota. Please clarify Authors: We used "plasticity" to express resilience of the microbiota. To avoid confusion, we replaced "plastic" by using "resilient".      Table   75 S1). Six throat microbial community types (or clusters) were identified using the Dirichlet 76 Multinomial Mixtures (DMM) modelling based on the lowest Laplace approximation (Fig. 1a) and 77 visualized by Nonmetric Multidimensional Scaling (NMDS) based on Bray-Curtis distance (Fig. 1b).

78
Thirteen of 19 specimens of healthy adults (H) formed an independent cluster defined as community 79 type H. The vast majority of the specimens of COVID-19 patients were divided into four clusters, 80 herein named community types I-IV (Fig. 1a). Other specimens from 6 COVID-19 patients were 81 clustered with those from 10 non-COVID-19 patients and two healthy controls. Because this cluster 82 has a significantly higher proportion of NP patients (55.6%, P<0.01) than COVID-19 patients (33.3%) 83 and healthy controls (11.1%) ( Supplementary Fig. S2), it was designated as community type NP. All 84 COVID-19-related community types, as well as the NP type, were significantly distant from the H 85 type. Community types III and IV were not only separated from the types I and II, but also from each 86 other (Fig. 1b). A decrease in alpha-diversity of the microbiome was observed from type I to IV, and 87 significantly lower richness and evenness were observed in community types III and IV, compared 88 with the H type (Fig. 1c). Similar decreasing trends of alpha-diversity were also observed when the 89 Margalef's indexes were used to control the effect of sample size (Supplementary Fig. S3) 15 . 90 To more directly demonstrate that the variation of throat microbial composition is an indicator of 91 COVID-19 disease stages, the community type-specific indicator taxa were identified based on the 92 top 30 microbial genera (Fig. 1d). The type H was characterized by bacterial genus Bacteroides (predominant taxa in the lung of healthy individuals) and unclassified Comamonadaceae, whereas 94 the NP type was marked by pro-inflammatory Enterobacteriaceae members. In contrast, the indicator 95 bacteria of four COVID-19-related community types were Alloprevotella in type I, Porphyromonas, 96 Neisseria, Fusobacterium and unclassified Bacteroidales in type II, Pseudomonas in type III, and 97 Saccharibacteria incertae sedis, Rothia and unclassified Actinomycetales in type IV (Fig. 1d).

98
Community type I contained Alloprevotella genus, as well as abundant Bacteroides and Prevotella 99 that typically present in the H type (Fig.1a). Some indicator bacteria substantially enriched in types 100 II and IV belong to opportunistic pathogenic bacteria that may be associated with human diseases  Table S1 and Fig. S4). Lower alpha-diversity with 112 enrichment of pro-inflammatory Enterobacteriaceae indicates that the type NP represents another 113 status of microbial dysbiosis.

114
According to indicator bacteria and alpha-diversity characteristics, the microbial community 115 types from I to IV may represent a progressive imbalance of the respiratory microbiome (Fig. 1c-d).

116
Among all throat specimens from COVID-19 patients, 47 (56.6%) belong to community type II 117 ( Supplementary Fig. S2), indicating that altered upper respiratory microbiome by COVID-19 was 118 mainly characterized by community type II. Longitudinal analysis showed that community types with 119 relatively lower alpha-diversity are more likely to have appeared in early specimens (Fig. 1e), but the 120 diversity did not significantly correlate with the time after symptom onset regardless of being 121 analyzed at the overall and individual levels ( Supplementary Fig. S5-S6). Among 22 COVID-19 122 adults who had specimens at two or more time points, over half (12, 54.5%) maintained a relatively 123 stable microbiome community types, and the others had community types altered over time. An 124 obvious throat microbiome recovery from types IV or II in early specimens to type I in late specimens 125 was observed in five patients (p17, p25, p13, p11 and p05) with 4 or more consecutive specimens 126 (Fig. 1e), accompanied with the restoration of throat microbiota, appearance of beneficial 127 commensals, and increased bacterial diversity (Supplementary Fig. S4). Conversely, an opposite 128 pattern was observed in four patients who had microbiome composition shift from early higher-129 diversity types (I or II) to later lower-diversity type (II-IV), implying a worsening of the throat 130 microbiome. In particular, the only severe case (p09) experienced a community type shift from type 131 I on day 10 to type IV on day 27, and sustained type IV to at least day 33 after symptom onset (Fig.   132 1e). Accompanied with this shift, opportunistic pathogenic bacteria Saccharibacteria incertae sedis 133 and Rothia were substantially enriched at late stage ( Supplementary Fig. S4). These indicate that the   Table S1). The gut microbial communities of COVID-19 patients formed three 141 distinct community types I-III ( Fig. 2a-b). The richness and evenness of the gut microbiome 142 decreased from type I to III (Fig. 2c). Indicator analyses showed that type I was primarily 143 characterized by healthy gut bacteria including Bacteroides genus and several known butyrate-144 producing bacteria (e.g. Faecalibacterium, Roseburia, Blautia, and Coprococcus) and one 145 opportunistic pathogenic bacterium (Finegoldia) (Fig. 2d) [24][25][26][27][28][29] . The indicators of type II mainly 146 contain various pathogenic or opportunistic pathogenic bacteria (e.g. Neisseria and Actinomyces). In 147 community type III, the gut microbiota was dominated by Pseudomonas, implying a severe dysbiosis. 148 We also used the community types I-III to examine the dysbiosis status of the gut microbiome.

149
A shift of the gut microbiome from the lower-diversity community type (II or III) towards a 150 higher-diversity type (I or II) was observed over time in 8/10 patients who had anal swabs at different 151 time points (Fig. 2e). Accompanied with the shift, a clear trend of increased bacterial diversity and 152 the relative abundance of beneficial commensals (e.g. Bacteroides and Faecalibacterium) was 153 observed in the gut microbiota from early to late stages of COVID-19 ( Supplementary Fig. S7), 154 suggesting a restoration of gut microbiota. Two patients maintained a stable microbiome community 155 types, and only one patient had an opposite shift of community type from higher-diversity type II to 156 lower-diversity community type III.

158
Association between the respiratory and gut microbiomes in COVID-19 159 Most paired throat and anal swabs showed the same or similar community type levels (Fig. 3). 160 In particular, the shift of microbiome community types over time appeared to match between the 161 throat and the gut in 7/8 patients who had two or more paired specimens at different time points (Fig.   162 3). Synchronous improvement of both the respiratory and gut microbiomes from early lower-diversity 163 community type towards late higher-diversity type occurred in six patients (p05, p17, p13, p11, p25 164 and p29). One patient (p33) experienced an improved respiratory microbiome but maintained an 165 unchanged gut community type up to day 24. One case (p07) had a worsen gut microbiome from day 166 24 to day 35 but maintained an unchanged respiratory community type. Because of no available anal 167 specimens, we were unable to assess whether the gut microbiota, like the respiratory microbiota, 168 shifted from higher-diversity type to lower-diversity type over time in the severe case (p09) (Fig. 1e).

169
Except for the duration of COVID-19, the upper respiratory and gut microbial community divergence 170 seemed not to be significantly associated with age, gender, antibiotics use, and detection of SARS-  Fig. S11 and Table S2). 177 We further selected the top indicator bacteria with > 0.5 indicator values from each community 178 type (Figs. 1d and 2d) and several major core functional bacteria (e.g. Faecalibacterium, 179 Lactobacillus and Bifidobacterium) in gut as the representative bacteria to assess their dynamic 180 changes in relative abundance over time (Fig. 4). In general, the relative abundances of 181 Bifidobacterium, Lactobacillus and/or Faecalibacterium appeared to be negatively associated with 182 the relative abundance of the opportunistic pathogens (e.g. Rothia and Neisseria), especially in the 183 gut microbiome. An obvious decrease in the relative abundance of opportunistic pathogenic bacteria 184 was accompanied by an increase in the relative abundance of resident commensals Bacteroides in gut 185 microbiome over time in five patients having three or more longitudinal samples (Fig. 4 and   186 Supplementary Fig. S4 and S7). Moreover, a substantially decreased abundance of Pseudomonas was 187 observed in both organs in another two patients (p23 and p29). The relative abundance of 188 Pseudomonas increased only in patient (p07) who experienced a worsening gut microbiome.  Table S2), suggesting the possibility of bacteria translocation.

201
To further investigate the association between the respiratory and gut microbiomes, we 202 performed co-occurrence network analysis using paired specimens from 13 patients. We constructed 203 a co-occurrence network consisting of a total of 153 co-occurred pairs with Pearson correlation |r| > 204 0.7 under FDR-adjusted P < 0.05 (Fig. 5). Bacteria in the same niche trended to have close co-205 occurrence relationship, and the cross-talks of microbial compositions between the upper respiratory 206 tract and the gut were also observed. In particular, a competitive relationship between Gut-type-II 207 and Gut-type-I was mediated by a significantly negative interaction between gut bacterial genera 208 Neisseria and Bacteroides (Fig. 5), which might determine the microbiome shift from Gut-type-II to 209 Gut-type-I during the COVID-19 disease progression. Furthermore, core resident commensals 210 Bacteroides appeared to mediate the cross-talk between Throat-type H and Gut-type-I, which might 211 modulate the restoration of throat and gut microbiota during COVID-19 27,32 .

213
Whether SARS-CoV-2 infection alters microbiota to affect COVID-19 disease progression is an 214 important question that needs answers. In this study, we made three major observations. First, the 215 upper respiratory and gut microbiota compositions of COVID-19 adults can be characterized by four 216 (I-IV) and three (I-III) community types, respectively, and these types reflect different levels of 217 balance between the more diverse microbiota (type I) and dysbiosis (type II-IV). Second, upper 218 respiratory and gut microbiome altered by COVID-19 are mainly characterized by community type 219 II, and the microbiome community types with lower alpha-diversity more likely appears in the early  Gut microbiota plays an important role in human health by shaping local immunity and remodeling 276 mucosal tissues 39 . It is relatively more stable and resilient than the respiratory microbiota, and it may 277 affect the latter by cross-talk between these two organs along the airway-gut axis 36,37 . In spite of 278 longer duration of SARS-CoV-2 shedding in the gut than in the respiratory tract, gut microbiota 279 appeared to have a synchronous change with the respiratory microbiota (Fig. 3). Although the 280 dynamic change of the microbiome was relatively divergent and independent of early microbiome 281 community types, synchronous restoration of both the respiratory and gut microbiomes from early 282 low diverse status towards late more diverse status was observed in 6 (75%) mild COVID-19 adult 283 patients who had two or more paired specimens at different time points. Age, gender and antibiotics 284 use seemed not to be linked to restoration of the microbiome, implying potential contributions from 285 other factors such as diet and genetic background.

286
As common bacteria in bovine raw milk, Pseudomonas was rarely detected in human 23 . Because 287 of specific antibiotics, it was not surprising that Pseudomonas-dominated bacterial community type 288 III was difficult to restore towards higher-diversity community types in the upper respiratory tract.

289
The identification of some opportunistic pathogenic bacteria (Neisseria,Porphyromonas,Rothia,290 Actinomycetales and Saccharibacteria) in more dysbiosis community types II and IV might imply a 291 need for microbiota-based personalized antibiotics treatment against these specific pathogens. As the 292 most common microbiome status in COVID-19 patients, the community type II represents a crucial   Fig S1).

415
"The ANOSIM statistic "R" compares the mean of ranked dissimilarities between groups to the mean 416 of ranked dissimilarities within groups. An R value close to "1.0" indicates dissimilarity between 417 groups, whereas an R value close to "0" indicates an even distribution of high and low ranks within 418 and between groups". The ANOSIM statistic R always ranges between −1 to 1. The positive R values 419 closer to 1 suggest more similarity within sites than between sites, and that close to 0 represent no 420 difference between sites or within sites 59 . ANOSIM p values that are lower than 0.05 imply a higher 421 similarity within sites. Richness (Observed OTUs/ASVs) and Pielou's or Species evenness for each 422 community type were calculated for estimating the difference of alpha-diversity. The analyses of 423 alpha-diversity, NMDS and ANOSIM were performed using R package "vegan" v2.5-6. Dynamic        The manuscript from Xu and colleagues has improved dramatically and most of my comments have been addressed. Please find below some suggestions that I believe will help to improve the clarity of the paper: Please add that all confirmed COVID-19 cases are hospitalized in China, even if they have no symptoms. This is not a common procedure in most countries, and being able to sample patients at the beginning of the infection is a massive advantage of this study.
Abstract (L38-39): the results of this study do not 'suggest that modulations of regional microbiota might help to improve the recovery of COVID-19 patients', as microbiome modulation experiments have not been performed here. I suggest rephrasing this sentence.
L46: This sentence indicates that only the elderly and people with comorbidities have severe COVID-19 symptoms, but we now know several cases of healthy individuals with severe reactions to the virus. Please rephrase.
L189: 'who experienced a worsening gut microbiome' -in terms of diversity? Or how do you define a worsening microbiome? L200-201: The LPS was a good addition but needs to be explained better. Please explain the association between LPS and your conclusions from these analysis.
L217: I don't think you can be so confident here. Maybe rephrase to 'these types possibly reflect different…' L219: The conclusion that 'COVID-19 are mainly characterized by community type II,' comes out of the blue here. It also gives the impression that community type II is the same in the gut and the respiratory microbiome. Maybe rephrase with more qualitative indications of what are the predominant features found in most COVID-19 patients. In order to make this study reproducible, please provide the analyses scripts/code as supplementary material. Very little information is given for the CLR transformation for example (e.g. what number was added (offset) to avoid log transformation with zeros?). These details can change the results, and therefore would ideally be provided with the R and/or Shell scripts.
Reviewer #2 (Remarks to the Author): I would like to thank the authors for meticulously addressing all of my questions and concerns. Although I still see important limitations of the present work, these are now more explicitly discussed and the authors have invested a lot of effort in solving the problems at hand with the best possible methodological approach. I support a publication of the manuscript in the present form.
Reviewer #3 (Remarks to the Author): I thank the authors for their time spent reviewing this manuscript and listening to suggestions, and apologise for my late response. I believe this manuscript has been significantly improved by acknowledging the limitations of their study more clearly and limiting the strength of their conclusions to better fit the data.
All line references are on "tracked changes" version. Major Comments: -BLAST scores (fig R2 and R4; lines 102-114; line 287) are used to identify select species of bacteria based on the samples with highest sequence similarity, yet the results do not reflect any level of selectivity in the results. E.g. in figure R4, the authors identify P. lactis as the species with highest sequence similarity, yet all other species listed also had 100% sequence identity with equal coverage and E value. It is not appropriate to claim a definitive identification based on these results, and the authors should rephrase the document to address the ambiguity. 16S sequencing does not have the sample level of specificity as metagenomics and results should be interpreted with more generality if the specificity is not available for a given taxa.
-Lines 150-157: the authors describe that 8/10 patients experienced a shift to higher diversity, but then list 2 patients who maintained stable communities and one who had an opposite shift. This is a total of 11 patients -please check these numbers. Similarly, in Lines 160-167, the authors describe 7/8 patients having synchronous changes in communities, but then describe 6 which had similar changes, and 2 which did not. Please check these numbers -Line 319-321: Please make the language less definitive. While the use of biomarkers is possible, the data presented here is not sufficient to suggest it. I suggest changing "can be" to "may potentially be". In fact, the whole concluding paragraph should be cautious not to overstate the therapeutic benefits given the limitations of this study as outlined in the rest of the discussion.