Diet and gut microbiome enterotype are associated at the population level in African buffalo

Studies in humans and laboratory animals link stable gut microbiome “enterotypes” with long-term diet and host health. Understanding how this paradigm manifests in wild herbivores could provide a mechanistic explanation of the relationships between microbiome dynamics, changes in dietary resources, and outcomes for host health. We identify two putative enterotypes in the African buffalo gut microbiome. The enterotype prevalent under resource-abundant dietary regimes, regardless of environmental conditions, has high richness, low between- and within-host beta diversity, and enrichment of genus Ruminococcaceae-UCG-005. The second enterotype, prevalent under restricted dietary conditions, has reduced richness, elevated beta diversity, and enrichment of genus Solibacillus. Population-level gamma diversity is maintained during resource restriction by increased beta diversity between individuals, suggesting a mechanism for population-level microbiome resilience. We identify three pathogens associated with microbiome variation depending on host diet, indicating that nutritional background may impact microbiome-pathogen dynamics. Overall, this study reveals diet-driven enterotype plasticity, illustrates ecological processes that maintain microbiome diversity, and identifies potential associations between diet, enterotype, and disease.

In previous analyses, enterotypes have been identified in the data by clustering analyses agnostic to metadata variables. For example, the original Arumugam et al paper use Partitioning Around Medoids (PAM) clustering to identify the enterotypes. See: https://enterotype.embl.de/enterotypes.html for a detailed explanation of the methods and R code. Although the data presented here in the PC plots strongly suggest that the samples do cluster into 2 groups, the PC axes alone (which do not explain all the variation in the underlying beta-diversity measures) are not sufficient to demonstrate this. I would recommend conducting the analyses outlined in the original Enterotype paper (link above). For example, it would be interesting to see what the optimal number of clusters is using Calinski-Harabasz indices. Because the enterotype categories fundamentally rely on the statistical methods employed, it would be best to replicate previous analyses as closely as possible. Then, it would be interesting to test whether clusters that emerge from the data without respect to metadata variables (e.g., diet) are associated with the metadata variables. I think this approach would be a more robust way of addressing the questions 1) do clusters exist in the data and 2) if yes, are the clusters associated with diet/infection status. The analyses already present in the paper could still be presented, as it is also of interest to test for associations between microbiota composition and metadata variables directly.
Fecal sampling: Please provide more information on fecal sampling. 8 hrs is a long window for samples to be sitting at ambient conditions before freezing during which time the composition of the microbial community is quickly changing. How did this time vary among samples? Is this variation associated with recovered composition? If this information is not available, this could be discussed as a potential caveat. Figure 1: If available, it would be interesting to include sample locations as points/stars/etc on the map (or as n='some number' for each sampling site), so that the spatial distribution of samples can be readily assessed by readers. Figure 2: It would be interesting to assess taxonomic/genus-level differences between enterotypes as identified by the Arumugam et al. analyses suggested above, in addition to the differences between feed regimes shown here. Figure 3: Similarly, it would be interesting to assess whether individuals shift between enterotypes (as identified by the Arumugam et al. analyses suggested above), and whether these shifts (if the exist) correspond to diet shifts. Figure 4: It was difficult to assess significance of the difference between positive and negative for each panel. Would it be possible to include stats in the figure and/or legend? As noted for Figures 2 and 3, it would be interesting here to test whether infection status is associated with enterotypes (as identified by the Arumugam et al. analyses suggested above).
Reviewer #2 (Remarks to the Author): Couch et al. have analyzed fecal microbiome composition collected over two years in African buffalo fed with supplemental diets. The framework of the study is interesting, the longitudinal data valuable, and the collected disease parameters useful. I don't have major criticisms of the paper, except a few things for the authors to consider (see below). However after reading through this mostly descriptive paper, I feel like I've not really learned anything particularly new about African buffalo gut microbiomes? For example, the study goes into detail showing that there are alpha and beta diversity differences between diet types, which is in itself not terribly surprising, but it doesn't describe clearly in what way they are different besides alpha and beta diversity measures? (see below) And how this is related to seasonal differences besides diet? In addition, I feel like the authors have a highly valuable dataset with collected disease parameters and body condition, but this section of the paper is unfortunately short and does not clearly outline what happens in the gut microbiome in individuals with disease.
Major comments I hope the authors are aware of the controversy surrounding the term Enterotypes. The initial paper describing enterotypes in human gut microbiomes has been extremely criticized, to the point where it is often used as an cautionary example of "how to not analyze microbiome data". I have personally no problem with using this word, and I believe it is up to the authors to decide how to best describe their results. But thinking how to best help the authors with their revision, it may be worth to at least consider if they still feel like this word is the best way to present the differences they find.
How precisely does the microbiome differ with the diet regimes? Which taxa are associated with which diet, and do the results agree with previous diet studies? Why does one of the diet types lead to lower diversity? How does this relate to seasonal changes? The authors have conducted Lefse analyses but do not really present these clearly, except refer to Figure 2a which is not explained particularly well. Is this presented somewhere else and I missed it?
A very interesting aspect of this paper is the association tests with diseases. I feel like this section could be stronger, however. What I've gathered from the presented results in figure 4 is that three diseases are associated with the gut microbiome. However, the authors could try to more clearly explain in what way the gut microbiome is different? Are there specific taxa associated with/without disease? What does it mean? Does it vary over the season or between the sexes? What about body condition and disease?
Minor comments L109: The words "each animal" has been accidentally repeated.
L157 & L160: All ASVs not identified to genus level were removed and the rest merged within genera. I'm curious as to the reasoning behind this approach? I have not encountered it previously. Do the authors not miss out on a large proportion of the microbiome data if they require such precise taxonomic assignment? Seeing how the current databases, Silva, GreenGenes, etc are populated with bacteria obtained primarily from model organisms such as humans and mice, a large proportion of unculturable bacteria from wildlife are likely not present with genus-level-specific information. Because of this concern, I'm interested to hear about the reasoning to only keep ASVs with genus information in a wild mammal. Related to this, a genus-specific approach also complicates the interpretation of the bar plots in Figure  2b. Do these proportions look very different when retaining the excluded genera? L176: PairwiseAdonis package is not cited. In addition, this package seems to be under development still? The developer writes: "This is still a developing version --results using interactions may not be right. Please validate." Are there any evidence and tests available that this function runs correctly as it says it might not be?
There's a large focus in the paper how the microbiome is associated with diet. However, diet differences are here directly associated with seasonal differences. Therefore, I'm curious how much seasonal changes play a role in the gut microbiome in addition to diet? For example there could be a large bacterial fluctuation present in the environment depending on whether it is dry or wet seasons, irrespective of diet. Several previous papers have found large associations between gut microbiome and season in wild mammals.
Figure 2: I'm not able to tell the differences between the brown, orange, and yellow colors in the bar plots. Consider using more distinct colors or fewer taxonomic groups in the legend. In addition, why only present 10 randomly selected samples from each feeding regime? I think all host individuals could be useful to present.
L367: The authors write that "These results imply that host population size and connectivity may be important for maintaining microbiome diversity within a population". My question is in what way have the authors analyzed associations with population size? I cannot find this approach in the paper. I appreciate the opportunity to read the manuscript submitted to Nature Communications by Couch et al. entitled, 'Diet drives gut microbiome enterotype shifts at the population level in wild African buffalo' and offer the following review: In this study, the authors describe a two-year study, with sampling every 2-3 months, to assess the impact of seasonal fluctuations in diet on gut microbiome composition, pathogen occurrence, and parasite burden in a population of wild African buffalo inhabiting Kruger National Park in South Africa. The sampling frequency (17 capture periods), sampling size (50-65 buffalo per capture period), and numerous types of samples collected and analyzed represent a considerable research effort.

General comments
My main critique is that I'm unsure if this research study, as it's currently presented, significantly advances our understanding of how dietary shifts influence microbiome composition in wild mammals and/or how pathogen/parasite infections influence or are influenced by microbiome composition. As the authors mention in their manuscript, it's well established in the microbiome literature that humans and other primates exhibit gut microbiome plasticity in response to seasonality in diet. How does this study contribute novel ecological insight, other than demonstrating seasonality in gut microbiome composition for wild buffalos? Secondly, though the authors found two gut microbial "enterotypes" representative of resource rich and resource poor dietary conditions, it's unclear if there was a formal statistical analysis to identify enterotypes or if microbiome samples collected during the hay/green vegetation or restricted feed periods were de facto assigned to separate enterotypes.
conditions? Could respiratory pathogen detection be confounded by environmental conditions and not necessarily be associated with gut microbiome composition?

Introduction
Line 71: I recommend including much more detail about the buffalo study population and sampling regime.
Line 84: It's unclear to me how respiratory pathogen infection would be directly influenced by changes in gut microbiome composition.

Methods
Line 96: Include the total number of capture periods here (according to Table 1, there are 17). How many individual animals were longitudinally sampled, for each type of sample? Line 109: "sedate" should be "sedated" and "opiod" should be "opioid" Line 157-158: What rRNA database was used to assign taxonomic classifications to ASVs? What percentage of ASVs in the dataset were not classified at the genus level? I am concerned that removing ASVs that are unclassified at the genus level is discarding a significant amount of diversity from the dataset, considering that rRNA taxonomy databases are biased towards human-associated bacteria. If the authors have not already done so, I recommend repeating some downstream analyses with the full diversity of ASVs included (i.e., ASV-level beta and alpha diversity across diet regimes, in addition to genus-level), especially if a large percentage of ASVs in the dataset are unclassified at the genus level.
Line 165: Are Bray-Curtis dissimilarities based on an ASV abundance table or a genus-level abundance  table? Line 168: PERMANOVA may be more appropriate because it has a more reliable Type I error rate compared to dbRDA (McArdle and Anderson, 2001). You can control for capture number using the "strata" argument.
Lines 174-176: Did the authors do a correction for multiple comparisons?
Line 186: Please include a bit more detail here so that readers do not need to look up the Flannery & Stagaman methods to follow this section. I recommend including an introductory sentence describing the aim of the CCA analysis and explicitly listing the different types of covariates considered, in addition to referring to table 2.
Lines 202 and 292: The equation has "otu_table" as the dependent variable. Should this actually be "genus table" (if genus level ASV counts were used) or ASV table (if individual ASV level counts were used)?

Results
Line 212: List the number of samples analyzed and specify the percentage of ASVs that were not classified at the genus level. If a large percentage of ASVs were not classified at the genus level and thus discarded, I recommend repeating downstream analyses (e.g., phylum-level LEfSe analysis) with all ASVs included and including results in the supplement.
Line 255: Does beta-diversity include both intra-individual and inter-individual pairwise sample comparisons?
Line 259: There is not information in the methods concerning how enterotypes were classified. Typically, a clustering analysis (e.g., PAM) is performed on the dissimilarity matrix to determine the optimal number of enterotypes (i.e., clusters). For example, see https://enterotype.embl.de/enterotypes.html or Hicks et al. (doi:10.1038/s41467-018-04204-w) "Identification of enterotype" under Statistical Analysis in "Methods" section. Did the authors subjectively decide that there is one enterotype for "restricted nutrition" and one enterotype for "high nutrition"? Given the amount of between-sample variation in the PCoA plot ( Fig. 3), there may be multiple enterotypes among the restricted diet microbiome samples.
Line 265: Include the statistical test associated with these p-values.
Line 298: Please include more detail. What was the direction of association for each diet regime and pathogen? Based on Figure 4, microbiome composition differences according to pathogen infection seem more pronounced in the restricted diet regime.

Discussion
See general comments at the beginning of my review.
Line 363: Does sociality (e.g., herd cohesion) change between resource rich and resource deficient time periods?  Tables   Table 1 Does each sample represent one individual or were multiple samples collected from the same individual during each capture period? Does number of samples refer to fecal samples or all types of samples? If 50-65 buffalo were captured for each sampling period, what was the decision process for which individuals/samples were included in the study? Define NDVI and include information for how NDVI were analyzed in the methods section. Table 2 Is there a difference between "incidence" and "status"? I recommend specifying that pathogens were dummy coded 1/0 for presence/absence or seroconverted/did not seroconvert (if this is correct). Does "burden" refer to the numbers of parasite worms and/or eggs counted for individual fecal samples? Supplement Figure S2: For acute and chronic respiratory pathogens, include the full names in the figure caption.

Reviewers' comments:
Reviewer #1 (Remarks to the Author): This study examines factors associated with the composition of the gut microbiota in wild African buffalo. Results include the identification of 'enterotype' clusters in the data associated with host diet, as well as associations between gut microbiota composition and respiratory pathogens. It is rare to see a study in which so many interesting data types (microbiome, diet, pathogen status, etc.) are analyzed, and the associations identified will be of broad general interest to the field. The manuscript is well-written and most of the analyses are well explained. However, there is one major issue regarding the identification of enterotypes, as it appears the authors have not tested for the emergence of microbiome clusters from the beta diversity results (revisions/additions suggested below). This issue should be addressed before publication.
In previous analyses, enterotypes have been identified in the data by clustering analyses agnostic to metadata variables. For example, the original Arumugam et al paper use Partitioning Around Medoids (PAM) clustering to identify the enterotypes. See: https://enterotype.embl.de/enterotypes.html for a detailed explanation of the methods and R code. Although the data presented here in the PC plots strongly suggest that the samples do cluster into 2 groups, the PC axes alone (which do not explain all the variation in the underlying beta-diversity measures) are not sufficient to demonstrate this. I would recommend conducting the analyses outlined in the original Enterotype paper (link above). For example, it would be interesting to see what the optimal number of clusters is using Calinski-Harabasz indices. Because the enterotype categories fundamentally rely on the statistical methods employed, it would be best to replicate previous analyses as closely as possible. Then, it would be interesting to test whether clusters that emerge from the data without respect to metadata variables (e.g., diet) are associated with the metadata variables. I think this approach would be a more robust way of addressing the questions 1) do clusters exist in the data and 2) if yes, are the clusters associated with diet/infection status. The analyses already present in the paper could still be presented, as it is also of interest to test for associations between microbiota composition and metadata variables directly.
We have included PAM clustering results and Calinski-Harabasz Index comparisons that demonstrate an optimal number of 2 clusters, consistent with our original manuscript. For the overwhelming majority of samples (85% of the 426 original samples), clusters align by dietary group -i.e. "high nutrition" (either green vegetation or supplemental) vs "restricted nutrition". The methods for this process have been added at lines [189][190][191][192][193][194] and results at lines 276-279. Our primary interest was in understanding microbiome variation within the context of environmental and dietary variation, therefore we opted to focus our downstream analyses around dietary groups rather than a priori data clusters.
Fecal sampling: Please provide more information on fecal sampling. 8 hrs is a long window for samples to be sitting at ambient conditions before freezing during which time the composition of the microbial community is quickly changing. How did this time vary among samples? Is this variation associated with recovered composition? If this information is not available, this could be discussed as a potential caveat.
We omitted from the original manuscript that samples were placed on ice within 15-30 minutes of collection. They were then frozen at -80 within 8 hours of collection. This detail has been added at line 137. Figure 1: If available, it would be interesting to include sample locations as points/stars/etc on the map (or as n='some number' for each sampling site), so that the spatial distribution of samples can be readily assessed by readers.
The vast majority of animals were corralled in a corner of the enclosure prior to capture and sampling, therefore sampling location would not be particularly informative for interpreting results of this study. This detail has been included in the caption for figure 1. Figure 2: It would be interesting to assess taxonomic/genus-level differences between enterotypes as identified by the Arumugam et al. analyses suggested above, in addition to the differences between feed regimes shown here.
As discussed above, enterotypes defined by PAM clustering were 85% aligned with diet, therefore comparing dominant genera between enterotypes did not reveal much additional information relative to what is already shown in figure 2. To informally illustrate this, we identified the top 10 most abundant genera within each of the PAM clusters. We found that 9 out of the top ten genera were shared between cluster 1 and the restricted nutrition group, 8 out of the top ten were shared between cluster 2 and the hay group, and 10/10 between cluster 2 and the green vegetation group. We have included envfit p-values on each panel, in addition to further explanation in the legend. Because we were primarily interested in how disease and microbiome composition interact across dietary regimes, we opted not to include the enterotype test suggested above. This decision was made partly for succinctness/clarity, and partly because of the criticisms and concerns about the validity of enterotypes in microbiome literature.

Reviewer #2 (Remarks to the Author):
Couch et al. have analyzed fecal microbiome composition collected over two years in African buffalo fed with supplemental diets. The framework of the study is interesting, the longitudinal data valuable, and the collected disease parameters useful. I don't have major criticisms of the paper, except a few things for the authors to consider (see below). However after reading through this mostly descriptive paper, I feel like I've not really learned anything particularly new about African buffalo gut microbiomes? For example, the study goes into detail showing that there are alpha and beta diversity differences between diet types, which is in itself not terribly surprising, but it doesn't describe clearly in what way they are different besides alpha and beta diversity measures? (see below) And how this is related to seasonal differences besides diet? In addition, I feel like the authors have a highly valuable dataset with collected disease parameters and body condition, but this section of the paper is unfortunately short and does not clearly outline what happens in the gut microbiome in individuals with disease.
We have restructured our narrative to emphasize what we believe are the most impactful findings of this study.

First, the unique circumstances of our study allowed us to separate season from diet and to demonstrate that gut microbiome plasticity associates with dietary fluctuations independent of seasonal environmental variation, which has not been possible in previous studies of the mammalian microbiome. The time frame of our study overlapped an extended period of drought, which resulted in dry season conditions continuing months beyond what is typical for the region.
During this time, buffalo were provided with supplemental feed, which resulted in gut microbiomes shifting to the wet season phenotype despite the extremely dry environmental conditions.

Third, as emphasized by this reviewer, our dataset provides valuable information on associations between the microbiome and disease. In addition to the associations we described between microbiome community and disease in our original submission, we have added results an analysis that explicitly links individual microbial taxa with disease.
Major comments I hope the authors are aware of the controversy surrounding the term Enterotypes. The initial paper describing enterotypes in human gut microbiomes has been extremely criticized, to the point where it is often used as an cautionary example of "how to not analyze microbiome data". I have personally no problem with using this word, and I believe it is up to the authors to decide how to best describe their results. But thinking how to best help the authors with their revision, it may be worth to at least consider if they still feel like this word is the best way to present the differences they find.
Yes, we are aware of the controversy surrounding this term, however we believe that in light of the common use of the term "Enterotype" within the microbiome research community, it is the most understandable way to describe the compositionally distinct clusters we identified in the dataset. In our analysis, we delineate the microbiome clusters based on diet regime rather than a priori defined enterotypes, therefore we do not believe our conclusions are overly reliant on controversial assumptions about the existence of enterotypes.
How precisely does the microbiome differ with the diet regimes? Which taxa are associated with which diet, and do the results agree with previous diet studies? Why does one of the diet types lead to lower diversity? How does this relate to seasonal changes? The authors have conducted Lefse analyses but do not really present these clearly, except refer to Figure 2a which is not explained particularly well. Is this presented somewhere else and I missed it?
We have attempted to clarify this by further discussion of taxonomic associations with diet at lines 411-427.
A very interesting aspect of this paper is the association tests with diseases. I feel like this section could be stronger, however. What I've gathered from the presented results in figure 4 is that three diseases are associated with the gut microbiome. However, the authors could try to more clearly explain in what way the gut microbiome is different? Are there specific taxa associated with/without disease? What does it mean? Does it vary over the season or between the sexes? What about body condition and disease?
As suggested, we have included results from generalized linear mixed models in table 4, which identified specific genera associated positively and negatively with infection. A discussion of these findings has been included at lines 440-453.
Minor comments L109: The words "each animal" has been accidentally repeated.
This has been corrected.
L157 & L160: All ASVs not identified to genus level were removed and the rest merged within genera. I'm curious as to the reasoning behind this approach? I have not encountered it previously. Do the authors not miss out on a large proportion of the microbiome data if they require such precise taxonomic assignment? Seeing how the current databases, Silva, GreenGenes, etc are populated with bacteria obtained primarily from model organisms such as humans and mice, a large proportion of unculturable bacteria from wildlife are likely not present with genus-level-specific information. Because of this concern, I'm interested to hear about the reasoning to only keep ASVs with genus information in a wild mammal.
Related to this, a genus-specific approach also complicates the interpretation of the bar plots in Figure 2b. Do these proportions look very different when retaining the excluded genera?
We used this approach because (a) relative abundances of unclassified ASVs were exceedingly low (median relative abundance = 2.34 e-07), therefore the bar plots in figure 2b look very similar (b) initial PAM clustering and diversity analyses were robust to the removal of these ASVs, with 98% of the samples falling into the same cluster regardless of whether ASV or genus-level clustering was used, and (c) removal of unidentified ASVs and merging by genus substantially decreased computational intensity.
L176: PairwiseAdonis package is not cited. In addition, this package seems to be under development still? The developer writes: "This is still a developing version --results using interactions may not be right. Please validate." Are there any evidence and tests available that this function runs correctly as it says it might not be?

According to the developer's website (https://github.com/pmartinezarbizu/pairwiseAdonis), the pairwise.adonis() function is fully functional. It is only pairwise.adonis2() that was still under development at the time of our original submission. We have included a citation at line 209.
There's a large focus in the paper how the microbiome is associated with diet. However, diet differences are here directly associated with seasonal differences. Therefore, I'm curious how much seasonal changes play a role in the gut microbiome in addition to diet? For example there could be a large bacterial fluctuation present in the environment depending on whether it is dry or wet seasons, irrespective of diet. Several previous papers have found large associations between gut microbiome and season in wild mammals.
As explained above, the unique circumstances of our study allowed us to separate season from diet and to demonstrate that gut microbiome plasticity associates with dietary fluctuations independent of seasonal environmental variation, which has not been possible in previous studies of the mammalian microbiome. The time frame of our study overlapped an extended period of drought, which resulted in dry season conditions continuing months beyond what is typical for the region. During this time, buffalo were provided with supplemental feed, which resulted in gut microbiomes shifting to the wet season phenotype despite the extremely dry environmental conditions. As shown in figure 2, the microbiome differences between wet season and feed-supplemented dry season samples were minimal, suggesting that the large associations between the gut microbiome and season are likely driven by dietary availability.
Figure 2: I'm not able to tell the differences between the brown, orange, and yellow colors in the bar plots. Consider using more distinct colors or fewer taxonomic groups in the legend. In addition, why only present 10 randomly selected samples from each feeding regime? I think all host individuals could be useful to present.
We have added all samples from all feeding regimes to the figure. We will change the color palette once we have received the editor's suggestions for best color alternatives for printing.
L367: The authors write that "These results imply that host population size and connectivity may be important for maintaining microbiome diversity within a population". My question is in what way have the authors analyzed associations with population size? I cannot find this approach in the paper.
While we did not directly assess the effects on population size, a potential connection is implied by our results. We found that the increase in population-level beta diversity during dietary restriction maintains population-level gamma diversity. This suggests that a large, wellconnected host population could be important for recolonizing and restoring individual-level alpha diversity following dietary restriction. This explanation has been added at lines 463-469.
Reviewer #3 (Remarks to the Author): I appreciate the opportunity to read the manuscript submitted to Nature Communications by Couch et al. entitled, 'Diet drives gut microbiome enterotype shifts at the population level in wild African buffalo' and offer the following review: In this study, the authors describe a two-year study, with sampling every 2-3 months, to assess the impact of seasonal fluctuations in diet on gut microbiome composition, pathogen occurrence, and parasite burden in a population of wild African buffalo inhabiting Kruger National Park in South Africa. The sampling frequency (17 capture periods), sampling size (50-65 buffalo per capture period), and numerous types of samples collected and analyzed represent a considerable research effort.

General comments
My main critique is that I'm unsure if this research study, as it's currently presented, significantly advances our understanding of how dietary shifts influence microbiome composition in wild mammals and/or how pathogen/parasite infections influence or are influenced by microbiome composition. As the authors mention in their manuscript, it's well established in the microbiome literature that humans and other primates exhibit gut microbiome plasticity in response to seasonality in diet. How does this study contribute novel ecological insight, other than demonstrating seasonality in gut microbiome composition for wild buffalos? Secondly, though the authors found two gut microbial "enterotypes" representative of resource rich and resource poor dietary conditions, it's unclear if there was a formal statistical analysis to identify enterotypes or if microbiome samples collected during the hay/green vegetation or restricted feed periods were de facto assigned to separate enterotypes.
The reviewer raises two main concerns: (1) lack of clarity regarding the impact and significance of this work, and (2) a need for more robust analysis to identify the existence of host enterotype. We have attempted to address (1) by emphasizing the following novel insights from this study: First, the unique circumstances of our study allowed us to separate season from diet and to demonstrate that gut microbiome plasticity associates with dietary fluctuations independent of seasonal environmental variation, which has not been possible in previous studies of the mammalian microbiome. The time frame of our study overlapped an extended period of drought, which resulted in dry season conditions continuing months beyond what is typical for the region. During this time, buffalo were provided with supplemental feed, which resulted in gut microbiomes shifting to the wet season phenotype despite the extremely dry environmental conditions. Second, our study demonstrates a potential ecological mechanism for the seasonal microbiome plasticity that is observed in wild mammals. While seasonal microbiome changes have been described previously, our large sample size and longitudinal design enabled us to demonstrate that the increase in population-level beta diversity during the dietary restriction maintains population-level gamma diversity despite seasonal loss of individual-level alpha diversity. This finding is significant because it offers an explanation for how microbiome diversity could be maintained over time in social mammals. While individual-level alpha diversity is lost during dietary restriction, individuals can later be recolonized by the microbes that are still present in other members of the population.
Third, as emphasized by this reviewer, our dataset provides valuable information on associations between the microbiome and disease. In addition to the associations we described between microbiome community and disease in our original submission, we have added results an analysis that explicitly links individual microbial taxa with disease.

Introduction
Line 71: I recommend including much more detail about the buffalo study population and sampling regime.

More detail has been added (lines 68-81)
Line 84: It's unclear to me how respiratory pathogen infection would be directly influenced by changes in gut microbiome composition.
This has been clarified at lines 96-98: "Changes to microbiome-pathogen relationships could be mediated by resource-driven variation in the immune system, similar to environmentally-driven changes observed in predator-prey dynamics (Bastille-Rousseau et al. 2017). Additionally, changes in host diet could alter competition for resources between commensal and pathogenic microbes, or enable facilitative interactions (DuBowy 1988)."

Methods
Line 96: Include the total number of capture periods here (according to Table 1, there are 17). How many individual animals were longitudinally sampled, for each type of sample?

Information on the number of captures and number of animals sampled are now included on line 107-111.
Line 109: "sedate" should be "sedated" and "opiod" should be "opioid" This has been corrected.
Line 157-158: What rRNA database was used to assign taxonomic classifications to ASVs? What percentage of ASVs in the dataset were not classified at the genus level? I am concerned that removing ASVs that are unclassified at the genus level is discarding a significant amount of diversity from the dataset, considering that rRNA taxonomy databases are biased towards human-associated bacteria. If the authors have not already done so, I recommend repeating some downstream analyses with the full diversity of ASVs included (i.e., ASV-level beta and alpha diversity across diet regimes, in addition to genus-level), especially if a large percentage of ASVs in the dataset are unclassified at the genus level.
Silva v 132 was used for classification. We opted to exclude unclassified genera because (a) relative abundances of unclassified ASVs were exceedingly low (median relative abundance = 2.34 e-07), (b) initial clustering and diversity analyses were robust to the removal of these ASVs, with 98% of the samples falling into the same cluster regardless of whether ASV or genus-level clustering was used, and (c) removal of unidentified ASVs and merging by genus substantially decreased computational intensity. Line 168: PERMANOVA may be more appropriate because it has a more reliable Type I error rate compared to dbRDA (McArdle and Anderson, 2001). You can control for capture number using the "strata" argument. This is an intelligent suggestion, however, we opted to use dbRDA rather than PERMANOVA because the latter is sensitive to differences in beta dispersion. Permutation tests demonstrated significant differences in beta dispersion between diet regimes, thus undermining our PERMANOVA results.
Lines 174-176: Did the authors do a correction for multiple comparisons?

Yes, we have added this information at line 210.
Line 186: Please include a bit more detail here so that readers do not need to look up the Flannery & Stagaman methods to follow this section. I recommend including an introductory sentence describing the aim of the CCA analysis and explicitly listing the different types of covariates considered, in addition to referring to table 2.

More detail has been provided at lines 222-228.
Lines 202 and 292: The equation has "otu_table" as the dependent variable. Should this actually be "genus table" (if genus level ASV counts were used) or ASV table (if individual ASV level counts were used)?
Yes, we have made the suggested change.

Results
Line 212: List the number of samples analyzed and specify the percentage of ASVs that were not classified at the genus level. If a large percentage of ASVs were not classified at the genus level and thus discarded, I recommend repeating downstream analyses (e.g., phylum-level LEfSe analysis) with all ASVs included and including results in the supplement.
Initial PAM clustering results were robust to the removal of these ASVs, with 98% of the samples falling into the same cluster regardless of whether all ASVs or known genera were used. Phylum-level LEfSe analysis of all ASVs classified to the phylum level identified 8/9 differentially abundant phyla among known genera. This information has been included at lines 273-274 and 297-301. Line 241: Should P>0.05 be P<0.05?

Yes, this has been corrected
Line 255: Does beta-diversity include both intra-individual and inter-individual pairwise sample comparisons?
Yes, this has been clarified.
Line 259: There is not information in the methods concerning how enterotypes were classified. Typically, a clustering analysis (e.g., PAM) is performed on the dissimilarity matrix to determine the optimal number of enterotypes (i.e., clusters). For example, see https://enterotype.embl. de/enterotypes.html or Hicks et al. (doi:10.1038/s41467-018-04204w) "Identification of enterotype" under Statistical Analysis in "Methods" section. Did the authors subjectively decide that there is one enterotype for "restricted nutrition" and one enterotype for "high nutrition"? Given the amount of between-sample variation in the PCoA plot (Fig. 3), there may be multiple enterotypes among the restricted diet microbiome samples.

We compared Calinski-Harabasz index values for 2 -20 clusters, which identified an optimum number of 2 clusters in both the ASV-level and genus-level datasets (lines 276-277).
Line 265: Include the statistical test associated with these p-values.
These results are based on permutation tests for homogeneity of multivariate dispersions. This information has been added at line 324-325.
Line 298: Please include more detail. What was the direction of association for each diet regime and pathogen? Based on Figure 4, microbiome composition differences according to pathogen infection seem more pronounced in the restricted diet regime.
Directionality is difficult to interpret in CCA results, therefore we included results from generalized linear mixed models assessing relationships between each bacterial genera and each of the three diseases of interest. These results are included in table 4 and at lines 370-376.

Discussion
See general comments at the beginning of my review.
The discussion has been revised substantially to emphasize the novel contributions of our study, as described above.
Line 363: Does sociality (e.g., herd cohesion) change between resource rich and resource deficient time periods?
Social dynamics likely change between feed regimes, however, we did not explicitly measure social interactions in this study.

Figures
General comment: Please include figures of the time series for host covariates and pathogen occurrence/burden over the course of the study (i.e., Fig. S2 expanded to show values for individual capture periods).
The requested figures (S3 and S4) have been added to the supplementary materials. Tables   Table 1 Does each sample represent one individual or were multiple samples collected from the same individual during each capture period?
We have clarified in the legend: "Each individual was sampled only once at each capture period." Does number of samples refer to fecal samples or all types of samples?
We have clarified in the column heading: "number of fecal samples".
If 50-65 buffalo were captured for each sampling period, what was the decision process for which individuals/samples were included in the study?
We have clarified in the methods and in the figure legend that all microbiome samples were included in alpha and beta diversity analyses and envfit analysis, but that only samples with complete covariate datasets were used in the CCA.
Define NDVI and include information for how NDVI were analyzed in the methods section.
NDVI has been defined in the table legend, and analysis has been included at lines 121-127 in the methods section. Table 2 Is there a difference between "incidence" and "status"? I recommend specifying that pathogens were dummy coded 1/0 for presence/absence or seroconverted/did not seroconvert (if this is correct).

Yes, pathogens are recorded as either presence/absence (chronic) or seroconversion status (acute). This information has been added to the table legend.
Does "burden" refer to the numbers of parasite worms and/or eggs counted for individual fecal samples?
Yes, burden refers to eggs per gram in individual fecal samples. This information has been added to the table legend.