Lamont J. WilkinspersName, Manoj MongapersName, Aaron W. MillerpersName (2019date). “Defining Dysbiosis for a Cluster of Chronic Diseases..” Scientific Reportstitle·j.
Object of study — the nature of dysbiosisterm·dom across a cluster of co-morbid chronic diseasesterm
Research question — whether diseases driven by dysbiosisterm·dom share metrics such as antibioticterm·key use, diversity change and common taxa
Contribution — evidence that a cluster of chronic diseasesterm shares an association with antibioticterm·key use and a loss of gut microbial diversityterm·key
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | increasing prevalence of many chronic diseases — “Many chronic diseases are on a trend of increasing prevalence.” | |
| claim-type | existence — “Furthermore, each disease state was associated with a loss of microbial diversity in the gut.” | |
| method — evidence | — “To address these questions, we conducted a meta-analysis of MWAS data from multiple sources for a cluster of disease states that included USD, diabetes, cardiovascular disease, obesity, and kidney disease.” | |
| object of study — evidence | — “Here we use multiple independent datasets to determine the nature of dysbiosis for a cluster of chronic diseases that includes urinary stone disease (USD), obesity, diabetes, cardiovascular disease, and kidney disease, which often exist as co-morbidities.” | |
| research question — evidence | — “We hypothesize that if the diseases in question are driven in part by dysbiosis, that there will be common metrics of dysbiosis that include antibiotic use, loss or gain of microbial diversity, and microbial taxa associated with disease.” | |
| contribution — evidence | — “Importantly, our results show that for the cluster of chronic diseases that span USD, diabetes, kidney disease, obesity, and cardiovascular disease, there is a statistically significant association with antibiotic use and a loss of microbial diversity.” | |
| mechanism | antibiotic use Dysbiosis driven by environmental factors such as antibiotic use is postulated to shift the microbiome in ways that increase inflammation and chronic disease onset. — “It has been postulated that dysbiosis driven by environmental factors such as antibiotic use is shifting the microbiome in ways that increase inflammation and the onset of chronic disease.” | mechanisms[0] |
| sense | dysbiosis (abstract) - the loss of bacteria that promote health or the gain of bacteria that promote disease — “Dysbiosis can be defined through the loss or gain of bacteria that either promote health or disease, respectively.” | senses[0] |
| effect direction | dysbiosis —causes→ chronic disease — increases — “While a number of genetic and environmental factors contribute to the manifestation of chronic disease, an emerging hypothesis postulates that dysbiosis, an imbalance in the composition and metabolic capacity of our microbiota, increases the risk of developing chronic disease 10 – 13 .” | concept_edges[0] |
| relation | dysbiosis —causes→ chronic disease — While a number of genetic and environmental factors contribute to the manifestation of chronic disease, an emerging hypothesis postulates that dysbiosis, an imbalance in the composition and metabolic capacity of our microbiota, increases the risk of developing chronic disease 10 – 13 . — “While a number of genetic and environmental factors contribute to the manifestation of chronic disease, an emerging hypothesis postulates that dysbiosis, an imbalance in the composition and metabolic capacity of our microbiota, increases the risk of developing chronic disease 10 – 13 .” | concept_edges[0] |
| weight | dysbiosis —causes→ chronic disease — 0.6 — “While a number of genetic and environmental factors contribute to the manifestation of chronic disease, an emerging hypothesis postulates that dysbiosis, an imbalance in the composition and metabolic capacity of our microbiota, increases the risk of developing chronic disease 10 – 13 .” | concept_edges[0] |
| relation | loss of function dysbiosis —contrasted-with→ gain of function dysbiosis — In that meta-analysis, it was found that some diseases such as enteric diarrheal disease, Clostridium difficile infection, and IBD were associated with loss of function dysbiosis, whereas diseases such as colorectal cancer, autism spectrum disorder, liver diseases, and Parkinson’s disease were associated with gain of function dysbiosis 64 . — “In that meta-analysis, it was found that some diseases such as enteric diarrheal disease, Clostridium difficile infection, and IBD were associated with loss of function dysbiosis, whereas diseases such as colorectal cancer, autism spectrum disorder, liver diseases, and Parkinson’s disease were associated with gain of function dysbiosis 64 .” | concept_edges[1] |
| weight | loss of function dysbiosis —contrasted-with→ gain of function dysbiosis — 0.8 — “In that meta-analysis, it was found that some diseases such as enteric diarrheal disease, Clostridium difficile infection, and IBD were associated with loss of function dysbiosis, whereas diseases such as colorectal cancer, autism spectrum disorder, liver diseases, and Parkinson’s disease were associated with gain of function dysbiosis 64 .” | concept_edges[1] |
| relation | urinary stone disease —same-as→ USD — Cardiovascular disease, obesity, diabetes, urinary stone disease (USD), asthma, and inflammatory bowel disease (IBD) are all on the rise, among others 1 – 6 . — “Cardiovascular disease, obesity, diabetes, urinary stone disease (USD), asthma, and inflammatory bowel disease (IBD) are all on the rise, among others 1 – 6 .” | concept_edges[2] |
| weight | urinary stone disease —same-as→ USD — 0.9 — “Cardiovascular disease, obesity, diabetes, urinary stone disease (USD), asthma, and inflammatory bowel disease (IBD) are all on the rise, among others 1 – 6 .” | concept_edges[2] |
| relation | metagenome-wide association studies —same-as→ MWAS — In the last ten years, there has been an explosion of metagenome-wide association studies (MWAS) that seek to determine if the microbiome contributes to disease processes. — “In the last ten years, there has been an explosion of metagenome-wide association studies (MWAS) that seek to determine if the microbiome contributes to disease processes.” | concept_edges[3] |
| weight | metagenome-wide association studies —same-as→ MWAS — 0.9 — “In the last ten years, there has been an explosion of metagenome-wide association studies (MWAS) that seek to determine if the microbiome contributes to disease processes.” | concept_edges[3] |
| relation | operational taxonomic units —same-as→ OTUs — Each individual harbors thousands of unique microbial operational taxonomic units (OTUs) 28 , 29 . — “Each individual harbors thousands of unique microbial operational taxonomic units (OTUs) 28 , 29 .” | concept_edges[4] |
| weight | operational taxonomic units —same-as→ OTUs — 0.9 — “Each individual harbors thousands of unique microbial operational taxonomic units (OTUs) 28 , 29 .” | concept_edges[4] |
| replication | original | |
| limitation | Longitudinal studies are needed to confirm that diversity loss causes chronic disease risk. — “However, longitudinal studies are needed to confirm the causative effect of diversity loss for chronic disease risk.” | limitations[0] |
| limitation | Health status rests on self-reported diagnoses, which may bias the results. — “Health status for each defined term were based on self-reported medical diagnoses which may bias results 41 .” | limitations[1] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | In every chronic disease state, individuals were significantly more likely than healthy individuals to have taken antibiotics in the last year. — “For all chronic disease states, individuals were significantly more likely to have taken antibiotics in the last year compared to healthy individuals (Fig. 1A ).” | content_tags.findings[0] |
| quantity | 1468 — — “The entire combined dataset spanned a total of 1468 samples (Table 1 ).” | content_tags.quantities[0] |
| key-terms | dysbiosis, microbiome, antibiotics, chronic disease, microbial diversity, urinary stone disease, obesity, diabetes, operational taxonomic units, meta-analysis |
| concept | dysbiosis — an imbalance in the composition and metabolic capacity of our microbiota — “While a number of genetic and environmental factors contribute to the manifestation of chronic disease, an emerging hypothesis postulates that dysbiosis, an imbalance in the composition and metabolic capacity of our microbiota, increases the risk of developing chronic disease 10 – 13 .” | concepts[0] |
| concept | chronic disease — not defined in this paper's own words; long-lasting illness such as diabetes, obesity or kidney disease — “While a number of genetic and environmental factors contribute to the manifestation of chronic disease, an emerging hypothesis postulates that dysbiosis, an imbalance in the composition and metabolic capacity of our microbiota, increases the risk of developing chronic disease 10 – 13 .” | concepts[1] |
| concept | gain of function dysbiosis — pathogens and their functions can be acquired or opportunistically overgrow to promote disease — “First, pathogens and their functions can be acquired or opportunistically overgrow to promote disease, termed gain of function dysbiosis.” | concepts[2] |
| concept | loss of function dysbiosis — health-protective bacteria and their functions may be lost or suppressed, which then promotes the onset of disease — “Second, health-protective bacteria and their functions may be lost or suppressed, which then promotes the onset of disease, termed loss of function dysbiosis.” | concepts[3] |
| concept | urinary stone disease — not defined in this paper's own words; the formation of stones in the urinary tract — “Cardiovascular disease, obesity, diabetes, urinary stone disease (USD), asthma, and inflammatory bowel disease (IBD) are all on the rise, among others 1 – 6 .” | concepts[4] |
| concept | USD — urinary stone disease — “Cardiovascular disease, obesity, diabetes, urinary stone disease (USD), asthma, and inflammatory bowel disease (IBD) are all on the rise, among others 1 – 6 .” | concepts[5] |
| concept | metagenome-wide association studies — not defined in this paper's own words; studies relating the gut microbiome to disease across whole microbial genomes — “In the last ten years, there has been an explosion of metagenome-wide association studies (MWAS) that seek to determine if the microbiome contributes to disease processes.” | concepts[6] |
| concept | MWAS — metagenome-wide association studies — “In the last ten years, there has been an explosion of metagenome-wide association studies (MWAS) that seek to determine if the microbiome contributes to disease processes.” | concepts[7] |
| concept | operational taxonomic units — not defined in this paper's own words; clusters of similar gene sequences used as stand-ins for microbial species — “Each individual harbors thousands of unique microbial operational taxonomic units (OTUs) 28 , 29 .” | concepts[8] |
| concept | OTUs — operational taxonomic units — “Each individual harbors thousands of unique microbial operational taxonomic units (OTUs) 28 , 29 .” | concepts[9] |
Full text — tagged
Defining Dysbiosisterm·dom for a Cluster of Chronic Diseasesterm.
Abstract The prevalence of many chronic diseasesterm has increased over the last decades. It has been postulated that dysbiosisterm·dom driven by environmental factors such as antibioticterm·key use is shifting the microbiometerm·key in ways that increase inflammation and the onset of chronic diseaseterm. Dysbiosisterm·dom can be defined through the loss or gain of bacteria that either promote health or disease, respectively. Here we use multiple independent datasets to determine the nature of dysbiosisterm·dom for a cluster of chronic diseasesterm that includes urinary stone diseaseterm·dom (USDterm·dom), obesityterm·key, diabetesterm·key, cardiovascular disease, and kidney disease, which often exist as co-morbidities. For all disease states, individuals exhibited a statistically significant association with antibioticsterm·key in the last year compared to healthy counterparts. There was also a statistically significant association between antibioticterm·key use and gut microbiota composition. Furthermore, each disease state was associated with a loss of microbial diversityterm·key in the gut. Three genera, Bacteroides, Prevotella , and Ruminococcus , were the most common dysbiotic taxa in terms of being enriched or depleted in disease populations and was driven in part by the diversity of operational taxonomic unitsterm (OTUsterm) within these genera. Results of the cross-sectional analysis suggest that antibiotic-driven loss of microbial diversityterm·key may increase the risk for chronic diseaseterm. However, longitudinal studies are needed to confirm the causative effect of diversity loss for chronic diseaseterm risk.
Subject terms: Metagenomics, Microbial ecology Introduction Many chronic diseasesterm are on a trend of increasing prevalence. Cardiovascular disease, obesityterm·key, diabetesterm·key, urinary stone diseaseterm·dom (USDterm·dom), asthma, and inflammatory bowel disease (IBD) are all on the rise, among others 1 – 6 . These chronic inflammatory diseases often exist as co-morbidities with common physiological manifestations, which both compound the burden on patients and suggests common origins 7 – 9 . While a number of genetic and environmental factors contribute to the manifestation of chronic diseaseterm, an emerging hypothesis postulates that dysbiosisterm·dom, an imbalance in the composition and metabolic capacity of our microbiota, increases the risk of developing chronic diseaseterm 10 – 13 . Dysbiosisterm·dom can contribute to the onset of chronic diseaseterm in one of three general ways. First, pathogens and their functions can be acquired or opportunistically overgrow to promote disease, termed gain of function dysbiosisterm·dom. Gain of function dysbiosisterm·dom leads to infectious diseases such as cholera or streptococcal pharyngitis, but can also lead to chronic inflammation 14 – 16 . Second, health-protective bacteria and their functions may be lost or suppressed, which then promotes the onset of disease, termed loss of function dysbiosisterm·dom. Loss of function dysbiosisterm·dom has been linked to chronic diseasesterm such as IBD, USDterm·dom, obesityterm·key, and others 17 – 26 . Finally, some combination of loss and gain of function dysbiosisterm·dom may be required for the onset of disease, such as with recurrent Clostridium difficile infection 27 .
The complexity of the microbiometerm·key presents unique challenges to our understanding of the role of dysbiosisterm·dom in disease. Each individual harbors thousands of unique microbial operational taxonomic unitsterm (OTUsterm) 28 , 29 . Compounding this issue is the considerable inter-individual variability in the composition of the microbiometerm·key 30 , 31 . Given the hyper-variable nature of high-throughput 16S rRNA data combined with the high levels of inter-individual variability, any two random populations of individuals are likely to harbor a proportion of OTUsterm unique to each population, with no relevance to any phenotype. To help restrict false-discoveries associated with this issue, false-discovery rate corrections are commonly integrated into microbiometerm·key studies 32 . Despite the limitations of microbiometerm·key studies, many clinical studies use the differential abundance of OTUsterm between healthy and disease populations at a single time-point to determine if dysbiosisterm·dom contributes to the disease, which can lead to erroneous results when it comes to the nature of dysbiosisterm·dom and microbial taxa involved. Thus, a number of questions remain surrounding conclusions that dysbiosisterm·dom contributes to the manifestation of disease. First, is the differential abundance of OTUsterm between two populations driven by inter-individual variability of the two populations or disease status? Second, what features of microbial taxa contribute to them being identified as dysbiotic? Finally, are there taxa consistently found to be dysbiotic across chronic diseaseterm states, indicative of common, dysbiosis-driven disease processes?
To address these questions of defining dysbiosisterm·dom, we performed an independent analysis of microbiometerm·key data from multiple sources that spanned a cluster of chronic diseasesterm that included USDterm·dom, cardiovascular disease, obesityterm·key, diabetesterm·key, and kidney disease. Specifically, we quantified the use of antibioticsterm·key in disease populations compared to healthy controls, which can lead to a loss in microbial diversityterm·key relevant to disease processes 33 – 40 , quantified metrics of loss or gain of microbiota diversity associated with each disease state relative to stochastically defined populations, determined the similarities and differences in microbial taxa enriched and depleted in disease populations, and determined which of these taxa were differentially abundant more than expected given their genus-level diversity in samples. These diseases often exist as comorbidities in patients and have all been linked to the microbiometerm·key to some degree. We hypothesize that if the diseases in question are driven in part by dysbiosisterm·dom, that there will be common metrics of dysbiosisterm·dom that include antibioticterm·key use, loss or gain of microbial diversityterm·key, and microbial taxa associated with disease.
Results Datasets For cardiovascular disease, obesityterm·key, diabetesterm·key, kidney disease, and healthy counterparts, 16S rRNA sequences from the stool and associated sample metadata were drawn from the April 26 th , 2017 version of American Gut Project data (AGP), which has an extensive list of metadata that includes antibioticterm·key history and the presence or absence of several disease states 41 . Samples for sequencing were collected between December 2012 and April 2017 from individuals from a global population. Healthy individuals were defined by the “I do not have this condition” entry for diabetesterm·key, cardiovascular disease, and kidney disease, as well as “normal” for body mass index (“BMI_CAT” in the metadata). Those with “diagnosed” for cardiovascular disease, diabetesterm·key, or kidney disease, or with “obese” for BMI_CAT were assigned to the appropriate chronic diseaseterm state. Health status for each defined term were based on self-reported medical diagnoses which may bias results 41 . A random number generator was used to assign random numbers to each sample that fit one of the defined health states for the AGP data and the samples with the lowest 300 randomly assigned numbers for each health status were extracted for further analysis, except for kidney disease which only had 111 entries. For kidney disease, we extracted samples with the lowest 100 numbers for further analysis. For USDterm·dom, which is not represented in the AGP metadata, 16S rRNA sequences and associated sample metadata were downloaded from all publicly available metagenomic data from clinical USDterm·dom studies published at the time of analysis 18 , 20 , 42 , 43 . Samples for USDterm·dom studies were collected between 2015–2019 and originated in the United States, Canada, India, or China. Samples were defined as healthy or USDterm·dom based on the criteria used in the original study. The entire combined dataset spanned a total of 1468 samples (Table 1 ).
Characteristics of datasets used in the meta-analysisterm·mth.
Association of antibioticsterm·key and chronic diseaseterm For all chronic diseaseterm states, individuals were significantly more likely to have taken antibioticsterm·key in the last year compared to healthy individuals (Fig. 1A ). Furthermore, antibioticterm·key use in the last year was significantly associated with microbiota composition as assessed by a weighted UniFrac analysis followed by a post-hoc PERMANOVA of the regularized AGP datasets (Fig. 1B ) 20 , 44 .
The effect of antibioticsterm·key on chronic diseaseterm and the microbiota. ( A ) Antibioticterm·key use within the last year for individuals with or without chronic diseaseterm. For diabetesterm·key, cardiovascular disease, kidney disease, obesityterm·key, and their healthy counterparts, antibioticterm·key history was derived from the subset of AGP samples randomly selected for this study (N = 300 for each group except kidney disease which had 100 samples; Table 1 ). Only one study on the microbiometerm·key of USDterm·dom patients included metadata associated with antibioticterm·key use (N = 43 healthy individuals and 24 individuals with USDterm·dom; Table 1 ). Proportions of antibioticterm·key use were compared between chronic diseaseterm states and healthy populations with a relative risk ratio followed by a post-hoc Fisher’s exact test, which was Holm’s corrected for multiple comparisons. *p < 0.05; **p < 0.01; ***p < 0.001 compared to the healthy population. ( B ) PCoA plot based on a weighted UniFrac analysis the microbiometerm·key composition from the AGP data. Community composition based on antibioticterm·key use was compared by PERMANOVA with 999 permutations. p = 0.006 .
Nature of dysbiosisterm·dom in chronic diseaseterm The differential abundance of OTUsterm between healthy and disease populations was used to determine if each of the disease states were associated with a depletion or enrichment of microbial diversityterm·key compared to healthy controls, with the number of OTUsterm enriched or depleted in disease quantified for each pairwise comparison (Fig. 2 ). The average fold difference in OTUsterm enriched in healthy cohorts vs. disease cohorts reveals that for each disease state, there was significantly more OTUsterm enriched in healthy cohorts than disease cohorts, indicative of a loss of microbial diversityterm·key in the gut (Fig. 3 ). No significant bias for group 1 or 2 was detected when health status was stochastically assigned to samples through random assignment (see methods).
An example of differential abundance analysis for each of the disease states. Each dot represents an OTUterm. Grey dots are OTUsterm that do not exhibit significant differential abundance, while red dots are differentially abundant OTUsterm. ( A ) Stochastic metadata; ( B ) USDterm·dom; ( C ) Cardiovascular disease; ( D ) Diabetesterm·key; ( E ) Obesityterm·key; ( F ) Kidney disease.
Average fold difference in the number of OTUsterm enriched in either the healthy group/stochastic group 1 or in the disease group/stochastic group 2. Positive values reflect greater enrichment in healthy group/stochastic group 1, whereas negative values reflect greater enrichment in disease group/stochastic group 2. Significance was determined with a one-sample t-test against an expected value of 1. *p < 0.05; **p < 0.01; ***p < 0.001.
Strength of dysbiotic taxa and commonalities across chronic diseaseterm states The proportion of independent healthy x disease comparisons in which a particular genus had at least one OTUterm that was differentially abundant was plotted on a heatmap (Fig. 4 ). The heatmap reveals that the most common dysbiotic genera were the Coprococcus, Prevotella , and Bacteroides for OTUsterm enriched in the healthy cohorts. For OTUsterm enriched in disease cohorts, the Bacteroides, Ruminococcus , and Blautia genera were most common. Hierarchal clustering reveals statistically significant similarities between diabetesterm·key and kidney disease when considering potential health protective bacteria lost. When considering potential pathogenic bacteria acquired, obesityterm·key and USDterm·dom exhibit a statistically significant cluster, which also clusters with cardiovascular disease. Diabetesterm·key again clustered with kidney disease with statistical significance. Additionally, from the heatmaps, it is apparent that each of the diseases is associated with a loss of diverse genera more so than the gain of microbial genera, largely corroborating analyses based on antibioticterm·key use and the number of dysbiotic OTUsterm (Fig. 4 ).
Heatmaps showing the most common dysbiotic genera for each disease. Genera were counted for each independent population comparison had at least one dysbiotic OTUterm associated with it. The proportion of populations each genera showed up in is plotted. ( A ) Genera depleted in the disease populations (potential probiotic bacteria lost); ( B ) Genera enriched in disease populations (potential pathogenic bacteria). Hierarchal cluster analysis shows clustering of disease states with the approximately unbiased alpha levels (AU) and bootstrap probability (BP) provided for each cluster (AU/BP). AU values > 95 are considered significant and are bolded.
Impact of genus diversity on dysbiotic OTUsterm When the total number of OTUsterm identified in each genus is plotted against the number of dysbiotic OTUsterm identified in each genus, there is a significant correlation between the two factors, reflecting the fact that dysbiosisterm·dom is driven in part by the number of constituent OTUsterm in each genus (Fig. 5 ). However, the number of genera determined to be dysbiotic more than expected given genus level diversity ranged from one (USDterm·dom) to 14 (kidney disease). Of these, the Bacteroides genus was more likely to be dysbiotic for all diseases and Corynebacterium for all but USDterm·dom. The Anaerococcus genus was more likely to be dysbiotic than expected for three of the five disease states (Table 2 ).
Total genus diversity vs. dysbiotic OTUsterm per genus. Correlations were calculated with a Spearman’s rank order correlation (r). ( A ) Cardiovascular disease; ( B ) Obesityterm·key; ( C ) Diabetesterm·key; ( D ) Kidney disease; ( E ) USDterm·dom.
Microbial genera that were identified as dysbiotic more than expected based on the number of unique OTUsterm identified per genus.
Discussion In the last ten years, there has been an explosion of metagenome-wide association studiesterm (MWASterm) that seek to determine if the microbiometerm·key contributes to disease processes. Importantly, these studies have revealed that there is both considerable diversity in the microbiometerm·key with hundreds to thousands of unique bacterial species in the gut alone 28 , 29 and a high amount of inter-individual variability 30 , 31 . Given these unique characteristics of the microbiometerm·key, studies on any one population of individuals can be driven simply by the individual variability and are thus unreliable. Despite these limitations, one leading hypothesis has emerged that postulates that the loss of diversity from our microbiometerm·key is increasing the risk of developing chronic diseaseterm 45 – 47 . Several questions remain about defining the nature of dysbiosisterm·dom for any one particular disease. First, are MWASterm results driven by inter-individual variability and study design, or are there consistent responses by disease? Second, are different diseases consistently associated with the loss of beneficial bacteria or the gain of harmful bacteria? Finally, is dysbiosisterm·dom driven by a general loss or gain of microbial taxa or are there specific taxa that are more important than others? To address these questions, we conducted a meta-analysisterm·mth of MWASterm data from multiple sources for a cluster of disease states that included USDterm·dom, diabetesterm·key, cardiovascular disease, obesityterm·key, and kidney disease.
Some of the most striking results of the current study is that every disease in the cluster of chronic diseasesterm exhibited a statistically significant association with past year antibioticterm·key use and that antibioticsterm·key had a statistically significant association with gut microbiota composition up to a year after use, both of which supports the loss of function dysbiosisterm·dom hypothesis (Fig. 1 ). Further evidence that a loss of microbial diversityterm·key increases risk of developing the chronic diseasesterm studied here comes from the differential abundance analyses, in which individuals from each of the disease states exhibited a statistically significant propensity for the reduced presence of bacteria in their gut compared to their healthy counterparts (Fig. 3 ).
Past studies have shown both direct and indirect links between antibioticterm·key use and the risk of developing USDterm·dom 20 , 35 , 48 – 51 and obesityterm·key 37 – 39 , 52 , 53 . For diabetesterm·key, the literature is mixed with some studies concluding that antibioticsterm·key increase the risk of diabetesterm·key 54 – 57 , some that conclude repeated antibioticterm·key use but not single antibioticterm·key use increases diabetesterm·key risk 58 , and some that conclude antibioticsterm·key protect against diabetesterm·key 54 , 59 , 60 . For cardiovascular disease, animal studies show that antibioticsterm·key lead to a reduction in microbial activity that promotes cardiovascular disease 61 . However, outcomes using antibioticterm·key therapy to treat cardiovascular disease in a clinical setting to date have been mixed 62 , 63 . In addition to the specific diseases above, another meta-analysisterm·mth of MWASterm studies that spanned multiple unrelated diseases has been published. In that meta-analysisterm·mth, it was found that some diseases such as enteric diarrheal disease, Clostridium difficile infection, and IBD were associated with loss of function dysbiosisterm·dom, whereas diseases such as colorectal cancer, autism spectrum disorder, liver diseases, and Parkinson’s disease were associated with gain of function dysbiosisterm·dom 64 .
Among the taxa that were most consistently found to be depleted in disease individuals were the Bacteroides, Coprococcus, Prevotella, Ruminococcus , and Sutterella (Fig. 5 ). However, subsequent analysis suggests that the association of some of these taxa with dysbiosisterm·dom is closely tied to their number of constituent OTUsterm in the gut, indicative of a non-specific response. Some taxa, primarily Bacteroides, Corynebacterium , and Anaerococcus were identified as dysbiotic more than expected given their diversity, indicative of a more specific physiological interaction between these taxa and disease (Table 2 ). Past studies have concluded that Bacteroides either has a health-protective 65 – 67 or health-antagonistic response 68 – 70 . Results from the current analysis show that this genus is strongly associated with both health and disease in terms of the number of independent populations it was found to be associated with (Fig. 4 ) and thus it is likely that some OTUsterm within the genus provide more of a protective effect and others more of a detrimental health impact. The Corynebacterium and Anaerococcus genera were only associated with the healthy individuals in our meta-analysisterm·mth. Corynebacterium is most often associated with diphtheria, an illness that primarily affects the respiratory system 71 . The number of independent healthy populations associated with Corynebacterium here suggests that this genus of bacteria may play a more beneficial role in the context of the gut microbiota. Anaerococcus sp . are commonly associated with the normal flora of the skin, mouth, and gut, but are also often recovered from clinical specimens such as vaginal discharge and chronic infectious wounds 72 .
Numerous MWASterm studies have demonstrated statistically significant interactions between the microbiometerm·key and disease processes. However, questions remain about the consistency of results obtained from single studies and whether there are common dysbiotic origins for related diseases. Importantly, our results show that for the cluster of chronic diseasesterm that span USDterm·dom, diabetesterm·key, kidney disease, obesityterm·key, and cardiovascular disease, there is a statistically significant association with antibioticterm·key use and a loss of microbial diversityterm·key. Furthermore, several of the dysbiotic taxa are shared between different disease states suggesting that there are some common dysbiotic associations for these diseases. However, as the current data represents a retrospective analysis of data from a single time-point, prospective, longitudinal clinical studies are required to understand the underlying mechanisms between antibioticsterm·key, loss of function dysbiosisterm·dom, and the onset of chronic diseaseterm.
Materials and Methods Data processing Raw sequencing data for all studies were demultiplexed and quality-controlled with default parameters in QIIME 73 . An open reference strategy was used to assign OTUsterm with 97% homology compared to a reference database composed of the Greengenes V13.8 dataset 74 – 76 . Datasets were combined prior to OTUterm assignment to allow for direct comparisons. OTUsterm that did not match any of sequences in the reference database were then classified de novo , which first clusters unclassified sequences based on homology, picks a representative sequence, and assigns taxonomy based on a 94% similarity threshold to the reference database 77 . Chloroplasts, mitochondria, and chimeric sequences were removed from datasets as well as those that had <10 representations across each dataset as previously described 74 , 75 . The final OTUterm count across all datasets was 35,582. Datasets were regularized with a negative binomial Wald test, executed by the DESeq2 algorithm, which fits the data to a negative binomial distribution, to account for differences in sequencing depth in each sample, but does not account for the compositional nature of the data 78 , 79 .
Association of antibioticsterm·key and chronic diseaseterm To determine if antibioticsterm·key were associated with chronic diseaseterm, we quantified the proportion of individuals who had taken antibioticsterm·key in the last year for cardiovascular disease, obesityterm·key, diabetesterm·key, and kidney disease compared to their healthy counterparts in the AGP data. For USDterm·dom, only one of the four studies included in our meta-analysisterm·mth had metadata associated with antibioticterm·key use 20 . Proportions were compared with a relative risk ratio followed by a post-hoc, Fisher’s exact test, using a Holm’s correction for multiple hypotheses.
In order for antibioticterm·key use to have a meaningful impact on chronic diseaseterm, we expect that antibioticsterm·key would exhibit a statistically significant association with microbiota composition. Therefore, we also conducted a weighted UniFrac beta-diversity analysis on all regularized data from the AGP dataset, which were all collected and processed by the same group, followed by a post-hoc PERMANOVA against antibioticterm·key use in the past year.
Defining dysbiosisterm·dom for each disease state To determine the nature of dysbiosisterm·dom for each disease state, while considering the multi-dimensional and variable nature of microbiometerm·key data, we divided the samples into the following independent populations. For USDterm·dom data, each study was considered an independent population and the healthy vs. USDterm·dom groups were compared against each other within each study. Four populations were available for USDterm·dom and their healthy counterparts. For AGP data, each category of health states was sub-divided into three independent sets of 100 samples by assigning samples to one of three groups through the lowest, middle, and highest 100 random numbers previously assigned. For kidney disease, only one subset was available. Subdivisions provided three independent populations for AGP healthy individuals, cardiovascular disease, obesityterm·key, and diabetesterm·key. One population was available for kidney disease. Additionally, random numbers were reassigned to AGP data independent of metadata. The random numbers generated were used to assign three independent populations each for stochastic groups 1 & 2. Groups with stochastic metadata were used to determine if our definition of dysbiosisterm·dom for each disease state were driven by disease or some other stochastic process.
Following subdivisions, differential abundance analysis, calculated as the log fold-change divided by standard error of negative binomial regularized data (lfcse), was conducted on all relevant pairwise comparisons of independent populations using the DESeq2 algorithm 78 . All p-values for differentially abundant OTUsterm were false-discovery rate corrected, following the Benjamini-Hochberg method 80 . Differential abundance analysis was chosen to define dysbiosisterm·dom because it gives the specific OTUsterm enriched in both healthy and disease populations rather than net differences in diversity or community composition as with alpha- and beta-diversity respectively. Thus, this metric is suitable for identifying situations in which a combination of loss and gain of function dysbiosisterm·dom may contribute to disease processes. For USDterm·dom data, since samples and data were collected independently by four different research groups, differential abundance analysis between healthy and USDterm·dom populations within each study was conducted. For AGP data, since all samples were collected, processed, and sequenced by the same group, each independent healthy population was compared to each disease population. This resulted in nine pairwise comparisons for cardiovascular disease, diabetesterm·key, obesityterm·key, and stochastic groups 1 & 2, along with three pairwise comparisons for kidney disease.
From the results of independent differential abundance analyses, the fold difference in OTUsterm between the healthy and disease populations was calculated as the number of OTUsterm enriched in the healthy group divided by the number of OTUsterm enriched in the disease group. Absolute values were compared with a one-sample t-test against an expected value of 1, indicative of no difference, and p-values were Holm’s corrected for multiple comparisons. For purposes of plotting fold difference, raw values below 1 were inverted and made negative.
Strength of dysbiotic taxa and commonalities across chronic diseaseterm states To determine the consistency of dysbiotic taxa identified and if there were common dysbiotic genera across the different chronic diseasesterm, differentially abundant OTUsterm for each comparison were first reduced to genera, which allows for greater comparison between each disease state but comes with the limitation that physiologically disparate OTUsterm can exist within the same genus. The OTUsterm that were not classified at least to the genus level were removed from further analysis. We counted the number of times a microbial genus had at least one OTUterm that was enriched in either the healthy or disease populations across all pairwise comparisons for each disease. Genera enriched in the healthy populations were considered “potentially beneficial” whereas those enriched in the disease populations were considered “potentially pathogenic”. The proportion of independent analyses that each genus appeared in was plotted in a heatmap in R statistical software using the heatmap.2 function with dendograms generated by the hclust function to show clustering by disease state. The bootstrap probability and approximately unbiased alpha levels were calculated by the pvclust package with 1000 bootstraps 81 .
Influence of genus diversity on dysbiotic genera Next, we sought to determine if dysbiosisterm·dom was driven by a general loss or gain of bacteria in the gut or if some microbial genera had a greater association with disease than expected by the number of constituent OTUsterm within the genus. To do this, for each disease, we plotted the total number of OTUsterm detected per genus against the number of dysbiotic OTUsterm detected per genus and determined if there was a Spearman correlation significantly different than 0 with the cor.test function in R statistical software. Next, genera in which the number of dysbiotic OTUsterm was greater than three standard deviations above what was expected based on genus diversity were considered to be highly associated with dysbiosisterm·dom, following the three-sigma rule 82 .
Acknowledgements This study was funded with seed funds from Lerner Research Institute and grants from the Research Program Committee at The Cleveland Clinic Foundation and the Endourology Society Summer Student Scholarship Program. The funding bodies had no role in the design of the study.
Author Contributions A.W.M., M.M. and L.J.W. contributed to the design of the study. L.J.M. and A.W.M. conducted data analysis. All authors contributed to writing the manuscript and have approved the final version.
Data Availability Sequence reads for USDterm·dom are available at the Sequence Read Archive under Accession #’s SRP140641, SRP140933, PRJNA382644, PRJNA304689. American gut project data is available at the European Bioinformatics Institute under accession # PRJEB11419.
Competing Interests The authors declare no competing interests.
Footnotes References Associated Data Data Availability Statement Sequence reads for USDterm·dom are available at the Sequence Read Archive under Accession #’s SRP140641, SRP140933, PRJNA382644, PRJNA304689. American gut project data is available at the European Bioinformatics Institute under accession # PRJEB11419.