Composition of the adult digestive tract bacterial microbiome based on seven mouth surfaces, tonsils, throat and stool samples

Nicola Segata, Susan Kinder Haake, Peter Mannon, Katherine P. Lemon, Levi Waldron, Dirk Gevers, Curtis Huttenhower, Jacques Izard•View original
OverviewBalancedalloy voice
If you want to understand when and how the gut goes wrong, you first need to know what "right" looks like. That’s the simple but powerful idea behind the Human Microbiome Project’s deep dive into the healthy digestive tract. Segata and colleagues took a tour from mouth to colon and asked a basic question: in people who feel fine, how do the communities of microbes arrange themselves, and what can they do? The answer isn’t a single "healthy microbiome" but a set of recurring neighborhoods, each with its own residents and jobs. They built this map at scale. More than 200 healthy adults were enrolled at two distant U.S. centers. Ten digestive tract sites were sampled in each person, from seven oral surfaces and two spots in the oropharynx to stool as a proxy for the colon. All told, they analyzed 2,105 samples, which gives you statistical power to tell signal from noise. And they did it gently. Protocols were tuned to minimize disturbing the native microbiota, so you’re seeing communities as close to "at rest" as feasible in a clinical study. Under the hood, two lenses came into focus. One zoomed in on "who’s there" using 16S ribosomal RNA profiling on the 454 FLX Titanium platform. Reads went through a mothur-based curation pipeline and the Ribosomal Database Project, or RDP, classifier, with a conservative 80 percent confidence threshold for taxonomic calls. After quality control, 209 subjects remained, 147 of whom had all ten digestive sites. The other lens looked at "what can they do," using whole-genome shotgun sequencing in a subset of 98 people and the HMP Unified Metabolic Analysis Network, or HUMAnN, pipeline to reconstruct metabolic pathways from short Illumina reads. That functional layer matters. Reads were mapped to Kyoto Encyclopedia of Genes and Genomes, or KEGG, Orthology groups using a translated search with stringent filters, then stitched into pathways with MinPath and consistency checks against the community’s taxonomic makeup. This gives relative abundances of genes and modules across sites. And a quick reality check: non-bacterial passengers were tiny. Archaea were about 0.04 percent in stool and below detection in oral sites. Non-human eukaryotes peaked at 0.34 percent in buccal mucosa and were under a tenth of a percent elsewhere. One nice sanity test tied genes to bugs: arabinofuranosyltransferases aftA and aftB — enzymes for cell envelope building — only appeared on tooth surfaces and tracked tightly with Corynebacterium, with a Spearman correlation around 0.76 and a p-value well below ten to the negative fifteen. When they stepped back, a simple geography emerged. Ten sites fell into four reproducible community types grounded in the balance between two big bacterial phyla, Firmicutes and Bacteroidetes. Group 1 clustered the buccal mucosa, keratinized gingiva, and hard palate. Group 2 gathered saliva, tongue dorsum, palatine tonsils, and throat. Group 3 was the tooth surfaces — subgingival and supragingival plaque. Group 4 was stool. These weren’t fuzzy tendencies; they were tight clusters. Within-group Bray-Curtis similarity sat around 0.58 for Group 1, 0.51 for Group 2, 0.49 for Group 3, and 0.53 for stool. In every case, within-group similarity was far higher than between-group similarity, with p-values below ten to the negative twenty. Let’s meet the residents. Group 1 — those firm, non-shedding oral surfaces — was dominated by Streptococcus, averaging about 47 percent with a pretty wide spread across people, and Gemella hovered near 5 percent. Group 2 spread the wealth more evenly. Here you saw Veillonella near 10 percent, Prevotella around 12 percent, and notable appearances from Porphyromonas and Neisseria, each only a few percent on average but consistently enriched by biomarker analysis. Group 4, the stool, flipped the script: Bacteroidetes took more than 60 percent of the pie in many individuals, with Firmicutes around a third or less, and Proteobacteria and Actinobacteria were minor players. Plaque — Group 3 — deserves its own spotlight. Those tooth surface communities are distinct from other oral habitats despite sharing some family-level lineages. Actinobacteria were relatively enriched there, and classic periodontal genera like Porphyromonas and Tannerella were common in the mouth yet low or sporadic in stool. Porphyromonas averaged around 3 percent across upper sites and was detectable in about a quarter of stool samples; Tannerella showed up in roughly 97 percent of upper tract sites but barely 3 percent in stool. It’s a reminder that presence isn’t destiny — context and abundance matter. Diversity followed the habitat. If you look at alpha diversity using the inverse Simpson index — a way of saying how many different kinds of organisms are present and how evenly they’re represented — Group 1 sites were on the low end, below about 5.3. That fits with one genus, Streptococcus, taking nearly half the reads. Group 2 ranged higher, from roughly 7.3 in tonsils to about 10.6 in saliva. Plaque sat near 9.6 to 9.8. And stool, despite its complexity, averaged around 4.6, often because a few Bacteroides species are extremely abundant. The big picture: each habitat has a signature level of diversity and a characteristic spread across people, and those signatures are reproducible. Now to the jobs. Across all four groups, there was a shared metabolic core — the basic tools of life show up everywhere — but the accessories differed by neighborhood. Think of sugar transport as the grocery delivery system. In oral sites, the phosphotransferase system for small sugars was prominent: transporters for mannose and fructose, and even galactosamine, were common. Plaque went a step further with extra capacity for trehalose, alpha-glucosides, and cellobiose. The gut focused on different groceries: transporters for lactose and arabinose, systems for bringing in oligogalacturonides, and a toolkit for breaking down big, sulfated polysaccharides like dermatan, chondroitin, and heparan sulfate. These differences line up with what’s on offer — dietary fibers and host glycans in the colon, saliva-fed simple sugars and host-derived substrates in the mouth. Gas metabolism drew an even sharper line. Genes for making and using molecular hydrogen clustered in the oral cavity and were nearly absent in stool. One hydrogenase complex, often called hydrogenase-four, and a related methyltransferase carried strong oral signals. Likely contributors included Veillonella and Selenomonas. In fact, an unclassified Pasteurellaceae group in the mouth tracked almost perfectly with hydrogenase-four. The correlation was above 0.78 and p-values were vanishingly small. Hydrogen sulfide told a complementary story: the gut was enriched for cystathionine-beta-lyase, a route to hydrogen sulfide, and plaque highlighted methionine-gamma-lyase. Those plaque hydrogen sulfide genes rode alongside Treponema and Fusobacterium — correlations in the 0.7 to 0.8 range. That makes biological sense in a niche linked to periodontal risk. You can also read signatures of energy and byproduct management. Stool carried strong representation of beta-glucosidase, which helps crack plant glycosides, and a robust Embden-Meyerhof pathway, the classic glycolysis route. Ammonia and methane metabolism modules were prominent too, consistent with known colon physiology. The mouth, by contrast, was rich in transporters for putrescine, a small amine that, when imbalanced, contributes to halitosis. Metal handling wove through both sites with local accents: uroporphyrinogen synthase tended to be higher in stool and inversely tracked with Clostridiales, while protoporphyrinogen oxidase leaned oral and aligned with Prevotella. Even the tongue and plaque shared an uptick in coproporphyrinogen oxidase. Export systems tied to metal resistance, like MtrCDE and HrtAB, were broadly present, with hints of enrichment depending on the niche. And in the gut, pyruvate:ferredoxin oxidoreductase — often linked to anaerobic energy metabolism and sometimes flagged alongside antibiotic resistance dynamics — stood out. The taxonomy-function links weren’t just anecdotes. Biomarker discovery used a method called Linear Discriminant Analysis Effect Size, or LEfSe. It combines nonparametric tests with linear discriminant analysis to pick out clades and pathways that distinguish groups, with a conservative alpha of 0.05 and an effect size filter. That’s how features like Streptococcus in Group 1 and Lachnospiraceae and Ruminococcaceae in stool emerged as robust discriminators, not just hunches. And multidimensional scaling of Bray-Curtis distances showed the same story visually: samples from the same habitat cluster together, and samples from different habitats do not. One of the most intriguing threads is how "environmental" phyla have set up shop in us. Candidate phylum TM7 popped up in at least one upper digestive site in about 85 percent of people and in stool in roughly 14 percent. SR1 was detectable in the upper tract of about two-thirds of subjects and barely in stool. Synergistetes, another group once thought of as external, appeared in upper sites in close to 60 percent of people and in under 10 percent of stools. Lower down, Verrucomicrobia — often Akkermansia — was present in the colon of about 41 percent of participants, and Lentisphaerae in around 15 percent. These aren’t curiosities; they’re part of the normal cast. And yes, some genera with scary reputations are common in health. Treponema was present in the upper tract of 96 percent of subjects and reached a couple of percent on average in subgingival plaque, yet it didn’t signal disease here. Porphyromonas hovered around three percent across many oral sites; Tannerella was nearly ubiquitous in the upper tract and rare in stool. The lesson Segata and colleagues emphasize is nuance: potential pathogens at low, stable abundance in the right neighborhood can be part of a healthy system. There’s also traffic between neighborhoods. A surprisingly long list of genera appeared in both oral sites and stool in nearly half the cohort — Bacteroides and Faecalibacterium, sure, but also Prevotella, Veillonella, Streptococcus, Roseburia, Coprococcus, and more. Saliva and the upper aerodigestive tract likely seed the distal gut continuously. Most of the time, that’s background noise. Sometimes, especially when the gut is perturbed, it may matter. All of this sits on a foundation of careful measurement. The team made conservative calls — from read curation to confidence thresholds — and they used first-visit samples by default to avoid overcounting the same person. They checked for cross-talk between species and pathways, made their data public, and designed analyses that can be repeated and extended. Even small details, like the trace 0.04 percent archaeal signal in stool and negligible levels in the mouth, help calibrate expectations for future studies. So what do we walk away with? A baseline. Four distinct, reproducible communities from mouth to colon. A shared metabolic core with local accents in sugar handling and gas metabolism that match the habitat. Tight links between certain genes and the taxa that carry them. And a caution that presence alone is not proof of pathology. That’s the canvas on which disease paints. If you’re thinking ahead, the next moves are obvious and exciting but need restraint. Layer these baselines onto longitudinal data to watch communities bend before disease shows up. Bring in host genetics to see when the same microbes do different things in different people. And be precise about function: the differential distribution of hydrogen and hydrogen sulfide genes hints at immunomodulatory roles that are ripe for targeted testing. Segata’s map doesn’t answer those questions. It tells you where to look, and how to know when what you’re seeing is a true departure from healthy.

If you want to understand when and how the gut goes wrong, you first need to know what "right" looks like. That’s the simple but powerful idea behind the Human Microbiome Project’s deep dive into the healthy digestive tract. Segata and colleagues took a tour from mouth to colon and asked a basic question: in people who feel fine, how do the communities of microbes arrange themselves, and what can they do?

The answer isn’t a single "healthy microbiome" but a set of recurring neighborhoods, each with its own residents and jobs.

They built this map at scale. More than 200 healthy adults were enrolled at two distant U.S. centers. Ten digestive tract sites were sampled in each person, from seven oral surfaces and two spots in the oropharynx to stool as a proxy for the colon.

All told, they analyzed 2,105 samples, which gives you statistical power to tell signal from noise. And they did it gently. Protocols were tuned to minimize disturbing the native microbiota, so you’re seeing communities as close to "at rest" as feasible in a clinical study.

Under the hood, two lenses came into focus. One zoomed in on "who’s there" using 16S ribosomal RNA profiling on the 454 FLX Titanium platform. Reads went through a mothur-based curation pipeline and the Ribosomal Database Project, or RDP, classifier, with a conservative 80 percent confidence threshold for taxonomic calls.

After quality control, 209 subjects remained, 147 of whom had all ten digestive sites. The other lens looked at "what can they do," using whole-genome shotgun sequencing in a subset of 98 people and the HMP Unified Metabolic Analysis Network, or HUMAnN, pipeline to reconstruct metabolic pathways from short Illumina reads.

That functional layer matters. Reads were mapped to Kyoto Encyclopedia of Genes and Genomes, or KEGG, Orthology groups using a translated search with stringent filters, then stitched into pathways with MinPath and consistency checks against the community’s taxonomic makeup. This gives relative abundances of genes and modules across sites.

And a quick reality check: non-bacterial passengers were tiny. Archaea were about 0.04 percent in stool and below detection in oral sites. Non-human eukaryotes peaked at 0.34 percent in buccal mucosa and were under a tenth of a percent elsewhere.

One nice sanity test tied genes to bugs: arabinofuranosyltransferases aftA and aftB — enzymes for cell envelope building — only appeared on tooth surfaces and tracked tightly with Corynebacterium, with a Spearman correlation around 0.76 and a p-value well below ten to the negative fifteen.

When they stepped back, a simple geography emerged. Ten sites fell into four reproducible community types grounded in the balance between two big bacterial phyla, Firmicutes and Bacteroidetes. Group 1 clustered the buccal mucosa, keratinized gingiva, and hard palate.

Group 2 gathered saliva, tongue dorsum, palatine tonsils, and throat. Group 3 was the tooth surfaces — subgingival and supragingival plaque. Group 4 was stool.

These weren’t fuzzy tendencies; they were tight clusters. Within-group Bray-Curtis similarity sat around 0.58 for Group 1, 0.51 for Group 2, 0.49 for Group 3, and 0.53 for stool. In every case, within-group similarity was far higher than between-group similarity, with p-values below ten to the negative twenty.

Let’s meet the residents. Group 1 — those firm, non-shedding oral surfaces — was dominated by Streptococcus, averaging about 47 percent with a pretty wide spread across people, and Gemella hovered near 5 percent. Group 2 spread the wealth more evenly.

Here you saw Veillonella near 10 percent, Prevotella around 12 percent, and notable appearances from Porphyromonas and Neisseria, each only a few percent on average but consistently enriched by biomarker analysis. Group 4, the stool, flipped the script: Bacteroidetes took more than 60 percent of the pie in many individuals, with Firmicutes around a third or less, and Proteobacteria and Actinobacteria were minor players.

Plaque — Group 3 — deserves its own spotlight. Those tooth surface communities are distinct from other oral habitats despite sharing some family-level lineages. Actinobacteria were relatively enriched there, and classic periodontal genera like Porphyromonas and Tannerella were common in the mouth yet low or sporadic in stool.

Porphyromonas averaged around 3 percent across upper sites and was detectable in about a quarter of stool samples; Tannerella showed up in roughly 97 percent of upper tract sites but barely 3 percent in stool. It’s a reminder that presence isn’t destiny — context and abundance matter.

Diversity followed the habitat. If you look at alpha diversity using the inverse Simpson index — a way of saying how many different kinds of organisms are present and how evenly they’re represented — Group 1 sites were on the low end, below about 5.3. That fits with one genus, Streptococcus, taking nearly half the reads.

Group 2 ranged higher, from roughly 7.3 in tonsils to about 10.6 in saliva. Plaque sat near 9.6 to 9.8. And stool, despite its complexity, averaged around 4.6, often because a few Bacteroides species are extremely abundant.

The big picture: each habitat has a signature level of diversity and a characteristic spread across people, and those signatures are reproducible.

Now to the jobs. Across all four groups, there was a shared metabolic core — the basic tools of life show up everywhere — but the accessories differed by neighborhood. Think of sugar transport as the grocery delivery system.

In oral sites, the phosphotransferase system for small sugars was prominent: transporters for mannose and fructose, and even galactosamine, were common. Plaque went a step further with extra capacity for trehalose, alpha-glucosides, and cellobiose. The gut focused on different groceries: transporters for lactose and arabinose, systems for bringing in oligogalacturonides, and a toolkit for breaking down big, sulfated polysaccharides like dermatan, chondroitin, and heparan sulfate.

These differences line up with what’s on offer — dietary fibers and host glycans in the colon, saliva-fed simple sugars and host-derived substrates in the mouth.

Gas metabolism drew an even sharper line. Genes for making and using molecular hydrogen clustered in the oral cavity and were nearly absent in stool. One hydrogenase complex, often called hydrogenase-four, and a related methyltransferase carried strong oral signals.

Likely contributors included Veillonella and Selenomonas. In fact, an unclassified Pasteurellaceae group in the mouth tracked almost perfectly with hydrogenase-four. The correlation was above 0.78 and p-values were vanishingly small.

Hydrogen sulfide told a complementary story: the gut was enriched for cystathionine-beta-lyase, a route to hydrogen sulfide, and plaque highlighted methionine-gamma-lyase. Those plaque hydrogen sulfide genes rode alongside Treponema and Fusobacterium — correlations in the 0.7 to 0.8 range. That makes biological sense in a niche linked to periodontal risk.

You can also read signatures of energy and byproduct management. Stool carried strong representation of beta-glucosidase, which helps crack plant glycosides, and a robust Embden-Meyerhof pathway, the classic glycolysis route. Ammonia and methane metabolism modules were prominent too, consistent with known colon physiology.

The mouth, by contrast, was rich in transporters for putrescine, a small amine that, when imbalanced, contributes to halitosis. Metal handling wove through both sites with local accents: uroporphyrinogen synthase tended to be higher in stool and inversely tracked with Clostridiales, while protoporphyrinogen oxidase leaned oral and aligned with Prevotella. Even the tongue and plaque shared an uptick in coproporphyrinogen oxidase.

Export systems tied to metal resistance, like MtrCDE and HrtAB, were broadly present, with hints of enrichment depending on the niche. And in the gut, pyruvate:ferredoxin oxidoreductase — often linked to anaerobic energy metabolism and sometimes flagged alongside antibiotic resistance dynamics — stood out.

The taxonomy-function links weren’t just anecdotes. Biomarker discovery used a method called Linear Discriminant Analysis Effect Size, or LEfSe. It combines nonparametric tests with linear discriminant analysis to pick out clades and pathways that distinguish groups, with a conservative alpha of 0.05 and an effect size filter.

That’s how features like Streptococcus in Group 1 and Lachnospiraceae and Ruminococcaceae in stool emerged as robust discriminators, not just hunches. And multidimensional scaling of Bray-Curtis distances showed the same story visually: samples from the same habitat cluster together, and samples from different habitats do not.

One of the most intriguing threads is how "environmental" phyla have set up shop in us. Candidate phylum TM7 popped up in at least one upper digestive site in about 85 percent of people and in stool in roughly 14 percent. SR1 was detectable in the upper tract of about two-thirds of subjects and barely in stool.

Synergistetes, another group once thought of as external, appeared in upper sites in close to 60 percent of people and in under 10 percent of stools. Lower down, Verrucomicrobia — often Akkermansia — was present in the colon of about 41 percent of participants, and Lentisphaerae in around 15 percent. These aren’t curiosities; they’re part of the normal cast.

And yes, some genera with scary reputations are common in health. Treponema was present in the upper tract of 96 percent of subjects and reached a couple of percent on average in subgingival plaque, yet it didn’t signal disease here. Porphyromonas hovered around three percent across many oral sites;

Tannerella was nearly ubiquitous in the upper tract and rare in stool. The lesson Segata and colleagues emphasize is nuance: potential pathogens at low, stable abundance in the right neighborhood can be part of a healthy system.

There’s also traffic between neighborhoods. A surprisingly long list of genera appeared in both oral sites and stool in nearly half the cohort — Bacteroides and Faecalibacterium, sure, but also Prevotella, Veillonella, Streptococcus, Roseburia, Coprococcus, and more. Saliva and the upper aerodigestive tract likely seed the distal gut continuously.

Most of the time, that’s background noise. Sometimes, especially when the gut is perturbed, it may matter.

All of this sits on a foundation of careful measurement. The team made conservative calls — from read curation to confidence thresholds — and they used first-visit samples by default to avoid overcounting the same person. They checked for cross-talk between species and pathways, made their data public, and designed analyses that can be repeated and extended.

Even small details, like the trace 0.04 percent archaeal signal in stool and negligible levels in the mouth, help calibrate expectations for future studies.

So what do we walk away with? A baseline. Four distinct, reproducible communities from mouth to colon.

A shared metabolic core with local accents in sugar handling and gas metabolism that match the habitat. Tight links between certain genes and the taxa that carry them. And a caution that presence alone is not proof of pathology. That’s the canvas on which disease paints.

If you’re thinking ahead, the next moves are obvious and exciting but need restraint. Layer these baselines onto longitudinal data to watch communities bend before disease shows up. Bring in host genetics to see when the same microbes do different things in different people.

And be precise about function: the differential distribution of hydrogen and hydrogen sulfide genes hints at immunomodulatory roles that are ripe for targeted testing. Segata’s map doesn’t answer those questions. It tells you where to look, and how to know when what you’re seeing is a true departure from healthy.

More in Dentistry →

Related lectures