Oral Microbiota Community Dynamics Associated With Oral Squamous Cell Carcinoma Staging

Chia‐Yu Yang, Yuan‐Ming Yeh, Haiying Yu, Chia-Yin Chin, Chia‐Wei Hsu, Hsuan Liu, Po-Jung Huang, Songnian Hu, Chun‐Ta Liao, Kai‐Ping Chang, Yu‐Liang Chang•View original

OverviewBalancedalloy voice

Picture this: two people with the same tumor on the tongue. One is at an early stage, while the other is at an advanced stage. Surgeons, radiologists, and oncologists will see different treatment paths. But what about the microbes living in their mouths? Do those communities change in a way that tracks with the cancer’s march from early to late? And if they do, could that shifting ecosystem help us spot trouble earlier or understand what’s happening on the ground? Oral squamous cell carcinoma, or OSCC, is the major form of oral cancer, making up more than 90 percent of cases. It’s common in parts of Asia, including Taiwan, where it ranks as the fourth most common cancer in men. Survival hasn’t budged enough; five years after diagnosis, only about fifty to sixty percent of patients are still alive. We know the classic risks: cigarettes, alcohol, and betel quid. But chronic inflammation and microbial imbalance—dysbiosis—keep popping up in cancer stories, including head and neck cancers. Older culture-based studies hinted at more anaerobes on tumors. Names like Porphyromonas gingivalis and Fusobacteria came up again and again. Then came sixteen S ribosomal RNA gene sequencing, which doesn’t need to grow microbes in a dish, and suddenly the oral microbiome was a map we could read. Still, one piece was missing: a stage-by-stage view in humans that asked, not just “who’s there,” but “how does the community reorganize as OSCC progresses?” That’s the gap Yang and colleagues set out to fill in a study published in Frontiers in Microbiology. They didn’t follow one person over time—that would be ideal—but they built a cross-sectional snapshot across the disease arc, including two hundred forty-eight people. They had fifty-one healthy controls, forty-one patients at stage one, sixty-six patients at stages two and three grouped together, and ninety at stage four. Everyone did a standardized oral rinse, and then the team profiled the bacteria using a familiar workhorse: sixteen S rRNA sequencing of the V3 to V4 region on an Illumina MiSeq, paired-end three hundred base reads. If you’ve worked in microbiome pipelines, the rest will sound like a greatest hits album. They used the QIAamp DNA Microbiome Kit for extraction, clustered reads into operational taxonomic units at 97 percent similarity, assigned taxonomy with Greengenes and cross-checked with oral reference databases, and built diversity and community-structure analyses using Bray–Curtis distances and ordination plots. For differential abundance, they leaned on LEfSe, and for functional inference, PICRUSt mapped predicted genes to Kyoto Encyclopedia of Genes and Genomes pathways. To turn microbes into markers, they used multivariate logistic regression and receiver operating characteristic curves, reporting area under the curve values with confidence intervals. Before we dive into the names and numbers, let’s step back for the big picture. The oral microbiome was not static. It shifted with stage, and by the time patients reached stage four, the community looked and behaved differently from health. That pattern showed up at multiple levels: in measures of diversity, in which phyla dominated, and in the rise and fall of specific species. The result isn’t just a list; it’s a trajectory. Start with diversity. Think of Shannon diversity as a way to capture richness and evenness in one number. In late-stage OSCC, that number was higher than in healthy mouths, and the difference was striking—statistically rock-solid with a p-value below 0.001. Richness measures like Chao1 and observed species nudged upward only a little. That tells you the story isn’t simply “more bugs.” It’s a rebalancing—a community spreading its bets more evenly among certain players as the disease advances. Now look at the whole-community layout. When the team compared samples by their overall composition—using Bray–Curtis distances and principal coordinates analysis—stage four didn’t just blur into health. It pulled away. On hierarchical clustering, stage four samples tended to sit together, forming their own neighborhood. And as you moved from stage one to stages two and three to stage four, the neighborhood shifted progressively, not randomly. That gradient matters; it suggests the microbiome isn’t just different in cancer; it tracks with how far the cancer has gone. Zoom out to the big phyla and who dominates. In healthy controls, Firmicutes led the pack—about three-fifths of the community—followed by Proteobacteria, with Actinobacteria and Bacteroidetes making smaller showings. Fusobacteria were bit players in health, roughly three percent. With cancer progression, that last name—Fusobacteria—kept rising. By stage four, it landed around eight percent, and that wasn’t a wobble; the increases were significant from stage one onward. Meanwhile, Bacteroidetes and Actinobacteria ebbed as disease advanced. Firmicutes and Proteobacteria, despite their dominance, stayed relatively steady across stages. So the headline at this level is simple: late-stage OSCC brings more Fusobacteria and less of certain commensals. At the genus level, the familiar oral cast was all there—Streptococcus, Haemophilus, Veillonella, Neisseria, and Rothia—but their billing changed. In healthy mouths, Streptococcus was the star, roughly a third of reads. In stage four, it dropped to just over a quarter. Fusobacterium—the genus that includes several periodontal pathogens—rose. And a handful of other genera moved in quieter but consistent ways: Haemophilus and Actinomyces trended down in late disease, while Veillonella and Neisseria jostled in the top ranks without dramatic swings. The key is the pattern: increases in Fusobacterium and decreases in several commensal genera, in step with progression. Species-level resolution sharpened that picture. Using LEfSe to sift signal from noise, Yang’s team highlighted five species enriched in stage four: Fusobacterium periodonticum, Parvimonas micra, Streptococcus constellatus, Haemophilus influenzae, and Filifactor alocis. Several species were depleted in late-stage disease, including Streptococcus mitis, Haemophilus parainfluenzae, Porphyromonas pasteri, Veillonella parvula, and Actinomyces odontolyticus. A couple of numbers tell the story. Fusobacterium periodonticum climbed from about 1.7 percent at stage one to 3.3 percent at stage four. Parvimonas micra jumped from well under one percent in health to roughly 3.7 percent in stage four. On the flip side, Streptococcus mitis fell from about 32 percent in healthy controls to under 20 percent in stage four, and Porphyromonas pasteri slipped from around 3.4 percent to under one percent. Those are not rounding errors. That’s a community reconfiguring, with potential implications for inflammation, tissue invasion, and metabolic cross-talk in the tumor microenvironment. One more twist at the species level: a few operational taxonomic units surfaced only in stage four, including Neisseria elongata, Eikenella corrodens, an Oribacterium oral taxon, and Dialister pneumosintes. Think of these as rarer or more context-dependent residents that appear when the ecosystem changes enough to welcome them. It’s a hint that late-stage disease might open ecological niches that weren’t there before. Turning those shifts into diagnostic signal, Yang and colleagues built a simple, biologically grounded panel. Three species: one up and two down. Elevated Fusobacterium periodonticum, combined with reduced Streptococcus mitis and Porphyromonas pasteri, separated stage four from health with an area under the receiver operating characteristic curve of 0.956, and the 95 percent confidence interval—from 0.925 to 0.986—was tight. For a single bacterium, Parvimonas micra did well on its own, with an area under the curve of 0.883. Fusobacterium periodonticum alone hit 0.864. But the trio together carried the load. That’s important: it says a small, interpretable set of microbes can carry strong discriminative power, at least in this cohort. Of course, composition is only half the tale; function matters too. PICRUSt let the team infer metabolic potential from the sixteen S profiles. As OSCC advanced, the predicted metagenomes leaned harder into carbohydrate and energy metabolism. Pathways linked to methane metabolism and oxidative phosphorylation were enriched with progression. Meanwhile, pathways for building blocks—amino acids like valine, leucine, isoleucine, aromatic amino acids—and folate biosynthesis were relatively stronger in health. That pivot suggests late-stage communities may be wired for higher energy turnover and different carbon processing, which could dovetail with the metabolic demands of inflamed, hypoxic tumor niches. It’s an inference, not a direct measurement, but it’s a cohesive one. A couple of technical notes help interpret what we’re hearing. The sequencing depth was substantial—tens of millions of reads across the cohort—and after quality control the data clustered into four hundred twenty-four operational taxonomic units. A core set of eighty-nine operational taxonomic units popped up in ninety percent of samples, and forty-seven were shared across all groups. Those core microbes are the bedrock of the oral ecosystem. The stage-specific changes ride on top of that, which makes them more interesting: the ecosystem keeps its backbone, but its proportions and peripheries adapt with disease. Now, the guardrails. The healthy controls were not matched to cancer patients for age, gender, or oral health status. That matters because age and periodontal health can nudge the microbiome on their own. The samples were oral rinses, which give a mouth-wide average and can pull in bacteria from plaque and mucosa; they may not reflect what’s sitting right on the tumor. DNA extraction and sequencing choices can tilt which taxa you see and how abundant they look. And the cross-sectional design means we can’t say the microbes caused the cancer, or even that they accelerated it. We can say they track with it, and that tracking is reproducible across several measures. So what do we do with this? First, we have a progression-resolved map: more Fusobacteria as OSCC advances, fewer commensals like Streptococcus mitis and Haemophilus parainfluenzae, and a stage four community that’s measurably distinct in composition and predicted function. Second, we have a compact microbial signature with excellent discriminative performance for late-stage disease in this cohort. Third, we have a hypothesis about function—that late-stage ecosystems are geared toward carbohydrate and energy metabolism—ready to be tested with shotgun metagenomics or metabolomics. The restraint is healthy here. Yang and colleagues don’t sell these microbes as causes, cures, or screening tools tomorrow morning. They lay out a blueprint: validate the panel in a new, matched cohort. See if the signal holds up when you add clinical risk factors like tobacco, alcohol, and betel quid. Sample tumor-adjacent sites to compare with whole-mouth rinses. And follow people over time to watch the ecosystem shift before, during, and after treatment. That’s where causality and clinical utility begin to come into focus. If you’re imagining this as a future saliva test, you’re not wrong to be curious. An area under the curve just under 0.96 for distinguishing stage four from health is eye-catching. But the real win might be earlier: if a stage-resolved signature emerges at stage one or two, and if it adds value beyond what a clinician can see and feel, then we have something that could change decision-making. Right now, the late-stage separation tells us the microbiome is deeply entwined with the disease environment. That’s not a biomarker in a bottle yet. It is, however, a clear signal that the neighborhood changes as the tumor grows—and that you can hear that change in the data. Underneath all the names and percentages is a simple narrative. An inflamed, stressed tissue environment shifts the balance of residents. Pathobionts—organisms that live peacefully until conditions tilt—find an opening. Keystone commensals lose ground. The energy needs of the ecosystem change, and so do the tools the microbes bring to the job. Yang and colleagues captured that shift across a human population with careful sequencing and conservative statistics. It’s a snapshot, but it’s a revealing one. And for a disease that still takes too many lives, insights like this—grounded in data, tied to stage, and open about their limits—are the kind that move a field forward.

But what about the microbes living in their mouths? Do those communities change in a way that tracks with the cancer’s march from early to late? And if they do, could that shifting ecosystem help us spot trouble earlier or understand what’s happening on the ground?

Oral squamous cell carcinoma, or OSCC, is the major form of oral cancer, making up more than 90 percent of cases. It’s common in parts of Asia, including Taiwan, where it ranks as the fourth most common cancer in men. Survival hasn’t budged enough; five years after diagnosis, only about fifty to sixty percent of patients are still alive.

We know the classic risks: cigarettes, alcohol, and betel quid. But chronic inflammation and microbial imbalance—dysbiosis—keep popping up in cancer stories, including head and neck cancers. Older culture-based studies hinted at more anaerobes on tumors.

Names like Porphyromonas gingivalis and Fusobacteria came up again and again. Then came sixteen S ribosomal RNA gene sequencing, which doesn’t need to grow microbes in a dish, and suddenly the oral microbiome was a map we could read. Still, one piece was missing: a stage-by-stage view in humans that asked, not just “who’s there,” but “how does the community reorganize as OSCC progresses?”

That’s the gap Yang and colleagues set out to fill in a study published in Frontiers in Microbiology. They didn’t follow one person over time—that would be ideal—but they built a cross-sectional snapshot across the disease arc, including two hundred forty-eight people. They had fifty-one healthy controls, forty-one patients at stage one, sixty-six patients at stages two and three grouped together, and ninety at stage four.

Everyone did a standardized oral rinse, and then the team profiled the bacteria using a familiar workhorse: sixteen S rRNA sequencing of the V3 to V4 region on an Illumina MiSeq, paired-end three hundred base reads. If you’ve worked in microbiome pipelines, the rest will sound like a greatest hits album. They used the QIAamp DNA Microbiome Kit for extraction, clustered reads into operational taxonomic units at 97 percent similarity, assigned taxonomy with Greengenes and cross-checked with oral reference databases, and built diversity and community-structure analyses using Bray–Curtis distances and ordination plots.

For differential abundance, they leaned on LEfSe, and for functional inference, PICRUSt mapped predicted genes to Kyoto Encyclopedia of Genes and Genomes pathways. To turn microbes into markers, they used multivariate logistic regression and receiver operating characteristic curves, reporting area under the curve values with confidence intervals.

Before we dive into the names and numbers, let’s step back for the big picture. The oral microbiome was not static. It shifted with stage, and by the time patients reached stage four, the community looked and behaved differently from health.

That pattern showed up at multiple levels: in measures of diversity, in which phyla dominated, and in the rise and fall of specific species. The result isn’t just a list; it’s a trajectory.

Start with diversity. Think of Shannon diversity as a way to capture richness and evenness in one number. In late-stage OSCC, that number was higher than in healthy mouths, and the difference was striking—statistically rock-solid with a p-value below 0.001.

Richness measures like Chao1 and observed species nudged upward only a little. That tells you the story isn’t simply “more bugs.” It’s a rebalancing—a community spreading its bets more evenly among certain players as the disease advances.

Now look at the whole-community layout. When the team compared samples by their overall composition—using Bray–Curtis distances and principal coordinates analysis—stage four didn’t just blur into health. It pulled away.

On hierarchical clustering, stage four samples tended to sit together, forming their own neighborhood. And as you moved from stage one to stages two and three to stage four, the neighborhood shifted progressively, not randomly. That gradient matters; it suggests the microbiome isn’t just different in cancer; it tracks with how far the cancer has gone.

Zoom out to the big phyla and who dominates. In healthy controls, Firmicutes led the pack—about three-fifths of the community—followed by Proteobacteria, with Actinobacteria and Bacteroidetes making smaller showings. Fusobacteria were bit players in health, roughly three percent.

With cancer progression, that last name—Fusobacteria—kept rising. By stage four, it landed around eight percent, and that wasn’t a wobble; the increases were significant from stage one onward. Meanwhile, Bacteroidetes and Actinobacteria ebbed as disease advanced.

Firmicutes and Proteobacteria, despite their dominance, stayed relatively steady across stages. So the headline at this level is simple: late-stage OSCC brings more Fusobacteria and less of certain commensals.

At the genus level, the familiar oral cast was all there—Streptococcus, Haemophilus, Veillonella, Neisseria, and Rothia—but their billing changed. In healthy mouths, Streptococcus was the star, roughly a third of reads. In stage four, it dropped to just over a quarter.

Fusobacterium—the genus that includes several periodontal pathogens—rose. And a handful of other genera moved in quieter but consistent ways: Haemophilus and Actinomyces trended down in late disease, while Veillonella and Neisseria jostled in the top ranks without dramatic swings. The key is the pattern: increases in Fusobacterium and decreases in several commensal genera, in step with progression.

Species-level resolution sharpened that picture. Using LEfSe to sift signal from noise, Yang’s team highlighted five species enriched in stage four: Fusobacterium periodonticum, Parvimonas micra, Streptococcus constellatus, Haemophilus influenzae, and Filifactor alocis. Several species were depleted in late-stage disease, including Streptococcus mitis, Haemophilus parainfluenzae, Porphyromonas pasteri, Veillonella parvula, and Actinomyces odontolyticus.

A couple of numbers tell the story. Fusobacterium periodonticum climbed from about 1.7 percent at stage one to 3.3 percent at stage four. Parvimonas micra jumped from well under one percent in health to roughly 3.7 percent in stage four.

On the flip side, Streptococcus mitis fell from about 32 percent in healthy controls to under 20 percent in stage four, and Porphyromonas pasteri slipped from around 3.4 percent to under one percent. Those are not rounding errors. That’s a community reconfiguring, with potential implications for inflammation, tissue invasion, and metabolic cross-talk in the tumor microenvironment.

One more twist at the species level: a few operational taxonomic units surfaced only in stage four, including Neisseria elongata, Eikenella corrodens, an Oribacterium oral taxon, and Dialister pneumosintes. Think of these as rarer or more context-dependent residents that appear when the ecosystem changes enough to welcome them. It’s a hint that late-stage disease might open ecological niches that weren’t there before.

Turning those shifts into diagnostic signal, Yang and colleagues built a simple, biologically grounded panel. Three species: one up and two down. Elevated Fusobacterium periodonticum, combined with reduced Streptococcus mitis and Porphyromonas pasteri, separated stage four from health with an area under the receiver operating characteristic curve of 0.956, and the 95 percent confidence interval—from 0.925 to 0.986—was tight.

For a single bacterium, Parvimonas micra did well on its own, with an area under the curve of 0.883. Fusobacterium periodonticum alone hit 0.864. But the trio together carried the load.

That’s important: it says a small, interpretable set of microbes can carry strong discriminative power, at least in this cohort.

Of course, composition is only half the tale; function matters too. PICRUSt let the team infer metabolic potential from the sixteen S profiles. As OSCC advanced, the predicted metagenomes leaned harder into carbohydrate and energy metabolism.

Pathways linked to methane metabolism and oxidative phosphorylation were enriched with progression. Meanwhile, pathways for building blocks—amino acids like valine, leucine, isoleucine, aromatic amino acids—and folate biosynthesis were relatively stronger in health. That pivot suggests late-stage communities may be wired for higher energy turnover and different carbon processing, which could dovetail with the metabolic demands of inflamed, hypoxic tumor niches. It’s an inference, not a direct measurement, but it’s a cohesive one.

A couple of technical notes help interpret what we’re hearing. The sequencing depth was substantial—tens of millions of reads across the cohort—and after quality control the data clustered into four hundred twenty-four operational taxonomic units. A core set of eighty-nine operational taxonomic units popped up in ninety percent of samples, and forty-seven were shared across all groups.

Those core microbes are the bedrock of the oral ecosystem. The stage-specific changes ride on top of that, which makes them more interesting: the ecosystem keeps its backbone, but its proportions and peripheries adapt with disease.

Now, the guardrails. The healthy controls were not matched to cancer patients for age, gender, or oral health status. That matters because age and periodontal health can nudge the microbiome on their own.

The samples were oral rinses, which give a mouth-wide average and can pull in bacteria from plaque and mucosa; they may not reflect what’s sitting right on the tumor. DNA extraction and sequencing choices can tilt which taxa you see and how abundant they look. And the cross-sectional design means we can’t say the microbes caused the cancer, or even that they accelerated it.

We can say they track with it, and that tracking is reproducible across several measures.

So what do we do with this? First, we have a progression-resolved map: more Fusobacteria as OSCC advances, fewer commensals like Streptococcus mitis and Haemophilus parainfluenzae, and a stage four community that’s measurably distinct in composition and predicted function. Second, we have a compact microbial signature with excellent discriminative performance for late-stage disease in this cohort.

Third, we have a hypothesis about function—that late-stage ecosystems are geared toward carbohydrate and energy metabolism—ready to be tested with shotgun metagenomics or metabolomics.

The restraint is healthy here. Yang and colleagues don’t sell these microbes as causes, cures, or screening tools tomorrow morning. They lay out a blueprint: validate the panel in a new, matched cohort.

See if the signal holds up when you add clinical risk factors like tobacco, alcohol, and betel quid. Sample tumor-adjacent sites to compare with whole-mouth rinses. And follow people over time to watch the ecosystem shift before, during, and after treatment. That’s where causality and clinical utility begin to come into focus.

If you’re imagining this as a future saliva test, you’re not wrong to be curious. An area under the curve just under 0.96 for distinguishing stage four from health is eye-catching. But the real win might be earlier: if a stage-resolved signature emerges at stage one or two, and if it adds value beyond what a clinician can see and feel, then we have something that could change decision-making.

Right now, the late-stage separation tells us the microbiome is deeply entwined with the disease environment. That’s not a biomarker in a bottle yet. It is, however, a clear signal that the neighborhood changes as the tumor grows—and that you can hear that change in the data.

Underneath all the names and percentages is a simple narrative. An inflamed, stressed tissue environment shifts the balance of residents. Pathobionts—organisms that live peacefully until conditions tilt—find an opening.

Keystone commensals lose ground. The energy needs of the ecosystem change, and so do the tools the microbes bring to the job. Yang and colleagues captured that shift across a human population with careful sequencing and conservative statistics.

It’s a snapshot, but it’s a revealing one. And for a disease that still takes too many lives, insights like this—grounded in data, tied to stage, and open about their limits—are the kind that move a field forward.

Related lectures