Oral LPS activity may represent a functional marker of microbial inflammatory burden across periodontal and cerebrovascular disease states, with LPS-derived features providing complementary but not independently sufficient discriminatory information when combined with demographic and behavioural variables in machine learning models.
Key Findings
Results
Integrated oral LPS activity increased across disease groups and was highest in participants with both periodontitis and stroke.
Participants were stratified by periodontal status and cryptogenic ischaemic stroke from the SECRETO Oral study.
Log-transformed mean oral LPS activity was evaluated across disease states.
The gradient of LPS activity followed a pattern reflecting combined disease burden, with co-occurrence of periodontitis and stroke associated with the highest LPS levels.
LPS activity was quantified using a recombinant Factor C assay.
Results
The subgingival-to-salivary LPS activity ratio differed across disease states, indicating altered niche distribution of microbial inflammatory activity.
The subgingival-to-salivary LPS activity ratio was evaluated as a distinct variable alongside raw LPS activity measures.
Differences in the ratio across disease groups suggest that the relative contribution of subgingival versus salivary compartments to overall oral LPS burden varies by disease state.
This ratio was included as an LPS-derived feature in machine learning models.
The finding implies that oral niche-specific microbial activity, not just total LPS burden, may be relevant to systemic disease associations.
Results
Machine learning models combining demographic, behavioural, and LPS-derived features discriminated disease groups better than models using either feature set alone.
An exploratory machine-learning framework was used to evaluate feature combinations.
LPS-derived features provided complementary, but not independently sufficient, discriminatory information for disease group classification.
Models were compared using demographic and behavioural variables alone, LPS-derived features alone, and combined feature sets.
The superiority of combined models suggests that LPS activity adds information beyond standard clinical and behavioural predictors but cannot replace them.
Background
Taxonomic composition alone does not capture the host-relevant inflammatory activity of microbial products, motivating the use of functional LPS activity measures.
The study was motivated by the premise that oral dysbiosis may contribute to systemic inflammation but that taxonomic composition alone is insufficient to capture this.
LPS activity was proposed as a functional marker of microbial inflammatory burden rather than a compositional marker.
The recombinant Factor C assay was used to quantify LPS activity as a functional readout of endotoxin levels.
Both salivary and subgingival compartments were sampled to provide a more complete picture of oral LPS activity.
Conclusions
The study was conducted as an exploratory analysis requiring validation in independent cohorts with paired microbiome and endotoxin data.
Participants were drawn from the SECRETO Oral study, which includes individuals with cryptogenic ischaemic stroke.
The authors explicitly note that findings are exploratory and require validation in independent cohorts.
Paired microbiome and endotoxin data were identified as necessary components for future validation studies.
The machine learning framework was described as exploratory rather than confirmatory.
What This Means
This research suggests that measuring the inflammatory activity of bacterial byproducts called lipopolysaccharides (LPS) in saliva and beneath the gumline may help distinguish between different states of gum disease and stroke risk. The study found that people with both periodontitis (severe gum disease) and cryptogenic ischaemic stroke (a type of stroke with no clear cause) had the highest levels of oral LPS activity, and that the balance of LPS activity between subgingival (under the gum) and salivary compartments also varied by disease group. These patterns suggest that the mouth's bacterial environment may contribute differently to inflammation depending on a person's overall disease burden.
Using a machine learning approach, the researchers found that combining LPS measurements with demographic and lifestyle information (such as age, sex, and smoking status) produced better disease classification than using either type of information alone. However, LPS measures on their own were not sufficient to reliably distinguish disease groups, meaning they add useful information but cannot replace conventional clinical variables. This points to oral LPS activity as a potential functional biomarker — one that reflects what bacteria are actually doing in terms of producing inflammatory signals, rather than simply which bacteria are present.
This research matters because it moves beyond simply cataloguing which bacteria live in the mouth and instead tries to measure the actual inflammatory impact of those bacteria on the body. If validated in larger and independent studies, this approach could eventually help identify individuals at higher risk of both severe gum disease and stroke, and might offer a non-invasive way to monitor microbial-driven inflammation. The authors emphasize that these findings are preliminary and need confirmation before any clinical conclusions can be drawn.
Dong A, Xie Z, Manzoor M, Leskelä J, Putaala J, Könönen E, et al.. (2026). Machine Learning Stratification of Periodontal and Cerebrovascular Disease Status Using Salivary and Subgingival Lipopolysaccharide Activity.. Journal of cellular and molecular medicine. https://doi.org/10.1111/jcmm.71337