Background: Shifts in the oral microbiome from a healthy state to a dysbiotic one are a key driver in oral diseases such as caries and periodontitis. While shotgun metagenomics has enabled taxonomic profiling of these diseases, a comprehensive, multifaceted analysis comparing their taxonomic, functional, resistance, and mobile genetic element profiles is needed. Furthermore, a lack of sophisticated in vitro models that capture host-biofilm interactions limits the ability to study the longitudinal progression of oral dysbiosis.
Aims: This project aimed to 1) conduct a comprehensive bioinformatic analysis of public
metagenomic data to define and compare the taxonomic, functional, resistome, and
mobilome signatures of dental caries and periodontitis compared to healthy controls. 2)
Utilise machine learning to determine the predictive power of these signatures for disease classification. 3) Validate a novel host-biofilm co-culture model to observe the microbial and functional changes during an induced dysbiotic shift over 14 days.
Materials and Methods: Publicly available shotgun metagenomic data for caries and
periodontitis cohorts were downloaded and processed through a custom bioinformatic
pipeline, including assembly, gene prediction, and annotation against custom and public
databases. This generated counts for taxonomic, functional, antibiotic resistance gene
(ARG), and horizontal gene transfer (HGT) profiles. These counts were used to train
classical and deep learning models for disease classification. Concurrently, a stratified
epithelial collagen gel model was co-cultured with a complex, multi-species oral biofilm
derived from healthy human donors. Dysbiosis was induced using a high serum medium,
and samples were collected over a 14-day period for metagenomic and metatranscriptomic
sequencing.
Results: Bioinformatic analysis of public data revealed distinct taxonomic and functional
profiles for caries and periodontitis compared with health. The resistome and mobilome
analyses identified differentially abundant ARGs and HGT events associated with each
disease state. Machine learning models accurately classified periodontitis samples from
healthy samples but failed to distinguish between caries and healthy samples. The in vitro
co-culture model successfully demonstrated a longitudinal shift in both microbial
composition and gene expression following the dysbiotic media use, containing key markers of a known periodontitis state.
Conclusion: The findings suggest that dental caries and periodontitis possess unique
profiles beyond taxonomy to the functional, resistome, and mobilome level. These
signatures are robust enough to serve as predictive biomarkers for disease classification in periodontitis. Furthermore, the developed host-biofilm model provides a novel platform for investigating mechanisms driving the shift from oral health to disease and for testing new therapeutics.
