Pooled analysis of oral microbiome profiles defines robust signatures associated with periodontitis

Abstract

Oral microbial dysbiosis has been associated with periodontitis in studies using 16S rRNA gene sequencing analysis. However, this technology is not sufficient to consistently separate the bacterial species to species level, and reproducible oral microbiome signatures are scarce. Obtaining these signatures would significantly enhance our understanding of the underlying pathophysiological processes of this condition and foster the development of improved therapeutic strategies, potentially personalized to individual patients. Here, we sequenced newly collected samples from 24 patients with periodontitis, and we collected available oral microbiome data from 24 samples in patients with periodontitis and from 214 samples in healthy individuals (n = 262). Data were harmonized, and we performed a pooled analysis of individual patient data. By metagenomic sequencing of the plaque microbiome, we found microbial signatures for periodontitis and defined a periodontitis-related complex, composed by the most discriminative bacteria. A simple two-factor decision tree, based on Tannerella forsythia and Fretibacterium fastidiosum, was associated with periodontitis with high accuracy (area under the curve: 0.94). Altogether, we defined robust oral microbiome signatures relevant to the pathophysiology of periodontitis that can help define promising targets for microbiome therapeutic modulation when caring for patients with periodontitis.

Importance: Oral microbial dysbiosis has been associated with periodontitis in studies using 16S rRNA gene sequencing analysis. However, this technology is not sufficient to consistently separate the bacterial species to species level, and reproducible oral microbiome signatures are scarce. Here, using ultra-deep metagenomic sequencing and machine learning tools, we defined a simple two-factor decision tree, based on Tannerella forsythia and Fretibacterium fastidiosum, that was highly associated with periodontitis. Altogether, we defined robust oral microbiome signatures relevant to the pathophysiology of periodontitis that can help define promising targets for microbiome therapeutic modulation when caring for patients with periodontitis.

Keywords: dysbiosis; microbiome; periodontitis; signature.

Fig 1

(A) Principal coordinate analysis of the matrix of species distances for periodontitis sites and healthy sites. Box plots shown along each axis represent the median and interquartile range and indicate the distribution of samples along the given axis. Each point represents a single sample and is colored by site. PERMANOVA values, R² values, and P values are shown. (B and C) Alpha diversity of periodontitis sites and healthy sites, depicted by the species richness and the Shannon index at species level. ^#P refers to P values calculated using limma linear models including study ID. ^$P refers to P values calculated using the Mann-Whitney U test adjusted on study ID. The lower and upper hinges of violin plots correspond to the 25th and 75th percentiles, respectively. The midline is the median. The upper and lower whiskers extend from the hinges to the largest (or smallest) value no further than ×1.5 interquartile range from the hinge, defined as the distance between the 25th and 75th percentiles. (D and E) Alpha diversity of periodontitis sites and healthy sites, depicted by the pathway richness and the Shannon index at pathway level. ^#P refers to P values calculated using limma linear models including study ID. ^$P refers to P values calculated using the Mann-Whitney U test adjusted on study ID. (F) Most common dominant bacteria in the oral microbiome in healthy individuals. (G) Most common dominant bacteria in the oral microbiome in individuals with periodontitis. (H) Triad of Socransky’s red complex in healthy subjects. (I) Triad of Socransky’s red complex in individuals with periodontitis.

Fig 2

(A) Significant differences at species level when comparing healthy and diseased sites using differential abundance methods and plotted with SIAMCAT. We used DESeq2 with the poscounts estimator and with apeglm, limma with TMM values, ANCOM-BC, Maaslin2, and beta-binomial regression with corncob. We keep only features that we found significant in the five methods. P values were corrected for multiple hypothesis testing using the Benjamin-Hochberg procedure, and a false discovery rate < 0.05 was defined as the significance threshold. (B) Machine learning association analysis between taxonomic features (species abundance) of the microbiome and periodontitis showed consistent association. Area under the curve-receiver operating characteristic (AUC-ROC) curves are computed using Lasso models trained using 100 times-repeated fivefold-stratified cross-validations. Shaded areas represent AUC-ROCs from each individual machine learning model. (C) Significant differences at pathway level when comparing healthy and diseased sites using differential abundance methods and plotted with SIAMCAT. We used DESeq2 with the poscounts estimator and with apeglm, limma with TMM values, ANCOM-BC, Maaslin2, and beta-binomial regression with corncob. We keep only features that we found significant in the five methods. P values were corrected for multiple hypothesis testing using the Benjamin-Hochberg procedure, and a false discovery rate < 0.005 was defined as the significance threshold. (D) Machine learning association analysis between functional profiles (pathways abundances) and periodontitis showed consistent association. AUC-ROC curves are computed using Lasso models trained using 100 times-repeated fivefold-stratified cross-validations. Shaded areas represent AUC-ROCs from each individual machine learning model.

Fig 3

(A) Prediction matrix for microbiome-based prediction of diet within each data set (values on the diagonal), across pairs of cohorts (one cohort used to train the model and the other for testing). We report the AUC-ROC values obtained from Lasso models on species-level relative abundances. Values on the diagonal refer to the median AUC-ROC values of 100 times-repeated fivefold-stratified cross-validations. Off-diagonal values refer to AUC-ROC values obtained by training the classifier on the cohort of the corresponding row and applying it to the cohort of the corresponding column. (B–D) AUC-ROC for microbiome-based prediction of periodontitis in the leave-one-dataset-out setting (training the model on all but one dataset and testing on the left-out cohort) with species. We report the AUC-ROC values obtained from Lasso models on species-level relative abundances. (E) Prediction matrix for microbiome-based prediction of diet within each data set (values on the diagonal), across pairs of cohorts (one cohort used to train the model and the other for testing). We report the AUC-ROC values obtained from Lasso models on pathway-level relative abundances. Values on the diagonal refer to the median AUC-ROC values of 100 times-repeated fivefold-stratified cross-validations. Off-diagonal values refer to AUC-ROC values obtained by training the classifier on the cohort of the corresponding row and applying it to the cohort of the corresponding column. (F–H) AUC-ROC for microbiome-based prediction of periodontitis in the leave-one-dataset-out setting (training the model on all but one dataset and testing on the left-out cohort) with pathways. We report the AUC-ROC values obtained from Lasso models on pathway-level relative abundances.

Fig 4

(A) Random-effects model via the combining_mv function in the MetaVolcanoR R package on the random data sets on the species profiles. Random-effects P values obtained from each of these methods were corrected for multiple hypothesis testing using the Benjamin-Hochberg procedure. (B) Random-effects model via the combining_mv function in the MetaVolcanoR R package on the random data sets on the pathway profiles. Random-effects P values obtained from each of these methods were corrected for multiple hypothesis testing using the Benjamin-Hochberg procedure.

Fig 5

(A) ROC of the periodontitis risk index applied to predict type 2 diabetes and mucositis, based on species abundances. (B) ROC of the periodontitis risk index applied to predict type 2 diabetes and mucositis, based on pathway abundances.