Integrative microbiome- and metatranscriptome-based analyses reveal diagnostic biomarkers for peri-implantitis

Abstract

Peri-implantitis is a severe biofilm-associated infection affecting millions worldwide. This cross-sectional study aimed to identify taxonomic and functional biomarkers that reliably indicate peri-implantitis by utilizing paired data from full length 16S rRNA gene amplicon sequencing (full-16S) and metatranscriptomics (RNAseq) in 48 biofilm samples from 32 patients. Both full-16S and RNAseq analyses revealed significant differences between healthy and peri-implantitis samples, with a shift toward anaerobic Gram-negative bacteria in peri-implantitis. Metatranscriptomics identified enzymatic activities and metabolic pathways associated with peri-implantitis and uncovered complex peri-implant biofilm ecology related to amino acid metabolism. Integrating taxonomic and functional data enhanced predictive accuracy (AUC = 0.85) and revealed diagnostic biomarkers including health-associated Streptococcus and Rothia species and peri-implantitis-associated enzymes (urocanate hydratase, tripeptide aminopeptidase, NADH:ubiquinone reductase, phosphoenolpyruvate carboxykinase and polyribonucleotide nucleotidyltransferase). Thus, biofilm profiling at taxonomic and functional levels provides highly predictive disease biomarkers, laying the foundation for novel diagnostic and personalized treatment approaches for peri-implant disease.

Fig. 1. Graphical overview of the experimental design.

CAP canonical analysis of principal co-ordinates, RFE recursive feature elimination, AUC-ROC area under the receiver operating characteristic curve. Implant icon—modified from Zheng et al..

Fig. 2. Microbial class level community composition shows distinct community abundances in healthy and peri-implantitis samples.

a Relative abundance of taxonomic classes averaged for healthy and peri-implantitis groups in full-16S and RNAseq. The error bars show the standard errors of the means. b Full-16S-based bacterial class level relative abundance in each sample. c RNAseq-based transcriptionally active class level relative abundance in each sample. d Dot plots showing the correlation between the relative abundance and relative activity within biofilms with exact correlation values plotted for each class on a barplot. nMDS ordination plots based on Bray–Curtis dissimilarity matrices for class level taxa in e Full-16S and f RNAseq.

Fig. 3. Distinct separation between diagnosis groups based on full-16S species-level taxa.

a nMDS plots based on Bray–Curtis dissimilarity matrices of Full-16S data, aggregated to species level features of the training and validation set samples. b Canonical analysis of principal co-ordinates (CAP) constrained ordination plot based on species data illustrating the CAP axis that maximizes the separation of the diagnosis groups. c Vector overlays on nMDS plot shows the species with highest correlation to the MDS1 axis, with vector length indicating strength and direction representing the nature of the association. Species are colored by bacterial classes. d Results of leave one out cross-validation of CAP analysis showing proportions of correct allocations of observations to diagnosis groups across species and genus level taxonomic features in training and the two test datasets (Germany—species & genus and Italy—genus). e Heatmap showing the log (x + 1) normalized relative abundances for 16S taxonomic features with the highest correlation (cut-off >0.4 or < −0.4) to CAP1 axis. Clustering of species level taxa is based on Spearman Rank correlation matrix.

Fig. 4. Distinct metabolic pathways were expressed in healthy and peri-implantitis samples.

a Global overview of metabolic pathways generated using KEGG annotation of the RNA based ECs and their expression in health and peri-implantitis. b nMDS plots based on Bray–Curtis dissimilarity matrices of log (x + 1) transformed RNA data, aggregated to EC4 level. c CAP constrained ordination plots of EC4 level data illustrating the CAP axis that maximizes the separation of the diagnosis groups.

Fig. 5. Amino acid metabolic pathways as the most differentially expressed pathways between diagnosis groups.

a Volcano plot of differentially expressed ECs in health and peri-implantitis based on the LDA correlation cut-off ±0.4 and p value ≤ 0.001. b Specific metabolic pathways identified using the BRITE classification of ECs, revealed amino acid metabolic pathways as the most abundant differentially expressed pathways between the diagnosis groups. c nMDS ordination plot based on Bray–Curtis matrix of normalized EC counts associated with different amino acid anabolic and utilization pathways. (PERMANOVA p value ≤ 0.01) d Heat map showing the abundances of RNA-based reads grouped to amino acid metabolic pathways. Samples clustered based on the Bray–Curtis similarities. p values are obtained by Mann–Whitney U test with FDR correction, calculated for two diagnosis groups.

Fig. 6. System-level perspective of amino acid ecology within peri-implant biofilms.

a Proposed bacterial metabolic pathways for amino acids production and utilization based on metatranscriptomic profiles obtained in this study. Accompanied bars indicate the average relative abundances of major bacterial classes contributing to the respective amino acid activities. Gray boxes indicate amino acids with both biosynthetic and utilization pathways. Metabolic end-products with potential roles in inflammation and bacterial virulence are shown in red text. b Lysine as a model illustrating the high-resolution view of the curated pathways based on RNA ECs. Color scale indicates the LDA correlation (CAP model based) of each enzyme to the CAP1 axis that best separates the diagnosis groups. Heat map showing the log (x + 1) normalized relative expression levels for lysine anabolic and utilization activities within the important bacterial classes. c Normalized abundances of peptidases in healthy and peri-implantitis samples. p values are obtained by Mann–Whitney U test with FDR correction, calculated for two diagnosis groups. *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. 7. Predicted amino acid activities of major bacterial classes.

a Levels of amino acid anabolic and catabolic activities in diverse taxa visualized as network comprised of 8 amino acids. b nMDS plot showing amino acid activities of the most abundant genera in both diagnosis groups. Vectors represent amino acid pathways, where vector direction indicates association with specific genera, and length reflects the strength of the association. See Supplementary Fig. 8 for more details.

Fig. 8. Random Forest (RF) predictive modeling results.

a Results of AUC-ROC and LOOCV accuracies comparing RF models based on different inputs for diagnosis prediction. b Top biomarker candidates for health and peri-implantitis identified by different input datasets. c Results of RFE showing the optimum number of features from “Species + EC” combination. d Normalized abundances of top biomarkers between diagnosis groups. * indicates p values obtained by Mann–Whitney U test with FDR correction, calculated for two diagnosis groups.