Shotgun Metagenomics Identifies in a Cross-Sectional Setting Improved Plaque Microbiome Biomarkers for Peri-Implant Diseases

Abstract

Aim: This observational study aimed to verify and improve the predictive value of plaque microbiome of patients with dental implant for peri-implant diseases.

Materials and methods: Patients were included in one of the following study groups according to the health status of their dental implants: (a) healthy, (b) affected by mucositis and (c) affected by peri-implantitis. From each patient, submucosal plaque microbiome samples were collected from the considered dental implant and from a contralateral healthy implant/tooth. After shotgun metagenomic sequencing, the plaque microbiome was profiled taxonomically and functionally with MetaPhlAn 4 and HUMAnN 3, respectively. Taxonomic and functional profiles were fed into machine-learning models, which were then evaluated with cross-validation to assess the extent to which the plaque microbiome could be used to pinpoint peri-implant diseases.

Results: Shotgun metagenomics sequencing was performed for a total of 158 samples spanning 102 individuals. Four-hundred and forty-seven prokaryotic species were identified as part of the peri-implant microbiome, 34% of which were currently uncharacterized species. At the community level, the peri-implant microbiome differed according to the health status of the implant (p ≤ 0.006 for all pairwise comparisons) but this was site-specific, as healthy contralateral sites showed no discriminating microbiome features. Peri-implantitis microbiomes further showed lower inter-subject variability than healthy plaque microbiomes (p < 0.001), while mucositis-associated microbiomes were in the middle of the continuum between health and peri-implantitis. Each health condition was associated with a strong signature of taxonomic and functional microbiome biomarkers (log₁₀ LDA score ≥ 2.5), 30% and 13% of which represented uncharacterized microbial functions and unknown species, respectively. Distinct Fusobacterium nucleatum clades were associated with implant status, highlighting the subspecies F. nucleatum's functional and phenotypic diversity. Machine-learning models trained on taxonomic or functional plaque microbiome profiles were highly accurate in differentiating clinical groups (AUC = 0.78-0.96) and highlighted the extent to which the peri-implant microbiome is associated with peri-implant clinical parameters (AUC = 0.79-0.87).

Conclusions: Plaque microbiome profiling with shotgun metagenomics revealed consistent associations between microbiome composition and peri-implant diseases. In addition to pointing to peri-implant-associated microbes, warranting further mechanistic studies, we showed high-resolution plaque microbiome evaluation via metagenomics as an effective tool. Its utility within protocols for clinical management of peri-implant diseases should be explored in the future.

Keywords: microbiome; mucositis; peri‐implant diseases; peri‐implantitis; shotgun metagenomics.

FIGURE 1

Overview of the taxonomic composition of the peri‐implant microbiome. (a) Comparison between the amount of unknown species‐level genome bins (uSGBs, red) and known species‐level genome bins (kSGBs, black); kSGBs are further divided between those already detected in Ghensi et al. (2020) and those detected for the first time in this work. (b) Abundance profiles across samples of the 25 most abundant SGBs (according to their 25th percentile stratified by study condition). uSGBs are reported in red and represent new unknown species. (c) Ordered prevalence of the SGBs with > 10% prevalence, divided into kSGBs and uSGBs.

FIGURE 2

Taxonomic and functional clustering of peri‐implant microbiome samples according to disease state. (a) Multidimensional scaling (MDS) ordination plot of healthy, mucositis and peri‐implantitis sites based on the Bray–Curtis beta‐diversity between microbiome samples. p‐values were obtained by PERMANOVA. (b) and (c) report the same type of ordination plots but based on functional abundance profiles (UniProt90 gene family relative abundances) and on the taxonomic profiles of the contralateral samples from implants/teeth.

FIGURE 3

Plaque microbiome biomarkers across disease states. The heatmap (a) reports the relative abundances (log scale) and the effect sizes (the LDA score from LEfSe) of the top 10 species‐level genome bins (SGBs) with the highest effect sizes that characterize each study condition: Health SGBs (health vs. peri‐implantitis, green effect size bars); mucositis SGBs (mucositis vs. health or peri‐implantitis, grey effect size bars): and peri‐implantitis SGBs (peri‐implantitis vs. health, red effect size bars). The p‐values of the abundance test performed by LEfSe are shown inside the bar. The rectangular boxes represent the results of the prevalence test (Fisher exact test) between healthy and peri‐implantitis biomarkers: The colour of the box corresponds to the study condition in which the associated species is more prevalent, while the p‐values are reported as stars (white boxes are used to indicate species with a non‐significant difference in prevalence between healthy and peri‐implantitis samples). Arrows point to unknown SGBs. Panel (b) specifically highlights the abundance patterns and associations of the five subspecies of F. nucleatum .

FIGURE 4

Microbiome‐based machine‐learning classification of disease states. Evaluation of the performance of the random forest classifier in classifying samples into the study conditions based on taxonomic features (SGB relative abundances, a–c) and functional features (UniRef90 relative abundances, d–f). ROC curves and AUC values are averaged across testing folds. Grey areas represent the interval of ±1 standard deviation away from the mean.

FIGURE 5

Association between plaque microbiome and peri‐implant clinical parameters using machine learning. Receiver operating characteristic (ROC) curves and mean area under the ROC curve (AUC) values for the cross‐validation performance of the random forest classifier based on SGB relative abundances for predictingars shown: Different thresholds of bleeding on probing (BOP) positive sites (a), different thresholds of suppurating positive sites (b), different thresholds of bop and/or SUP sites (c), different thresholds of mPPD (d) and combined clinical signs of health (g). The scatterplots show the significant correlations between SGB relative abundances and mPPD (e, f, h, i); Spearman correlation coefficients and R ² values are reported.