We recruited a Japanese cohort of 22 healthy individuals (age: 43.7 ± 11.8 years; 10 female and 12 male), 10 elderly individuals with more than 26 natural teeth (age: 83.4 ± 2.3 years; 5 female and 5 male) and 24 subjects with periodontal disease (age: 49.3 ± 12.0 years; 9 female and 15 male) (Figure 1; Supplementary Table 1). Subjects in the elderly group did not receive treatment for caries or periodontal disease in the prior 10 years, and oral conditions were well controlled by long-term maintenance by dentists. Subjects’ mouth-rinsed water, feces, anthropometric information (age and sex) and oral health indicator (number of teeth, rate of bleeding on probing (BOP) and rate of probing pocket depth (PPD) no less than 4 mm (PPD ≧4 mm) data were collected (Table 1). Elderly subjects had significantly higher BOP and PPD > 4 mm rates than healthy subjects but significantly lower rates than subjects with periodontal disease (Table 1).

We conducted oral microbiome analysis by sequencing the amplified 16S rRNA gene. There was no significant difference in the alpha diversity of the oral microbiome among the healthy group, periodontal disease group, and elderly group (Figure 2A; Kruskal–Wallis test; p = 0.195). Principal coordinate analysis (PCoA) and permutational multivariate analysis of variance (PERMANOVA) of the oral microbiome were performed to determine whether the oral microbiome composition of elderly individuals was distinct from that of healthy adults or periodontal disease patients (Figures 2B–D). PCoA of weighted UniFrac distance showed distinct separation between healthy and elderly subjects but not between elderly individuals and subjects with periodontal disease. PERMANOVA showed that oral microbiome composition variance was significantly explained by the subject group for the elderly and healthy subjects but not for the elderly individuals and subjects with periodontal disease (Table 2; healthy vs. elderly R² = 27.9%, Bonferroni corrected p = 0.015; periodontal disease vs. elderly R² = 6.2%, Bonferroni corrected p = 1.000), supporting the PCoA results (Figures 2B,C).

PCoA and PERMANOVA of the Bray–Curtis dissimilarity of the oral metabolome showed that the oral metabolome profiles of the elderly subjects were different from those of the individuals in the other groups (Figure 2D). PERMANOVA showed that the grouping significantly explained the oral metabolome composition variance for the healthy vs. elderly and elderly vs. periodontal disease groups (Table 2; healthy vs. elderly R² = 29.8%, Bonferroni corrected p = 0.006; periodontal disease vs. elderly R² = 21.6%, Bonferroni corrected p = 0.015).

As PERMANOVA revealed a distinct elderly oral microbiome and metabolome, differential abundance analysis was performed to identify oral microbiome and metabolome features that were associated with the elderly individuals (Figure 3). Genus level analysis results revealed 10 and 8 genera with significant differences in relative abundance (p < 0.05) from healthy vs. elderly (Figure 3A) and healthy vs. periodontal disease group comparisons (Figure 3C), with 5 genera showing significant differences in abundance in both comparisons. Among them, Neisseria, Haemophilus, and Bergeyella were enriched in the healthy group, and Prevotella and Megasphaera were depleted in the healthy group. False discovery rate (FDR) corrections (q < 0.05) reduced the significant features to 6 and 2 genera (enriched in the elderly group: Prevotella, Veillonella, Oribacterium, Stomatobaculum and Megasphaera; depleted in the elderly group: Haemophilus; enriched in the periodontal disease group: Treponema and Filifactor). Some features were observed in the elderly group, which may be influenced by the higher periodontal disease-related score in that group. Therefore, analysis of covariance (ANCOVA) between the periodontal disease and elderly groups was performed using the periodontal disease score (rate of PPD no less than 4 mm) as a confounding factor (Supplementary Table 2). The results showed that three microbes significantly more prevalent in the elderly group than in the healthy group were also significantly enriched by ANCOVA (Oribacterium, Stomatobaculum, and Selenomonas), and one of those less prevalent in the healthy group was also significantly less prevalent according to ANCOVA (Haemophilus).

The differential abundance analysis for each metabolite revealed significant differences in the levels of 114 metabolites between the healthy and elderly groups (Supplementary Table 2; Figure 3B) and in the levels of 20 metabolites between the healthy and periodontal disease groups (enriched in periodontal disease group; 2-hydroxy-4-methylpentanoate, 2-hydroxyglutarate + citramalate, 3-ureidopropionate, 4-hydroxymandelate, agmatine, Ala-Ala, CMP, fumarate, His, N1,N8-diacetylspermidine, octylamine, Phe, Phe-Phe, phenylpyruvate, threonate, urocanate, o-succinylhomoserine; depleted in the periodontal disease group; biotin, desthiobiotin, isobutylamine; Figure 3D), with 10 overlapping metabolites (2-hydroxy-4-methylpentanoate, 2-hydroxyglutarate + citramalate, agmatine, Ala-Ala, His, N1,N8-diacetylspermidine, Phe, Phe-Phe, threonate, urocanate; Supplementary Table 2). All 10 overlapping metabolites increased in abundance compared to that in the healthy group. After FDR correction, there were only 70 significant differences between the healthy and elderly groups. The oral metabolites of elderly individuals and subjects with periodontal disease showed partially similar enrichment patterns compared to those of healthy subjects. Except one metabolite (ethanolamine phosphate), 113 metabolites were significantly enriched, indicating that many metabolites were enriched in the elderly group (Supplementary Table 2). For example, oral SCFAs, such as butyrate and propionate, were enriched in both elderly individuals and subjects with periodontal disease compared to those in healthy subjects, but this enrichment was not significant in subjects with periodontal disease. As for bacteria, ANCOVA was performed using the periodontal disease score as a confounding factor. The results showed that for 82 metabolites, there was a significant increase in abundance after adjusting for periodontal disease score (Supplementary Table 2; p < 0.05). This indicates that many metabolites are enriched in the elderly group due to factors beyond the effects of periodontal disease.

We analyzed the gut microbiome and metabolome as well as oral samples. There was no significant difference in the alpha diversity of the gut microbiome among the healthy group, periodontal disease group, and elderly group (Figure 4A; Kruskal–Wallis test; p = 0.834). PCoA and PERMANOVA were also performed on the gut microbiome weighted/unweighted UniFrac distance and metabolome Bray–Curtis distance profiles of the healthy and elderly groups (Figures 4B–D; Table 3). Periodontal disease subjects were excluded from the PERMANOVA due to insufficient fecal samples (n = 4). PERMANOVA of the gut metabolome composition of the elderly and healthy subjects showed significant differences, but not for the gut microbiome (Table 3).

Similar to the compositional analysis results, differential abundance analysis between the gut microbiome and metabolome of elderly and healthy subjects showed minimal differences, with no microbes or metabolites showing significant differences in abundance after FDR correction. Before FDR correction, two genera (Megasphaera and Catenibacterium) and 4 metabolites (Gly-Gly, 10-hydroxydecanoate, propionate, and butyrate) were enriched in the elderly group (Supplementary Table 3; Figure 5).

Oral-gut transmission of microbes, especially in patients with gastrointestinal diseases, has been indicated by several studies (Schmidt et al., 2019; Kitamoto et al., 2020; Park et al., 2021), including reports demonstrating higher oral-gut microbiome similarity in elderly subjects than in healthy adults (Iwauchi et al., 2019). Therefore, to identify putative oral-gut transmission in our cohort, oral-gut association analyses were performed on the subjects’ oral and gut microbiomes and metabolomes.

First, we examined how many of the same amplicon sequence variants (ASVs) were detected in both the oral cavity and gut as an indicator of potential oral-gut transmission. To exclude the effect of the number of ASVs for each individual, the ratio of ASVs detected simultaneously in the oral and gut to the total number of ASVs in both the oral and gut microbiome was calculated for each individual (referred to as “shared ASV ratio”). As the however, only a few ratio of ASVs were detected in both the oral cavity and gut, indicating minimal oral-gut transmission (Figure 6A; Supplementary Table 4). In addition, we compared shared ASV ratio between the healthy and elderly groups, but no significant difference was detected (Wilcoxon rank-sum test; p = 0.807).

Next, correlation analysis was performed on metabolites that were detected in both the oral cavity and gut to further validate putative oral-gut transmission. Since most microbes were detected in only one of the two environments, correlation analysis was performed only for metabolites. Correlations were observed for some metabolites for each group. No metabolites showed significant correlations in both groups, but positive correlations were found for four metabolites in the elderly group (anserine + homocarnosine, homoserine, 3-methylhistidine and isopropanolamine) and for two metabolites in the healthy group (diethanolamine and gamma-amino butyric acid (GABA)) (Figure 6B; Supplementary Table 5).

This trial was approved by the clinical trial ethics review committee of the Chiyoda Paramedical Care Clinic and reported to https://www.umin.ac.jp/ (identifier: UMIN000031334). All participants understood the study objective and provided written informed consent, which we affirmed. All experiments were performed in accordance with approved guidelines.

Participants were recruited at the Hiyoshi Oral Health Clinics and grouped into healthy, periodontal disease and elderly groups based on the following criteria. The healthy group criteria were (a) between 20 and 79 years old, (b) all PPD less than 4 mm, (c) BOP tooth surface less than 10%, (d) no dental caries, (e) no denture or implants, (f) nonsmoker, (g) no orthodontic treatment and (h) no antibiotic consumption during the prior 6 months. The periodontal disease group criteria were as follows: (a) between 20 and 79 years old, (b) with bleeding from PPD ≧ 4 mm, (c) no carious pit, (d) no denture or implants, (e) nonsmoker, and (f) no antibiotic consumption during the prior 6 months. The elderly group criteria were (a) between 80 and 100 years old, (b) more than 26 original teeth, and (c) did not receive treatment for caries or periodontal disease during the prior 10 years. All participants were instructed not to eat, drink, or brush their teeth for at least 1 h prior to sampling. The sample size for the elderly group was set to ten. Preliminary power calculations were not performed, as there were no reference studies. However, it was assumed to have a large effect size for the elderly group, and the sample size was set to ten in accordance with previous studies (Whitehead et al., 2016).

Participants’ oral samples were collected at Hiyoshi Oral Health Clinics, as previously described (Jo et al., 2019). Specifically, participants were instructed to wash their mouths with 3 mL of sterilized water, which was collected for microbiome and metabolite extraction. All samples were stored in a freezer within 10 min of collection and kept at −80°C until further processing.

Participants’ oral health conditions were examined after sample collection. The numbers of original, decayed, missing and filled teeth were recorded as indicators of participants’ caries history. PPD and BOP at four sites (mesiobuccal, distobuccal, mesiolingual, and distolingual) of all teeth were measured with a periodontal pocket probe as indicators of periodontal disease.

DNA extraction from mouth-rinsed water and fecal samples was performed as previously described (Jo et al., 2019; Hashimoto et al., 2023). PCR was conducted using universal primers (27Fmod and 338R) for 16S rRNA gene sequencing, as previously described. PCR was performed using Ex Taq polymerase (Takara Bio, Shiga, Japan) and approximately 20 ng of template DNA.

Thermal cycling was performed in a Veriti Thermal Cycler (Life Technologies Japan, Tokyo) with the following cycling conditions: initial denaturation at 96°C for 2 min, followed by 25 cycles of denaturation at 96°C for 30 s, annealing at 55°C for 45 s, and extension at 72°C for 1 min, with a final extension at 72°C for 10 min. PCR amplicons were purified using AMPure XP magnetic purification beads (Beckman Coulter, Brea, CA, USA) and quantified using a Quant-iT PicoGreen dsDNA Assay Kit (Life Technologies Japan). After quantification, mixed samples were prepared by pooling approximately equal amounts of each amplified DNA and sequenced using a MiSeq Reagent Kit V3 (300 bp × 2 cycles) and a MiSeq sequencer (Illumina, CA, USA) according to the manufacturer’s instructions.

The extraction of metabolites from mouth-rinsed water and fecal samples was also previously described (Maruyama et al., 2022). Frozen saliva samples were thawed and centrifuged at 13,000 × g for 5 min at 4°C. The supernatant was transferred to a 5-kDa-cutoff filter (Human Metabolome Technologies, Tsuruoka, Japan) to remove proteins of sizes greater than 5 kDa. Prior to CE-TOFMS analysis, a 45 μL aliquot of the filtrate was added to 5 μL of Milli-Q water containing reference compounds (200 mmol/L each of methionine sulfone, D-camphor-10-sulfonic acid, 3-aminopyrrolidine, and trimesic acid). CE-TOFMS-based metabolome profiling was performed using an Agilent 7,100 capillary electrophoresis system (Agilent Technologies, Waldbronn, Germany), an Agilent 6,224 TOF LC/MS system (Agilent Technologies, Santa Clara, CA), an Agilent 1,200 series isocratic HPLC pump, a G1603A Agilent CE-MS adapter kit, and a G1607A Agilent CE-electrospray ionization (ESI)-MS sprayer kit. In anionic metabolite analysis, the ESI sprayer was replaced with a platinum needle instead of an initial stainless-steel needle. Other conditions of the CE–ESI–MS sprayer were the same as received. The metabolome analysis conditions were the same as those described previously (Sugimoto et al., 2010). Data analysis was performed using the metabolome analysis software MasterHands as previously described (Hirayama et al., 2009).

For 16S rRNA gene analysis, QIIME2 (version 2019.10) was used (Bolyen et al., 2019). In the analytical pipeline, adapters were trimmed by cutadapt (options: --p-front-f AGRGTTTGATYMTGGCTCAG --p-front-r TGCTGCCTCCCGTAGGAGT --p-discard-untrimmed) (Martin, 2011). Trimmed sequence data were processed by using the DADA2 pipeline for quality filtering and denoising (options: --p-trunc-len-f 275 --p-trunc-len-r 230) (Callahan et al., 2016). The filtered output sequences were assigned to taxa by using the “qiime feature-classifier classify-sklearn” command with the default parameters.

SILVA LTP (version 132) was deduplicated prior to classifier training. The identified duplicate sequences were compared to LTP_2020, and those not included in LTP_2020 were also removed from the SILVA LTP database used in this study. The deduplicated SILVA LTP was combined with the extended HOMD (eHOMD) 16S rRNA RefSeq (version 15.22) database and then processed by using the “qiime feature-classifier extract-reads” command (options: --p-f-primer AGRGTTTGATYMTGGCTCAG --p-r-primer TGCTGCCTCCCGTAGGAGT --p-min-length 150 --p-max-length 450) prior to classifier training. The taxon classifier was trained by using the “qiime feature-classifier fit-classifier-naive-bayes” command with the default parameters. UniFrac (weighted and unweighted) distance was calculated from rarified samples’ ASVs using the “qiime diversity core-metrics-phylogenetic” command (options: --sampling-depth 7,000). PERMANOVA was performed with scikit-bio. Other statistical analyses were performed with in-house Python scripts (version 3.7.6). For pairwise comparison of the abundances of microbes and metabolites, the Wilcoxon rank sum test [scipy version 1.5.2 (Virtanen et al., 2020)] with Benjamini–Hochberg FDR correction (statsmodels version 0.10.0) was used. For the comparison, microbial taxa with a mean relative abundance below 0.001 were excluded. When comparing the abundance of bacteria and metabolites between elderly and periodontal disease group, ANCOVA was performed after performing ranking (pingouin version 0.5.3). Rate of PPD no less than 4 mm was used as a confounding factor. In addition, metabolites detected in 50% of the saliva and fecal samples were included in the oral-gut metabolite correlation analysis.