Participants were recruited from faculty members and graduate students working at the Osaka Dental University hospital, as well as from dentists who were acquainted with an author of this study. Ten healthy volunteers (6 males and 4 females; age: 27–60 years [39.8 ± 3.1 (mean ± SD)]) were enrolled in the study and were anonymized randomly as A–G, O, Q, and R ( Table 1). The inclusion criteria were as follows: healthy men and women over 20 years of age. The exclusion criteria were as follows: daily smoking, treatment with local or systemic antibiotics within 1 month prior to the study, and allergy to kiwifruit. The exclusion criteria of one month for antibiotic treatment was set based on previous reports on the robustness and resilience of salivary microbiome ^{17, 18} . For example, change in microbiome caused by exposure to clindamycin lasted up to 1 month in saliva ¹⁸ . According to the medical questionnaire, (1) none of the participants were undergoing or planning treatment for dental caries or periodontal disease, (2) there were no participants who were suffering from diabetes, chronic kidney disease, lung diseases, malignant tumors, etc., or who were visiting hospitals or taking medication, and (3) none of the participants experienced frequent thirst. The method and objective of this study were explained to the participants, who provided written informed consent before participating. The Osaka Dental University Medical Ethics Committee approved this study (approved on 3/31/2015; approval number 110864) and the investigations were conducted following the rules of the Declaration of Helsinki. The committee did not consider the study to be interventional in nature and therefore is not a clinical trial.

Two types of oral care tablets for tongue cleaning were tested in the current study. One type contained actinidin, a cysteine protease extracted from kiwifruit (“protease tablet”) and the other did not (“plain [placebo] tablet”). Both tablets were provided by Ezaki Glico Co., Ltd (Osaka, Japan). The protease tablets were identical to those marketed as BREO EX (Ezaki Glico Co.). Tablet composition was described previously ¹² . To use the tablets, the participants placed one tablet on the dorsum of the tongue and waited until it dissolved naturally. One tablet takes approximately 5–7 min to completely dissolve.

The study design is illustrated in Figure 1. The tongue tablet experiment was a placebo-controlled double-blind crossover study. The 10 participants were randomly divided into 2 groups of 5 participants each, by using computer-generated random numbers. All participants performed an initial tongue cleaning (by brushing) at the beginning of the study. The participants were asked not to eat, drink, or perform oral cleaning before each sampling. After a washout period of 10 days during which the participants did not perform any tongue cleaning, they collected their saliva and tongue coating into separate containers in the morning immediately after waking up (sample D1). Then, the participants in each group took tablets, with or without the protease. The participants and the researchers who analyzed the data were not informed about the tablet types given to the participants. The participants were asked to use the tablet three times on the day of the experiment—in the morning (between 9–12 am), in the afternoon (1–4 pm), and in the evening (7–10 pm)—taking one tablet each time. The following morning, the participants collected their saliva and tongue coating separately immediately after waking up (sample D2). After a washout period of 10 days, the participants took the other type of tablet that they had not previously received, and collected the saliva and tongue samples as before. Control experiments (no tablet usage) were conducted with the same participants, after they conducted treatment using the tablets. The duration between the tablet treatments and the control experiment ranged from 10–60 days, depending on the participant. In these experiments, after an initial tongue cleaning and a 10-day washout period, all participants collected samples on two consecutive days (D1 and D2) without taking any tablet in between.

The saliva and tongue-coating samples were collected immediately after the participants woke up, in the morning of the day of tongue cleaning by tablet (D1) and the next morning (D2). The participants first collected 3 mL of saliva in a 25-mL sterile plastic tube. The tongue coating was collected by scrubbing the tongue with a swab and then soaking the tip of the swab in 0.6 mL of phosphate-buffered saline (PBS(-); Wako Pure Chemical Industries, Ltd., catalogue number 166-23555) to suspend the coating. Because this collection method involves scrubbing the tongue with a swab, the tongue coating was collected from the left half of the tongue for D1 and from the right half of the tongue for D2, to minimize the carryover effects of scrubbing. The collected samples were maintained at 4°C for up to 1 day and transported to the laboratory. The saliva samples (3 mL) were homogenized by repetitive pipetting. Then, 0.5-mL aliquots were transferred into sterile tubes. The saliva (0.5 mL) and tongue-coating (0.6 mL) samples were then centrifuged at 10,000 × g for 4 min. The supernatant was discarded and the pellet was stored at −20 °C until DNA extraction. All samples were frozen no later than on the day of D2 sampling.

Bacterial DNA was extracted from the pellets via chemical and mechanical lysis using a QIAamp UCP Pathogen Mini kit (QIAGEN, catalogue number 50214), as previously described ^{19, 20} . Briefly, thawed pellets were immediately suspended in 0.5 mL of ATL buffer containing the optional DX reagent, transferred to a Pathogen Lysis Tube S, and then homogenized using a Mixer Mill MM 301 (Retsch) for 3 min at a vibrational frequency of 30 Hz. The manufacturer’s protocol was followed thereafter to complete the DNA purification. DNA was eluted in 50 μL of the AVE buffer (QIAGEN). DNA concentration was determined using a Quantus fluorometer (Promega) and a Qubit dsDNA BR Assay kit (Thermo Fisher Scientific, catalogue number Q32850). DNA was stored at –80 °C until use.

Bacterial 16S ribosomal DNA amplification and library construction were performed according to the 16S Metagenomic Sequencing Library Preparation guide supplied by Illumina (part No. 15044223_B), as previously described ¹⁹ . The V3–V4 region of the 16S ribosomal RNA gene was amplified by polymerase chain reaction (PCR) with a thermal cycler MJ-Mini (Bio-Rad Laboratories), using primers 341F (5’-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3’) and 806R (5’-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGGACTACHVGGGTWTCTAAT-3’) (custom-synthesized by Invitrogen), and Premix Ex Taq Hot Start Version (Takara Bio, catalogue number RR030A). The thermal cycling conditions were initial denaturation at 98 °C for 10 s, followed by 25 cycles at 98 °C for 10 s, 55 °C for 30 s, and 72 °C for 1 min (the first PCR step). The underlined nucleotides served as primer sequence parameters to extract the V3–V4 region for feature classifier training (see next section). The amplicons were purified using AMPure XP beads (Beckman Coulter, catalogue number A63880). Sequencing adapters containing 8-bp indices were incorporated at the 3’- and 5’-ends of the purified amplicons during a second PCR step. The amplicons were again purified using the AMPure XP beads, and then quantified using a Quantus fluorometer (Promega) and a Qubit dsDNA HS Assay kit (Life Technologies, catalogue number Q32851). After pooling equimolar amounts of the amplicons, 5% of an equimolar amount of PhiX DNA (PhiX Control v3, Illumina, catalogue number FC-110-3001), was added. The obtained library was pair-end sequenced at 2 × 250 bp using a MiSeq Reagent Kit v2 (Illumina, catalogue number MS-102-2001) and the Illumina MiSeq platform. Sequencing was performed over seven independent runs at the Oral Microbiome Center (Takamatsu, Japan), followed by demultiplexing. Raw nucleotide sequences are available at DDBJ/EMBL-EBI/NCBI database under the accession number DRA010849.

Demultiplexed paired-end sequences were processed using QIIME 2 (v.2020.2) and its associated plugins ²¹ in a Docker container. Sequences obtained from independent Miseq runs were denoised separately using DADA2 (via q2-dada2) ²² applying previously-optimized parameters ¹⁴ (trim-left-f = 20; trim-left-r = 20; trunc-len-f and trunc-len-r were set between 241 and 248 depending on the sequence quality; other parameters followed the default settings, including chimera-method = “consensus”). The resulting exact amplicon sequence variants (ASVs) were merged (via q2-feature-table). For taxonomy assignment to each ASV, a naïve Bayes taxonomy classifier trained (via q2-feature-classifier) ²³ on the V3–V4 region of the 16S rRNA sequences in the expanded human oral microbiome database ( eHOMD; v.15.2) ²⁴ was used. All ASVs were aligned using MAFFT ²⁵ and used to construct a phylogeny with FastTree 2 (via q2-phylogeny) ²⁶ . Sample metadata format was validated using the cloud-based tool Keemei ²⁷ .

Alpha diversity was assessed by calculating the number of observed features (ASVs) and the Shannon index (via q2-diversity), after samples were subsampled without replacement (rarefied) to 40,000 sequences per sample. The non-parametric Kruskal–Wallis test was used to test for significant differences in alpha diversity between sample groups ( P < 0.05).

Beta diversity was computed based on the unweighted UniFrac distance ²⁸ (via q2-diversity) and visualized as three-dimensional principal coordinate analysis (PCoA) plots using EMPeror (via q2-emperor) ²⁹ . Permutational analysis of variance (PERMANOVA) ³⁰ was used to test the significant difference in bacterial composition among samples ( P < 0.05). Differential abundance of bacterial taxonomic groups was tested using the analysis of composition of microbiomes (ANCOM) (via q2-composition) ³¹ .

Clustered operational taxonomic unit (OTU) table was also used to calculate the beta diversity and differential abundance of specific taxonomic groups. Open-reference clustering was chosen for a high-quality taxonomic assignment to a curated database ³² . Here, the entire ASVs were clustered into OTUs by open-reference picking (via q2-vsearch) ³² using the V3–V4 region of 16S rRNA sequences in the eHOMD v.15.2 as a reference, with 99% identity threshold. A phylogenetic tree was constructed as described above. The R package treeio (v.1.12.0) ³³ was used to change the OTU names within the phylogenetic trees, as per the requirement for the Rhea package (the software disallows OTU names starting with a number). Multidimensional scaling (MDS) based on generalized UniFrac distance ³⁴ was performed to examine the difference in microbial composition among samples, using Rhea pipeline (v.1.1.3) ³⁵ . For differential abundance analysis, “Serial group comparisons” in Rhea was performed (abundance_cutoff = 0.2; prevalence_cutoff = 0.3; max_median_cutoff = 1) with significance cutoff of P < 0.05 in Kruskal-Wallis test.

Box plots that show the alpha diversities were generated using R packages ggplot2 (v.3.3.2) and ggpubr (v.0.4.0), with data retrieved from QIIME 2 artifacts using qiime2R (v.0.99.31). Heatmaps and box plots that show the relative abundance of microbial taxonomies, and a Venn diagram that show core OTUs, were generated using ampvis2 (v.2.6.5) ³⁶ . R (v.4.0.2) and RStudio (v.1.3.959) were used for all analyses. Software, plugins, and R packages used in this study are listed in Table 2.

The study design is illustrated in Figure 1. Ten healthy adults participated in the study, with three different treatments tested: two different types of oral tablets (with or without protease), and a negative control (no tablet). For the tablet treatments, the saliva and tongue coating were collected before (D1) and after (D2) the intervention. Overall, 116 samples were collected (10 participants, three treatments, two sites [the saliva and tongue], and two sampling time points [D1 and D2], with four samples excluded because of insufficient amount of extracted DNA). The sample metadata are provided as underlying data (Table S1) ³⁷ .

DNA was extracted from each sample and the V3–V4 region of the 16S rRNA gene was PCR-amplified. The amplicons were paired-end sequenced using the Miseq platform. After quality control and error correction using DADA2, 11,260,102 reads corresponding to 5342 ASVs were obtained. Per-sample median was 89,923, with a maximum of 186,864 and a minimum of 46,422. Open-reference clustering, using the curated 16S rRNA sequences in the eHOMD v.15.2 database as the reference (at 99% identity threshold), grouped the sequences into 1210 OTUs. Either the full or clustered table was analyzed further, depending on the type of analysis performed, as described. The clustered OTU table is provided as underlying data (Table S2) ³⁸ .

We first analyzed the microbiome of the saliva and tongue coating, to determine the baseline for the study. In total, 30 D1 samples (10 participants, three independent treatments) of the saliva and tongue coating were analyzed. Alpha diversity of the tongue microbiome was significantly lower than that of the salivary microbiome, using both the number of observed ASVs (254 ± 53 in saliva and 175 ± 37 in tongue; P = 8.9e-7, Kruskal–Wallis test) ( Figure 2a) and Shannon index (6.0 ± 0.4 in saliva and 5.4 ± 0.3 in tongue; P = 2.0e-7, Kruskal–Wallis test) as measures ( Figure 2b). This was consistent with a previous report ³ . The difference in diversity is probably associated with the tongue acting as a specialized niche for specific microorganisms, and the saliva containing a mixture of microbiota from different sites in the oral cavity. Interestingly, the number of observed ASVs varied among the individuals, ranging from approximately 170 to 325 in the saliva ( P = 0.027, Kruskal–Wallis test), and from approximately 125 to 225 in the tongue coating ( P = 0.022) ( Figure 2c), suggesting a difference in oral microbiome among individual.

We next assessed the differences in bacterial composition among samples (beta diversity) ( Figure 3). A significant difference between the salivary and tongue microbiomes was detected both in PCoA, based on unweighted UniFrac distances using the full ASV table ( P = 0.001, PERMANOVA) ( Figure 3a), and MDS, based on generalized UniFrac distances ³⁴ using the clustered OTU table ( P = 0.003, PERMANOVA) ( Figure 3c).

The data also revealed a significant difference between individual microbiomes ( Figure 3b and 3d; P = 0.001, PERMANOVA using the clustered OTU table). As indicated by the plots in Figure 3, the similarity of the salivary and tongue microbiomes within an individual was greater than the similarity of the salivary or tongue microbiomes between individuals. This suggests there is stability in an individual’s oral microbiome, at least within the relatively short time period of the study (several weeks). This observation is consistent with earlier studies that highlight the stability of an individual’s oral microbiome ³⁹ .

The abundances of bacterial taxonomic groups at the genus or species levels in the salivary and tongue microbiomes, determined by the analysis of D1 samples from the three treatments and based on the clustered OTU table are summarized in Figure 4a and 4b. The eight most abundant genera were common to the salivary and tongue microbiomes, accounting for nearly 80% of both microbiomes (78.2% in the saliva and 80.9% in the tongue). These genera were Prevotella (18.4% and 23.5%, respectively), Veillonella (9.0% and 12.6%, respectively), Neisseria (11.6% and 9.9%, respectively), Haemophilus (10.9% and 8.6%, respectively), Streptococcus (9.4% and 6.1%, respectively), Alloprevotella (8.2% and 5.9%, respectively), Porphyromonas (6.1% and 6.7%, respectively), and Fusobacterium (4.6% and 7.6%, respectively) ( Figure 4a). The abundance of bacterial taxa at species level is shown in Figure 4b. Prevotella melaninogenica HMT-469 was most abundant in both microbiomes (8.1% in the saliva and 11.8% in the tongue), followed by Streptococcus oralis subsp. dentisani HMT-398 (7.2%) and Haemophilus parainfluenzae HMT-718 (7.0%) in the saliva, and by Fusobacterium periodonticum HMT-201 (7.4%) and H. parainfluenzae HMT-718 (7.4%) in the tongue ( Figure 4b).

Five OTUs were differentially abundant in the salivary and tongue microbiomes ( P < 0.05, Kruskal–Wallis test) ( Figure 4c). Among them, S. oralis subsp. dentisani HMT-398 (7.2% in the saliva and 2.7% in the tongue) and Neisseria mucosa HMT-682 (1.8% and 0.2%, respectively) were more abundant in the saliva, whereas F. periodonticum HMT-201 (3.5% and 7.4%, respectively), P. melaninogenica HMT-469 (8.1% and 11.8%, respectively), and Prevotella histicola HMT-298 (1% and 2.1%, respectively) were more abundant in the tongue ( Figure 4b and 4c).

To identify the core members of the oral microbiome, we focused on OTUs that were present in ≥95% of the saliva or tongue D1 samples. Seven OTUs were present in ≥95% of both, the saliva and tongue samples. These were F. periodonticum HMT-201, Saccharibacteria (TM7) [G-1] sp. HMT-352, S. oralis subsp. dentisani HMT-398, P. melaninogenica HMT-469, Granulicatella adiacens HMT-534, Campylobacter concisus HMT-575, and H. parainfluenzae HMT-718 ( Figure 5).

Further, we identified site-specific core OTUs, detected in ≥95% samples from one site but not from the other. The saliva-specific core OTUs were Veillonella parvula HMT-161, Porphyromonas pasteri HMT-279, Prevotella nanceiensis HMT-299, Gemella haemolysans HMT-626, and Alloprevotella sp. HMT-914. The tongue-specific core OTUs were Catonella morbi HMT-165, Oribacterium sinus HMT-457, Peptostreptococcaceae [XI][G-1] sulci HMT-467, Solobacterium moorei HMT-678, and Rothia mucilaginosa HMT-681 ( Figure 5).

Using the above data as the base line, we finally assessed the effect of taking oral tablets (with or without protease) on the salivary and tongue microbiomes. Alpha diversity in D1 and D2 samples was not significantly different between any treatments ( Figure 6a). In the control (no tablet) treatment, whereas the observed number of ASVs seemed to slightly increase in the saliva and to slightly decrease in the tongue, they were not statistically significant ( Figure 6a). This indicates that some fluctuation of the oral microbiome may occur naturally. Further, MDS analysis indicated that the beta diversity between D1 and D2 samples was not significantly different in any treatment, for either the salivary or tongue microbiome ( P > 0.7, Kruskal–Wallis test) ( Figure 6B).

We next examined whether any bacterial species were specifically impacted by oral tablet usage. Both ANCOM using the full ASV table or Kruskal–Wallis test using the clustered OTU table revealed that no OTU was differentially abundant before (D1) or after (D2) tablet use, in any of the treatments (no tablet, protease tablet, and plain tablet). The OTU abundance in each treatment group is summarized in Figure 7a–c. Although according to a recent study oral tablet use decreases the abundance of Fusobacterium nucleatum on the tongue of healthy young adults ¹⁶ , we did not detect any significant decrease of OTUs that correspond to F. nucleatum. Further, in the current study, whereas 7.6% of all OTUs from the tongue microbiome were assigned to the genus Fusobacterium ( Figure 4a), the majority of them were classified as F. periodonticum (7.5% of total) and only <0.1% of all OTUs was assigned to F. nucleatum at species level.

Raw nucleotide sequences are available at DDBJ/EMBL-EBI/NCBI database under the accession number DRA010849.

Figshare: Table_S1_sample-metadata.tsv for "Inter-site and interpersonal diversity of salivary and tongue microbiomes, and the effect of oral care tablets". https://doi.org/10.6084/m9.figshare.13289618.v1 ³⁷

Table_S1_sample-metadata.tsv: Columns indicate sample ID, participant, sampling date (D1 or D2), treatment, Miseq run number, and sampling body site (saliva or tongue) of each sample. The “treatment” column indicates, no tablet (E2), protease tablet (E3), or plain tablet (E4) treatments.

Figshare: Table_S2_clustered-OTU-table.tsv for "Inter-site and interpersonal diversity of salivary and tongue microbiomes, and the effect of oral care tablets". https://doi.org/10.6084/m9.figshare.13291535.v1 ³⁸

Table_S2_clustered-OTU-table. After open-reference clustering, OTU table was constructed from the BIOM file using QIIME 2. Number of reads for each OTU in each sample are indicated, together with the bacterial taxonomy assigned to each OTU. OTU IDs are identical to matching HOMD Refseq IDs, except for those which did not match the database.

Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).