Ethics approval for this study was granted by the Ethics Committee of the Peking University Health Science Center (approval number: PKUSSIRB-201631135). All participants were individually informed and signed informed consent forms in accordance with the Declaration of Helsinki.

From January 2017 to August 2017, 13 patients with AgP and 13 patients with ChP were recruited from Peking University School and Hospital of Stomatology, and 10 PH subjects were selected as controls. The clinical inclusion criteria for PH and the diagnostic criteria for ChP and AgP have been previously described in detail (Shi et al., 2018). Exclusion criteria included systemic diseases, use of antibiotics or any other drugs that were known to affect periodontal conditions within the past 3 months, periodontal treatment in the past 6 months, a remaining tooth with caries, pregnancy or lactation, and smoking.

Full-mouth periodontal indices, including probing depth (PD), attachment loss (AL), and bleeding index (BI) (Mazza et al., 1981), were recorded at six sites per tooth by one practitioner (WH). His reliability was calibrated as described by Shi et al. (2018). All patients fulfilled the categorization of generalized Stage III or IV, Grade C periodontitis, according to the new periodontitis classification system (Tonetti et al., 2018).

One week after the full-mouth periodontal examination, saliva samples were collected in the morning (around 8 a.m. to 9 a.m.). All subjects were restrained from food for 8 h and oral hygiene for 12 h before sampling. Each subject was instructed to allow saliva to be collected in the mouth, and let the saliva drain into a sterile graduated tube for 10 min. About 3 ml of unstimulated whole saliva was collected. After centrifugation, the saliva was immediately frozen at –80°C.

The bacterial DNA of saliva samples was extracted using the QIAamp DNA Mini Kit (QIAGEN Sciences, USA). Polymerase chain reaction amplification was performed using specific primers, 338F (5′, ACTCCTACGGGAGGCAGCAG, 3′) and 806R (5′, GGACTACNNGGGTATCTAAT, 3′), targeting the V4–V5 region of the bacterial 16S rRNA gene. Thermal cycling conditions were as follows: initial denaturation at 95°C for 3 min, followed by 24 cycles of denaturation at 95°C for 30 s, annealing at 56°C for 30 s, elongation at 72°C for 45 s, and finally at 72°C for 10 min. Purified amplicons were pooled in equimolar and paired-end sequenced using the Illumina MiSeq PE300 platform (Illumina, San Diego, USA). The datasets generated and analyzed during the current study are available from the NCBI Sequence Read Archive database (Accession Number: SRP173111; SRP228020).

Gas chromatography-mass spectrometry (GC-MS) was performed as described in a previous study (Shi et al., 2020). Briefly, 300 μL of methanol was used for extraction, following the addition of 20 μL of 2-chloro-l-phenylalanine (0.3 mg/mL) as an internal standard. All samples were analyzed with an Agilent Gas Chromatography Mass Spectrometer (7890A/5975C GC-MS System, Agilent, CA, USA). The GC system employed a chromatographic column (HP-5MS) (30 m × 0.25 mm × 0.25 μm, Agilent J & W Scientific, Folsom, CA, USA).

The sequences were analyzed using the Quantitative Insights Into Microbial Ecology (QIIME) software package, and alpha and beta diversity values were determined. After the OTU table was rarified, alpha diversity and beta diversity calculations were performed. Analysis of similarity (ANOSIM), permutational multivariate analysis of variance (Adonis), and multi-response permutation procedure (MRPP) were used to examine the differences in the microbial community structure between the groups. The composition and abundance depicted in the taxonomy profile were visualized by GraphPad PRISM^® software (version 4.0). Normality tests for each group of data were conducted. The Kruskal–Wallis test, Mann–Whitney test, and Student’s t-tests were performed using SPSS 20.0.

The raw data obtained in .D format from the platform were converted to .CDF format by using ChemStation analysis software (version E.02.02.1431, Agilent, CA, USA), and then imported into Chroma TOF software (version 4.34, LECO, St Joseph) for preprocessing. The remaining steps were carried out as described in the literature (Shi et al., 2020). Unsupervised principal component analysis (PCA) and (orthogonal) partial least-squares analysis [(O) PLS-DA] were then applied to determine the differences in metabolic profiles between the groups. In the (O)PLS-DA analysis, variable importance in the projection (VIP) value greater than 1 and p-values (two-tailed Student’s t-test) less than 0.05 were considered to be different variables. Metabolites were then identified by searching in a self-built database of the Majorbio I-Sanger Cloud Platform,¹ which integrated the data of the Human Metabolome Database, KEGG Compound Database, and Lipid Maps Structure Database.

The correlation between microbiomes and metabolomics was analyzed by the Spearman method. A heat map of Spearman’s rank correlation coefficient was used to illustrate the relationship between microbial communities and metabolites.

The demographic characteristics and clinical parameters are presented in Table 1. A total of 36 subjects, including 13 patients with AgP, 13 patients with ChP, and 10 PH, were included in this study. PD, AL, and BI values were significantly higher in the group with periodontitis (AgP and ChP) than in the PH subjects (p < 0.05). The mean age in the ChP group was significantly higher than in the other two groups.

Sequences were clustered to 888 OTUs based on 97% sequence similarity. A total of 12 phyla, 30 classes, 54 orders, 92 families, and 173 genera were detected. The rarefaction curves suggested that the sequencing depth was adequate (Supplementary Figure 1).

In the AgP, ChP, and PH groups, the five most abundant phyla were Firmicutes, Bacteroidetes, Proteobacteria, Fusobacteria, and Actinobacteria, which represented more than 97% of the total sequences (Figure 1A). At the genus level, Streptococcus, Prevotella, Neisseria, Fusobacterium, Haemophilus, Veillonella, Actinomyces, Rothia, Porphyromonas, Gemella, Peptostreptococcus, Granulicatella, Leptotrichia, Parvimonas, and Capnocytophaga accounted for more than 85% of the sequencing results in all saliva samples, though there were some differences in the precise proportions among the AgP, ChP, and PH groups (Figure 1B).

Then we compared the relative abundance of microbiota between samples from periodontitis and healthy subjects. The abundance of Fusobacteria, Spirochetes, and Synergistetes increased significantly in the periodontitis group. The abundance of Gracilibacteria_GN02 was significantly higher in AgP than in ChP and PH (Figure 1C). At the genus level, 43, 31, and 5 genera were distributed differently between AgP vs. PH, ChP vs. PH, and AgP vs. ChP, respectively (Supplementary Tables 1–3). Most significant differences were observed in the low-abundance genera. Leptotrichia, Treponema, Peptococcus, Catonella, Desulfobulbus, Peptostreptococcaceae_[XI] ([G-2], [G-3] [G-4], [G-6], and [G-9]), Bacteroidetes_[G-5], TM7_[G-5], Dialister, Eikenella, Filifactor, Fretibacterium, Lachnospiraceae_[G-8], Mollicutes_[G-1], and Slackia increased significantly both in the ChP and AgP groups when compared to the PH group. Agrobacterium, GN02_[G-2], GN02_[G-1], Mobiluncus, and Peptostreptococcaceae_[XI][G-6] showed significant differences between the AgP and ChP groups.

To illustrate the richness, evenness, and species diversity of the microbial community, α diversity indices, including the Chao index, ACE index, Shannon index, and Simpson index, were calculated. The results showed that the alpha diversity of subjects with periodontitis was greater than that observed in healthy individuals (Figure 2). The β diversity calculations based on Bray-Curtis, Jaccard, weighted, and unweighted UniFrac distances showed significant compositional shifts between periodontitis and healthy groups (Figure 3). ANOSIM, Adonis, and MRPP were used to further confirm that there was no significant difference between the AgP and ChP groups (Table 2).

Multivariate analyses between different groups were performed in PCA, PLS-DA, and OPLS-DA. In the plot of PCA scores, we noted a trend of divergence in the saliva sample among the three groups. The PLS-DA model demonstrated satisfactory modeling and predictive abilities. The OPLS-DA model demonstrated a distinct separation between the metabolite profiles of these groups (Figures 4–6). The dissimilarity tests, including ANOSIM, Adonis, and MRPP, indicated significant differences in metabolites among the three groups (Table 3).

There were 13, 14, and 15 metabolites that were distributed differently between AgP vs. PH, ChP vs. PH, and AgP vs. ChP, respectively (Figure 7). Among these compounds, the levels of isoleucine, serotonin, hydrocinnamic acid, 4-hydroxycinnamic acid, serine, proline, stearic acid, palmitic acid, aspartate, and glycolic acid were significantly higher in AgP, while the levels of hypoxanthine, beta-alanine, and oleic acid were significantly higher in PH. Compared with PH, subjects with ChP are characterized by lower levels of urea, aminomalonate, and phosphoethanolamine, and higher levels of allantoin, isoleucine, hydrocinnamic acid, serotonin, threonine, 3-(4-hydroxyphenyl)propionic acid, 4-hydroxycinnamic acid, phenylacetic acid, 4-aminobutyric acid, serine, and lysine. Compared with ChP, the levels of biphenyl, stearic acid, palmitic acid, urea, and conduritol-beta-expoxide increased significantly in AgP, while the levels of thymine, fumaric acid, allantoin, beta-alanine, 2-hydroxyhexanoic acid, uracil, 3-aminoisobutyric acid, 4-aminobutyric acid, lysine, and citric acid were significantly decreased.

Following the enrichment analysis of the significantly different metabolites, the key pathways with the highest correlation between metabolite differences were identified based on both raw p and impact values.

When the AgP and PH groups were compared, several pathways containing more types of differential metabolites were detected, including protein digestion and absorption (including beta-alanine, proline, isoleucine, aspartate, and serine), biosynthesis of antibiotics (including serine, proline, isoleucine, glycolic acid, and aspartate), biosynthesis of amino acids (including serine, proline, isoleucine, and aspartate), aminoacyl-tRNA biosynthesis (including serine, proline, isoleucine, and aspartate), and fatty acid biosynthesis (including palmitic acid, stearic acid, and oleic acid) (Figure 8A).

When the ChP and PH groups were compared, several pathways containing more types of differential metabolites were detected, including aminoacyl-tRNA biosynthesis (including serine, isoleucine, lysine, and threonine), biosynthesis of amino acids (including serine, isoleucine, lysine, and threonine), and biosynthesis of valine, leucine, and isoleucine (including isoleucine and threonine) (Figure 8B).

When the ChP and AgP groups were compared, several pathways containing more types of differential metabolites were identified, including pyrimidine metabolism (including uracil, urea, thymine, and beta-alanine); alanine, aspartate, and glutamate metabolism (including citric acid, fumaric acid, and 4-aminobutyric acid); beta-alanine metabolism (including uracil, beta-alanine, and 4-aminobutyric acid); citrate cycle (including citric acid and fumaric acid); and arginine and proline metabolism (including urea and 4-aminobutyric acid) (Figure 8C).

Spearman correlation analysis was performed between the differential metabolites, the microbial diversity was analyzed at the phylum and genus levels, and the related network was constructed (| correlation r| > 0.25, p < 0.05).

At the phylum level, by comparing AgP and PH groups, we found that Actinobacteria and Fusobacteria were the most closely related to the metabolites (Supplementary Figure 2). At the genus level, Rothia, Streptococcus, Corynebacterium, Enterococcus, Lautropia, and Solobacterium were the genera with the most correlations with differential metabolites, which had significant correlations with 7, 6, 5, 5, 5, and 5 types of metabolites, respectively (Figure 9). Analysis from the perspective of metabolites showed that serotonin, oleic acid, proline, and hydrocinnamic acid had the most significant correlations with microorganisms, which were significantly correlated with 24, 20, 19, and 17 types of bacterial genera, respectively.

The comparison of ChP and PH groups revealed that Firmicutes, Actinobacteria, and Fusobacteria were most closely related to the metabolites (Supplementary Figure 3). At the genus level, Streptococcus, Fusobacterium, and Gemella were the genera with the most correlations with differential metabolites, which had significant correlations with 13, 8, and 8 types of metabolites, respectively (Figure 10). Lysine, aminobutyric acid, hydrocinnamic acid, and phenylacetic acid had the most significant correlations with microorganisms, which were significantly correlated with 25, 17, 15, and 13 types of bacterial genera, respectively.

The comparison of ChP and AgP groups showed that Actinobacteria, Synergistetes, Fusobacteria, and Proteobacteria were the most closely related to the metabolites (Supplementary Figure 4). At the genus level, Rothia, Enterococcus, Rhizobium, and Haemophilus are the genera with the most correlations with differential metabolites, which had significant correlations with 8, 7, 6, and 6 types of metabolites, respectively (Figure 11). Urea, beta-alanine, 3-aminoisobutyric acid, and thymine had the most significant correlations with microorganisms, which were significantly correlated with 23, 22, 21, and 20 types of bacterial genera, respectively.