Preliminary characterization of the oral microbiota of Chinese adults with and without gingivitis

The six subjects were healthy, non-smoking adults in age ranging from 21 to 39 years (Table 1). Group assignment was based on the frequency of bleeding on probing (BOP). Three subjects were assigned to the healthy group (H) based on a BOP frequency ≤5 and three subjects to the unhealthy (gingivitis) group (U) based on a BOP frequency of ≥ 20. Group H consisted of three women and Group U included two men and one woman. Bleeding indices of the individual subjects were shown in Table 1.

Five samples (each from a different oral site; see Methods) from each individual were collected and analyzed. Barcoded 16 S-rDNA amplicon sequencing using 454 Titanium (average read length of 400 bp) yield a total of 494,988 raw reads, resulting in a dataset of 258,385 reads (after stringent quality assessment and control measures; Methods). The number of reads per sample ranged from 4,405 to nearly 13,562, with an average of 8,612 (Table 2).

Clustering the unique sequences into operational taxonomic units (OTUs) at a 3% genetic distance resulted in 464~737 different "species level" phylotypes per microbiome (Table 2). For all of the oral microbial communities analyzed, the number of OTU detected was very close to the total number of OTU estimated by Chao1 and ACE richness indicators. The average level of Good's coverage was over 97% in all samples, indicating that about three new phylotypes would be expected for every 100 additional sequenced reads. This level of coverage suggested that the 16 S rDNA sequences identified represent the majority of bacterial members present in saliva and plaque samples in the current study. The individual rarefaction curves showed a similar pattern of increasing diversity that has not yet reached saturation (Figure 1). Comparisons of the rarefaction curves in the healthy (H) and gingivitis (U) populations for the sampled sites of saliva and plaque showed that the two host-groups displayed similar richness of bacterial OTUs at 97% identity level.

To determine whether the microbiota from the saliva and plaque distinguished healthy hosts and those with gingivitis, multivariate analyzes were applied to compare the overall structure of microbiota from each oral ecological niche based on FastUnifrac-derived and thetaYC-based distance matrices. FastUniFrac [20] allows pairwise comparisons of the evolutionary distances between two microbial communities and measures similarities among microbial community-structures. In the PCoA analysis, under a weighted UniFrac scheme, segregation between H and U groups was observed (p < 0.001) when all samples or when only plaque samples were considered, but not when only saliva samples were considered (Figure 2A). Moreover, regardless of their host-group affiliation, saliva microbiota formed a distinct cluster from the plaque microbiota (p = 0.034) (Figure 2B, C). The thetaYC-based PCoA analysis showed consistent results (Figure 3A, B). When only plaque samples were considered, the structural segregation between "H" and "U" groups was more discriminating (p = 0.001) (Figure 3C) than when all samples were included (p = 0.005) (Figure 3A). Therefore, gingivitis- and healthy-gingival-microbiomes can be distinguished based on the distinct community structures of plaque microbiomes, but not the salivary microbiomes.

Bacterial phyla and genera were identified and quantified through taxonomic assignment against reference databases using MOTHUR, which reveal their relative abundance in all of the plaque and saliva microbiota (Figure 4; only phyla were shown). All sequences were found distributed in 11 bacterial phyla that include six predominant phyla commonly encountered in the oral cavity: Firmicutes, Proteobacteria, Bacteroidetes, Actinobacteria, Fusobacteria and TM7 [6, 21]. The relative abundance of all plaque phyla detected in each of the two host groups suggested that significant differences for most of the phyla were found between hosts with or without gingivitis, except for Firmicutes, Proteobacteria and Spirochaetes (Figure 5). Among those gingivitis-associated phyla, Actinobacteria and Bacteroidetes were gingivitis-depleted while the remaining five phyla were gingivitis-enriched (Figure 5).

A total of 70 genera were identified in the oral microbiota from the five sampled sites (Additional file 1). The most frequently detected taxa at the genus level (the 12 most abundant genera that each represents at least 2% of oral microbiome) were Streptococcus, Neisseria, Leptotrichia, Actinomyces, Prevotella , Veillonella, Rothia, Fusobacterium, Lautropia, Selenomonas, Haemophilus, Granulicatella.

Statistically, 26 genera were found differentially distributed between gingivitis plaques (Group U) and healthy-gum plaques (Group H) (Table 3). Five genera (Streptococcus, Veillonella, Prevotella , Lautropia and Haemophilus) were significantly more abundant in healthy-gum plaque microbiota than those from gingivitis plaque, while the remaining 21 genera were found to be statistically less abundant.

As the above comparisons on the relative abundance of microbial taxa were derived from only those reads with confidence value above 0.8 (Methods), essentially all those reads without any reliable phylogenetic assignments (0.33~27.86% of the total reads in each sample) had been masked. Therefore, we further compared, between the two host-groups, the relative abundance of all species-level OTUs in plaque microbiota, independently of their phylogenetic identity assignments. In total, 98 OTUs (accounted for 5.38% of all OTUs) were found differentially distributed at the significance cutoff of 0.05 (both p-value and q-value of Metastats) in not only each of the four plaque sites but in all the plaque sites. Interestingly, among those 98 OTUs, 36 of them were gingivitis-depleted and the remaining 62 were gingivitis-enriched (Additional file 2). Consensus taxonomy of the OTUs was interrogated by MOTHUR based on oral "CORE" 16 S rDNA Gene Database (Methods). Twenty-four out of the 36 gingivitis-deplete OTUs while 57 out of the 62 gingivitis-enriched OTUs were supported by the taxonomy-assignment based results. Thus results from the two methodologies are largely consistent. These gingivitis-depleted or gingivitis-enriched OTUs represent a novel set of potential organismal markers for evaluation and prognosis of gingivitis, although their validity and significance remain to be further tested in larger human populations.

The abundances, in gene copies per ng of DNA, of two genera (Streptococcus and Fusobacterium) in all of the plaque and saliva samples were determined with qPCR. The relative abundance of these two genera as determined by pyrosequencing showed a positive correlation to the qPCR-based gene copies per ng of DNA (Streptococcus: r = 0.554; p < 0.002; Fusobacterium: r = 0.813; p < 0.001) (Additional file 3).

All oral samples were collected at the Hai Tai He Chang Clinical Research Center in Beijing with approval from P&G Beijing Technical Center (China) Institutional Review Board and in accordance with the World Medical Association Declaration of Helsinki (1996 amendment). ICH Guidelines for Good Clinical Practice (GCPs) were followed. Healthy subjects aged 18 years or older who had a minimum of 18 natural teeth were recruited from the Beijing area. Voluntary informed consent was provided. Individuals meeting the following criteria were excluded: current participation in another clinical study; use of antibiotic, anti-inflammatory or anticoagulant therapy within 30 days prior to examination; self-reported pregnancy or lactation; diabetes; a history of hepatitis or blood disorders such as hemophilia or leukocythemia; the presence of orthodontic appliances or removable partial dentures; significant oral pathology, such as advanced periodontal disease, hard or tissue tumors, or other conditions considered significant by the study director. Gingivitis was assessed using Mazza Gingival Index (MGI) as defined by Mazza in 1981 [55]. Specifically, probing was performed by a dentist on the mesiobuccal and the distolingual sites of each tooth, for a maximum of 56 sites. BOP (Bleeding on probing) frequency and mean MGI were recorded for each subject. The MGI is similar to the Loe and Silness Gingival index; both are validated indices for describing gingivitis [55]. The merit, however, of using MGI is that it combines measurements that address both the signs of inflammation as well as the degree of the severity of bleeding. Scores range from 0-5, with 0 assigned for normal appearing and healthy gingiva up to a score of 5 for spontaneous bleeding (without provocation). Five individuals with healthy gums and another five with extensive gingivitis were enrolled. Subjects were assigned to the healthy group (H) if there were less than ≤5 bleeding sites and to the unhealthy (gingivitis) group (U) when the frequency of bleeding sites was ≥ 20. No randomization among groups was performed. Two subjects from each group did not return for follow-up examinations and were excluded from further analyses. In the end, a total of six subjects (three in each group) completed the full study.

Samples of dental plaque and saliva were collected in the morning, 12 hours after evening tooth brushing. No oral hygiene or intakes of food and drink were allowed in the morning before sampling. Five samples were collected from each subject: supragingival dental plaque from anterior teeth (3~4 upper incisors), denoted A-sup; subgingival plaque [2 mm below gingival margin] from the same teeth, denoted A-sub; supragingival plaque from posterior teeth (2~3 upper molars), denoted P-sup; subgingival plaque from the same teeth, denoted P-sub; and saliva, denoted S. In the healthy group, plaque samples were collected from non-bleeding sites; in the unhealthy (gingivitis) group, incisor plaques were collected from non-bleeding sites and molar plaque were collected from bleeding sites. For unhealthy subjects, there are both non-bleeding incisor sites and bleeding molar sites for collecting plaque. We have considered the possibility that samples from the bleeding sites might not represent a complete picture of the microbiome of unhealthy gum. Therefore, plaque samples from both non-bleeding and bleeding sites were collected in our study.

Dental plaque samples were collected with sterile Gracey curettes and then removed from the curettes with a cotton-tipped swab. The tip of the swab was then placed into 0.4 mL lysis buffer (20 mM pH 8.0 Tris, 2 mM EDTA, 1.2% Triton X-100) and vortexed for 30 s. To collect salivary samples, subjects rinsed the mouth with 10 mL 0.9% saline buffer for 1 min and expectorated into a 50 ml tube. All samples were stored under -70°C before total genomic DNA extraction.

The amplification mix contained 12.5 ul of Gotaq Hotstart polymerase 2 × mix (Promega, USA), a 1 ul of each primer (5 pM), 1 ul genomic DNA (0.1-10 ng μl^-1) and 9.5 ul H₂O in a total volume of 25 μl. Cycling conditions were an initial denaturation at 95°C for 2 min, 25 cycles at 94°C for 30 s, at 56°C for 25 s, and at 72°C for 25 s, followed by a final 5 minute extension at 72°C. Samples were processed via separate PCR reactions (ABI StepOnePlus™ Real-Time PCR Systems) and then pooled. Each sample was amplified using one specific barcoded primer. To assess quality, the PCR product for each sample was subjected to electrophoresis (1.2% agarose, 5 V cm^-1, for 40 min). Gels were stained with a buffer containing SYBR Gold Nucleic Acid Gel Stain (Invitrogen, USA); DNA fragments of approximately 500 bp were excised from the gel and further purified using Qiagen MiniElute kit. Concentrations of DNA in purified PCR products were further analyzed with PicoGreen (Invitrogen, USA). The amplicons were pooled into a single tube in equimolar ratios. Pyrosequencing of the 16 S PCR-amplicons was carried out on Genome Sequencer FLX Titanium (Roche, USA) where, on average, 400 bp-long reads were produced.

The sequences generated from pyrosequencing were mainly analyzed with MOTHUR [56] for preprocessing, identification of operational taxonomic units (OTU), taxonomic assignment and community-structure comparisons. To minimize the effects of random sequencing errors and avoid overestimates of the phylogenetic diversity [57], relatively stringent quality-based trimming of the reads was performed. First, the 454-reads were removed if they were < 150 bp, had an average quality score < 35 in each 50-bp window rolling along the whole read, had an ambiguous base call (N), had any homopolymers of more than eight bases or did not contain the primer sequence; reads were then sorted by the tag sequences. To reduce sequencing noise from pyrosequencing data, we performed the pre-clustering step [58] with the "pre.cluster" script in MOTHUR [56]. We also removed chimeric sequences detected by UCHIME [59].

The trimmed reads were assigned to clusters using UCLUST http://​www.​drive5.​com/​uclust/​. An in-house perl script was used to convert UCLUST output into a format recognized by MOTHUR [56] http://​www.​mothur.​org/​ for further analysis. Reads were assigned to OTUs (species-level). Calculation of coverage percentage (Good), species richness estimators (ACE and Chao1) and rarefaction analysis were performed using MOTHUR [56]. The relative abundance of OTUs with 97%-identity between pair-wise samples or between groups of samples were compared.

For taxonomic assignments, we used the "classify.seqs" script in MOTHUR [56] to classify all trimmed reads based on Naive Bayesian method with oral "CORE" [21] taxonomy sequences as the reference database. The confidence score threshold was set to 0.8, such that those with bootstrap value below 0.8 were assigned as unclassified. Relevant abundances of the bacterial taxa at the phylum and genus level were calculated and compared.

The OTUs defined by a 3% distance level were phylogenetically classified using the "classify.otu" script in MOTHUR [56] with oral "CORE" database [21] and a taxonomy file describing the complete taxonomic information of each sequence in the database from domain to species (using a 51% confidence threshold). The consensus taxonomy for each OTU was obtained in this step.

FastUnifrac-based community structure comparisons were performed [20]. In each sample, representative sequences from each OTU were chosen by selecting the longest sequence based on UCLUST. Each sequence was assigned to its closest relative in a phylogeny of the Greengenes core set [60] using BLAST's megablast protocol. The resulting sample ID mapping file and category mapping file were used as inputs to the unweighted and weighted FastUniFrac [20]. FastUniFrac allows pairwise comparisons of distances between two microbial communities in terms of the fraction of evolutionary history that separates the organisms. A distance (a measurement of the similarity in community structure between two microbiota) was computed for each pair of samples, both within a single population and across the two populations, to create a matrix of pairwise distances among all samples. These distances were then clustered to reduce dimensionality using PCoA [61]. PCoA is a multivariate statistical technique for finding the most important axes along which the samples vary. The principal coordinates (PC), in descending order, describe of the degree of variation that each of the axes in the new space explains. ThetaYC-based community structure comparisons were performed in parallel with MOTHUR [56]. ThetaYC (D θ Y C = 1 - ∑ i = 1 S T a i b i ∑ i = 1 S T ( a i - b i ) 2 + ∑ i = 1 S T a i b i) measures the dissimilarity between the structures of two communities [62], where S _T is the total number of OTUs in communities A and B, a _i is the relative abundance of OTU i in community A, b _i is the relative abundance of OTU i in community B. A matrix of pairwise thetaYC-based distances among all samples was created for clustering and PCoA analysis.

Quantitative PCR assays on selected species were performed to test the degree of correlation with 16 S rDNA pyrosequencing data. Two genera, Streptococcus and Fusobacterium, were frequently identified based on our taxonomy assignments of the reads. Therefore, we chose two pairs of primers and probes targeting these two genera to perform the quantitative assays for comparisons to the pyrosequencing data.

Genus-specific primers and TaqMan probes were used, as listed in Additional file 4. The oligonucleotide probes were labeled with the fluorescent dyes 6-carboxyfluorescein (FAM) at the 5' end and 6-carboxytetramethylrhodamine (TAMRA) at the 3' end. The specificities of the probe and primer sets for their target DNA were tested in duplicate with the TaqMan Universal PCR Master Mix. The optimized concentrations of the forward primer, the reverse primer, and the fluorogenic probe in the 20-μl reaction volume were selected to be 300 nM, 300 nM, and 200 nM, respectively. Amplification and detection by quantitative PCR were performed with the StepOnePlus™ Real-Time PCR Systems (Applied Biosystems, Foster City, CA, USA). For each quantitative PCR, 20 μl reaction mixtures containing 2-μl sample DNA, forward primer, reverse primer and TaqMan probe at the optimized concentrations (as described above) were placed in each well of a 96-well plate. Following the fast TaqMan thermocycling protocol, reaction conditions were set at 95°C for 20 seconds, followed by 40 cycles of 95°C for 1 second and 58°C for 20 seconds. Standard curves for each organism were plotted in duplicate for each primer-probe set using the Ct (the cycle number at which the threshold fluorescence was reached) values, which were obtained by amplifying successive 10-fold dilutions of a known concentration of bacterial DNA (Streptococcus mutans UA159 and Fusobacterium nucleatum subsp. nucleatum ATCC25586). Copy-numbers of the target genes (tuf-elongation factor Tu and 16 S rDNA) in standard samples were calculated by the genome sizes (S. mutans 2.0 Mb and F. nucleatum 2.2 Mb) and the copy-number per genome (one copy of tuf gene per cell of S. mutans and five copies of 16 S rDNA gene per cell of F. nucleatum [46, 63]. One ng of S. mutans genomic DNA contains 4.63 × 10⁵ copies of tuf gene while 1 ng F. nucleatum genome DNA contains 2.10 × 10⁶ copies of 16 S rDNA gene. Based on these assumptions, the absolute copy number of a target gene was determined by referring Ct value to a standard cure measured on the same plate. The relative abundance of these bacteria in the 30 different oral specimens was normalized by the absolute quantity of DNA in the clinical samples.

AMOVA (Analysis of Molecular Variance) were used to test whether two communities from H and U populations have the same centroid [64, 65]. HOMOVA (Homogeneity of Molecular Variance) was employed to test whether the genetic diversity are similar between the communities from the H and U populations [65, 66].

Relative abundance of OTUs and phylotypes were reported as mean ± SEM. Due to the small sample sizes of these oral-site-specific datasets, features that are differentially distributed (i.e. abundant) between populations were statistically detected using Metastats [67] via a web interface http://​metastats.​cbcb.​umd.​edu/​detection.​html. Frequency data of OTUs and phylotypes were converted into a Feature Frequency Matrix as the input to this analysis tool. To exclude the extremely sparsely-sampled features (OTUs/phylotypes), tests were applied only if the total number of observations of a feature (OTU/phylotype) in either population is greater than the total number of subjects in the population (i.e. the average number of observations across subjects for a given feature is greater than one). Metastats was performed using 1000 permutations to compute p-values in statistical tests. We set p-value threshold of significance as 0.05. To control the FDR (False Discovery Rate) within the entire set of tests, we only took those features whose q-values and p-values were both below 0.05 into considerations. Levels of confidence were denoted as: *: 0.01 < p < 0.05; **: p < 0.01.

In validating the pyrosequencing results, the relative abundance of selected genera (Streptococcus and Fusobacterium) as measured via 16 S-amplicon pyrosequencing was compared to the corresponding gene copy number as determined by qPCR. The Shapiro-Wilk statistics of the variables were statistically significant. The degrees of correlation between the two measured parameters were determined from the Spearman's nonparametric correlation coefficient, r. Statistical analyses were performed with R (version 2.13.1). All reported p values were two-sided, at a 95% confidence level.