Site- and Taxa-Specific Disease-Associated Oral Microbial Structures Distinguish Inflammatory Bowel Diseases

Abstract

Background

The gut and oral microbiome have independently been shown to be associated with inflammatory bowel disease (IBD). However, it is not known to what extent gut and oral microbial disease markers converge in terms of their composition in IBD. Further, the spatial and temporal variation within the oral microenvironments of IBD remain to be elucidated.

Patients and Methods

We used a prospectively recruited cohort of patients with IBD (n = 47) and unrelated healthy control patients (n = 18) to examine the spatial and temporal distribution of microbiota within the various oral microenvironments, represented by saliva, tongue, buccal mucosa, and plaque, and compared them with stool. Microbiome characterization was performed using 16S rRNA gene sequencing.

Results

The oral microbiome displayed IBD-associated dysbiosis, in a site- and taxa-specific manner. Plaque samples depicted a relatively severe degree of dysbiosis, and the disease-associated dysbiotic bacterial groups were predominantly the members of the phylum Firmicutes. Our 16S rRNA gene analyses show that oral microbiota can distinguish patients with IBD from healthy control patients, with salivary microbiota performing the best, closely matched by stool and other oral sites. Longitudinal profiles of microbial composition suggest that some taxa are more consistently perturbed than others, preferentially in a site-dependent fashion.

Conclusions

Collectively, these data indicate the potential of using oral microbial profiles in screening and monitoring patients with IBD. Furthermore, these results support the importance of spatial and longitudinal microbiome sampling to interpret disease-associated dysbiotic states and eventually to gain insights into disease pathogenesis.

Introduction

Inflammatory bowel disease (IBD) is a lifelong condition characterized by intestinal ulceration, pain, rectal bleeding, loss of quality of life, and a need for bowel surgery. Its increasing prevalence has been documented within the developed world.^1-5 Crohn disease (CD) and ulcerative colitis (UC) are the 2 classical forms of IBD. Both CD and UC share many clinical and extraintestinal manifestations and hence it is often difficult to make an accurate diagnosis, particularly at the earliest stages of disease. Although most patients with IBD respond to standard-of-care clinical treatment, some patients rapidly progress to complicated disease behaviors such as perforated bowel, stricturing because of fibrosis, and/or penetrating fistulas.⁶

Endoscopic evaluation combined with histopathological examination of the mucosal biopsy is the gold standard for diagnosis or disease monitoring in IBD. However, these invasive procedures are associated with high cost and relatively low patient acceptance rate and are not ideal for disease monitoring and assessing response to therapy. Serologic studies have been proposed to help diagnose and monitor IBD but suffer from a low sensitivity and specificity. Therefore, there is a compelling need for the identification of novel, noninvasive, cost-effective, robust, and reproducible biomarkers for accurate diagnosis, treatment selection, and disease monitoring.

Although the exact mechanism is not known, the pathogenesis of IBD has been attributed to a dysregulated immune response to alterations in the gut microbial composition in genetically susceptible individuals.^{7, 8} Genes and susceptibility genetic loci implicated in IBD have been shown to be enriched for pathways involving bacterial recognition or host response to microbial infections, suggesting a microbial contribution to disease pathogenesis.⁹ Screening for perturbations in fecal microbial composition, referred to as microbial dysbiosis, has emerged as a promising noninvasive approach for IBD screening.^{10, 11} We and others have previously shown that pretreatment stool microbial dysbiosis is present in IBD and that a fecal microbial dysbiosis index could be used as a screening tool to diagnose IBD and differentiate CD from UC and therapy responders from nonresponders.^{9, 10} Furthermore, a recent study by Ananthakrishnan et al¹² showed the potential of the gut microbiome in predicting response to anti-integrin therapy in IBD. Although the mechanistic framework underlying the therapeutic potential of gut microbiota has not yet been fully elucidated, there is ample experimental evidence suggesting a causal role for gut microbial dysbiosis in IBD susceptibility and progression.¹³ However, it is not clear whether this dysbiosis is specific to gut microbial community or is a systemic phenomenon in IBD.

Studies have begun to reveal oral microbial alterations in IBD.^14-19 A pediatric study that included 40 patients with CD and 43 control patients without IBD reported a significant decrease in the overall diversity of tongue microbiota in CD.¹⁶ Phylum-level analysis of salivary microbiota revealed an increased abundance of Bacteroidetes and a reduced abundance of Proteobacteria in patients with IBD.¹⁶ Furthermore, perturbed salivary microbial communities in IBD showed statistically significant associations with inflammatory cytokines such as IL-1β and IL-8 and lysozyme levels.¹⁶ Despite limited information, these studies have shown oral microbial differences in IBD, implicating the potential of the oral microbiome in diagnosing and monitoring patients with IBD. However, these studies were limited to a specific region in the oral cavity. Surprisingly, although several microbial changes have been reported independently from 1 or more oral sites, there is no consistent pattern of dysbiosis between patients with IBD and healthy control patients that could, at least partly, be attributed to baseline differences in composition and diversity based on the region of the mouth sampled.

The oral microbiota includes a large repertoire of approximately 700 bacterial species or phylotypes with spatially patterned composition between the oral microenvironments; however, less is known about which anatomical location within the oral cavity is more indicative of IBD. Further, it is not known to what extent oral and gut microbial disease markers converge in terms of their composition in patients with IBD. Furthermore, the longitudinal trajectory of disease-associated microbial changes has not been thoroughly investigated. Given the dynamic nature of microbiota and the relapsing-remitting behavior of IBD, understanding the longitudinal trajectory of microbiota is crucial to elucidate the mechanistic basis of microbial contributions to IBD and for the advancement of microbiome-based diagnostic and therapeutic interventions. Here we used a prospectively recruited cohort of patients with IBD and unrelated healthy control patients to examine (1) spatial (characterizing microbial profiles while retaining information on various oral microenvironments, represented by saliva, the tongue, buccal mucosa, and plaque) and temporal (longitudinal changes in microbial composition within the same individual) dynamics of the oral microbiota in IBD, (2) the concordance and divergence between oral and gut microbiota in IBD, and (3) the predictive potential of the oral and gut microbiota in assessing the presence of the disease.

PATIENTS AND METHODS

Study Population

Patients with IBD were recruited from the Children’s Healthcare of Atlanta inpatient wards and outpatient pediatric IBD clinics. Criteria to participate in the study included CD or UC diagnosis confirmed by colonoscopy and/or magnetic resonance enterography, willingness to participate, and ability to maintain close follow-up. The control population was composed of unrelated, age- and sex-matched healthy individuals who volunteered to participate upon request. Exclusion criteria included patients who were on or had a recent history (within the preceding month) of antibiotic treatment at the time of enrollment and patients who had oral infections or manifestations relevant to IBD or any oral diseases.

As illustrated in Table 1, a total of 65 patients (30 with CD, 17 with UC, and 18 healthy control patients) ranging in age from 5.1 to 19.3 years (median age 14.6 years) were enrolled in the study between January 2015 and February 2017. Of the 65 patients, 44 (25 with CD, 9 with UC, and 10 healthy control patients) were followed longitudinally at regular intervals for up to a maximum period of 88 weeks, which yielded up to 6 follow-up samples over time. Of the 47 patients with IBD, 26 (55%) were treatment-naïve at the time of enrollment (17 with CD, 9 with UC; Table 1). Patients with suspected diagnosis of IBD based on symptoms and laboratory work were approached for participation in the new-onset portion of the study. These patients did not have a prior IBD diagnosis or a prior history of immunomodulator therapy or biologic therapy. Meanwhile, the remaining patients with IBD (n = 21; 45%) were a priori established for 1 of the 2 forms of IBD (13 with CD, 8 with UC) and were on concomitant therapy or had a prior history of immunomodulator and/or biologic therapy at the time of enrollment. The IBD diagnoses were conducted according to the Paris Classification²⁰ at the time of enrollment.

Demographic and phenotypic data were collected on each patient enrolled via patient interview and chart review at the time of sample collection. Abbreviated Pediatric Crohn’s Disease Activity Index^{21, 22} or Pediatric Ulcerative Colitis Activity Index²³ scores were obtained at all clinical visits. Medical treatment was not affected by participating in this study. Although no patient was on antibiotic therapy during enrollment, a small number of patients (n = 6) reported short courses of antibiotic usage during the course of the follow-up period. All participants and families provided informed consent and assent for specimen collection and analysis under the study protocol approved by the Institutional Review Board of Children’s Healthcare of Atlanta.

Specimen Collection and Processing

Oral microbiota samples were collected using DNA swabs (Isohelix, United Kingdom) from 4 anatomically different regions within the oral cavity—saliva, the tongue, plaque, and buccal mucosa. Locations sampled included the dorsum of the tongue (for tongue samples), the buccaneers surface of the central and lateral incisors at the gum line (plaque samples), and the inside of the cheek (buccal mucosa). All samples were immediately stored at –80°C until further processing. Bacterial genomic DNA was extracted from all the oral samples using the Biostic Bacteremia DNA Isolation Kit (MO BIO Laboratories Inc., Carlsbad, CA) according to the manufacturer’s guidelines. All samples from the same patient were processed together to minimize batch effects. Fecal specimens were collected along with oral samples from patients whenever possible. Each fecal sample was collected in a Para-Pak vial (Meridian Bioscience Inc., Cincinnati, OH) that contained no additional additive. Fecal specimens were stored at –20°C until they were aliquotted into smaller workable units and then stored at –80°C. We extracted DNA from fecal samples using the MagAttract PowerMicrobiome DNA/RNA Kit (Qiagen, Valencia, CA) according to the manufacturer’s instructions.

16S rRNA Gene Sequencing and Curation

The V4 region of the 16S rRNA gene was polymerase chain reaction–amplified and sequenced on an Illumina MiSeq platform using a 2 × 250-bp paired-end protocol adapted from the Human Microbiome Project.^{24, 25} The forward and reverse primer sequences are provided in Supplementary Table 1. The obtained sequences were curated using the mothur pipeline (v1.38).^{26, 27} Briefly, paired-end reads were merged into contigs, screened for quality following the mothur MiSeq standard operating procedure, and then aligned to the SILVA 16S rRNA gene sequence database. Aligned sequences were then screened for chimeras using the VSEARCH algorithm.²⁸ Sequences were classified using a naive Bayesian classifier trained against a 16S rRNA gene training set provided by the Ribosomal Database Project.²⁹ Sequences were then clustered into operational taxonomic units (OTUs) using a 97% similarity cutoff with the average neighbor clustering algorithm.

16S rRNA Gene Sequencing Data Analysis

The OTU-based overall microbial diversity was estimated by calculating 3 alpha-diversity indices: Shannon, Simpson, and alpha. The OTU-based overall richness was determined by calculating the Chao1 richness estimate. Differences in overall microbial community structure were visualized by calculating Bray-Curtis dissimilarity measures between all pairs of samples. The significant differences in principal coordinate analysis plots were analyzed using PERMANOVA < 0.05. All the available samples from each patient per site were used for estimating the overall diversity, richness, and community structural differences. To statistically test for the individual microbial member level differences in the relative abundances of taxa at different taxonomic levels, phyla, class, order, or family, between groups, we used the metagenomeSeq,³⁰ a Bioconductor package that uses a zero-inflated Gaussian distribution mixture model. Cumulative sum scaling using default settings was used to normalize the data set before fitting the model. Differential abundance analysis was carried out with respect to the baseline samples that were collected on the first visit. In patients for whom the first visit sample was unavailable, we used the first available sample from each patient. We included age, sex, ethnicity, and anti-tumor necrosis factor (TNF) treatment status as covariates in the model. The P values presented for the differential abundance analyses were obtained from 10,000 permutations, which were then corrected for multiple hypothesis testing using the false-discovery rate (FDR) method.

Random Forest

To ascertain whether the oral microbiome could distinguish patients with IBD from healthy control patients, we used a random forest classifier, which is a decision tree–based algorithm.³¹ We used the same dataset as in differential abundance tests in this analysis—ie, from each site, we only selected those samples that were collected on the first visit. If the sample at the first visit was missing, then we selected the sample from the next visit. Details on the number of samples per group and per site used for the random classifier and the area under the curve (AUC) statistics are provided in Table 1. Data from each site were divided into training (two-thirds of the dataset) and test (one-third) datasets. The random forest classifier was run on the training set, with 10,000 random trees, and the predicted model was then used on the test data set to obtain the AUC. We created 100 such random splits for the training and test datasets and used the random forest classifier to obtain the AUC prediction for each site.

Data Availability

The high-throughput sequence data from the 16S rRNA gene sequencing have been deposited in the National Center for Biotechnology Information (NCBI) Sequence Read Archive database with the project number PRJNA416207. All other data supporting the findings of this study are available within the article and its Supplementary Information Files or from the corresponding author upon request.

RESULTS

16S rRNA Gene Data Processing

We sequenced bacterial 16S ribosomal RNA gene using the Illumina MiSeq platform with primers targeting the V4 variable regions. Using this approach, we generated a data set consisting of a median of at least ~33,000 reads per sample for each site (Supplementary Fig. 1). Of these, sequences that passed the quality control criteria were sorted into OTUs. All the samples across the habitats and over time were then rarefied to 6575 reads per sample to minimize the effects of uneven sampling, which resulted in a dataset with 768 samples and 4259 OTUs. The rarefaction curves for all the sites, collectively and individually, are shown in Supplementary Fig. 2. To ensure robustness, we applied 3 separate filters to this dataset, which resulted in 3 independent datasets for downstream analyses. Our first filter retained OTUs that were present in at least 1% of the total samples (n = 768), yielding 753 OTUs. The second filter retained OTUs present in at least 5% of the total samples, which resulted in a dataset consisting of 268 OTUs, and the third filter retained OTUs that were present in at least 1% of the total samples and had a minimum total read count of 50 for all samples. Using the third approach, we generated a dataset consisting of 462 OTUs. All the downstream analyses were performed on all 3 datasets; however, the results presented below pertain to the dataset with 753 OTUs, unless stated otherwise.

Site-Specific Microbial Composition Within the Oral Habitat

To examine beta diversity, we assessed the differences in the overall microbial community structure across all the habitats and over time from both patients and control patients using a non-phylogeny-based Bray-Curtis dissimilarity metric. A relatively small Bray-Curtis distance implies that 2 communities are similar where the majority of the species is shared. Consistent with the previous notion, Bray-Curtis distance–based principal coordinates analysis revealed strong primary clustering by habitat, rather than by disease status or over time, reflecting the high heterogeneity of the sampled habitats (Fig. 1A). Within the oral microenvironments, buccal samples showed a clear separation from tongue samples (Fig. 1A and Supplementary Fig. 3), suggesting that anatomical regions within the oral cavity were akin to microbial “islands,” possessing distinct bacterial communities that persisted temporally. This observation is consistent with previous reports from the Human Microbiome Project (HMP) and other studies showing that most oral bacterial taxa are habitat specialists.^{32, 33} On the other hand, microbial diversity between saliva and plaque seemed to be nominal, and the 2 sites were almost indistinguishable from one another on the plotted ordination axes (Fig. 1A and Supplementary Fig. 3).

Richness, Diversity, and Relative Abundance of Oral Microbiota

Microbial diversity and richness vary by anatomic site,³⁴ partly in response to the local environment and biology of each body habitat. We evaluated spatial trends in the structure of the bacterial communities both at global parameters and at an individual microbial member level, across the 4 oral sites (saliva, tongue, plaque, and buccal mucosa). When we compared these trends across the sites using an analysis of variance followed by the Tukey test (number of samples per site are presented in Table 1), we identified statistically significant differences in the overall diversity of buccal microbiota as measured by the Shannon diversity index (P < 1 × 10^–7, compared to the other 3 oral sites); however, no significant differences were noted among the other 3 oral sites (Fig. 1B). Meanwhile, we did not observe any significant difference between the oral sites in overall richness as measured by the Chao1 index (Fig. 1C). As expected, all 4 oral sites showed significant differences in both overall diversity and richness when compared with stool (Fig. 1B and C).

Further, we noted changes in the composition of individual microbial members, between habitats (Fig. 2). The human microbiota is typically dominated by the 4 bacterial phyla, Firmicutes, Bacteroidetes, Proteobacteria, and Actinobacteria.^34-36 Our analysis of oral microbiota at the phylum level showed Firmicutes (46%), Bacteroidetes (12%), Proteobacteria (25%), and Actinobacteria (10%) in saliva; Firmicutes (42%), Bacteroidetes (18%), Proteobacteria (21%), and Actinobacteria (13%) in the tongue; Firmicutes (35%), Bacteroidetes (11%), Proteobacteria (21%), and Actinobacteria (22%) in plaque; and Firmicutes (53%), Bacteroidetes (9%), Proteobacteria (25%), and Actinobacteria (7%) in buccal mucosa (Fig. 2A). Although the same 4 phyla dominated the microbiota in stool—Firmicutes (72%), Bacteroidetes (5%), Proteobacteria (3%), and Actinobacteria (17%)—we noted a pronounced reduction in terms of the relative abundance of Firmicutes and an increase in Proteobacteria in oral samples compared to stool (Fig. 2A). This shift in Firmicutes and Proteobacteria in the oral microbiota is interesting in particular because it is in line with the shift seen in the intestinal microbiota of patients with severe UC, which is characterized by a decline in Firmicutes and an increase in Proteobacteria when compared to mild or moderate UC.³⁷ When we observed the members of these 2 phyla, we found that most members (6/7) of the phylum Proteobacteria had an increased abundance in the oral cavity compared to stool, which is directionally consistent with the shift seen at the phyla level (Fig. 2B). Whereas members belonging to the class Clostridia (5/6) of the phylum Firmicutes showed directional consistency, members of the class Bacilli showed polarizing shifts (Fig. 2B).

Site- and Taxa-Specific Oral Microbial Dysbiosis in IBD

To assess overall differences in the microbial community structure in the patients with IBD and the control patients, we calculated measures of alpha- and beta-diversity in all 4 profiled oral sites and in the fecal microbiota. As shown in Fig. 3A, beta-diversity entropy measured using a Bray-Curtis dissimilarity depicted statistically significant differences between patients with IBD (CD or UC) and healthy control patients in a site-specific manner (permutational multivariate analysis of variance [PERMANOVA] < 0.05). Similarly, we noted site-specific microbial differences in alpha-diversity; in agreement with the previous notion, IBD was associated with a reduced diversity in stool as indicated by the alpha index (P = 0.003) and reduced richness as indicated by the Chao1 index (P = 0.001; Supplementary Fig. 4A). Interestingly, we noted a similar trend in terms of overall richness in saliva as measured by the Chao1 index (P = 0.082), whereas no such trend was found for any other oral sites including the tongue, plaque, and buccal mucosa (P > 0.1; Supplementary Fig. 4B). However, note that our findings from both stool and saliva were not robust to other alpha-diversity measures (Supplementary Fig. 4).

Next, we surveyed oral microbial samples for IBD-associated changes at an individual microbial member level, with respect to the first available sample, adjusting for age, sex, ethnicity, and anti-TNF treatment status. Because anti-TNF therapy may skew microbiota composition, we used anti-TNF status as a covariate in addition to age, sex, and ethnicity. Antibiotic usage was not included as a covariate in our differential abundance analysis because no patient was reported to be on antibiotics with respect to the first available sample. At the population level, we noted significant differences or pronounced shifts in the relative abundance of several bacterial members at different taxonomic levels between the patients with IBD and the healthy control patients. For instance, at the phyla level, Actinobacteria, Bacteroidetes, and Spirochaetes showed a trend for enrichment in patients with IBD across all 4 oral sites, whereas Fusobacteria, Firmicutes, and Proteobacteria were among the phyla that showed a trend for depletion in IBD (Supplementary Fig. 5). Similar but significant associations were previously reported between salivary Bacteroidetes, Proteobacteria, and IBD.¹⁶

At an increased resolution, interestingly, we observed IBD-associated oral microbial dysbiosis in a site- and taxa-specific manner. For example, the relative abundance of the order Bacillales in the phylum Firmicutes showed a significant reduction in the patients with IBD compared to the healthy control patients in both plaque (P = 0.015) and buccal mucosa (P = 0.019); however, this difference did not reach significance in the other 2 oral sites (Fig. 3B). Similarly, the IBD-associated depletion of Carnobacteriaceae, a family in the phylum Firmicutes, was confined to plaque (P = 0.007) and saliva (P = 0.013) samples (Fig. 3B). Overall, based on the number of bacterial groups that showed significant differences (FDR < 0.05) between the patients with IBD and the control patients, the degree of the disease-associated dysbiosis was relatively severe in plaque compared to the other 3 oral sites (Fig. 3B).

On the other hand, microbes that depicted IBD-associated dysbiosis are predominantly members of the phylum Firmicutes (Fig. 3B). Notably, perturbed Firmicutes and Actinobacteria abundances have long been implicated in IBD. For instance, IBD has been shown to be associated with an overall depletion of Firmicutes in disease-relevant intestinal mucosal biopsies and in stool samples,^9,38 whereas Actinobacteria have been reported to be substantially more abundant in patients with IBD compared to healthy control patients.³⁹ In agreement with this trend, we found a depletion of the members of the phylum Firmicutes and enrichment of the members of the phylum Actinobacteria (family Actinomycetaceae and genus Actinomyces) in the oral microbiota of patients with IBD (Fig. 3B).

Meanwhile, as expected, several bacterial groups in the fecal microbiota showed a correlation with the disease phenotype. However, to our surprise, most of the IBD-associated microbial signal from stool was either lost or trended in the opposite direction in the oral samples (Supplementary Fig. 6). We performed these differential abundance analyses on the other two datasets, with 462 OTUs and 268 OTUs, and obtained similar results (data not shown). Collectively, these data highlight the importance of site- and taxa-specific dysbiosis in IBD. Statistical significance was evaluated with a permutation test (number of permutations = 10,000), which was then corrected for multiple comparisons using the FDR.

Oral Microbiota Can Differentiate Patients With IBD From Healthy Control Patients

To ascertain whether the oral microbiota could distinguish patients with IBD from healthy control patients, we used a random forest classifier on the first available oral sample from each patient and compared its prediction accuracy to stool. Details of the number of samples per group and per body site are given in Table 1. Our classifier for IBD in oral samples attained an average AUC ranging from 0.65 to 0.73 depending on the location, suggesting that oral microbial composition across all 4 profiled sites can distinguish patients with IBD from healthy control patients (Table 2). When the classifier was compared among the sites, saliva performed best, closely matched by the stool samples. In 100 random splits of the data between the training and test sets, our classifier for IBD attained an average AUC of 0.73 for saliva vs an average AUC of 0.67 for stool (Table 2). One such split of the data is shown in Fig. 4. The buccal mucosa (AUC = 0.70), plaque (AUC = 0.67), and tongue (AUC = 0.65) were also comparable to stool in classification accuracy (Table 2). Furthermore, despite the limited sample size, we were able to make a classifier from the oral microbiota that sorted both CD and UC samples from samples from healthy control patients (Table 2). Though we did not notice any significant difference, in alpha-diversity or beta-diversity or at an individual microbial member level between patients with newly diagnosed and established IBD, note that a significant proportion of our patients with established IBD were on concomitant therapy during the enrollment sampling. However, medication should have had systemic effects on microbial composition and hence we assume that no significant bias was introduced because we were only interested in comparing the diagnostic utility of oral microbiota to that of stool. Nevertheless, our results suggest that oral samples, saliva in particular, can differentiate patients with IBD from healthy control patients and may be used as a surrogate to diagnose or monitor the presence of IBD. This finding was robust to the choice of the datasets, with 462 OTUs and 268 OTUs (Supplementary Table 2).

Longitudinal Trajectory of Oral Microbiota

Cross-sectional studies have shown IBD to be associated with site-specific dysbiosis, reduced diversity, and species richness^9,40-43; however, the longitudinal trajectory of these disease-associated changes has not been thoroughly investigated. We examined the longitudinal trajectories of the relative abundance of individual microbial members across sites from patients with at least 3 samples over time (n = 19; 11 with CD, 5 with UC, and 3 healthy control patients). We observed 2 general patterns: global stability and global variability. The global stability group included microbial members whose relative abundance remained fairly stable during the course of the follow-up period across individuals, irrespective of disease status. Microbial organisms belonging to the phyla Bacteroidetes, Fusobacteria, and Proteobacteria were among the globally stable group, across sites, including saliva, the tongue, and plaque; however, this pattern seemed to be disrupted in buccal samples (Figs. 5A-C and Supplementary Figs. 7-9). On the other hand, the global variability group included members whose relative abundances displayed inter- and intraindividual variability patterns, intermittently disappearing and reappearing over time. Among these were the phyla Firmicutes, SR1, and Actinobacteria (Figs. 5D-F and Supplementary Figs. 7-9). Interestingly, a previous analysis of samples over time reported Firmicutes as being more temporally dynamic within the gut microbiomes of individuals.³² Collectively, the longitudinal trajectory findings from the oral sites support the view that the composition of some microbial organisms is more consistent over time, whereas others exhibit relatively frequent transitions across patients, irrespective of disease status.

Discussion

This is the first investigation to characterize the spatial and temporal dynamics of the oral microbiota as it relates to IBD. To obtain an integrated view of the spatial and temporal distribution of the oral microbiota, we examined bacteria from 4 anatomically different sites within the oral cavity. Our findings confirm that although a candidate set of bacterial members was shared between all oral sites, each site harbored a characteristic microbiota differing in both composition and diversity that persisted longitudinally. We further noticed that the oral microbiota in IBD featured site- and taxa-specific dysbiosis. Interestingly, the oral microbiota—the salivary microbial structure in particular—performed similar if not better than the fecal microbiota in distinguishing patients with IBD from healthy control patients.

The primary goal of this study was to define and understand the spatial and temporal dynamics of oral microbial composition in IBD to subsequently test whether site-specific oral microbial dysbiosis can be used as a surrogate marker of IBD in lieu of or in addition to stool. Although it is not a perfect representation of the gut microbiome,¹¹ fecal sampling has been widely used to characterize the microbial landscape and its contribution to disease development in IBD. Reduced overall microbial diversity and richness, inferred from stool, has long been recognized as a hallmark of IBD and its 2 main subtypes, CD and UC. Interestingly, we did not notice any significant difference in terms of microbial diversity or richness between patients with IBD and healthy control patients in any of the oral microenvironments, suggesting that disease-associated effects on alpha-diversity are relatively less pronounced in oral habitats than in stool. However, note that although our observation is largely supported by previous reports on plaque, buccal, and salivary microbiotas,^{14, 16, 17} Docktor et al¹⁵ reported reduced diversity in tongue samples obtained from patients with CD but not UC, compared to healthy control patients.

At the population level, oral samples reflected dysbiosis in patients with IBD and healthy control patients, surprisingly, in a site- and taxa-specific manner. Although some of these changes could be attributed to IBD (either as causal to or a consequence of disease), we cannot rule out the role of various other factors with localized effects, including nutrient and environmental conditions such as pH and temperature, sugar intake, oral hygiene, saliva flow, tooth eruption, and the level of plaque development,¹⁹ which confound or result in pleiotropic associations of IBD and the microbiome. As reviewed elsewhere,⁴⁴ because of vulnerability to confounding and reverse causation, distinguishing microbial changes that are causal to disease from those that are a consequence has always been a challenge. Our current observations implicating a potential role for confounding or horizontal pleiotropy in IBD-microbial associations, particularly in oral samples, suggest that disentangling this intricately complex relationship between the microbiome and IBD into a simple cause-effect will be even more challenging. Nevertheless, as proposed elsewhere⁴⁴ for the first time and supported by the work published by Sanna et al⁴⁵ on delineating causal effects of microbial features in metabolic diseases, future studies with increased sample size and available genome-wide summary statistics could leverage genetic associations and the concept of Mendelian randomization to pinpoint microbial changes that are potentially causal to IBD.

Furthermore, disease-associated microbial changes that are consistent across all habitats could provide insights into disease pathogenesis. Our efforts to this end were, to an extent, thwarted because of the usage of 16S rRNA gene sequencing rather than shotgun metagenomics with deep sequencing to infer species- and strain-level taxonomic classifications. Nevertheless, at the phyla level, Actinobacteria, Bacteroidetes, and Spirochaetes were enriched in IBD across all 4 oral sites, whereas Fusobacteria, Firmicutes, and Proteobacteria were among the phyla that were depleted in patients with IBD, in agreement with previous findings from 1 or more of these oral microenvironments.^{15, 16} In contrast, the phyla Fusobacteria and Proteobacteria were significantly enriched in IBD fecal microbiotas, whereas Actinobacteria showed a trend of depletion. We also noted striking differences between oral and stool microbiotas in relation to IBD at other taxonomic levels, including class, order, and family. These observations are particularly noteworthy in that previous investigations of IBD have consistently found compositional changes of oral-resident microbes in intestinal and/or stool samples from patients with IBD compared to healthy control patients, giving rise to the oral–gut axis hypothesis. However, the directional relationship between the dysbiosis of oral-resident microbes in intestinal vs oral habitats in patients with IBD remains to be addressed. Our finding that IBD-associated microbial signals from oral habitats were either lost or trended in the opposite direction in stool suggests a negative correlation between the 2 habitats that warrants future investigation.

Similarly, identifying individual microbes that remain fairly consistent over time in control patients while they exhibit transitions in patients with IBD would make them very good candidates for future microbiome-based functional studies. Our findings indicate that the composition of some taxa are more consistently perturbed than others, suggesting that drawing conclusions based on single time point microbiome features in case-control studies is problematic, especially when aiming for the identification of disease-specific microbial candidates. We noticed patterns of global stability or variability across patients irrespective of their disease state. Future studies are warranted to identify taxa that are potentially pathogenic by selecting for those that stay stable over time in control patients but exhibit striking shifts in patients with IBD with respect to changes in disease flares, severity, and treatment effects. Our efforts to make any such biologically meaningful observations were thwarted by the limited number of healthy control patients with longitudinal follow-up samples.

Collectively, our data are compatible with a differential effect of IBD depending on taxa and sample type. As expected from previous work,^{32, 33} we noted anatomical location as the strongest driver of microbial composition within the oral cavity, which supports the view that it is critical to define and understand baseline spatial differences with respect to the region of the oral cavity profiled to interpret disease-associated dysbiotic states and eventually to gain insights into disease etiology. Nevertheless, we showed for the first time that the oral microbiota has discriminatory power for classifying patients with IBD from healthy control patients regardless of location and, surprisingly, that the salivary microbiota performed even better than stool in that the latter was widely believed to hold the potential of a noninvasive diagnostic approach. This finding is particularly noteworthy in light of previous research suggesting that intestinal pathobionts that exacerbate IBD may reside in the oral cavity.^{11, 46}

There is great interest in using features of the microbiome as a means to diagnose and improve different aspects of human health. Given the fact that oral samples are significantly easier to obtain and process than fecal samples and are less invasive than intestinal biopsies, an opportunity is created to use an oral microbial sampling approach to diagnose and monitor patients with IBD. It would be of tremendous importance in the future to investigate whether oral microbiota can discriminate CD from UC and its usefulness in monitoring or predicting treatment effects. The identification of oral microbial biomarkers that predict changes in disease flares and the risk of developing disease-associated complications may help identify patients at high risk and may facilitate pre-emptive treatments.

Our study has several limitations. We did not have any measures pertaining to diet, oral health status, duration since the last oral hygiene, and stages of dentition, all of which are particularly relevant to oral microbial composition. We do acknowledge that our population size and study design to include patients with established IBD were suboptimal, although we did not notice any significant difference in alpha-diversity or beta-diversity or at the individual microbial-member level between newly diagnosed patients with IBD vs patients with established IBD. Although we were primarily interested in comparing the structural composition and diagnostic potential of the microbiota of various sites within the same patients, it is possible that treatment may have had site-specific effects on microbial composition, which could potentially influence our findings.

Conclusions

Nevertheless, our findings highlight the proposition that oral microbial surveillance can serve as a diagnostic marker to discriminate patients with IBD from healthy control patients. Given the fact that obtaining oral samples is relatively easier than obtaining stool and intestinal biopsies, an opportunity exists to perform microbiome-based studies in larger cohort sizes, preferentially in a longitudinal fashion. Our findings also highlight the importance of understanding baseline spatial differences within the oral microenvironments to interpret disease-associated changes. Given the differences and directional inconsistency between stool and oral microbiotas, fueled by our previous observation that stool is not a perfect reflection of intestinal microbial community structure,⁹ it would be of value to perform a side-by-side investigation of intestinal biopsy, stool, and oral microbial features in future studies of the role of the microbiome in IBD.

Author contributions: H.K.S., J.W., C.S., S.A., M.E.Z., D.J.C., D.T.O., P.C., and S.K. designed the study. J.W., A.D., K.L., J.P., A.K., and N.H. collected the samples. A.D., K.L., and A.K. processed the samples. H.K.S. and P.C. performed statistical analyses, and S.V assisted with the analysis. H.K.S. and S.K. interpreted the results and drafted the manuscript, and M.E.Z., D.J.C., D.T.O., and P.C. assisted with results interpretation and writing. All authors discussed the results and commented on the article.

Supported by: This work was primarily supported by The Marcus Foundation (to S.K.), with further support from DK087694 (S.K.) and DK098231 (S.K.) from the National Institutes of Health.

Acknowledgments

We are grateful to Claudio Fiocchi and Melanie Schirmer for their support and helpful comments on the manuscript. We also would like to thank AKESOgen, Inc., in Norcross, Georgia, for assistance with library preparation and 16S rRNA gene sequencing.

References

Author notes