The human microbiome plays a critical role in human health and disease1. In particular, the oral microbiome is associated with not only the health of the mouth, but also risk of other chronic diseases (e.g., cardiovascular disease2,3 hypertension4, type 2 diabetes5, and cancer6,7). Understanding of the factors (e.g., dietary intake, smoking behavior, medication use, etc.) affecting the composition of the oral microbiome is critical to understanding these observed associations with disease outcomes.

Over 700 different species of bacteria have been identified in the oral cavity8 with, on average, more than 250 different species in any one individual mouth9. The diversity of the oral microbiome in relation to oral health is complex. For example, previous data shows that the alpha-diversity of the microbiome in supragingival plaque samples (where cariogenic pathogens reside), decreases with the severity of caries10. Differently, the alpha-diversity in subgingival plaque samples, (where periodontal pathogens reside), increases with increasing severity of periodontal disease11,12, and such a relationship was observed in this cohort with the microbiome of our subgingival plaque samples13.

Diet has been shown to be associated with both caries and periodontal disease14 and hypothesized to influence the microbial composition and diversity of the saliva and gingival crevicular fluid15. Fermentable carbohydrates (simple sugars and starch) are significant sources of bacterial energy metabolism and are broken down by both bacterial enzymes and by endogenous processes in the oral cavity15. There is evidence that fermentable carbohydrates are essential to development of dental caries16. However, the association of carbohydrate intake with periodontal disease is less well studied17,18,19,20,21. Few studies have examining habitual intake of dietary carbohydrates in relation to the diversity and composition of the oral microbiome22,23,24.

We studied the association between habitual dietary carbohydrate intake and the subgingival plaque oral microbiome in a cohort of 1204 postmenopausal women, using data from the Buffalo Osteoporosis and Periodontal Disease (OsteoPerio) Study, a cohort study ancillary to the Women’s Health Initiative (WHI) Observational Study (OS)25. The OsteoPerio Study used 16S rRNA gene sequencing of oral plaque samples to identify and measure the relative abundance of the oral bacteria found26. We hypothesized that the alpha-diversity (within-subject diversity [number of species]) of the oral microbiome would be associated with intake of total carbohydrates, GL, starch, disaccharides (lactose, maltose, sucrose) and monosaccharides (fructose, galactose, and glucose) and that the beta-diversity (between group diversity) of the oral microbiome would differ across quartiles of intake in all carbohydrates and glycemic load (GL).


Study design

The OsteoPerio Study is an ongoing prospective cohort26, and ancillary to the WHI, a national study focused on health outcomes of postmenopausal women25. The OsteoPerio study was originated to examine the association between osteoporosis and loss of bone in the oral cavity27. Study participants were recruited from the WHI clinical center in Buffalo, NY between 1997 and 2001; 1,342 women participated in the baseline exam (Supplemental Fig. 1)26. Women were excluded if they had fewer than 6 teeth, bilateral hip replacement, a history of non-osteoporotic bone disease, a recent 10 years history of cancer, or if they were treated for serious diseases26. There were 1222 women with sequenced subgingival microbiome and dietary data at baseline, and of these, 18 women were excluded because their self-reported energy intakes were > 5000 or < 600 kcals, leaving a sample of 1204 women. All participants provided informed consent, and the study protocol was approved by the University at Buffalo’s Health Sciences Institutional Review Board. All experiments were in agreement with relevant guidelines regarding Human Subjects Research.

Assessment of dietary carbohydrate intake

Dietary intake was assessed as part of the WHI OS participant’s year 3 visit, which coincided with the OsteoPerio baseline exam26. A modified Block food frequency questionnaire (FFQ), with 122 main questions and 4 summary questions, was administered to participants asking them to recall usual consumption during the last 3 months28. The WHI FFQ has been validated in a study conducted among 113 women in the WHI comparing the FFQ to mean intake from four, 24-h recalls and one 4-day food record28. The energy-adjusted Pearson correlation coefficients for total carbohydrates and total fiber were 0.63 and 0.65, respectively28. Our main exposures include intake of total carbohydrates, GL, total fiber, soluble fiber, insoluble fiber, starch, disaccharide intake (lactose, maltose, sucrose) and monosaccharide intake (fructose, galactose, and glucose). GL reflects both the amount of carbohydrate in a food in addition to its influence on blood sugar. In this study, total carbohydrate including fiber intake, rather than available carbohydrate intake, was used to estimate the GL29. Carbohydrate intake is presented as the percent of calories from carbohydrate consumed, or in the case of fiber intake and GL, as grams per 1,000 kcals consumed. All analyses used these energy-adjusted variables.

Subgingival plaque samples and sequencing

A dental examiner performed an oral examination wherein subgingival plaque samples were taken with paper points from 12 index teeth (or substitutes), as described previously30. Paper points were inserted into subgingival pockets of a tooth’s mesiobuccal surface, with samples taken from maxillary and mandibular teeth and stored in freezers at − 80 °C.

The composition and diversity of the oral subgingival microbiome were assessed by 16S ribosomal DNA (rDNA) sequencing with the Illumina MiSeq platform as previously described26. Briefly, bacterial DNA was isolated from subgingival samples (maxillary and mandibular samples pooled) with the DSP Virus/Pathogen Mini Kit in QIAsymphony SP automated system (Qiagen, Valencia, CA). Before DNA extraction, an enzymatic pretreatment was performed for more efficient isolation of Gram-positive bacteria. Metagenomic DNA was subsequently amplified for the 16S rRNA gene hypervariable V3–V4 region with negative (extraction reagents and microbial DNA free water) and positive (subgingival plaque pools and Zymogen mock DNA standard) controls. Three hundred base paired-end sequencing (2 × 300) was performed using the MiSeq Reagent Kit V3 on the Illumina MiSeq. Paired-end sequences were joined using Paired-End read merger (PEAR version 0.9.6). The joined sequences were then filtered for quality with the Fastx-Toolkit (V.0.013) to isolate the Illumina paired-end reads that had 90% of their bases measured up to a score of at least Q3031. This score means that only 1 out of every 1000 bases may be incorrect32. Only participant samples that had a minimum at least 3,000 reads were included in our analytic sample31. Following quality filtering, sequences were clustered at 97% identity against the Human Oral Microbiome Database (HOMD) version 14.533 with Basic Logical Alignment Search Tool (BLAST) aiming at the species level34. Finally, in the raw OTU tables, any OTU that had a frequency count of < 0.02% of the total reads was removed from the sample31.

Assessment of additional covariates

Participant characteristics including height, weight, and blood pressure were measured in the OsteoPerio clinic by trained examiners. Either as part of the broader WHI OS25 or the OsteoPerio Study26, data were collected on women’s age, race/ethnicity, education, medical and oral history, lifestyle and health behaviors, dietary supplement intake, and use of medications, including antibiotics, in the last 30 days.

Statistical analysis

The subgingival microbiome was analyzed using Compositional Data Analysis techniques35,36 to avoid spurious correlations arising from compositional structure in the data. We used the centered log2-ratio (CLR) transformation, which represents the abundance of taxa relative to the geometric mean of the sample, and is defined by the formula CLR(x) = log2(x/g(x)), where g(x) is the geometric mean of the vector x37. We added 1 to all counts because of the existence of some zero values. This removes the zeros and keeps proportions of non-zero counts close to their natural values. Since we are using a logarithm base 2, a CLR transformed abundance of 3 represents a species with 23 times greater abundance than the average within the sample. Hereinafter, the CLR transformed relative abundance of each OTU is referred to as “relative abundance”. Measures of relative abundance are considered primary endpoints for this analysis, along with measures of alpha- and beta-diversity.

A correlation matrix across all carbohydrate variables was computed. Mean carbohydrate intake was described by the level of participant characteristics. T-tests and ANOVAs were used to test for significant mean differences across characteristics. We examined the association between carbohydrate intake and three indices of alpha-diversity: observed OTU count, the Chao-1 Index38,39 (both representing species richness), and the Shannon Index (representing species evenness)40,41. We regressed each alpha-diversity measure on each carbohydrate variable and GL to examine the intra-individual microbial diversity in relation to carbohydrate intake. We also tested differences in the beta-diversity of the microbiome by carbohydrate intake by examining measures of Euclidean distance within and between quartile groups of each carbohydrate intake and GL variable using a PERMANOVA test. We visualized the associations by graphing the samples according to the top two principal components explaining variance in our 245 OTUs, color-coding the points by quartile, and drawing 95% content ellipses.

We also regressed each OTU’s relative abundance measure on continuous measure of total carbohydrate intake, GL, and subtype of carbohydrate. We present crude models and models adjusted for age, race and ethnicity, frequency of flossing, frequency of brushing, frequency of dental visits, smoking status, pack-years of smoking, and antibiotic use. We also considered models further adjusted for body mass index (BMI) and diabetes status, which may be in the causal pathway between carbohydrate intake and the composition of the subgingival microbiome. Data was missing for smoking status (n = 1), frequency of flossing (n = 5), and pack-years of smoking (n = 27) therefore adjusted models have 1,172 rather than 1,204 participants. Crude and adjusted beta-coefficients, associated standard errors, and p-values for each carbohydrate variable and OTU association are presented. The beta-coefficients represent the difference in the relative abundance of a specific OTU for each one-unit increase in carbohydrate/GL intake. A Bonferroni correction for the p-values was used to account for multiple comparisons (0.05 divided by 245). In exploratory analyses, we also repeated our analyses for total carbohydrate intake further adjusted for sucrose. In this way, we explored to what extent the associations with total carbohydrate intake were explained by simple rather than complex carbohydrate intake.

In exploratory analyses, we examined which food groups explained the greatest between person variation in carbohydrate or GL intake. Only carbohydrate variables found to be significantly associated with microbiome relative abundance were examined. We used forward stepwise regression with an inclusion criteria p-value of 0.10 and an exclusion p-value of 0.05 to identify significant contributing foods groups.


We examined a correlation matrix of all carbohydrate variables and GL. The strongest correlations (≥ 0.70) were seen between total carbohydrates and GL, total fiber, and soluble fiber; between all fiber types (total, soluble, and insoluble); and between fructose and glucose (Supplemental Table 1).

With the exception of antibiotic use, all participant characteristics were associated with at least some of the carbohydrate components (Tables 1 and 2). There was greater mean soluble fiber, fructose, and glucose intake in older compared to younger women. Sucrose intake was highest in Non-Hispanic Black/African Americans and lowest in Hispanic/Latinas. Fructose and glucose intake was highest in Non-Hispanic Black/African Americans and lowest in Non-Hispanic Whites. The mean intake of total carbohydrate, total fiber, soluble fiber, insoluble fiber, and galactose was greater in women with a post-college education compared to those with less education. Intake of total carbohydrates, GL, total fiber, soluble fiber, insoluble fiber, fructose, galactose, and glucose intake was higher in those with a low compared to high BMI. Never-smokers had the highest intake of total carbohydrate, GL, total fiber, soluble fiber, insoluble fiber, fructose, and glucose, followed by former smokers, and current smokers. Dietary sucrose and glucose intake were  lower in women reporting diabetes compared to those with no history of diabetes.

Table 1 Mean energy-adjusted total carbohydrate intake, glycemic load, fiber intake, and starch intake by category of participant characteristics (n = 1,204).
Table 2 Mean Energy-adjusted Disaccharide and Monosaccharide Intake by Category of Participant Characteristics (n = 1,204).

Mean total carbohydrate intake, total fiber, soluble fiber, and insoluble fiber intake were higher in those who brushed more compared to less frequently. Mean total carbohydrate intake, GL, total fiber, insoluble fiber, lactose, and galactose intake were higher in participants who flossed more compared to less frequently. Dental visits were associated with total and insoluble fiber intake, with higher fiber intake in those who had visited the dentist more as compared to less frequently.

There were 122,631 read pairs generated per sample, 120,032 per sample after merging pair-end sequences, 91,165 reads per sample used for OTU-calling, and 86,972 reads per sample that remained in the OTU table. We identified 245 OTUs in the subgingival plaque samples. Firmicutes was the most abundant phylum identified, accounting for more than 45% of reads within the dataset, followed by Bacteroidetes (17.2%) and Fusobacterium (13.5%). The most abundant species identified were Veillonella dispar and Veillonella parvula, two species from the phylum Firmicutes neither of which ferment carbohydrates.

As intake of total carbohydrate, GL, lactose, and sucrose increased, all three alpha-diversity measures decreased (Table 3), and as starch intake increased, OTU count decreased. Adjustment of the total carbohydrate model for sucrose intake attenuated the associations with alpha-diversity measures (data not shown); however associations were still statistically significant. Microbial beta-diversity was found to be statistically significantly different by quartile of total carbohydrates, fiber (total, soluble, and insoluble), GL, sucrose, and galactose intake (PERMANOVA p < 0.05). Supplemental Fig. 2 plots the associations between the top two OTU principal components among study participants for GL that had the smallest p-value for PERMANOVA.

Table 3 Beta-coefficients and standard errors (SE) associated p-values for regression of alpha-diversity measures on carbohydrate intake (n = 1204).

We examined continuous intake of total carbohydrates, GL, and carbohydrate subtypes in relation to the relative abundance of all 245 OTUs (Table 4). The beta-coefficients, standard errors, and p-values for each association examined are shown with no adjustment (crude), adjustment for age, race and ethnicity, frequency of flossing, frequency of brushing, frequency of dental visits, smoking status, pack-years of smoking, and antibiotic use (Model 1); and with further adjustment for BMI, and diabetes status (Model 2). In adjusted models, after correction for multiple comparisons, there were significant associations between intake of total carbohydrates, GL, lactose, and sucrose and the relative abundance of at least one OTU. The relative abundance of Streptococcus mutans was positively associated with total carbohydrates, GL and sucrose intake in all models. The association between Streptococcus mutans and total carbohydrate intake in Model 2 was not statistically significant after further adjustment for sucrose intake (data not shown). We also observed a positive association between GL and Sphingomonas HOT 006 in Model 1, and in Model 2 we observed a positive association between GL and both Sphingomonas HOT 006 and Scardovia wiggsiae. In Models 1 and 2, we observed a negative association between lactose and Aggregatibacter segnis, and a negative association between sucrose and both TM7_[G-1] HOT 3436 and Leptotrichia HOT 223. A positive association between sucrose and Streptococcus lactarius was observed only in Model 2. Results for all dietary carbohydrate variables and OTUs associated at a p-value of < 0.05 are presented in Supplemental Table 2.

Table 4 Linear regression of the relative abundance of oral OTUs on carbohydrate intake and glycemic load with beta-coefficients (ß), standard errors (SE), and associated p-value for each dietary variable. (n = 1,204).

Exploratory analyses (Supplemental Table 3) identified which food groups, from a list of 122 food groups on the FFQ, explained at least 80% of the variation in total carbohydrate, GL, lactose, and sucrose intake. Twenty-four of 122 food groups were identified. We summarized these foods into the following descriptive groups: (1) grains and baked-goods; (2) starchy vegetables and fruit, and cooked tomatoes; (3) sugary drinks; (4) added sugar, candy, frozen desserts, and pudding-type desserts; (5) and dairy products.


In this analysis of postmenopausal women, we observed that intake of total carbohydrates, GL, starch, lactose and sucrose were negatively associated with the alpha-diversity of our microbiome measures; increased intake was associated with lower intra-individual diversity. We also observed differences in the diversity of the oral microbiome across level of intake of total carbohydrates, fiber, GL, sucrose, and galactose (beta-diversity). Intake of total carbohydrates, GL, and sucrose were positively associated with the relative abundance of Streptococcus mutans, a bacteria with an expansion of carbohydrate metabolizing genes42,43. We also observed a positive association between GL and the relative abundance of both Sphingomonas HOT 006 and Scardovia wiggsiae, and between sucrose and Streptococcus lactarius. We observed a negative association between lactose and Aggregatibacter segnis, and between sucrose and both TM7_[G-1] HOT 346 and Leptotrichia HOT 223. To best of our knowledge, this is one of the first epidemiologic studies to examine associations between habitual carbohydrate intake and subgingival, rather than salivary, microbiome samples; we found that carbohydrate intake is associated with the subgingival microbiome.

Minimal research on habitual carbohydrate intake and the oral microbiome has been conducted in humans. In a study in children, there were significant differences in the relative abundance of 18 species from the biofilm of occlusal surfaces by fermentable carbohydrate consumption assessed using an FFQ23. They did not identify Streptococcus mutans as one of the18 species. They did identify Aggregatibacter segnis, which we observed as associated with lactose intake, and Rothia mucilaginosa, which we identified as related to total carbohydrate intake in crude analyses. In a study of Danish adults (aged 20 to 81 years), there were no significant differences in salivary bacterial species by intake of energy-adjusted carbohydrates or the proportion of carbohydrates from sugar22. In a three week carbohydrate intervention study of 21 athletes, no significant differences in salivary microbial composition were observed24.

Our analysis does not fully capture associations between carbohydrate intake and all medias of the oral microbiome because we exclusively examined the microbiome in subgingival plaque samples unlike the previous studies of habitual carbohydrate intake that examined the microbiome of the saliva or biofilm of occlusal surfaces22,23,24. We found no evidence of Lactobacillus, known to be associated with caries risk44, in our subgingival plaque samples after we filtered out low abundance OTUs. Lactobacillus, highly abundant in saliva, is not found as frequently in the subgingival microbiome45. Several studies have identified Streptococcus mutans in subgingival microbiome samples, similar to our study46,47 Carbohydrates are likely accessible to a different composition of bacteria in subgingival, anaerobic conditions, compared to the salivary environment48. Despite likely differences in the microbiome of the saliva and subgingival plaque, studies have detected periodontal pathogens in both mediums, concluding that there is some overlap between these two microbiomes49,50. Likely the previous studies’ use of salivary or occlusal surface samples, and differences in participants’ ages may explain differences in our results and previous study findings.

As expected, we found a number of associations between sucrose intake and the subgingival microbiome. Sucrose can be broken down into glucose and fructose and taken up by the bacteria, or it can be cleaved inside the bacterial cell by bacterial enzymes15. Starches can be broken down by human salivary amylase or by bacterial amylases. Certain Streptococci such as Streptococcus gordonii and Streptococcus mitis can bind amylase to metabolize starch, while other bacteria, such as Streptococcus mutans, have enzymes of their own capable of metabolizing starch15. Once broken down into simple sugars, sucrose can be transported into the bacterial cell for energy production15. Experimental studies show that increasing sugar and fermentable carbohydrate intake increases prevalence of caries51 and that frequent sucrose consumption is associated with decreased species diversity, and increased relative abundance of certain Streptococcus spp. in the oral biofilm52. Our results support the existing evidence that certain fermentable carbohydrates (e.g., sucrose) promote the growth of cariogenic oral bacteria, such as Streptococcus mutans16,53 We also observed that increased carbohydrate intake was associated with decreased alpha-diversity similar to other studies23,52,54.

The association of carbohydrate intake with periodontal disease, rather than caries, is less well studied17,18,19,20,21. There is evidence of associations between increased carbohydrate intake and increased gingival bleeding17,55 and positive associations between diets high in percent of calories from carbohydrates and rates of periodontal disease21. Leptotrichia spp., which we observed to be positively associate with sucrose intake, has been shown to be associated with gingivitis in some studies12,56. The other bacteria we identified as associated with carbohydrate intake or GL have not been previously appreciated as contributing to periodontal disease in the literature12 or in this cohort13.

There is evidence that fiber intake is associated with decreased risk of periodontal disease progression markers18,20,57. In a previous study, the oral microbiome (from extracted mice jaws) of mice fed sugar and fiber pellets compared to mice fed sugar pellets alone was lower in Streptococcus, Staphylococcus, Lactobacillus, and Enterococcus, as well as greater in alpha-diversity58. This suggests fiber consumption may result from mechanical disruption of oral microbiome by fiber. We did not find any significant differences in alpha-diversity or the relative abundance of any of the measured bacterial species by differing fiber intake. It may be that any effect of fiber on the oral microbiome is less important in a cohort of women who frequently brush their teeth.

The relationship between carbohydrate intake and the relative abundance of bacteria is not just defined by whether a certain bacterium has the metabolic capability to utilize a carbohydrate. If two types of sugar are available, some bacteria may preferentially utilize one sugar over the other, as they possess regulatory mechanisms for carbohydrate metabolism59. Therefore, we may not see strong relationships with certain types of sugar if both are present and bacteria prefer one over the other. Additionally, bacteria can uptake sugars that have been cleaved by other bacteria or by salivary amylase60. Therefore, even if bacteria do not possess the metabolic capability to cleave a certain sugar, they may still be able to utilize its components, which is why we may see a relationship with a certain type of carbohydrate even if the bacteria cannot metabolize it.

We also identified top contributing food sources of total carbohydrate and GL, sucrose, and lactose in a cohort of postmenopausal women. Our findings suggest that attention to dental hygiene should occur after consumption of these foods (e.g., baked goods, added sugar, candy, milk, etc.). This is in alignment with the American Dental Associations’ guidelines on Diet and Nutrition which states “that oral health depends on proper nutrition and healthy eating habits, and necessarily includes avoiding a steady diet of foods containing natural and added sugars, processed starches and low pH-level acids…”61. A recent dietary intervention (n = 11 adults, average age 32 years) showed that milk and yogurt consumption, as compared to sucrose intake, resulted in less growth of cariogenic bacteria. Continued research needs to be conducted to better understand the influence of carbohydrate-containing foods, which also contain other nutrients, on the oral microbiome62.

Our study has several limitations. Because it was cross-sectional, we cannot make any assumptions about temporality or causality. FFQs, although useful in that they assess habitual dietary intake, are prone to social desirability bias and often underestimate energy intake63. We adjusted for energy intake in an attempt to minimize measurement error and underestimation of energy intake64. The measure of relative abundance is also not without its shortcomings65. Because relative abundance relies on the proportion of the bacteria rather than their absolute number, the measure may induce spurious correlations65. However, this limitation is minimized here by adopting Compositional Data Analysis techniques, such as the use of the CLR transformation. Another limitation is that we were unable to examine our oral microbial compositions by anterior versus posterior teeth or by teeth in the upper (maxillary) versus lower (mandibular) jaw arches. This is because we stored plaque samples from all maxillary teeth together and from all mandibular teeth together and then combined these plaque samples prior to sequencing them for bacterial DNA. We did not do an internal assessment of the reliability of our results. We also corrected for multiple testing for 245 OTUs, but did not further correct for multiple testing across our 11 carbohydrate variables and GL.

The age distribution of our participants could be considered a limitation. However, the postmenopausal age range gave us an opportunity to examine these effects in a subpopulation where the association between carbohydrates and the oral microbiome has not been previously studied. Findings may be different in samples with different ages; a broader age group might have allowed for examination of how more varied intake of carbohydrates might affect the oral microbiome over the lifespan.

Despite the limitations of this study, it has important strengths. It is the first study to examine carbohydrate intake and the subgingival microbiome in a sample consisting exclusively of postmenopausal women. We examined many subtypes of carbohydrate, and GL, in order to better understand which carbohydrate components have the strongest associations with the subgingival microbiome. We were able to control our analyses for potential confounding factors including oral hygiene, smoking, and antibiotic use. The selection of our participants into the OsteoPerio study is another strength—they were not selected based on disease status or dietary intake. This would have made our results less generalizable.

In conclusion, our findings suggest that total carbohydrate and GL, as well as intake of the disaccharides sucrose and lactose, are inversely associated with bacteria alpha-diversity in the subgingival microbiome. Furthermore, the beta-diversity of the microbiome varied by total carbohydrates and GL, but also by certain carbohydrate subtypes (sucrose, galactose, and fiber); and we observed that intake of the total carbohydrates, GL, sucrose and lactose to be significantly associated with the relative abundance of specific OTUs estimating bacterial species. Further study of food group intake and dietary patterns will contribute to our understanding of the extent to which the oral microbiome varies in association with carbohydrate consumption and the extent to which these differences are associated with periodontal disease, oral health, and the influence of oral health on systemic health.