Women were recruited by advertisements in student magazines, university notice boards, and social media and included between September 2017 and January 2018 at Rigshospitalet, Copenhagen, Denmark. All data were collected and managed using REDCap electronic data capture tools, hosted at the Capital Region of Denmark. The study is approved by The Regional Committee on Health Research Ethics (H-17017580) and the Data Protection Agency in the Capital Region of Denmark (2012-58-0004). All participants gave oral and written consent to participate. All participants were asked to phone the clinic when spotting/beginning of bleeding. They were then scheduled for a hospital visit on cycle day 1–3. An ultrasound scan was performed to confirm that the participant was not pregnant and to confirm the menstruation with shedding of the endometrium. Additionally, hormone levels (plasma estradiol and progesterone) were measured. The two following visits were booked (day 8–12) and (day 18–22) every time with an ultrasound scan to confirm cycle phase and absence of pregnancy. One of the inclusion criteria was to have a regular cycle (median 26 days). The subjects were divided into 3 groups: group 1 (n = 43) that did not use any hormonal contraceptives, group 2 (n = 41) that used combined oral contraception (COC: estrogen + progestin) for at least half a year prior to the start of the study and group 3 (n = 19) using levonegestrel intra-uterine system (LNG-IUS). Women were excluded from the study if they were pregnant or had an intention to become pregnant during the course of the study or had oligomenorrhea or irregular menstrual cycles or spotting. Subjects were also excluded if they had any systemic disease or medical condition or been treated by antibiotics in the past three months prior to the beginning of the study or during its duration.

Participants reported data concerning their life habits and health history. The food frequency questionnaire was based on four-week recall, with frequencies given on a 9-point scale from “0 times in the past four weeks” to “>3 times/day for the past four weeks”. Frequency of free-sugar consumption was derived from the sum of the frequencies for the following food items: chocolate milk, juice, soda with sugar, ice-cream, biscuits and cookies, sweet bread and rolls, dry cake, cake with filling and candy (including chocolate, licorice, jelly and other candy). On quantitative analyses, sugar consumption was divided into low (up to and including the first quartile), high (above the third quartile), or intermediate. In relation to oral health, participants were asked whether they had been to a dentist or dental hygienist during the 3 months prior to answering the questionnaire. Smoking was coded as “daily smoker”, “occasional smoker”, “former smoker”, and “never smoker”.

Women were followed during a full menstrual cycle including three hospital visits. The first hospital visit was at cycle day (CD) 1–3. The second visit was CD 8–12 and the third CD 18–22. The patients were fasting 30 min before saliva collection, including drinking, chewing gum or chewing tobacco, and smoking. Whole saliva samples (2 ml) were collected using a SalivaGene Collector (STRATEC Molecular GmbH, Germany) containing lyophilized DNA stabilization buffer, according to the instructions of the manufacturer, and were frozen at -80°C. Blood samples were drawn at every hospital visit. Blood was collected in 9 ml EDTA tubes, left until separated and spinned for 15 min at 3,000 rpm and plasma was aliquoted and frozen at -80°C. Plasma estradiol and progesterone were measured using the standard automated system (Cobas^® 8000 by Roche Diagnostics).

Saliva aliquots of 600 µl were shipped to CoreBiome (OraSure, Bethlehem, PA, USA) where they were extracted with MO Bio PowerFecal (Qiagen, Hilden, Germany) automated for high throughput on QiaCube (Qiagen), with bead-beating in 0.1 mm glass bead plates. Three spaced negative controls and one positive control were included in each extraction. All negative extraction controls had undetectable amounts of DNA, and all positive controls were also approved. DNA concentration (for samples and controls) was quantified using Quant-iT Picogreen dsDNA Assay (Invitrogen, ThermoFisher Scientific, Carlsbad, CA, USA). Libraries were prepared using an adapted Nextera (Illumina Inc, San Diego, CA, USA) procedure and sequenced on an Illumina NextSeq using single-end 150 bp reads with a NextSeq 500/550 High Output v2 kit. Reads were processed with CoreBiome’s BoosterShot shallow shotgun sequencing technology. The raw sequencing reads are available from the European Nucleotide Archive under project PRJEB37731, samples SAMEA6662389-SAMEA6662857.

Human reads were removed by mapping to the hg19 release of the human genome using BBTools (available at https://sourceforge.net/projects/bbmap/). Because BoosterShot technology is optimized for fecal samples, taxonomy was reannotated the reads using Kraken2 (26) with confidence set to 0.5 and Bracken, based on the Human Oral Microbiome Database v9.0.3 (27). Functional annotation based on KEGG modules was used as provided by CoreBiome.

All statistical analyses were performed in R v. 3.5.2. Alpha-diversity was calculated as the observed number of species as well as Simpson’s inverted index. Comparisons for the same individual across time were calculated as paired t-tests, while comparisons between individuals were calculated using Welch’s t-test. Differences between the three contraception groups were calculated with Pearson’s chi-square test or Fishers Exact test for count data and Kruskal-Wallis test for continuous data. Differences in variation were quantified using Levene’s test of equality of variance, using the Brown-Forsythe variant with R package lawsat (v3.2). Correlations were calculated with Pearson’s product moment. All tests were performed with a 95% confidence interval and a significance cutoff of 0.05. Multiple testing correction was conducted with the Benjamini-Hochberg procedure where applicable. Beta-diversity was calculated on Bray-Curtis distances and clustered on complete linkage. The relative impact of metadata factors on beta-diversity dispersion assessed through Permanova. Alpha- and Beta-diversity analyses were calculated with package Vegan (v2.5-3) and graphs were generated with packages RColorBrewer (v1.1-2) and Vioplot (v0.2). Associations between specific taxa and the metadata were calculated in Maaslin2, treating the contraceptive and phase of the cycle as fixed effects and individual’s identities, smoking, and sugar intake as random effects. A minimum abundance of 0.1% in at least six samples was required to keep a taxon in the analysis. Furthermore, clustering of species was done according to the “color complex” classification of Socransky et al., (1998). (i.e., six-color cluster-lists of bacteria according to their frequency of detection and levels in periodontitis and health) and according to the core microbiome classification data from Abusleme et al. (health-associated core species and periodontitis-associated core species in the subgingival microbiome) (Abusleme et al., 2013). Correlations between color groups and metadata were calculated initially as multivariate ANOVA, and where differences were found, investigated as beta-regressions treating contraceptive, phase of the cycle, smoking and sugar intake as fixed effects and individual’s identities as random effects, using R package glmmTMB (v0.2.2.0).

Demographic and clinical characteristics of the 103 study participants are categorized based on contraceptive use and summarized in Table 1. The participants reported normal menstrual cycles of approximately 26 days (range, 23 to 34 days). There were no significant differences (p > 0.05) in the age, BMI or frequency of tobacco or cannabis usage between the women in each contraceptive group (Table 1). For participants using COC, 25 of 41 reported a daily dosage of 150 µg levonorgestrel and 30 µg ethinylestradiol. The other participants had several different progestins, combined with a daily dosage of ethinylestradiol of 20–35 µg. A single participant was on a multiphasic pill, while three could not give the name of their pill. For participants with an IUS, three brand names were reported (Jaydess, Kyleena and Mirena, all produced by Bayer AB, Leverkusen, Germany), with four participants unable to name the brand of their device.

A few of the participants had been to a dental health professional during the three months prior to the beginning of the study, with no difference between groups (Supplementary Table 1). Self-rated health was overall high and not significantly different between groups (Supplementary Table 1). Sugar consumption did not vary throughout the menstrual cycle or between the contraceptive groups (Supplementary Table 1).

The total number of annotated sequence reads for the overall cohort of 103 women was 102,605,212 (median 242,834 reads per sample, IQR 106,839–482,885). There were 209 microbial taxa identified in the saliva samples. The taxonomic classification is presented in Supplementary Table 2. In brief, the OTUs were collectively represented by eight bacterial phyla, namely Actinobacteria (0–70%, median 4%), Bacteroidetes (0–66%, median 37%), Firmicutes (0–77%, median 31%), Proteobacteria (0–65%, median 20%), Fusobacteria (0–8.6%), Spirochaetes (0–1.4%), and candidate divisions SR1 (0–0.4%) and TM7 (0–14%). Fifty genera were identified, of which the most abundant genera across all samples were Haemophilus (0–48%, median 7%, specially H. parainfluenzae, 0–23%), Neisseria (0–53%, median 8%, specially N. flavescens, 0–21% and N. subflava, 0–23%), Prevotella (0–65%, median 34%, specially P. histicola, 0–24%, P. melaninogenica, 0–26%, P. pallens, 0–20% and oral taxon 313, 0–25%), Streptococcus (0–60%, median 13%, specially S. mitis, 0–60% and S. parasanguinis, 0–27%) and Veillonella (0–42%, median 11%, specially V. atypica, 0–29%) (Figure 1). In addition to these genera, a few samples presented with high levels of Rothia mucilaginosa (0–70%, median 1%) and R. dentocariosa (0–26%, median 0.2%).

To assess the relative effects of several metadata parameters on the structure of the microbiome, we ran a permutational analysis of variance (Permanova) including contraceptive usage, phase of the menstrual cycle, smoking status, free-sugar consumption, and individual identity. All factors are partially explanatory of the distance between samples (Table 2). While unspecific individual factors are dominating, contraception and cycle phase play a comparable role to well-known determinants of oral health, such as free-sugar consumption and smoking (Table 2). Therefore, the remaining analyses were adjusted for these factors when possible.

To assess whether any specific taxa differed in abundance according to female hormonal cycles and contraception, we ran Maaslin2 using contraceptive and cycle phase as fixed effects and individual identity, smoking and sugar intake as random effects (see the methods section for details). We found that Prevotella and Veillonella were increased in current smokers (daily and occasional). Additionally, four genera were found to differ in abundance throughout the menstrual cycle, namely Campylobacter, Haemophilus, Prevotella, and Oribacterium, the latter of which was over-represented by the species O. sinus (Figure 2). In addition, the genus Atopobium was found to be more abundant in IUS users (r = 0.0017, p = 0.047).

No significant differences in alpha-diversity (within-sample diversity) were observed as a result of smoking, although daily smokers had slightly higher richness (number of species) than never smokers (means 36 vs. 34, p = 0.06). Participants in the middle and high ranges of sugar consumption had higher richness and diversity than those with low sugar consumption, but this different was only significant for the middle range (mean richness: low 32.8, mid 39.5, high 38.0; mean diversity: low = 14.7, mid = 16.6, high = 15.7. p-values: richness 0.003, diversity 0.005). We next evaluated alpha-diversity (within-sample diversity) according to menstrual phases or contraceptive categories. No significant differences in alpha-diversity were observed across menstrual phases or between contraceptive methods. Higher richness (number of species) was observed in the luteal phase than in the menstrual phase (Figure 3A), but this did not prove to be significant by a narrow margin (p = 0.06), neither for the entire cohort nor for any of the contraceptive groups separately. There were no significant differences in diversity (Simpson’s inverted index; Figure 3B). Total DNA amounts in the samples did not correlate with either richness or diversity (Supplementary Table 3). While both contraceptive method and phase of the menstrual cycle were statistically correlated to beta-diversity (between samples; Table 2), there was no distinct pattern of clustering by either of these factors (Figure 4; Supplementary Figure 1 ).

The separation between samples was mainly driven by gradients in the abundances of a few species. A gradient was evident with high levels of N. subflava, N. mucosa, N. flavescens, R. mucilagniosa and H. parainfluenzae on one side, and high levels P. histicola, V. atypica and Prevotella taxon 313 marked the other side. A second gradient displayed P. pallens and P. melaninogenica on one extreme, and chiefly S. mitis on the other. Interestingly, P. pallens and P. melaninogenica are not related to smoking status. No linear combination of taxa covered a large amount of the variation, with the first two principal components covering only 16% of total variance. Only when including the first 12 principal components was 50% of the total variance covered (Supplementary Figure 1).

Furthermore, clustering of species was done according to the “color complex” classification (6 “color complex”—blue, green, yellow, purple, orange, and red—based on their frequency of detection) and according to the core subgingival microbiome classification (health-associated core species and periodontitis-associated core species in the subgingival microbiome) (28, 29). The blue, yellow, green, and purple complexes are associated with periodontal health, whereas the orange and red complexes are correlated with periodontitis (Socransky et al., 1998). The bacteria in the metagenomic dataset were grouped into the color complexes. The percentages of each of these complexes per sample were analyzed. The most prevalent and abundant groups in this cohort were the yellow (present in 298/300 women, median abundance 9.5%), orange (297/300, 10.7%), and purple complexes (287/300, 3.7%) (Figure 5).

Multivariate analysis of variance (MANOVA) revealed the yellow group to be associated to contraception and sugar intake, while the purple group was associated to smoking and sugar consumption. These two groups were analyzed further with a beta-regression, adjusting for subject as a random factor. This analysis found the yellow group to be increased in high and intermediate sugar consumption (high: r = 0.31, p = 0.039; intermediate: r = 0.25, p = 0.05). The purple group was found to be decreased in the IUS group (r = -0.43, p = 0.016) and vary across smoking groups, being decreased among daily smokers (r = -0.57, p = 0.0055), increased among former smokers (r = 0.43, p = 0.0052) and not changed for occasional smokers (r = -0.06, p = 0.71). Furthermore, in relation to the menstrual cycle, we observed: a) a higher abundance of the yellow complex during the menstrual phase, b) a tendency for higher abundance of the red and green complexes during the follicular phase, and c) a lower abundance of the blue complex during the follicular phase (Figure 5). The interplay between each metadata factor and each of the colored groups is depicted as a heatmap in Supplementary Figure 2.

When classified according to the two core microbiome groupings (health- or disease-associated), there were also no notable differences in abundances between phases of the menstrual cycle (Figure 6), or between groups of contraceptives (Supplementary Figure 3). The variance in the abundance of health-associated bacteria within the COC group was more than twice greater than that of the other two groups (Figure 6; Supplementary Figure 3; COC, 0.012; NHC, 0.005; IUS 0.004; Levene’s test, p = 10^-4). This difference was mostly driven by the dominance of the yellow complex, typically associated with periodontal health (Figure 5). Bacteria associated with caries were only found in low abundances (median 0, IQR 0–0.13%) and did not vary across the cycle (Supplementary Table 4). Supplementary Table 4 lists all observed bacterial species and their association with health and disease.

To examine the differential representation of particular microbial metabolic and biosynthetic pathways during the menstrual cycle or with different contraceptive use, the functional potential of each sample was assessed based on KEGG modules. These were submitted to Maaslin2 with sugar intake, smoking, cycle phase, and contraceptive as fixed effects and the individual as a random effect, as was done for the taxonomic annotation. A single module, pyrimidine deoxyribonucleotide biosynthesis, was increased for daily smokers, and a single module, Phosphatidylethanolamine biosynthesis, decreased with sugar consumption. 11 modules were found to be differentially abundant over the menstrual cycle (Figure 7; Supplementary Table 5). Amongst these are pathways involving co-factors, such as cobalamin biosynthesis and ascorbate degradation, as well as the biosynthesis of the neurotransmitter gamma-aminobutyric acid (GABA). Another 3 modules varied with contraceptive usage, namely the glyoxylate cycle, guanine biosynthesis and the NADH:quinone oxidoreductase step of oxidative phosphorylation (Figure 7). Overall, functional analysis indicated that more variation is attributed to the phase of menstrual cycle rather than contraceptive usage.

Because standard hormonal measurements only measure endogenous estradiol and progesterone, not the ethinylestradiol, progestin and levonorgestrel used in COC and IUS, respectively, direct associations between female reproductive hormones and the oral microbiome were only conducted for women not using hormonal contraception (Stanczyk and Clarke, 2010). The estradiol levels were higher in the follicular and luteal phase compared to the menstrual phase (Mean ± STDEV: 0.51 ± 0.24, 0.41 ± 0.31, 0.14 ± 0.05 nmol/L, respectively) (Supplementary Table 3). The progesterone concentrations were increased by 19-fold in the luteal phase compared to follicular and menstrual phases (Mean ± SDEV: 32.10 ± 20.45, 1.53 ± 1.21, 1.67 ± 2.78 nmol/L).

The MANOVA indicates that the red group is dependent on estradiol levels. The beta-regression adjusted for sugar intake and smoking did not converge, but visual inspection of the scatter plot between the red group and estradiol levels reveals that, for subjects where red group bacteria are present, they increase with increasing estradiol (Supplementary Figure 4). Adjusting for smoking and sugar intake, 24 species were correlated to estradiol and another 30 to progesterone, but only a single bacterial species was found to be significantly correlated to estradiol after correction for multiple testing, namely, Porphyromonas endodontalis (r = 0.0018, adj. p = 0.039)