Salivary microbiome responses to chlorhexidine and Coriandrum sativum essential oil mouthwashes in patients with periodontitis in a randomized 2-week clinical trial

Abstract

Objectives

To compare the short-term salivary microbiome effects of a Coriandrum sativum essential oil (CSEO) mouthwash, chlorhexidine (CHX) as a benchmark comparator, and water in patients with periodontitis.

Materials and methods

In this randomized three-arm trial, participants with periodontitis rinsed twice daily for 14 days with water, CSEO, or 0.12% CHX. Preclinical anti-Porphyromonas gingivalis and fibroblast cytotoxicity assays informed CSEO concentration selection. Shotgun metagenomic sequencing was performed on 54 saliva libraries, two per participant, from 27 participants sampled at baseline and week 2 (H₂O n = 9, CSEO n = 8, CHX n = 10). Alpha diversity, within-subject Bray-Curtis change, and differential abundance were analysed longitudinally.

Results

Both active agents inhibited P. gingivalis in vitro, with CHX showing greater potency. In the clinical trial, no statistically detectable difference between CSEO and water was observed for the primary species-level Bray-Curtis outcome (q = 0.413), alpha diversity (all q ≥ 0.259), or differential abundance. CHX showed greater species-level Bray-Curtis change than water (q = 0.020), lower species-level Simpson diversity (q = 0.014), and broad taxonomic restructuring.

Conclusions

Under the tested formulation and exposure conditions, CSEO did not produce a statistically detectable group-level salivary microbiome shift relative to water. CHX provided a benchmark for broad ecological perturbation. The CSEO null finding does not exclude modest or heterogeneous effects in a larger cohort.

Clinical Relevance

No clinical benefit can be inferred from the absence of a detectable CSEO-associated microbiome shift. Larger studies integrating clinical, inflammatory, and direct functional outcomes are needed.

Trial registration

ISRCTN79156900, retrospectively registered on 11 May 2026.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s00784-026-07198-8.

Introduction

Periodontitis is a chronic multifactorial inflammatory disease associated with dysbiotic plaque biofilms and progressive destruction of the tooth-supporting apparatus [1]. The current classification emphasizes that staging and grading should capture not only severity, but also complexity of management and risk of progression at the individual patient level [2]. This framework is especially relevant when considering adjunctive interventions that aim to modulate the oral ecosystem rather than simply reduce visible plaque. Contemporary periodontal therapy is organized as a stepwise approach that includes behavioural change, supragingival biofilm control, control of gingival inflammation, subgingival instrumentation, and supportive periodontal care [3]. Adjunctive rinses are therefore clinically relevant not as standalone periodontal treatment, but as short-term ecological tools that may complement mechanical care during active management.

Because adjunctive rinses are applied across the oral cavity rather than to individual periodontal pockets, their microbiological effects may extend beyond the subgingival niche to saliva, mucosal and tongue-associated communities, tooth surfaces, and supragingival biofilms. Subgingival plaque remains the most site-specific matrix for interrogating periodontal pocket dysbiosis and for determining whether an intervention directly modifies subgingival communities. Saliva, by contrast, provides a pragmatic whole-mouth readout of bacteria shed from multiple oral niches and is therefore suited to assessing short-term ecological perturbation after rinse exposure, including potential collateral effects on commensal taxa and functional microbial groups [4]. Accordingly, salivary findings in periodontitis patients should be interpreted as whole-mouth salivary community responses rather than direct measures of periodontal pocket ecology.

Among antiseptic adjuncts, chlorhexidine (CHX) remains the conventional benchmark in dentistry because of its broad-spectrum antibacterial activity and long clinical history [3, 5]. Yet its limitations are not confined to well-known adverse effects such as tooth and tongue discoloration and altered taste perception. Beyond these effects, CHX may also disturb the oral microbiome and its associated functions. For example, in a controlled human intervention study, 7 days of CHX rinsing caused a major shift in salivary microbiome composition, reduced salivary and plasma nitrite, lowered salivary pH and buffering capacity, and was followed by a trend toward increased systolic blood pressure [5]. These findings are biologically plausible because oral nitrate reduction is performed by taxa such as Veillonella, Rothia, and Actinomyces, directly linking community composition to nitric-oxide-related host physiology [6]. Community-level perturbation is therefore important in the oral cavity because health-relevant functions are distributed across common commensals rather than confined to any single organism. Consequently, from a microbiome-conscious periodontitis perspective, the relevant question is not only whether a rinse suppresses pathogenic bacteria, but what kind of ecological and functional disturbance accompanies that suppression.

Essential oil (EO) mouthrinses have a credible clinical and microbiological footprint in periodontal care [7, 8]. Randomized human studies have shown that EO rinses can reduce supragingival anaerobes and volatile sulfur compound (VSC) producing bacteria after single and repeated use, and can also reduce subgingival periodontopathogens after 14 days of rinsing [7, 8]. Clinical work also suggested that supragingivally delivered EO rinses may influence connected niches, as reductions in subgingival Porphyromonas gingivalis and Fusobacterium nucleatum were observed despite topical rinsing [8]. More recently, a randomized shotgun metagenomic study in patients with gingivitis reported that repeated EO mouthrinses shifted plaque communities toward a healthier state, suggesting that EO effects may include ecological reset rather than indiscriminate elimination of all detectable bacteria [9]. From a microbiome-focused perspective, this is especially compelling, given that some essential oils have shown selective activity (targeting pathogenic species more than commensals) in non-oral contexts [10].

Coriandrum sativum essential oil (CSEO) is a plausible candidate within this broader class of EOs because it has documented antibacterial activity across diverse bacteria, and it has also shown in vitro activity against oral species with MIC values ranging from 0.007 to 0.250 mg/mL [11, 12]. In oral biofilm models, CSEO has also shown activity against Candida-associated biofilm formation, further supporting its relevance as an oral care phytochemical rather than only a food or agricultural essential oil [13, 14]. At the same time, CSEO chemistry is not fixed. Its composition is influenced by a chain of biological, environmental, and technological variables, including genotype, cultivation environment, harvest timing, post-harvest processing, and extraction conditions [15]. As a result, both total yield and the relative abundance of major constituents such as linalool, camphor, geraniol, p-cymene, and limonene may differ substantially between preparations. This matters translationally, because biological effects should be attributed to a chemically characterized oil rather than generalized across nominally similar coriander products.

Against this background, the unresolved question was whether CSEO would produce a detectable salivary microbiome response relative to water and how any such response would compare with the broader effects of CHX [5, 9]. In the present study, we therefore used a paired baseline-to-week-2 design to evaluate the short-term effects of CSEO, CHX, and water on the whole mouth salivary microbiome response of patients with periodontitis, focusing on diversity, taxonomic composition, differential abundance, and exploratory oral health-related microbial guilds. CSEO concentration selection was informed by preclinical anti-P. gingivalis activity and fibroblast cytotoxicity dose-response assays. The study integrated preclinical dose-response work against P. gingivalis and in BALB/3T3 mouse fibroblasts with paired salivary shotgun metagenomics to compare a seed-derived CSEO-containing oral rinse with CHX and water over 2 weeks. We tested the hypothesis that, compared with water control, CSEO would induce a detectable but selective short-term shift in the salivary microbiome of periodontitis patients, with less broad ecological perturbation than CHX.

Materials and methods

Preclinical concentration-selection assays

Test products and tested concentration ranges

Before the clinical intervention, preclinical in vitro assays were performed to compare the antibacterial activity and mammalian cell tolerance of CSEO and CHX, and to guide selection of the CSEO concentration used in the human study. CSEO was supplied as a steam-distilled essential oil obtained from coriander seeds by the Institute of Environmental Solutions (Latvia), batch 394-M05-CorEO-2022. Gas chromatography–mass spectrometry characterization of this batch showed a linalool-dominant chemotype, with linalool as the major constituent (67.47%), followed by γ-terpinene (9.73%), α-pinene (8.34%), camphor (4.68%), limonene (2.93%), and smaller amounts of geranyl acetate and geraniol. The full GC-MS profile is provided in Online Resource 1 (Table S1). CHX was evaluated as a 0.12% chlorhexidine solution. For the clinical phase, the comparator rinse was an extemporaneously prepared 0.12% chlorhexidine solution dispensed in small batches by a local pharmacy.

Fresh CSEO working stocks for the antibacterial assay were prepared on the day of each experiment by dilution in the same sterile growth medium used for the assay and vigorous vortexing for 30 s to obtain a temporary dispersion. No surfactants or added organic solvents were used in the antibacterial assay. For the fibroblast assay, CSEO stock solutions were prepared in dimethyl sulfoxide (DMSO) and subsequently diluted in complete culture medium. For antibacterial testing against P. gingivalis, CSEO was assayed in two-fold serial dilutions yielding final in-well concentrations of 0.039–40 mg/mL. For the 24 h fibroblast assay, CSEO and CHX were tested across log-scaled concentration series of 0.01–10 mg/mL and 0.001–1.0 mg/mL, respectively.

Antibacterial activity against Porphyromonas gingivalis

Antibacterial activity was assessed against Porphyromonas gingivalis ATCC 33,277, selected as a clinically relevant periodontal pathobiont and a canonical target organism for periodontitis-oriented antimicrobial screening. The assay workflow was carried out at the Department of Oral Biology, University at Buffalo, School of Dental Medicine, using a broth microdilution dose-response format screening pipeline.

P. gingivalis was cultured anaerobically at 37 °C for 48 h in Brain Heart Infusion (BHI)-based medium supplemented with L-cysteine, yeast extract, vitamin K, and hemin, adjusted to pH 7.2. Before assay setup, cultures were standardized spectrophotometrically to OD600 = 1.0 in fresh medium and diluted to the working inoculum used for plate inoculation. Based on pilot titration experiments, a relatively high in-well inoculum was used for P. gingivalis to ensure reproducible growth during 48 h anaerobic incubation. The final in-well inoculum was approximately 10^7 CFU/mL.

Dose-response testing was carried out in sterile 96-well microtiter plates. Briefly, 100 µL of each test dilution was dispensed into the wells, followed by 100 µL of standardized bacterial suspension, yielding a final volume of 200 µL per well. Each plate included medium-only sterility controls, inoculated growth controls without test product, and product-only wells at matching concentrations to account for intrinsic turbidity. Plates were incubated anaerobically at 37 °C for 48 h using BD GasPak EZ anaerobe sachets. Endpoint absorbance at 600 nm was measured on a FlexStation 3 multi-mode microplate reader controlled by SoftMax Pro software (Molecular Devices, San Jose, CA, USA). To correct for optical interference from the essential oil, background-corrected growth values were calculated by subtracting the mean OD600 of product-only wells from the OD600 of inoculated wells at the corresponding concentration. Corrected values were then normalized to the maximum mean bacterial growth observed in the assay to generate dose-response curves. Each concentration was tested in triplicate wells, and each compound–organism combination was repeated on at least two independent experimental days starting from fresh cultures. Concentration-response curves were analysed in GraphPad Prism [version 10.6.0 (796), RRID:SCR_002798] using a four-parameter variable slope inhibitory dose-response model. Antibacterial potency was summarized as model-derived relative IC50 values with 95% profile-likelihood confidence intervals. Relative IC50 was defined as the concentration producing a fitted response halfway between the fitted top and bottom plateaus of the inhibitory dose-response curve.

Fibroblast cytotoxicity assay

To assess mammalian cell cytotoxicity, BALB/3T3 clone A31 mouse fibroblasts (ATCC CCL-163; American Type Culture Collection, Manassas, VA, USA, RRID:CVCL_0184) were used at post-thaw passages 3–18. Cells were maintained in Dulbecco’s modified Eagle medium (DMEM; Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum (Thermo Fisher Scientific) at 37 °C in a humidified 5% CO₂ atmosphere and were subcultured at least twice before experimental use. Cells were seeded in 96-well plates and exposed for 24 h to serial dilutions of CSEO (0.01–10 mg/mL) or CHX (0.001–1.0 mg/mL). CSEO test solutions were prepared in complete culture medium containing DMSO, whereas CHX was diluted in water. Each concentration and control condition was tested in triplicate in each of three independent experiments. After exposure, cells were stained with Hoechst 33,342 and SYTOX Red and imaged using an InCell 2200 high-content imaging system (GE HealthCare, USA). Hoechst-positive, SYTOX-Red-negative nuclei were quantified as viable cells using CellProfiler [version 4.2.5, RRID:SCR_007358]. CSEO cell counts were normalized to the DMSO vehicle control, whereas CHX cell counts were normalized to untreated cultures. The corresponding control mean was defined as 100%. Concentration-response curves were fitted in GraphPad Prism [version 10.6.0 (796)] using a four-parameter variable slope dose-response model. Model-derived cytotoxicity IC50 values and 95% profile-likelihood confidence intervals were extracted from the fitted relative cell-count curves.

Selection of intervention concentration

The preclinical dataset was used to guide selection of the CSEO concentration for the clinical intervention. Concentration choice was informed by three considerations: demonstrated anti-P. gingivalis activity, the comparative fibroblast dose-response profile from a conservative 24 h exposure assay, and practical suitability for short contact oral rinse exposure. Because the fibroblast assay involved continuous 24 h exposure, its cytotoxicity profile was interpreted conservatively relative to the much shorter mucosal contact expected during mouth rinsing. Based on the combined antibacterial and fibroblast dose-response data, 4 mg/mL CSEO was selected for the intervention phase. The comparator arm used standard 0.12% CHX.

Clinical intervention study

Study design, ethics approval, and trial registration

This study was a randomized, open label, three-arm, parallel-group clinical intervention study with paired saliva sampling at baseline and after 14 days of assigned mouthwash use. Ethical approval was granted by the RSU Ethics Committee on 15 April 2024 (approval No. 2-PĒK-4/356/2024), and all participants provided written informed consent before participation. The trial was retrospectively registered in the ISRCTN registry (ISRCTN79156900) on 11 May 2026, after completion of recruitment and follow-up.

The registered EOsPerio protocol included a three-arm 14-day mouthwash intervention with paired saliva collection for shotgun metagenomic profiling, salivary inflammatory marker profiling from the same samples, and self-reported rinse-use, adherence, and tolerability assessment. The registered primary outcome was salivary microbiome change measured by shotgun metagenomics at baseline and week 2. The present manuscript focuses on this completed short-term salivary microbiome component and includes baseline clinical characterization and self-reported rinse use, adherence, and tolerability data as contextual safety and acceptability information. Salivary inflammatory marker results relate to host-response outcomes and, although of interest, fall outside the scope of the present microbiome-focused manuscript.

Setting and recruitment

Participants were recruited from the RSU Institute of Stomatology between August 2024 and August 2025. Potential participants were identified from the institute’s register of patients with previously diagnosed periodontitis and were contacted by telephone to assess interest and preliminary eligibility. Screening focused on confirmation of periodontitis, willingness to participate, absence of recent dental or periodontal intervention, and absence of recent use of antibiotics or other products likely to influence the oral microbiome, specified below. Recruitment progressed more slowly than expected because the number of eligible patients was lower than anticipated, and the recruitment period concluded before the planned sample size of 60 participants could be achieved.

Patients or members of the public were not formally involved in the design, conduct oversight, outcome selection, or reporting strategy of this study. Participant involvement was limited to recruitment, informed consent, provision of saliva samples, and completion of post-intervention questionnaires on rinse-use experience, tolerability, and adverse effects.

Periodontitis diagnosis and baseline assessment

Periodontal diagnosis and stage/grade assignment were performed by clinicians at the RSU Institute of Stomatology according to the 2017 World Workshop classification [2]. Full-mouth periodontal assessment included periodontal charting and radiographic evaluation, with recording of periodontal pocket depth (PPD), clinical attachment loss (CAL), bleeding on probing (BoP), radiographic bone loss, and tooth loss attributable to periodontitis. Stage was determined from severity and complexity, considering CAL and radiographic bone loss, periodontitis-related tooth loss, and clinically relevant complexity features. Grade was assigned from direct evidence of progression when prior records were available or, otherwise, from indirect evidence using the established grading framework [2].

Eligibility criteria

Eligible participants were adults aged 18–60 years with Stage III or Stage IV periodontitis, able to understand and communicate with the clinician, and willing to provide written informed consent. Exclusion criteria included use of systemic antibiotics, probiotic lozenges, or antimicrobial oral products within the previous month, dental intervention or periodontal therapy within the previous month, pregnancy or breastfeeding, uncontrolled chronic systemic disease, active cancer or recent cancer treatment, and any other medical condition judged likely to substantially influence oral microbial composition or safe study participation. Smoking status was recorded but was not used as an exclusion criterion.

Sample size determination

At the protocol stage, the broader three-arm study was planned to recruit 60 participants (20 per arm). This target was informed by prospective microbiome-oriented sample-size planning using the power-analysis approach described by Mattiello et al., based on operational taxonomic unit frequency distributions with α = 0.05 and β = 0.20, and was adjusted for anticipated attrition [16]. The present manuscript reports the completed short-term paired saliva component only and therefore analyses the available complete paired microbiome dataset. No post hoc power calculation was performed [17].

Randomization, allocation procedure, and masking

A simple computer-generated random allocation sequence without stratification or blocking was generated before enrolment by a designated member of the study team. The sequence was used to prepare study bottles in advance, and allocation was implemented sequentially after baseline saliva collection by assigning each participant the next pre-prepared bottle in the sequence. Eligibility assessment and baseline sampling were completed before intervention assignment. Access to the allocation sequence at the time of assignment was restricted to personnel responsible for bottle preparation and intervention dispensing. The trial was classified as open label overall. CSEO and H₂O were dispensed in coded study bottles. In contrast, CHX was supplied in pharmacy-labelled bottles and had a recognizable organoleptic profile. Because the study products differed in taste and odour, full participant blinding was not feasible in this pragmatic short-term intervention. Sequencing files were processed using coded sample identifiers, and treatment group information was linked only at the statistical analysis stage.

Intervention products and administration

The CHX arm received an extemporaneously compounded 0.12% chlorhexidine mouthrinse prepared by a community pharmacy (Dzirciema Aptieka, Riga, Latvia). The CSEO arm received a rinse prepared from the same seed-derived, steam-distilled Coriandrum sativum essential oil batch used in the preclinical assays (batch 394-M05-CorEO-2022, supplied by the Institute of Environmental Solutions, Latvia). Full batch-specific GC–MS characterization of this oil is provided in Online Resource 1 (Table S1). For the clinical intervention, CSEO was diluted to 4 mg/mL in commercially available drinking water. No emulsifier was added, and participants were instructed to shake the bottle vigorously before each use. The control arm received commercially available drinking water. Participants were instructed to use 20 mL of the assigned rinse for 30 s twice daily, in the morning and evening after toothbrushing, for 14 consecutive days. Habitual oral hygiene practices, including the toothpaste used, were not standardized. Instead, participants were instructed to maintain their usual oral hygiene routine, as well as their usual diet and lifestyle during the intervention period, so that the assigned mouthwash represented the main protocol-directed change in daily oral care exposure.

Adherence and safety monitoring

At the week-2 visit, before follow-up saliva collection, participants were asked whether they had used the assigned rinse as instructed, twice daily for 14 days. Only participants who confirmed compliance with the protocol proceeded to follow-up saliva collection. Participants were then asked to complete the structured week-2 questionnaire. The questionnaire assessed rinse-use experience, tolerability, and self-reported adverse effects, including taste acceptability, perceived oral effects, ease of integration into the daily routine, willingness to use the product in the future, and adverse effects.

Saliva collection and sample handling

Unstimulated whole saliva was collected by passive drool method at baseline and after 14 days of intervention use. Both sampling visits were scheduled in the morning after an overnight fast, before breakfast and before toothbrushing, mouthwash use, or other oral hygiene procedures. At least 2 mL of saliva was collected into pre-chilled 15 mL Falcon tubes over approximately 5–10 min. Samples were placed on ice immediately after collection and transferred to − 80 °C storage within a few hours for downstream DNA extraction and sequencing.

DNA extraction, library preparation, and shotgun sequencing

Total genomic DNA was extracted from saliva using mechanical disruption in Lysing Matrix E tubes (MP Biomedicals, 2 mL tubes containing 1.4 mm ceramic spheres, 0.1 mm silica spheres, and one 4 mm glass bead) followed by phenol–chloroform extraction. Saliva aliquots were mixed with cell suspension solution (25 mM EDTA, 200 mM NaCl, pH 8.0), incubated at 80 °C for 20 min, and mechanically homogenized using an MP Biomedicals bead homogenizer for 40 s at a speed setting of 6.0. Samples were then lysed with SDS and Proteinase K for 1 h at 50 °C. After sequential extraction with equal volumes of buffered phenol and chloroform, phases were separated by centrifugation at 1,500 × g for 10 min at room temperature. DNA was precipitated with an equal volume of isopropanol, centrifuged at 13,400 × g for 15 min at 4 °C, washed with ice-cold 70% ethanol, air-dried, and resuspended in TE buffer. DNA concentration and spectrophotometric purity were assessed by NanoDrop spectrophotometry prior to library preparation. Sequencing libraries were prepared using the QIAGEN FX library preparation kit according to the manufacturer’s instructions. Libraries were sequenced on the Illumina NovaSeq 6000 platform using an SP500 flow cell in paired-end mode with a read length of 2 × 250 bp.

Bioinformatic processing and taxonomic profiling

Raw paired-end sequencing reads were initially assessed using FastQC. Adapter and quality trimming were performed with Cutadapt v4.7 [RRID:SCR_011841] using unconditional removal of 3 bases from the read extremities (--cut 3), trimming of terminal ambiguous bases (--trim-n), a quality-score threshold of 20 (-q 20), exclusion of reads containing more than 3 ambiguous bases (--max-n 3), and removal of reads shorter than 50 bp after trimming (-m 50). Quality-controlled reads were taxonomically classified with Kraken2 v2.1.3 [RRID:SCR_026838] in paired-end mode using default confidence settings against the Standard database containing UniVec, archaeal, bacterial, fungal, human, plant, plasmid, protozoan, and viral sequences (database build: 8 April 2025). Feature-level relative abundance estimates were re-estimated with Bracken v2.9 using an expected read length of 100 bp. For downstream microbiome analyses, abundance matrices were curated to remove host-associated and non-target eukaryotic clades, and final feature tables were analysed at the phylum, genus, and species levels.

Outcome definitions

The primary outcome was the between-group difference in within-subject species-level Bray–Curtis change magnitude from baseline to week 2. Secondary outcomes were changes in salivary alpha diversity, between-group differences in beta-diversity change magnitude, and taxon-level differential abundance across the intervention arms. Exploratory outcomes included guild-level analyses and within-arm taxonomic shifts from baseline to week 2.

Statistical analysis

All analyses were performed in R [version 4.5.2] using a complete-case paired design restricted to participants with both baseline and week-2 saliva samples. Analysis scripts were developed with the assistance of OpenAI Codex Desktop [version 26.324.21641, model GPT-5.4], under the supervision of a bioinformatician, and are available in an open-access GitHub repository. Alpha diversity was assessed at the phylum, genus, and species levels using Shannon and Simpson indices, with observed richness used descriptively. Simpson values were logit-transformed for mixed-model analyses. Longitudinal alpha-diversity analyses were performed using linear mixed-effects models with fixed effects for Group, Time, and Group x Time, adjustment for log-transformed total sequencing depth, and a random intercept for SubjectID. Within-group baseline-to-week-2 contrasts and H₂O-referenced difference-in-differences contrasts were obtained from estimated marginal means, with Benjamini–Hochberg correction applied within each family of tests.

For beta-diversity analyses, feature tables were filtered before Bray-Curtis computation using a minimum count threshold of 10 and a minimum prevalence of 10% across samples. The lowest 10% of features by interquartile-range variability were additionally removed. Bray-Curtis distances were then calculated on total-sum-scaled relative abundance tables. The primary beta-diversity endpoint was the within-subject Bray-Curtis distance between baseline and week 2 at the species level. Between-group differences in this change magnitude were assessed using Kruskal–Wallis tests followed by pairwise Wilcoxon tests against the H₂O group with Benjamini-Hochberg correction. Principal coordinates analysis, PERMANOVA, and tests of multivariate dispersion were used as supportive community-level analyses. Baseline differences in Bray-Curtis community composition among the H₂O, CSEO, and CHX groups were formally assessed at phylum, genus, and species levels using baseline-only PERMANOVA, with corresponding tests of multivariate dispersion. Within-arm time effects were assessed using PERMANOVA stratified by SubjectID.

For taxon-level differential abundance, rank-specific filtering thresholds were applied before modelling: prevalence ≥ 10% and mean relative abundance ≥ 0.1% at phylum level, prevalence ≥ 15% and mean relative abundance ≥ 0.05% at genus level, and prevalence ≥ 20% and mean relative abundance ≥ 0.01% at species level. A data-driven pseudocount was added, abundances were transformed using centred log-ratio methodology, and taxon-wise linear mixed-effects models were fitted with fixed effects for Group, Time, and Group x Time and a random intercept for SubjectID. Primary contrasts were H₂O-referenced difference-in-differences comparisons for CSEO versus H₂O and CHX versus H₂O, while CHX versus CSEO was treated as exploratory.

For the guild analysis, species-to-guild annotation was based on a literature-informed clinical-relevance framework adapted from Min et al., with categories restructured for the present study to emphasize red-complex taxa, periodontal-inflammation-associated taxa, malodorous taxa, acidogenic taxa, nitrate reducers, and other commensals. Because some species are relevant to more than one oral-health category, guild membership in the primary abundance analysis was non-exclusive [9]. Primary guild analyses used baseline-adjusted linear models of week-2 log-transformed guild abundance, with contrasts versus H₂O and Benjamini-Hochberg correction. Within-group paired change tests and delta-based sensitivity models were treated as secondary or exploratory analyses.

Baseline was handled differently across analyses because each microbiome outcome required a statistical approach suited to how that outcome was defined and compared. For alpha diversity and taxon-level differential abundance, baseline was incorporated through paired longitudinal Group × Time mixed-effects models, with H₂O-referenced difference-in-differences contrasts estimating treatment-specific change from baseline to week 2. For beta diversity, the primary endpoint was the within-subject Bray–Curtis distance between baseline and week 2, so baseline was incorporated in the change metric itself. For guild-level summaries, each guild yielded scalar baseline and week-2 abundance values. Week-2 guild abundance was therefore analysed using baseline-adjusted linear models. Diversity and distance metrics were calculated from the observed microbiome profiles. Baseline was then accounted for in the statistical analysis appropriate to each outcome, rather than by modifying the abundance matrices before metric calculation.

Baseline balance across the randomized arms was summarized using pairwise standardized mean differences (SMDs) for CSEO versus H₂O, CHX versus H₂O, and CSEO versus CHX. Continuous SMDs were calculated from group means and standard deviations, and binary SMDs from group proportions. SMDs were interpreted descriptively rather than as hypothesis tests. All nine baseline characteristics were examined individually in the sensitivity analyses, regardless of SMD magnitude.

To assess whether numerical baseline imbalances could influence the main diversity findings, exploratory single-covariate sensitivity analyses were performed. For the primary species-level Bray-Curtis change endpoint, linear models included treatment group and one baseline covariate at a time. For alpha diversity, longitudinal mixed-effects models included Group, Time, Group × Time, log-transformed sequencing depth, a random subject intercept, and one baseline covariate at a time. Covariates were age, sex, current nicotine use, periodontitis stage, periodontitis grade, BoP%, mean PPD, mean CAL, and plaque index. These analyses were supportive and were used to assess robustness of the main findings.

Sequencing QC fractions, including microbial read fraction and host-to-microbial read ratio, were summarized by group and timepoint. Exploratory Kruskal-Wallis tests were used to compare microbial read fraction and log10-transformed host-to-microbial read ratio across treatment groups at each timepoint and to compare baseline-to-week-2 changes across groups.

The original study protocol submitted to the RSU Ethics Committee is available from the corresponding author upon reasonable request. A separate formal statistical analysis plan was not prospectively archived. The statistical analysis approach used for the present manuscript is fully described in this section, and the corresponding analysis code is publicly available via the study GitHub repository.

Results

Preclinical antibacterial activity and cytotoxicity

Before the clinical intervention, in vitro testing showed concentration-dependent inhibition of Porphyromonas gingivalis by both CSEO and CHX, with CHX active at substantially lower concentrations. Fitted antibacterial dose-response models yielded relative IC50 values of 1.09 mg/mL for CSEO (95% CI 0.54–3.02) and 1.76 × 10⁻⁴ mg/mL for CHX (95% CI 1.61 × 10⁻⁴–2.07 × 10⁻⁴), confirming the markedly greater in vitro potency of CHX against P. gingivalis (Fig. 1A, C). In the 24 h BALB/3T3 fibroblast cell-count assay, model-derived cytotoxicity IC50 values were 0.351 mg/mL for CSEO (95% CI 0.310–0.417) and 0.0292 mg/mL for CHX (95% CI 0.0237–0.0357), indicating a higher fibroblast IC50 for CSEO under these assay conditions (Fig. 1B, C). These preclinical concentration-response profiles informed the choice of 4 mg/mL CSEO for the short contact rinse, while the 24 h fibroblast assay was interpreted conservatively because it does not directly simulate the 30 s oral rinse exposure. Standard 0.12% CHX was used as the comparator.

Fig. 1.

Preclinical in vitro dose-response analysis of CSEO and CHX against Porphyromonas gingivalis and in BALB/3T3 fibroblasts. Before the clinical trial, in vitro assays were performed to identify concentrations with antimicrobial activity and to assess mammalian cell cytotoxicity. (A) Growth of P. gingivalis after exposure to CSEO or CHX, measured by optical density (OD) and normalized to the maximum background-corrected mean OD observed in the assay. Lower values indicate stronger growth inhibition. (B) BALB/3T3 fibroblast viable cell count after 24 h exposure to CSEO or CHX. CSEO values were normalized to the DMSO vehicle control, whereas CHX values were normalized to untreated cultures. For both curves, the corresponding control mean was defined as 100%. The horizontal line marks 50% of the corresponding control. Points with error bars represent experimental measurements, and solid lines represent fitted dose-response curves. (C) Model-derived IC50 estimates with 95% profile-likelihood confidence intervals. *For the antibacterial assay, IC50 values are reported as relative IC50 estimates, defined as the concentration producing a fitted response halfway between the fitted top and bottom plateaus

Clinical study: participant flow and analysed cohort

Participant flow is summarized in Fig. 2. Recruitment proceeded more slowly than anticipated, and the available recruitment period ended after 38 participants had been randomized, below the planned target of 60. Of the 38 randomized participants, 27 contributed paired baseline and week-2 saliva samples eligible for microbiome analysis. The final complete-case analytical dataset comprised 27 participants and 54 saliva libraries, including 9 H₂O, 8 CSEO, and 10 CHX participants with paired samples. All downstream longitudinal analyses were restricted to this paired dataset.

Fig. 2.

CONSORT-style flow diagram for the paired salivary microbiome analysis cohort. The diagram summarizes randomized allocation, availability of paired baseline and week-2 saliva samples, and the final analytic set used for the primary microbiome outcome in the present manuscript. The planned recruitment target was 60 participants. Recruitment proceeded more slowly than anticipated and ended after 38 participants had been randomized. The implemented study component reported here consisted of baseline saliva sampling, a 2-week assigned mouthwash intervention, and post-intervention saliva sampling. The final paired analytic cohort comprised 27 participants (H₂O n = 9, CSEO n = 8, CHX n = 10)

Baseline demographic and periodontal characteristics, together with pairwise standardized mean differences (SMDs), are summarized in Table 1.

Table 1.

Baseline demographic and periodontal characteristics of the paired microbiome analysis cohort, with pairwise standardized mean differences (SMDs)

Characteristics	H2O (n = 9)	CSEO (n = 8)	CHX (n = 10)	SMD, CSEO vs. H2O	SMD, CHX vs. H2O	SMD, CSEO vs. CHX
Number of participants included in microbiome analysis	9	8	10	NA	NA	NA
Mean age, years (SD, range)	45.8 (6.1, 39–55)	46.9 (8.8, 32–59)	47.6 (9.0, 35–59)	0.145	0.236	−0.081
Sex, n (%)1
Female	6 (67%)	6 (75%)	5 (50%)	—	—	—
Male	3 (33%)	2 (25%)	5 (50%)	−0.184	0.343	−0.535
Smoking/nicotine status, n (%)2
Non-smoker	5 (56%)	7 (88%)	7 (70%)	—	—	—
Cigarette smoker	2 (22%)	1 (12%)	3 (30%)	—	—	—
Electronic cigarette user	2 (22%)	0 (0%)	0 (0%)	—	—	—
Current nicotine use	4 (44%)	1 (12%)	3 (30%)	−0.757	−0.302	−0.438
Periodontitis stage, n (%)
Stage III	7 (78%)	6 (75%)	8 (80%)	—	—	—
Stage IV	2 (22%)	2 (25%)	2 (20%)	0.065	−0.054	0.120
Periodontitis grade, n (%)
Grade B	1 (11%)	2 (25%)	2 (20%)	—	—	—
Grade C	8 (89%)	6 (75%)	8 (80%)	−0.367	−0.247	−0.120
Bleeding on probing, %, mean (SD)	61.7 (33.0)	70.5 (26.8)	64.9 (25.6)	0.294	0.110	0.214
Mean PPD, mm, mean (SD)	3.7 (0.7)	3.9 (0.7)	3.6 (0.7)	0.189	−0.161	0.338
Mean CAL, mm, mean (SD)3	4.3 (0.9)	4.5 (1.2)	4.3 (1.2)	0.172	0.007	0.142
Plaque index, %, mean (SD)	69.6 (37.1)	81.9 (27.3)	73.0 (24.5)	0.378	0.110	0.342

¹Percentages use the number of analysed paired participants in each group as the denominator. ²Current nicotine use includes cigarette smoking and electronic cigarette use and is therefore a derived summary category. ³CAL is summarized as absolute clinical attachment loss in millimetres. PPD, probing pocket depth; CAL, clinical attachment loss; BoP, bleeding on probing. SMDs compare the first-named group with the second-named group. Positive values indicate a higher mean or prevalence in the first-named group, whereas negative values indicate a lower mean or prevalence. For binary characteristics, one category was used to report each SMD: male sex, current nicotine use, Stage IV, and Grade C. Complementary categories are marked with a dash because their SMDs have the same absolute magnitude but opposite signs. For smoking-related SMDs, current nicotine use served as the binary measure. The individual smoking status categories are therefore also marked with a dash. NA indicates that an SMD was not applicable. SMDs are descriptive balance measures and are not hypothesis tests

The largest absolute pairwise SMD was observed for current nicotine use in CSEO versus H₂O (SMD = − 0.757), indicating lower prevalence in CSEO. Other comparatively large differences included lower male prevalence in CSEO than CHX (SMD = − 0.535), lower current nicotine use in CSEO than CHX (SMD = − 0.438), and higher plaque index in CSEO than H₂O (SMD = 0.378). All nine baseline characteristics shown in Table 1 were examined individually in the exploratory single covariate sensitivity analyses.

Sequencing summary and overall taxonomic composition

The final analysed microbiome dataset comprised 54 saliva libraries, corresponding to paired baseline and week 2 samples from 27 participants. Sequencing depth ranged from 7.85 to 58.41 million reads per sample, with a median of 19.86 million reads, corresponding to 1.24 × 10⁹ total reads across all libraries. Kraken2 classified a median of 97.1% of reads per sample (range 87.7–99.4%), leaving a median 2.9% unclassified fraction (range 0.60–12.3%). Reads assigned to Chordata, representing predominantly host-derived signal, accounted for a median of 85.3% of reads (range 33.4–96.7%), whereas microbial reads accounted for a median of 11.6% (range 2.7–63.6%). Low-abundance fungal and protozoan reads were consistently detected, with median fractions of 0.0156% and 0.136%, respectively, while viral reads remained below the reporting threshold used for the Pavian summary. Overall, classification performance was high, and microbial signal was sufficient for downstream taxonomic profiling.

Because the fraction of microbial versus host-derived reads varied across samples, we examined whether low microbial signal was concentrated in a specific treatment arm or timepoint. Median microbial read fractions at baseline/week 2 were 8.57%/13.3% for H₂O, 9.64%/12.36% for CSEO, and 11.0%/12.1% for CHX. Corresponding median host-to-microbial read ratios were 10.36/6.23, 9.61/7.18, and 7.84/6.99, respectively. Exploratory comparisons did not detect significant between-group differences in microbial fraction at baseline or week 2 (p = 0.821 and p = 0.747), nor in log10-transformed host-to-microbial ratio (p = 0.858 and p = 0.799). Baseline-to-week-2 changes in microbial fraction also did not differ significantly across groups (p = 0.333). The lowest microbial fraction was observed in one CHX week-2 sample (2.7%), but this sample still yielded approximately 0.62 million microbial reads, and low-microbial-fraction samples were not concentrated in one treatment group.

Across all groups and timepoints, the salivary microbiome was dominated by a relatively limited set of recurrent oral taxa (Fig. 3). At phylum level, Bacillota, Bacteroidota, Pseudomonadota, and Actinomycetota accounted for most of the community signal. At genus level, the most abundant taxa included Prevotella, Streptococcus, Veillonella, Neisseria, Haemophilus, Rothia, and Schaalia. At species level, dominant taxa included Haemophilus parainfluenzae, Prevotella melaninogenica, Rothia mucilaginosa, Veillonella atypica, Prevotella jejuni, Schaalia odontolytica, and Neisseria subflava. Baseline-only Bray-Curtis PERMANOVA did not detect statistically significant differences in salivary community composition among treatment arms at phylum (R² = 0.109, p = 0.202), genus (R² = 0.106, p = 0.178), or species (R² = 0.098, p = 0.170) levels. Tests of multivariate dispersion were also non-significant at phylum (p = 0.332), genus (p = 0.552), and species (p = 0.481) levels. Per-sample composition plots nevertheless illustrated substantial inter-individual heterogeneity (Online Resource 2, Fig. S1; Online Resource 3).

Fig. 3.

Group- and time-stratified salivary microbiome composition across taxonomic levels. Stacked bar plots show mean relative abundance of the most abundant taxa at the phylum (A), genus (B), and species (C) levels in saliva collected at baseline and week 2 from the H₂O control, Coriandrum sativum essential oil (CSEO), and chlorhexidine (CHX) experimental treatment groups. Taxa were ranked by overall mean relative abundance across the dataset; the top 15 species, top 12 genera, and top 8 phyla are shown, with all remaining taxa grouped as Others. Across taxonomic levels, mean community structure was dominated by a limited set of recurrent oral taxa

The clearest visual compositional reweighting was observed in the CHX arm at week 2, particularly at genus and species levels, with relative depletion of Prevotella-rich signal and enrichment of Haemophilus-, Neisseria-, and Capnocytophaga-associated fractions. By contrast, the CSEO arm remained visually closer to the water control.

Alpha diversity

Control-referenced longitudinal alpha-diversity analysis showed the clearest signal at species level in the CHX arm (Fig. 4). Relative to H₂O, CHX showed lower species-level Simpson diversity (q = 0.014), while the species-level Shannon contrast was borderline after FDR correction (q = 0.0503). Phylum- and genus-level CHX contrasts did not reach FDR significance (phylum Shannon/Simpson q = 0.512/0.168; genus Shannon/Simpson q = 0.699/0.902). CSEO did not differ from H₂O for either Shannon or Simpson diversity at any taxonomic resolution (all q ≥ 0.259). Overall, alpha-diversity results support a CHX-associated reduction most clearly at species-level Simpson diversity, whereas the CSEO–H₂O contrasts were non-significant. Complete alpha-diversity outputs are provided in Online Resource 3.

Fig. 4.

Control-referenced longitudinal shifts in salivary alpha diversity across taxonomic levels. Week 2 - baseline change scores for Shannon and Simpson alpha diversity calculated from phylum-, genus-, and species-level salivary microbiome profiles. Groups are H₂O control, Coriandrum sativum essential oil (CSEO), and chlorhexidine (CHX). Each point represents data obtained from one participant with paired samples. Box plots show median and interquartile range (IQR), whiskers indicate 1.5×IQR, and black diamonds denote group means. Positive values indicate increased alpha diversity at week 2, and negative values indicate reduced diversity. Labels above CSEO and CHX show Benjamini–Hochberg FDR-adjusted q values for H₂O-referenced difference-in-differences contrasts from longitudinal mixed-effects models accounting for repeated measures and sequencing depth. For Simpson diversity, q values were obtained from models fitted on the logit-transformed Simpson index. Plotted change scores remain on the original Simpson scale

Beta diversity and longitudinal change magnitude

The prespecified primary community-level endpoint was the within-subject species-level Bray-Curtis distance between baseline and week 2. Across taxonomic resolutions, CHX produced the largest longitudinal compositional shift (Fig. 5). Compared with H₂O, CHX showed significantly greater within-subject Bray-Curtis change at the species level (q = 0.020; Kruskal-Wallis p = 0.015) and genus level (q = 0.026, Kruskal-Wallis p = 0.024), whereas the phylum-level difference was not significant. CSEO did not differ from H₂O at any taxonomic level. These paired change analyses therefore identify CHX as the intervention associated with the strongest short-term restructuring of salivary microbial community composition.

Fig. 5.

Longitudinal changes in salivary beta diversity across taxonomic levels. Left column (panels A, C, E): within-subject Bray-Curtis distances between baseline and week 2 for phylum-, genus-, and species-level salivary microbiome profiles. Each point represents one participant with paired samples. Boxplots show the median and interquartile range (IQR), whiskers indicate 1.5 x IQR, and black diamonds denote group means. Kruskal–Wallis p values are shown within each panel. Labels above CSEO and CHX indicate Benjamini–Hochberg FDR-adjusted q values for pairwise comparisons versus the H₂O control. CHX showed greater longitudinal compositional change than H₂O at the species and genus levels (E and C, q = 0.020 and q = 0.026, respectively), whereas CSEO did not differ from H₂O and phylum-level differences were not significant (A). Right column (panels B, D, and F): Bray–Curtis PCoA plots stratified by timepoint for phylum, genus, and species levels, respectively. Data points are samples coloured by treatment group, and ellipses summarize within-group spread. Insets report PERMANOVA p values and p values from tests of homogeneity of multivariate dispersion. Baseline-only PERMANOVA did not detect statistically significant between-group differences at any taxonomic level, whereas week-2 PERMANOVA indicated significant group differences at phylum (p = 0.0239), genus (p = 0.0026), and species (p = 0.0013) levels, with significant dispersion heterogeneity at all three levels

Supportive PCoA analyses were consistent with the formal baseline PERMANOVA results reported above. By week 2, PERMANOVA indicated significant group differences at phylum (p = 0.0239), genus (p = 0.0026), and species (p = 0.0013) levels. These week-2 ordinations were also accompanied by significant heterogeneity of multivariate dispersion at all three taxonomic levels. Accordingly, the paired Bray-Curtis change analysis provides the most robust evidence for treatment-associated microbiome shifts, whereas the week-2 ordination separation should be interpreted as supportive rather than standalone evidence. Complete diversity results, including baseline-covariate sensitivity analyses, are provided in Online Resource 3.

Exploratory baseline-covariate sensitivity analyses supported the main diversity interpretation. Across models adjusted separately for age, sex, current nicotine use, periodontitis stage, periodontitis grade, BoP%, mean PPD, mean CAL, or plaque index, the primary species-level Bray-Curtis CSEO–H₂O contrast remained small and non-significant (estimates − 0.025 to -0.012; q = 0.682–0.856), whereas the CHX–H₂O contrast remained positive and significant (estimates 0.183–0.202; q = 0.004–0.012). Alpha diversity sensitivity analyses were directionally consistent with the primary longitudinal models, with CSEO remaining similar to H₂O and the most stable CHX-associated lower diversity signal observed for species-level Simpson diversity (Online Resource 3).

Differential abundance

Consistent with beta diversity analyses, taxon-level differential abundance modelling showed the strongest signal in the CHX-versus-H₂O contrast (Fig. 6). No significant or suggestive differential changes were detected for CSEO versus H₂O at phylum, genus, or species level. Likewise, no phylum-level feature met the significance threshold in the CHX contrast.

Fig. 6.

Chlorhexidine-associated genus- and species-level differential abundance relative to control. For each taxon, differential abundance was estimated using linear mixed-effects models of centred log-ratio transformed relative abundance, with fixed effects for Group, Time, and Group × Time and a random intercept for SubjectID. Plotted values represent CHX-versus-H₂O difference-in-differences in CLR units. Points denote effect estimates and horizontal lines 95% confidence intervals. Filled circles indicate BH-FDR q < 0.05 and open circles 0.05 ≤ q < 0.10, with q-values shown at right. Positive values indicate taxa that increased more (or decreased less) under CHX than under H₂O, whereas negative values indicate taxa that decreased more (or increased less) under CHX than under H₂O. The species panel shows the 20 significant species with the lowest q values, whereas the genus panel shows six significant and four suggestive genera

At genus level, CHX was associated with a significant increase in Capnocytophaga and significant decreases in Prevotella, Lachnoanaerobaculum, Pauljensenia, Schaalia, and Arachnia relative to H₂O. Additional decreases in Lancefieldella, Actinomyces, and Mogibacterium were borderline after BH-FDR correction (each q ≈ 0.050), while Candidatus Minimicrobia was suggestive only.

At species level, the dominant CHX-associated signal was depletion of multiple Prevotella taxa, including P. melaninogenica, P. scopos, P. jejuni, P. veroralis, P. fusca, and P. corporis, together with reductions in Peptostreptococcus equinus, Lachnoanaerobaculum gingivalis, Lachnoanaerobaculum umeaense, Actinomyces pacaensis, Pauljensenia hongkongensis, Schaalia odontolytica, Schaalia turicensis, Schaalia meyeri, Rothia dentocariosa, Corynebacterium durum, and several Streptococcus taxa. In the full primary results (Online Resource 4), CHX was also associated with enrichment of Capnocytophaga leadbetteri, Capnocytophaga sputigena, and Capnocytophaga sp. oral taxon 878. These species are not shown in Fig. 6 because the species panel was restricted to the top 20 displayed hits. Capnocytophaga endodontalis and Capnocytophaga haemolytica showed suggestive increases. Taken together, these findings indicate lineage-specific CHX-associated reshaping of the salivary microbiome rather than a uniform community-wide depletion.

Within-arm contrasts were directionally concordant with the primary between-group analyses and showed that the most extensive baseline-to-week-2 taxonomic changes occurred in the CHX arm, whereas H₂O and CSEO showed fewer and generally weaker within-arm shifts (Fig. S2–S4). Full differential abundance results are provided in Online Resource 4.

Guild-level analysis

Guild-level analyses were treated as exploratory. In the primary baseline-adjusted week-2 comparisons versus H₂O, no guild reached the prespecified BH-FDR significance threshold (Fig. 7). The strongest directional signal was a lower abundance of the Acidogenic guild in the CHX arm relative to H₂O (estimate ratio 0.19, approximately 81% lower, q = 0.064). A weaker downward trend was observed for the Malodorous guild under CHX (estimate ratio 0.52, approximately 48% lower, q = 0.440), whereas Periodontal inflammation and Nitrate reducers remained comparatively stable across groups. Apparent increases in the Red complex under both intervention arms were imprecise and not statistically significant. Other commensals also showed no statistically significant baseline-adjusted treatment effect, despite a numerical decrease in CSEO versus H₂O (approximately 31% lower, q = 0.236).

Fig. 7.

Exploratory guild-level relative abundance profiles and baseline-adjusted treatment comparisons across the intervention arms. Bars show mean guild relative abundance at baseline and week 2. Error bars show 95% confidence intervals. The y-axis shows guild relative abundance (%), calculated by summing the relative abundances of all species assigned to each guild. In the primary guild analysis, membership was non-exclusive. Species annotated to more than one guild contributed to each relevant guild total. Other commensals comprised commensal taxa not simultaneously annotated as periodontal-inflammation, malodorous, acidogenic, or nitrate-reducing taxa. Panel annotations report baseline-adjusted week 2 contrasts versus H₂O, expressed as percent differences derived from estimate ratios, together with BH-FDR-adjusted q values from ANCOVA models fitted to log10-transformed guild abundances

The descriptive guild heatmap (Online Resource 2, Fig. S5) showed that the most abundant mapped species were concentrated mainly in the Other commensals, Nitrate reducers, and Acidogenic categories, whereas Red complex, Periodontal inflammation, and Malodorous taxa were generally represented by lower-abundance species. For visualization, each species was displayed only once according to a predefined precedence rule (Red complex > Periodontal inflammation > Malodorous > Acidogenic > Nitrate reducers > Other commensals), whereas the primary guild analysis used non-exclusive guild membership. Thus, multiply annotated species contributed to more than one guild total in the boxplot analysis but appeared only once in the heatmap. Guild-based findings should therefore be interpreted as descriptive and hypothesis-generating rather than as definitive functional readouts. Full guild-level analysis results are provided in Online Resource 5.

Safety, adherence, and tolerability

All participants included in the paired microbiome analysis verbally confirmed protocol-compliant rinse use before week-2 saliva collection. This verbal confirmation was used as the adherence indicator. The subsequent structured questionnaire assessed rinse-use experience, tolerability, and adverse effects, but did not include a separate daily use diary or objective dose-tracking measure. Questionnaire responses are summarized in Online Resource 6 (Table S2). Questionnaire completion was complete in the CSEO and CHX arms (8/8 and 10/10, respectively). In the H₂O arm, questionnaire data were available for 6/9 participants, so H₂O questionnaire denominators refer to the six available responses. Taste ratings were mixed for both active rinses. In the CHX arm, 5/10 participants rated taste as neutral and 5/10 as unpleasant or very unpleasant. In the CSEO arm, 3/8 rated taste as pleasant or very pleasant, 1/8 as neutral, and 4/8 as unpleasant. No CSEO participant reported side effects, whereas 4/10 CHX participants reported adverse experiences, including taste changes and dryness (2/10), bitterness (1/10), and burning sensation with taste alteration (1/10). No side effects were reported among the available H₂O responses.

Perceived oral benefit was reported in both active-treatment arms. In the CHX arm, 5/10 participants reported slight improvement and 5/10 reported no change in gingival or oral sensation. In the CSEO arm, 4/8 reported improvement (2 slight, 2 marked) and 4/8 reported no change. Integration into daily routine was generally rated as easy across arms. Willingness for future use appeared highest in the CSEO group, where 6/8 participants answered “definitely yes,” compared with 3/10 in the CHX arm. In CHX, 6/10 answered “maybe” and 1/10 “no.” Overall, these descriptive data suggest that CSEO was well tolerated, whereas CHX was associated with more frequent unpleasant sensory effects.

Discussion

In this randomized 2-week intervention study in patients with periodontitis, the clearest microbiome signal was observed in the CHX arm. Compared with water, CHX produced larger within-subject Bray-Curtis shifts, showed lower alpha diversity most clearly for species-level Simpson diversity, and was associated with multiple genus- and species-level differential changes. In contrast, no statistically detectable group-level difference between CSEO and water was observed across the main salivary microbiome endpoints. Therefore, the hypothesis that CSEO would induce a detectable but selective salivary microbiome shift relative to water was not supported over the 14-day intervention. The preclinical assays confirmed dose-dependent inhibition of P. gingivalis by both agents. CHX was markedly more potent in vitro, whereas CSEO showed a higher fibroblast IC50 under the 24 h cytotoxicity assay. These preclinical profiles informed selection of the CSEO rinse concentration, with the fibroblast assay interpreted conservatively because clinical exposure was a short 30 s rinse.

This study has several important limitations. The analysed cohort was small, with 8–10 paired microbiome participants per arm. This was substantially below the planned recruitment target of 20 participants per arm. This limits statistical power, particularly for detecting modest or heterogeneous CSEO-associated effects. Consequently, the null CSEO findings should be interpreted as absence of a statistically detectable short-term salivary microbiome shift under the tested conditions, not as proof of absence of biological or clinical effect. The small group sizes also precluded reliable assessment of treatment-response heterogeneity or identification of responder subgroups. The intervention period was short, and the present report is based on saliva rather than subgingival plaque. The findings therefore primarily reflect whole mouth salivary ecological responses and cannot be assumed to represent periodontal pocket microbiome changes. Clinical periodontal outcomes and functional analyses were not included in this manuscript. Adherence was assessed by verbal self-report rather than objective compliance measures, and questionnaire completion was incomplete in the H₂O arm. Rinse-use and tolerability summaries should therefore be interpreted descriptively. In addition, toothpaste formulation and other home oral hygiene practices were not standardized, which may have contributed to inter-individual microbiome variability. While the paired design partly mitigates stable participant-level differences, unmeasured behavioural changes or between-group imbalance cannot be excluded in this small cohort. Stage IV distribution and mean PPD and CAL values were numerically similar across groups, but imbalances in sex distribution, current nicotine use, and plaque index were present. Exploratory single-covariate sensitivity analyses supported the robustness of the primary Bray-Curtis interpretation, but residual confounding cannot be excluded given the small cohort size. The metagenomic analysis did not incorporate spike-in standards, and host-derived read fractions were high and variable across saliva samples. The resulting abundance profiles are therefore compositional rather than quantitative and should be interpreted with caution, even though exploratory QC summaries did not indicate systematic concentration of low microbial signal in one treatment arm or timepoint. The open label design may have influenced self-reported adherence, tolerability, or participant behaviour during the intervention. Additionally, the guild framework was exploratory and structurally limited by overlapping species annotation. Finally, because the CSEO formulation was batch-specific and chemically characterized as a linalool-dominant coriander essential-oil chemotype, extrapolation to other coriander essential oils should remain cautious.

Nevertheless, several findings are informative. Our CHX findings are broadly consistent with previous work showing that CHX can perturb oral microbial ecology in a fairly wide-ranging way, rather than acting only through selective suppression of pathogens. For instance, as mentioned previously, Bescos et al. demonstrated in healthy volunteers that 7 days of CHX use caused a marked salivary microbiome shift, together with lower salivary pH, increased lactate and glucose, and reduced salivary and plasma nitrite availability [5]. In periodontal surgical patients, Bartsch et al. similarly observed that 4 weeks of 0.2% CHX reduced alpha diversity, increased Bray-Curtis dissimilarity, and favoured streptococci, alongside increases in low-abundance Capnocytophaga and Neisseria [18]. Comparisons across healthy volunteers, periodontal surgical patients, and the non-surgical periodontitis cohort studied here require caution because baseline community structure, inflammation, and ecological stability may differ among these cohorts. We therefore interpret concordance with prior CHX studies as evidence that CHX can exert broad ecological pressure across oral settings, not as evidence that identical taxon-level responses should be expected in periodontitis cohorts.

Within this cautious comparative frame, the increase in Capnocytophaga in our dataset is noteworthy but clinically ambiguous. Capnocytophaga species are common oral taxa and have been linked to plaque accumulation, halitosis and mild periodontal inflammation, although their relationship with periodontitis severity is complex [19]. Human oral Capnocytophaga species can also act as opportunistic pathogens, particularly in immunocompromised hosts [20]. Because our data are compositional and saliva-based, this relative increase cannot be interpreted as absolute proliferation or as direct evidence of worsened periodontal or systemic risk. Rather, it suggests that CHX created a broad ecological redistribution in which some taxa outside the classic red complex increased relatively as other salivary taxa were depleted.

This interpretation is consistent with the broader species-level pattern. The CHX signal was notable both for what it altered and for what it did not alter - specifically, we did not detect a pronounced, statistically clear reduction in the classic red-complex periodontopathogens. Instead, the dominant pattern involved shifts in a broader set of common salivary taxa, including multiple Prevotella species, Lachnoanaerobaculum, Schaalia, Rothia dentocariosa, Corynebacterium durum, several streptococci, and an increase in Capnocytophaga. Within the salivary ecosystem, CHX therefore appeared less as a narrowly targeted anti-periodontopathogen intervention and more as a broad ecological pressure that redistributed community structure. This point is important for interpreting the parallel reduction in diversity: a lower-diversity community after CHX should not automatically be read as a cleaner or more selective removal of disease-associated taxa.

A related interpretive issue is that taxonomic abundance and function do not necessarily move in parallel. Several factors may explain why strong red complex or guild-level signals were not detected. For example, saliva integrates microbes shed from multiple oral niches and is therefore less site-specific than subgingival plaque. Therefore, a short intervention may affect salivary ecology without producing a large, easily detectable signal in low-abundance subgingival pathobionts. The study was also small and short, which reduces sensitivity for rarer taxa. More importantly, prior work shows that antiseptic mouthwash can blunt oral nitrate reduction, lower salivary or plasma nitrite, and modestly increase systolic blood pressure, while a more recent CHX-versus-propolis study showed that CHX impaired nitrite-producing activity even when the relative abundance of nitrite-producing species increased [21]. In other words, functional nitrate reduction can worsen even when taxonomy does not show a simple numerical decline. This distinction is relevant to our guild analysis: a null guild-level result does not rule out biologically meaningful impairment of nitrate-related function. The same logic may apply more broadly to virulence-associated pathways. Our guild analysis should therefore be read as a biologically informed ecological approximation rather than a direct measurement of microbial function, and direct functional assays or metatranscriptomics would be more informative where mechanism is the primary interest.

Unlike CHX, no statistically detectable difference between CSEO and water was observed at the group level for alpha diversity, the magnitude of beta diversity change, or differential abundance. This null finding should be interpreted cautiously because it is consistent with either no effect or an effect too small or heterogeneous to detect in the present cohort. It should not be interpreted as microbiome protection or as beneficial preservation of the baseline community, particularly because the oral microbiome in periodontitis may already be dysbiotic. This finding also differs conceptually from the ecological shift toward a healthier state reported by Min et al. [9] after repeated essential oil mouthwash use, which involved a detectable change in microbial composition rather than an absence of detectable change. The clinical meaning of the present finding therefore remains uncertain. Among the possible explanations, insufficient effective in vivo exposure appears most consistent with the present data. Although the CSEO concentration was selected from in vitro dose-response assays, the clinical formulation was a non-emulsified aqueous rinse used for 30 s, whereas the antibacterial assay involved direct and prolonged exposure under controlled culture conditions. Limited dispersion and retention may therefore have reduced effective exposure below the level needed to produce a detectable compositional shift. Although growing evidence suggests that some EOs may exert anti-biofilm or anti-virulence effects at sub-MIC concentrations [22–24], these mechanisms were not assessed in the present study and remain speculative in relation to CSEO. They therefore cannot be inferred from the null compositional result observed here.

Taken together, the present findings show that CHX exerted a broad ecological pressure on the salivary microbiome, whereas no statistically detectable group-level difference between CSEO and water was observed. The main value of this study is therefore not to claim clinical superiority of either rinse, but to show that microbiome restructuring, tolerability, and in vitro antibacterial activity do not necessarily move in parallel. A rinse may show measurable in vitro activity and still produce only limited community-wide taxonomic change in vivo, while, on the other hand, it may produce a strong compositional signal without acting selectively on classic periodontitis-associated taxa. Larger and longer studies incorporating subgingival sampling, clinical periodontal outcomes, host-response markers, and direct functional readouts are needed to determine whether optimized CSEO formulations can produce clinically meaningful, anti-inflammatory, or functional effects and how these relate to microbiome change.

Conclusions

CSEO demonstrated measurable preclinical anti-P. gingivalis activity. In the clinical phase, CHX produced the strongest short-term restructuring of the salivary microbiome, whereas no statistically detectable group-level difference between CSEO and water was observed across the main compositional endpoints over 2 weeks. This null compositional finding does not establish microbiome protection or clinical benefit, nor does it exclude modest or heterogeneous effects that may have been undetectable in this small cohort. Larger, longer trials incorporating clinical periodontal, host-response, and functional microbiome endpoints are needed to evaluate CSEO and other EO formulations as adjuncts in periodontal therapy.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Alisa Kazarina designed the study, coordinated the project, performed formal analysis, interpreted the data, and drafted the manuscript. Anete Vaškevica, Dārta Elizabete Senkāne and Natalja Sergejeva contributed to participant recruitment, sample collection, and study logistics. Valdis Pirsko performed the cytotoxicity experiment and contributed to data analysis. Kristīne Vīksne performed DNA extraction and contributed to laboratory methodology. Pawel Zayakin contributed to bioinformatic analysis and data interpretation. Sigita Hušča prepared sequencing libraries. Lība Sokolovska contributed to experimental design and data interpretation. Ilva Nakurte and Laura Pastare performed essential oil analysis and interpretation. Stefan Ruhl supervised the project, contributed methodological guidance and interpretation, and critically revised the manuscript. All authors reviewed and approved the final manuscript.

Funding

This research is conducted under the Postdoctoral Grant “Oral microbiome and inflammatory dynamics in Stage III periodontitis: comparative effects of mouthwash formulations containing Latvian-grown Coriandrum sativum essential oil versus standard chlorhexidine regimen” (Grant No. RSU-PG-2024/1–0010 ), as part of the project “RSU Internal and RSU with LASE External Consolidation” (Project No. 5.2.1.1.i.0/2/24/I/CFLA/005), funded by the European Union Recovery and Resilience Facility and the budget of the Republic of Latvia.

Data availability

Raw shotgun metagenomic sequencing data generated in this study have been deposited in the BioStudies ArrayExpress collection under accession E-MTAB-16822. The dataset will be made publicly accessible upon publication. Additional supporting data are provided within the article and its Online Resources. Participant-level clinical covariate data are not publicly available because of participant privacy and ethical restrictions but may be available from the corresponding author on reasonable request, subject to applicable institutional and ethical requirements.

Code availability

Analysis code is available in the study GitHub repository: https://github.com/lexx-04/EOsPerio_microbiome_analysis. Analytical scripts were developed with assistance from OpenAI Codex Desktop under author supervision. All code, outputs, and statistical results were reviewed and validated by the authors before inclusion in the final analysis.

Declarations

Ethics approval

All procedures performed in this study involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Approval was granted by the RSU Ethics Committee on 15 April 2024 (No. 2-PĒK-4/356/2024).

Consent to participate

Written informed consent was obtained from all individual participants included in the study.

Consent for publication

Not applicable. The manuscript does not include identifiable individual participant data or identifiable images.

Clinical trial registration

Trial registration number: ISRCTN79156900, retrospectively registered on 11 May 2026.

Competing interests

Stefan Ruhl serves as an Editor-in-Chief of Clinical Oral Investigations. To avoid any potential conflict of interest, he should not be involved in the editorial assessment, peer review, or decision-making for this manuscript. The other authors declare no competing interests.