Oral squamous cell carcinoma is associated with altered salivary microbiome structure and reduced community evenness Authors: Miguel Ferriz-Jordán, David Hervás, Leticia Bagan, Concepción Gimeno, Alejandro Herreros-Pomares, José Bagán Published: 2026 ABSTRACT Oral squamous cell carcinoma (OSCC) remains a major global health burden, highlighting the need for improved non-invasive tools for early detection and disease monitoring. Because saliva is easily accessible and reflects the oral microenvironment, the salivary microbiome has emerged as a promising source of candidate biomarkers. In this study, we characterized the salivary microbiota of 113 individuals, including 52 patients with OSCC and 61 healthy controls, using 16S rRNA gene sequencing targeting the V3-V4 region. Microbial diversity and composition were analyzed using complementary bioinformatic and statistical approaches, including four differential abundance methods (MaAsLin2, ANCOM-BC2, LinDA, and ALDEx2). OSCC was associated with significant differences in overall microbial community structure, as shown by beta diversity analyses, together with reduced community evenness but no major loss of richness. Differential abundance analyses identified several taxa overrepresented in OSCC, including the genera Tannerella, Solobacterium, Dialister, and Bergeyella, as well as species such as Solobacterium moorei, Tannerella forsythia, and Prevotella nigrescens. In contrast, Leptotrichia was underrepresented in OSCC. These findings support the existence of an OSCC-associated salivary dysbiosis characterized by ecological restructuring rather than global diversity loss. The identified taxa represent exploratory candidate microbial signatures associated with OSCC that require validation in larger, independent, and longitudinal cohorts before any potential clinical application can be considered.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-54192-x. Introduction Oral cancer (OC) represents a major global health burden, with oral squamous cell carcinoma (OSCC) accounting for approximately 90% of all cases. Worldwide, OC ranks among the most common malignancies, with an estimated annual incidence exceeding 389,000 new cases and 188,000 deaths1. Over the past decade, both the incidence and mortality of OC have continued to increase2. The etiology of OC is multifactorial, with well-established risk factors including tobacco use, alcohol consumption, and betel quid chewing3,4. In addition, infection with oncogenic pathogens, such as high-risk human papillomavirus (HPV), has emerged as an additional etiological factor3,4. The human oral cavity, together with the aerodigestive tract, harbors one of the most complex and densely populated microbial ecosystems in the body. Current repositories such as the expanded Human Oral Microbiome Database (eHOMD) catalog 836 bacterial species as of March 2026, approximately 77% of which are culturable under laboratory conditions5. The resident oral microbiota interacts dynamically with the host, influencing local and systemic physiology through metabolic activity, immune modulation, and maintenance of epithelial homeostasis6. Consequently, disruption of this balanced community, known as dysbiosis, has been implicated in a wide range of systemic and local diseases, including atherosclerosis7, autoimmune disorders8, cancer9, bacteremia associated with endocarditis10, and diabetes11. Increasing evidence suggests that dysbiosis may contribute to the development of chronic periodontal disease12,13, proliferative verrucous leukoplakia (PVL)14,15, and OSCC16,17. Chronic exposure to proinflammatory bacteria can promote a persistent inflammatory microenvironment, characterized by oxidative stress, DNA damage, and activation of oncogenic signaling pathways18. Several bacterial taxa, including Fusobacterium nucleatum19, Porphyromonas gingivalis20, Prevotella intermedia21, and Capnocytophaga spp.22, have been repeatedly associated with OSCC and precancerous lesions, suggesting a potential etiological role or possible use as diagnostic biomarkers. Our group has established a research line investigating the role of the oral microbiome in oral potentially malignant disorders (OPMDs) and OC. In an initial study, we identified reduced microbial diversity and increased abundance of potentially oncogenic bacteria, such as Campylobacter jejuni, in tissue biopsies from patients with PVL23. Building on these findings, we subsequently compared the oral microbiome of tissue biopsies from patients with homogeneous leucoplakia, PVL, OSCC, and OSCC arising from PVL (PVL-OSCC). These analyses revealed distinct patterns of dysbiosis associated with disease progression, with cancer samples showing reduced microbial diversity and a microbiome enriched in members of the phylum Fusobacteriota24. We next sought to determine whether the dysbiosis detected in tissue biopsies could also be identified in saliva, a promising non-invasive biofluid for OC detection. Unlike tissue biopsies, saliva collection is simple, repeatable, and suitable for screening. Saliva contains exfoliated epithelial cells, extracellular vesicles, and microbial DNA that reflects the oral microenvironment25. Although no saliva-based diagnostic test for oral cancer has yet received FDA approval, largely due to the need for further clinical validation, saliva remains an attractive biospecimen for biomarker discovery because it can be collected easily, repeatedly, and non-invasively26. Recently, we conducted a pilot study analyzing saliva samples from 10 patients with oral leukoplakia, 28 patients with PVL, and 28 patients with OSCC, which suggested progressive microbial dysbiosis during malignant transformation27. However, this study did not include a healthy control group. In the present study, we analyze saliva samples of a cohort of 113 individuals, including 61 healthy controls and 52 patients with OSCC, using 16S rRNA gene sequencing. In order to ensure comparability with our previous studies, which targeted the V3-V4 region of the 16S rRNA gene, the same region was analyzed in the present saliva cohort. By applying multiple complementary statistical approaches, we aim to identify robust microbial signatures associated with OSCC. The present study is not intended to replace tissue-based analyses, but rather to evaluate whether tumor-associated microbial alterations previously identified at the tissue level can also be detected in saliva, thereby supporting its potential as a non-invasive surrogate for disease-associated microbial signatures. Our findings may support the development of saliva-based, non-invasive diagnostic strategies and help identify bacterial taxa potentially involved in oral carcinogenesis, thereby providing a foundation for future mechanistic studies. Results Participant characteristics and sequencing data summary This study included 113 participants, comprising 61 healthy controls and 52 patients diagnosed with OSCC. The clinicopathological characteristics of patients with OSCC are summarized in Table 1. The mean age of patients was 67.93 years (range 37.00-92.45), 55.8% were males, and 67.3% were diagnosed at an early stage of the disease (stages I and II).Table 1Clinicopathological characteristics of patients with OSCC.Mean age (standard deviation)67.93 (12.43)Number of casesPercentageGender Male2955.8 Female2344.2Location Buccal mucosa47.7 Floor mouth611.5 Gingiva1732.7 Palate713.5 Tongue1834.6Clinical form Erythroleukoplakia713.5 Exophytic815.4 Mixed11.9 Ulcerated3669.2TNM T1N0M02751.9 T2N1M0713.5 T2N0M0815.4 T3N1M011.9 T3N2M000.0 T3N0M011.9 T4aN1M0815.4Stage I2955.8 II611.5 III917.3 IV815.4Early-advanced stage Advanced1732.7 Early3567.3 Table 1Clinicopathological characteristics of patients with OSCC.Mean age (standard deviation)67.93 (12.43)Number of casesPercentageGender Male2955.8 Female2344.2Location Buccal mucosa47.7 Floor mouth611.5 Gingiva1732.7 Palate713.5 Tongue1834.6Clinical form Erythroleukoplakia713.5 Exophytic815.4 Mixed11.9 Ulcerated3669.2TNM T1N0M02751.9 T2N1M0713.5 T2N0M0815.4 T3N1M011.9 T3N2M000.0 T3N0M011.9 T4aN1M0815.4Stage I2955.8 II611.5 III917.3 IV815.4Early-advanced stage Advanced1732.7 Early3567.3 Clinicopathological characteristics of patients with OSCC. A total of 7,088,020 paired-end reads were obtained, with a mean of 62,726 reads per sample (range 31,566-115,113). Rarefaction curves relating sequencing depth to taxonomic richness showed a tendency to plateau in most samples, suggesting that sequencing depth was sufficient to capture the majority of bacterial diversity (Fig. 1). All detected taxa belonged to the domain Bacteria. In total, 12 phyla, 17 classes, 35 orders, 51 families, and 94 genera were identified. Additionally, 172 putative species-level taxa were detected.Fig. 1Rarefaction curves showing the relationship between sequencing depth (log-transformed read counts) and genus richness for each sample. Curves are stratified by group (controls and OSCC patients). The plateau observed in most samples indicates sufficient sequencing depth to capture the majority of bacterial diversity. Rarefaction curves showing the relationship between sequencing depth (log-transformed read counts) and genus richness for each sample. Curves are stratified by group (controls and OSCC patients). The plateau observed in most samples indicates sufficient sequencing depth to capture the majority of bacterial diversity. Taxonomic composition of the oral microbiota The composition of the oral microbiota was characterized in both OSCC patients and healthy controls. The relative abundances of bacterial taxa at the phylum, family, and genus levels are shown in Fig. 2.Fig. 2Stacked bar plots showing the relative abundance of the oral microbiota across samples from controls and OSCC patients. (A) Phylum-level composition including all identified phyla and unassigned reads. (B) Family-level composition showing the 10 most abundant families, together with unassigned reads and remaining low-abundance families grouped as “Other”. (C) Genus-level composition showing the 10 most abundant genera, along with unassigned reads and remaining genera grouped as “Other”. Stacked bar plots showing the relative abundance of the oral microbiota across samples from controls and OSCC patients. (A) Phylum-level composition including all identified phyla and unassigned reads. (B) Family-level composition showing the 10 most abundant families, together with unassigned reads and remaining low-abundance families grouped as “Other”. (C) Genus-level composition showing the 10 most abundant genera, along with unassigned reads and remaining genera grouped as “Other”. A high degree of inter-individual variability in microbial composition was observed across samples. At the phylum level, the microbial community was dominated by Firmicutes (mean relative abundance 34.1%), followed by Bacteroidota (28.1%), Pseudomonadota (22.6%), Fusobacteriota (10.6%), and Actinomycetota (3.0%). All other phyla accounted for less than 1% of the total relative abundance. At the family level, the most abundant taxa were Prevotellaceae (22.6%), Streptococcaceae (15.4%), Veillonellaceae (14.2%), Neisseriaceae (12.0%), Pasteurellaceae (8.2%), and Fusobacteriaceae (7.8%), followed by Porphyromonadaceae (4.4%), Leptotrichiaceae (2.8%), Micrococcaceae (2.3%), Carnobacteriaceae (1.4%), and Lachnospiraceae (1.3%). Remaining families were present at relative abundances below 1%. At the genus level, the most abundant genera were Prevotella (20.5%), Streptococcus (15.3%), Veillonella (13.5%), Neisseria (11.9%), Fusobacterium (7.8%), Haemophilus (6.9%), and Porphyromonas (4.4%), followed by Rothia (2.3%), Leptotrichia (2.1%), and Granulicatella (1.4%). All remaining genera individually accounted for less than 1% of total abundance. On average, 4.5% of reads could not be assigned at the genus level. Comparison of alpha and beta diversity between OSCC patients and controls Intra-sample microbial diversity (alpha diversity) was assessed using multiple indices and is shown in Fig. 3, stratified by clinical group. Across the entire cohort, the mean values were 3578.2 for Breakaway richness, 7.33 for the Shannon index, 0.9987 for the Simpson index, and 0.904 for Pielou’s evenness. No statistically significant differences between clinical groups were observed for Breakaway richness, Shannon, or Simpson indices. However, Breakaway richness showed greater variability in OSCC patients, with a tendency toward higher values in some samples. OSCC patients showed significantly lower evenness, as measured by Pielou’s index (p = 0.018). These results suggest that OSCC is not associated with major changes in overall richness, but rather with alterations in community structure reflected by reduced evenness.Fig. 3Alpha diversity of salivary microbiota in healthy controls (blue) and OSCC patients (red). Boxplots represent (a) Breakaway richness, (b) Shannon index, (c) Simpson index, and (d) Pielou’s evenness. A significant decrease in evenness was observed in OSCC patients (p = 0.018). Alpha diversity of salivary microbiota in healthy controls (blue) and OSCC patients (red). Boxplots represent (a) Breakaway richness, (b) Shannon index, (c) Simpson index, and (d) Pielou’s evenness. A significant decrease in evenness was observed in OSCC patients (p = 0.018). Inter-sample microbial diversity (beta diversity) was assessed using four distance metrics based on genus-level data: Bray-Curtis, Jaccard, unweighted UniFrac, and weighted UniFrac. The corresponding principal coordinates analyses (PCoA) are shown in Fig. 4. Significant differences in microbial community composition between healthy controls and OSCC patients were observed using PERMANOVA for the Bray-Curtis (p = 0.009), Jaccard (p = 0.007), and unweighted UniFrac (p < 0.001) distances. Although the weighted UniFrac analysis showed a tendency toward significance, the results were not considered robust, as variability across repeated analyses resulted in some p-values exceeding the significance threshold. The lack of robust differences in weighted UniFrac suggests that group differences are primarily driven by the presence or absence of taxa rather than shifts in dominant taxa. Overall, these findings indicate that the microbial community structure differs significantly between controls and OSCC patients.Fig. 4Principal coordinates analysis (PCoA) of beta diversity using (a) Bray-Curtis, (b) Jaccard, (c) unweighted UniFrac, and (d) weighted UniFrac distance metrics. Each point represents a sample, colored by clinical group (blue: controls; red: OSCC). Dashed ellipses indicate 95% confidence intervals. PERMANOVA p-values are shown in each panel. Principal coordinates analysis (PCoA) of beta diversity using (a) Bray-Curtis, (b) Jaccard, (c) unweighted UniFrac, and (d) weighted UniFrac distance metrics. Each point represents a sample, colored by clinical group (blue: controls; red: OSCC). Dashed ellipses indicate 95% confidence intervals. PERMANOVA p-values are shown in each panel. Differential abundance analysis of the oral microbiota in OSCC To identify specific taxa associated with the differences in overall microbiome composition between OSCC patients and healthy controls revealed by the PERMANOVA analyses, differential abundance analyses were performed using four complementary methods across multiple taxonomic levels: MaAsLin2, ANCOM-BC2, LinDA, and ALDEx2. Given the known methodological differences among these approaches, results from all methods were considered to provide a comprehensive overview of taxa associated with OSCC. The consistency of significant taxa across methods is summarized in Fig. 5 and Supplementary Fig. S1. The full results obtained by each method, including q-values and model coefficients or log fold changes, are provided in Supplementary Table S1, while taxa identified by at least one method are summarized in Table 2.Fig. 5Venn diagrams showing the number of taxa identified as significantly different between groups by each of the four statistical methods (MaAsLin2, ANCOM-BC2, LinDA, and ALDEx2), as well as their overlaps. Taxa are stratified by taxonomic level: (a) phylum, (b) family, (c) genus, and (d) species. Overlapping regions indicate taxa identified by multiple methods, whereas non-overlapping regions represent method-specific findings.Table 2Differentially abundant taxa between healthy controls and OSCC patients across taxonomic levels.PhylaFamiliesGeneraSpeciesOverrepresented in OSCCPseudomonadotaBacillotaTannerellaceaeWeeksellaceaeErysipelotrichaceaeFlavobacteriaceaeTannerellaDialisterBergeyellaSolobacteriumPrevotella nigrescensFusobacterium simiaeHaemophilus haemolyticusTannerella forsythiaLachnoanaerobaculum umeaenseSolobacterium mooreiHaemophilus influenzaeStreptococcus mitisDialister pneumosintesVeillonella tobetsuensisKingella denitrificansUnderrepresented in OSCC––Leptotrichia– Venn diagrams showing the number of taxa identified as significantly different between groups by each of the four statistical methods (MaAsLin2, ANCOM-BC2, LinDA, and ALDEx2), as well as their overlaps. Taxa are stratified by taxonomic level: (a) phylum, (b) family, (c) genus, and (d) species. Overlapping regions indicate taxa identified by multiple methods, whereas non-overlapping regions represent method-specific findings. Table 2Differentially abundant taxa between healthy controls and OSCC patients across taxonomic levels.PhylaFamiliesGeneraSpeciesOverrepresented in OSCCPseudomonadotaBacillotaTannerellaceaeWeeksellaceaeErysipelotrichaceaeFlavobacteriaceaeTannerellaDialisterBergeyellaSolobacteriumPrevotella nigrescensFusobacterium simiaeHaemophilus haemolyticusTannerella forsythiaLachnoanaerobaculum umeaenseSolobacterium mooreiHaemophilus influenzaeStreptococcus mitisDialister pneumosintesVeillonella tobetsuensisKingella denitrificansUnderrepresented in OSCC––Leptotrichia– Differentially abundant taxa between healthy controls and OSCC patients across taxonomic levels. Pseudomonadota Bacillota Tannerellaceae Weeksellaceae Erysipelotrichaceae Flavobacteriaceae Tannerella Dialister Bergeyella Solobacterium Prevotella nigrescens Fusobacterium simiae Haemophilus haemolyticus Tannerella forsythia Lachnoanaerobaculum umeaense Solobacterium moorei Haemophilus influenzae Streptococcus mitis Dialister pneumosintes Veillonella tobetsuensis Kingella denitrificans At the phylum level, Pseudomonadota and Bacillota were enriched in OSCC patients. At lower taxonomic levels, several taxa previously associated with oral dysbiosis and periodontal disease were found to be overrepresented in OSCC, including members of the families Tannerellaceae, Erysipelotrichaceae, and Flavobacteriaceae, as well as the genera Tannerella, Dialister, Bergeyella, and Solobacterium. At the species level, enrichments included Tannerella forsythia, Fusobacterium simiae, Prevotella nigrescens, Solobacterium moorei, Dialister pneumosintes, Haemophilus haemolyticus, Haemophilus influenzae, Lachnoanaerobaculum umeaense, Streptococcus mitis, Veillonella tobetsuensis, and Kingella denitrificans. Leptotrichia was the only genus that decreased in OSCC. No bacterial taxa were exclusively present in patients with OSCC or in healthy controls. Discussion This study evaluated the oral microbiome of 52 patients diagnosed with OSCC and 61 healthy controls. The OSCC cohort is clinically representative of a typical hospital-based oral cavity cancer series, with an older patient population, a slight male predominance, and a predominance of tumors arising in the tongue and gingiva, most frequently presenting as ulcerative lesions. Importantly, most cases were diagnosed at early stages (I-II, 67.3%), although a substantial fraction presented with advanced disease (III-IV, 32.7%), reflecting the heterogeneity of clinical presentation in OSCC. The taxonomic composition observed in this cohort is broadly consistent with the expected structure of the oral microbiota, being dominated by genera such as Prevotella, Streptococcus, Veillonella, and Neisseria, while also showing a high degree of inter-individual variability. This heterogeneity is a well-recognized feature of the oral ecosystem and highlights the importance of identifying disease-associated changes against a highly variable microbial background. Despite the absence of major differences in overall richness between OSCC patients and healthy controls, alpha diversity analyses revealed significantly lower evenness in OSCC, as measured by Pielou’s index, in agreement with previous reports28. This finding suggests that OSCC is not necessarily associated with a global loss of microbial diversity, but rather with an imbalance in community structure, in which certain taxa may become disproportionately represented. In this context, the trend toward greater variability in Breakaway richness among OSCC cases may further reflect the ecological instability of the tumor-associated oral microbiome. This interpretation is supported by beta diversity analyses, which consistently showed significant differences in overall microbial composition between OSCC patients and controls. The fact that these differences were detected using Bray-Curtis, Jaccard, and unweighted UniFrac distances indicates that the two groups differ not only in relative abundance patterns but also in the presence and absence of specific taxa. By contrast, the lack of robust significance for weighted UniFrac suggests that these differences are less strongly driven by dominant taxa and may instead be influenced by shifts in less abundant or less prevalent community members. We performed differential abundance analyses using complementary methods across multiple taxonomic levels, which revealed several taxa overrepresented in OSCC. At the species level, we identified Tannerella forsythia, Streptococcus mitis, Veillonella tobetsuensis, Prevotella nigrescens, Haemophilus haemolyticus, Haemophilus influenzae, and Solobacterium moorei as overrepresented in OSCC patients, among others. Tannerella forsythia is a well-known periodontal pathogen with strong pro-inflammatory potential, capable of promoting tissue degradation and creating a microenvironment conducive to tumor development29,30. Previous studies have also reported its association with early-stage head and neck squamous cell carcinoma31. Species belonging to the genera Streptococcus and Veillonella have also been associated with head and neck cancer, being proposed as discriminative taxa28. Enrichment of the genera Prevotella and Haemophilus in saliva from OSCC patients was reported in a pilot study a few years ago32. These authors32 also reported that Solobacterium moorei proportion increases with cancer progression in tumoral tissue. This microorganism has been consistently reported in different oral conditions, including dysbiosis, halitosis, and the production of volatile sulfur compounds33. In addition, a previous study including six patients with OSCC and four patients with oral leukoplakia already reported the enrichment of the genus Solobacterium in the saliva of patients with OSCC34. Beyond oral cancer, S. moorei has also been linked to colorectal cancer, where increased abundance has been observed not only in the oral microbiota35 but also in gut microbiota36. In both contexts, its levels appear to correlate with disease progression, and it has been proposed that elevated oral levels of S. moorei may contribute to cancer initiation, while also potentially influencing tumor progression. These findings suggest that S. moorei may play a broader role in cancer-associated microbial dysbiosis and could represent a promising target for future mechanistic and biomarker studies. Other genera identified in our analysis, including Dialister and Bergeyella, have also been associated with chronic inflammation and pathogenic biofilm formation29,37, further supporting the presence of a pro-inflammatory microbial environment in OSCC. Notably, Dialister has previously been proposed as a discriminative taxon for head and neck cancer in saliva28. Several less-characterized families identified in this study, including Weeksellaceae, Erysipelotrichaceae, and Flavobacteriaceae, have been only sparsely described in the context of OSCC. While some evidence links these taxa to cancer-related conditions or risk factors38,39 their precise role remains unclear, and further studies are required to determine whether they contribute to disease progression or reflect secondary ecological changes. Leptotrichia was the only taxon found to be significantly underrepresented in OSCC. Previous studies have reported a higher abundance of Leptotrichia in early-stage disease and associated its presence with improved survival outcomes31,40, suggesting a potential protective or context-dependent role. However, its behavior appears to vary depending on clinical and molecular factors such as HPV status and treatment strategy28. Several limitations of this study should be acknowledged. First, despite adjustment for age and sex, residual confounding is likely because several major determinants of the oral microbiome were not fully accounted for, including smoking, alcohol consumption, periodontal status, caries burden, recent dental procedures, oral hygiene habits, diet, socioeconomic factors, and HPV status. In addition, detailed information on systemic comorbidities such as diabetes mellitus, hypertension, or other chronic conditions was not systematically collected and therefore could not be included in the analysis. Imbalances in these variables may confound microbial differences attributed to OSCC. In addition, heterogeneity within the OSCC group, including tumor location, ulceration, and stage, may also influence the oral microbiota. Because the present analyses focused primarily on OSCC versus control comparisons, site- or stage-specific microbial signals may have been diluted. Second, the cross-sectional design precludes any inference about causality or temporality. Therefore, it remains unclear whether the observed dysbiosis precedes OSCC development, arises as a consequence of the tumor, or reflects a bidirectional interaction. This limitation is particularly relevant when interpreting disease-associated taxa, as cross-sectional microbiome studies cannot distinguish between microbial drivers, passengers, or secondary ecological changes associated with tumor presence. Third, saliva offers clear advantages as a non-invasive and clinically feasible sample type, but it is not tumor-specific. Rather than capturing only the tumor microenvironment, saliva reflects the ecology of the entire oral cavity. However, this lack of spatial specificity is counterbalanced by its ability to capture the overall oral ecosystem, including microorganisms potentially shed from tumor sites, making saliva a relevant matrix for the identification of disease-associated microbial patterns in a clinically accessible and non-invasive manner. Moreover, only a single saliva sample was analyzed per participant, precluding longitudinal assessment of the oral microbiome. As a result, intra-individual variability could not be assessed. The oral microbiome is known to fluctuate over time due to factors such as diet, circadian patterns, oral hygiene, or short-term environmental exposures. Therefore, the microbial profiles observed may reflect transient states rather than stable individual-specific patterns. This limitation restricts the ability to assess temporal dynamics and to determine whether the identified microbial alterations represent persistent features of OSCC or short-term fluctuations. Fourth, although this study has a considerable sample size, it is a single-center study, which may limit the generalizability of the findings. Validation in independent multicenter cohorts will be essential to confirm the reproducibility and robustness of the identified microbial signatures across different populations and clinical settings. Since most OSCC cases were diagnosed at early stages, the findings may also be less representative of advanced disease. Fifth, the use of 16S rRNA gene sequencing targeting the V3-V4 region imposes taxonomic and functional limitations. Although this region provides reliable classification at the genus level and allows comparison with existing literature, it does not always enable accurate species-level discrimination, particularly among closely related taxa. As a result, species-level assignments based on 16S data may be subject to ambiguity or misclassification and should therefore be interpreted with caution. Accordingly, species-level findings reported in this study should be considered indicative rather than definitive and require confirmation using higher-resolution approaches such as shotgun metagenomics. In addition, this approach does not provide functional or strain-level information, which would require alternative methods. The study was therefore limited to taxonomic profiling, without metagenomic, transcriptomic, metabolomic, or experimental validation of the inferred biological relevance of the identified taxa. Finally, although compositionality and zero inflation were addressed by applying four complementary statistical tools, overlap across methods was only partial. Thus, some differential abundance findings were method-dependent and should be interpreted as exploratory until independently replicated. Accordingly, the taxa identified in this study should not be interpreted as validated biomarkers, but rather as exploratory microbial signatures associated with OSCC. The main contribution of this study lies not only in confirming differences between OSCC patients and healthy controls, but in providing a robust and integrative characterization of these differences using a moderately sized saliva cohort, standardized sequencing depth, and a multi-method analytical framework. By combining four complementary approaches for differential abundance analysis, we aimed to prioritize reproducibility and methodological robustness, addressing a key limitation highlighted in previous studies, namely the inconsistency of reported microbial signatures across analytical pipelines. Within this framework, we identified a set of taxa repeatedly associated with OSCC across methods, including genera such as Tannerella, Solobacterium, Dialister, and Bergeyella. Notably, several of these taxa are linked to periodontal disease and pro-inflammatory oral environments, supporting the concept of OSCC-associated dysbiosis as an inflammation-driven ecological shift rather than a random compositional change. An additional relevant finding is the combination of reduced community evenness, in the absence of major changes in richness, together with consistent beta diversity differences driven primarily by presence/absence metrics. This pattern suggests that, in OSCC, the core oral microbiota is largely preserved but becomes imbalanced, with selective expansion of specific taxa and loss or reduction of others. Such ecological restructuring is consistent with a transition toward a dysbiotic, inflammation-associated microbial community. It may provide a basis for the future development of saliva-based diagnostic approaches. However, these findings should be considered exploratory and require validation in independent cohorts before any clinical application. Taken together, our results reinforce the potential of saliva as a non-invasive matrix for capturing disease-associated microbial signatures. Longitudinal studies following patients with oral potentially malignant disorders through malignant transformation are needed to distinguish microbial risk markers from disease-associated changes. In addition, future analyses should incorporate detailed clinical stratification, including tumor stage, anatomical site, nodal status, ulceration, periodontal condition, and lifestyle factors such as smoking and alcohol consumption. Although functional inference tools such as PICRUSt2 can provide insights into the potential metabolic capabilities of microbial communities, these approaches rely on predicted functions derived from taxonomic profiles and do not directly measure gene expression or activity. Therefore, we chose not to include functional predictions in this study in order to avoid overinterpretation. Integrative multi-omics approaches, combining metagenomics or metatranscriptomics with salivary metabolomics, may help elucidate the functional mechanisms underlying the observed microbial shifts. From a translational perspective, the development and validation of saliva-based microbial classifiers in large, independent, multicenter cohorts will be essential to assess their robustness and generalizability across populations and technical workflows. Methods Patients and samples This study included 113 individuals recruited at the Department of Stomatology and Maxillofacial Surgery of the General University Hospital of Valencia between 2023 and 2024. Participants were classified according to their oral status into two groups: Group I comprised 61 healthy controls, and Group II included 52 patients diagnosed with OSCC. Exclusion criteria were pregnancy, age under 18 years, history of previous cancer (primary tumor), presence of metastasis from another anatomical site, prior radiotherapy or chemotherapy to the head or neck region, and the presence of other oral conditions known to significantly alter salivary composition or the oral microbiome, including oral infections, autoimmune diseases affecting salivary glands (e.g., Sjögren’s syndrome), and the use of medications associated with reduced salivary flow or xerostomia (e.g., antidepressants, anticholinergics, antihypertensive agents). Participants were instructed to fast for at least 2 h prior to sample collection. None of the participants had received medication in the 3 months preceding sampling. No standardized periodontal examination, including periodontal probing or formal gingival assessment, was performed at the time of saliva collection. Unstimulated whole saliva was collected for 5 min into sterile 15-mL tubes using a funnel, following the protocol described by Navazesh41. To minimize blood contamination, participants were instructed not to brush their teeth within 45 min prior to sample collection. Samples visibly contaminated with blood were discarded and recollected. Saliva samples were not subjected to centrifugation prior to storage. Whole saliva was directly aliquoted and stored at − 80 °C. DNA extraction and 16S rRNA gene sequencing Genomic DNA was extracted from saliva samples using the GenElute™ Bacterial Genomic DNA Kit (Sigma-Aldrich, St. Louis, MO, USA) following the manufacturer’s instructions. Samples were lysed in buffer containing lysozyme and mutanolysin and incubated for 30 min at 37 °C. Subsequently, samples were treated with RNase for 2 min at room temperature, followed by Proteinase K incubation for 10 min at 55 °C. Prior to extraction, saliva samples were homogenized by gentle mixing to ensure representative recovery of microbial DNA. DNA concentration was measured by fluorometry using a Quantus™ Fluorometer (Promega, USA). The hypervariable V3-V4 region of the bacterial 16S rRNA gene was amplified using the universal primers 341F (5′-CCTACGGGNGGCWGCAG-3′) and 805R (5′-GGACTACHVGGGTATCTAATCC-3′). Library preparation consisted of a first PCR to amplify the target region followed by a second PCR to attach sample-specific indices. The resulting libraries were purified and assessed using the QIAxcel Advanced System (Qiagen, Germany) to verify fragment size and library quality prior to sequencing. Sequencing was performed on an Illumina MiSeq platform using paired-end reads (2 × 250 bp) and sequencing-by-synthesis chemistry. Raw sequencing data were generated in FASTQ format for downstream bioinformatic analysis. Bioinformatic analysis and taxonomic annotation Raw sequencing reads were processed using the DADA2 package (v1.36.0) in R (v4.5.1). During quality filtering, forward and reverse reads were truncated at position 245, and sequences with expected error ≥ 3 were discarded. The DADA2 denoising algorithm was applied to infer amplicon sequence variants (ASVs), after which paired-end reads were merged. Sequences of 20 bp or shorter were filtered. Chimeric sequences were removed using the removeBimeraDenovo function, setting the minFoldParentOverAbundance parameter to 4. Taxonomic assignment was performed up to the species level using the SILVA reference database (v138.2). Alpha diversity was assessed using the phyloseq (v1.52.0) and breakaway (v4.8.5) packages. Beta diversity analyses were performed using the vegan package (v2.7.1). Statistical analysis Prior to any statistical analysis, taxa that were present in one of the clinical groups and completely absent from the other were identified, after filtering for a prevalence threshold of 0.05. Age and sex were included as potential confounding variables in all analyses. Smoking status, alcohol consumption, and standardized periodontal parameters were not systematically collected at the time of saliva sampling and, therefore, could not be included as covariates. Continuous variables were summarized as mean ± standard deviation, while categorical variables were summarized as absolute and relative frequencies. Group differences between healthy controls and OSCC patients were evaluated using PERMANOVA with 9999 permutations. To improve robustness, the analysis was repeated 10 times, and the mean p-value was reported. A formal test for homogeneity of multivariate dispersions (e.g., PERMDISP) was not performed because our multivariable model included a continuous covariate, which precludes the straightforward definition of discrete group centroids required for such tests. However, the balanced nature of the comparison ensures that the PERMANOVA results remain reliable even in the presence of moderate dispersion differences42. Alpha diversity was evaluated using four indices reflecting microbial richness and evenness: Breakaway43, Shannon, Simpson, and Pielou indices. Differences in alpha diversity between groups were assessed using ordinal regression models implemented in the ordinal package (v2023.12.4.1). Beta diversity was assessed using principal coordinates analysis (PCoA) based on multiple distance metrics, including Bray-Curtis, Jaccard, unweighted UniFrac, and weighted UniFrac. Distance matrices were calculated using relative genus abundances. For UniFrac analyses, a phylogenetic tree was constructed using the DECIPHER (v3.6.0) and phangorn (v2.12.1) packages. The tree was inferred using the neighbor-joining method and optimized based on likelihood criteria using default parameters. Differential abundance analyses were performed to identify taxa associated with clinical groups at the phylum, family, genus, and species levels. Multiple complementary statistical approaches were applied, including MaAsLin244 (Maaslin2 package v1.22.0), ANCOM-BC245 (ANCOMBC package v2.12.0), LinDA46 (MicrobiomeStat package v1.2), and ALDEx247 (ALDEx2 package v1.42.0). Particularly in LinDA, we first performed the analysis using absolute count ASV data and subsequently repeated the analysis under a compositional data framework to validate the initial results. As robustness criteria across methods, we considered taxa significant only when they i) were significant after multiple-testing correction, using an FDR-adjusted q-value < 0.05, and ii) presented concordant directions of effect across methods. Taxa identified consistently across multiple methods were interpreted as having stronger supporting evidence. Statistical significance was set at α = 0.05 for both p-values and multiple-testing-adjusted q-values. Since each differential abundance method applies distinct statistical assumptions and approaches, special attention was given to handling compositionality and zero inflation. Detailed implementation is provided in the Supplementary Methods. For MaAsLin2, taxa with prevalence < 0.05 were filtered out. Compositionality was addressed using the centered log-ratio (CLR) transformation after the addition of pseudo-counts. For ANCOM-BC2, taxa with prevalence < 0.05 were filtered out. Compositionality was handled through a log-linear regression framework that explicitly models sampling bias and sequencing depth, without requiring external data transformation. A pseudo-count was added to handle zeros. For LinDA2, taxa with prevalence < 0.05 were filtered out. In the main analysis using absolute counts, an adaptive algorithm determines whether pseudo-count addition or zero imputation is applied. In both count-based and compositional analyses, the CLR transformation is applied after zero handling. A secondary compositional analysis was performed as a robustness check. Finally, for ALDEx2, compositionality was addressed using a Bayesian approach based on Monte Carlo sampling from a Dirichlet distribution, followed by CLR transformation to account for compositional constraints. All statistical analyses were performed in R (v4.5.1). Figures were generated using R packages ggplot2 (v 4.0.1), patchwork (v 1.3.2), and ggVennDiagram (v1.5.4). Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1