Multi-omics analysis reveals the association between intratumoral bacteria and somatic mutational signatures in oral squamous cell carcinoma

Abstract

Background

Oral squamous cell carcinoma (OSCC) poses significant challenges with its obscure etiology and poor prognosis. Intratumoral bacteria exert a significant influence on the progression of OSCC. Nonetheless, the precise composition and role of the intratumoral bacteria in OSCC patients remain obscure and the association between the intratumoral bacteria and somatic mutational signatures in OSCC patients has not been defined.

Methods

Paraffin-embedded tumors and adjacent normal tissues were collected retrospectively from OSCC patients. A five-region 16S rRNA gene sequencing (5R 16S rRNA-seq) approach (targeting V2, V3, V5, V6, and V8) was employed to characterize the intratumoral bacterial composition of OSCC patients, while whole-exome sequencing (WES) was utilized to identify the somatic mutational signatures of OSCC patients. Then, the multi-omics data were integrated to uncover the association between intratumoral bacterial composition and somatic mutational signatures in OSCC patients. Key findings were validated using fluorescence in situ hybridization (FISH) on tissue microarrays.

Results

We found significant differences in bacterial profiles between tumors and adjacent normal tissues. The diversity of the bacteria was less within tumor tissues and anaerobic bacteria and periodontopathogens (e.g., Dialister pneumosintes and Prevotella intermedia) were more prevalent in tumor tissues. And the presence of Dialister pneumosintes within tumor tissues is associated with poorer short-term disease-free survival in OSCC patients. The integrated analysis of the 5R 16S rRNA-seq and WES data revealed that tumor mutation burden was positively correlated with the α diversity of the intratumoral bacteria in OSCC patients.

Conclusion

This study utilized a multi-omics approach to elucidate the characteristics of the intratumoral bacteria and somatic mutational signatures in OSCC patients, revealing the intricate relationships between these elements. These findings pave the way for hypothesis-driven research aiming to further elucidate the role of the intratumoral bacteria in OSCC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-07500-4.

Introduction

Oral squamous cell carcinoma (OSCC) is a common malignancy that affects the oral cavity and accounts for approximately 90% of all oral malignancies [1]. In 2022, there were 389,485 newly diagnosed cases of oral cancer worldwide, with 188,230 reported fatalities [2]. The Global Cancer Observatory projects a 40% increase in the incidence of OSCC by 2040 along with a corresponding increase in mortality [3]. Despite recent advancements in detection, prevention, and treatment, factors such as late-stage diagnosis, recurrence, and metastasis persist, which contribute to a dismal 5-year survival rate for OSCC patients [3]. Consequently, elucidating the etiology and pathogenesis of OSCC in order to identify potential therapeutic targets is of paramount importance.

OSCC is strongly associated with various factors including tobacco use, alcohol intake, human papillomavirus infection, and genetic susceptibility [1, 4, 5]. Oral cavity is the beginning part of the digestive system, and chronic exposure to diverse internal and external stimuli induces genetic mutations in oral squamous epithelial cells. These mutated epithelial cells display atypical morphology and altered biological behavior, resulting in dysplasia of the oral mucosal epithelium that can potentially advance to OSCC. Extensive research has investigated the mutational landscape of OSCC through whole exome sequencing (WES) or whole genome sequencing. OSCC exhibits considerable heterogeneity in somatic mutational signatures; commonly altered genes include TP53, TTN, and FAT1 [4, 6–8]. The somatic mutational signatures are closely associated with the patients’ prognosis and treatment response [9]. For instance, tumor mutational burden (TMB), defined as the total number of somatic mutations per megabase of the genome, serves as a predictive biomarker for the effectiveness of immune checkpoint inhibitors in OSCC [10, 11]. Mutational signatures induced by various stimuli exhibit substantial variability [12], underscoring the importance of characterizing somatic mutations under specific conditions to enhance the management of OSCC.

Recent studies have examined intratumoral bacteria, confirming their presence in various solid tumors such as lung and breast cancer [13]. Furthermore, the intratumoral bacteria have been shown to induce DNA damage and promote somatic mutations, either directly or indirectly, which influences tumor progression [14–16]. The oral cavity represents the second largest microbial reservoir after the gastrointestinal tract. Disruptions in oral bacteria are strongly linked to the development of various oral diseases [17]. Several studies have confirmed the presence of intratumoral bacteria in OSCC [18–20], and the intratumoral bacteria may influence OSCC progression by modulating immune responses and metabolic processes [21–24]. Previous studies have elucidated the characteristics of the intratumoral bacteria in OSCC using traditional techniques such as 16S rRNA gene sequencing (16S rRNA-seq). Nevertheless, in tumor tissue samples with minimal microbial biomass, 16S rRNA-seq fails to precisely discern bacteria at the species level. Thus, it is imperative to delineate the intratumoral bacteria composition in OSCC with enhanced detection and resolution. As a result, no research has yet investigated the relationship between the intratumoral bacteria composition and somatic mutational signatures in OSCC.

Five-region 16S rRNA-seq (5R 16S rRNA-seq) is an improvement on conventional 16S rRNA-seq as it involves the amplification and sequencing of five regions (V2, V3, V5, V6, and V8) of the 16S rRNA gene [13]. The advantages of this technique include: (1) detection of bacteria in low-biomass microbial samples; (2) more accurate and comprehensive species identification compared to traditional 16S rRNA-seq (V3 and V4); and (3) compatibility with DNA samples that have undergone a certain degree of degradation, such as formalin fixed paraffin embedded (FFPE) samples [13]. In this study, we performed WES and 5R 16S rRNA-seq on tumor and adjacent normal tissues from OSCC patients. Subsequently, the data from WES and 5R 16S rRNA-seq were integrated and analyzed to elucidate the relationship between the intratumoral bacterial composition and mutational signatures in OSCC patients. Our study provides a theoretical basis for comprehending the role of the intratumoral bacteria in the initiation and progression of OSCC.

Materials and methods

Study population and sample collection

The inclusion criteria for participants in this study were: (1) Diagnosis of OSCC confirmed through clinical and histopathological examination and (2) Surgical resection of OSCC performed at the Stomatological Hospital of Chongqing Medical University and the Second Affiliated Hospital of Chongqing Medical University between 2022 and 2023. The exclusion criteria for participants were: (1) Use of immunosuppressants or antibiotics within three months prior to surgery, (2) Receipt of other anti-tumor therapies before surgery, and (3) Presence of recurrent OSCC. A total of 196 paraffin-embedded tumor specimens and 132 paraffin-embedded adjacent normal tissues were retrospectively collected from OSCC patients who underwent surgical excision. Pathological and clinical data were gathered, and postoperative survival outcomes were assessed through outpatient visits or telephone follow-ups, continuing until mortality or June 2024. The study was approved by the institutional ethical review boards of the Stomatological Hospital of Chongqing Medical University (Ethics No. 2024-087) and the Second Affiliated Hospital of Chongqing Medical University (Ethics No. 2024-064) for anonymous data analysis, and informed consent was waived for the retrospective study. This study was registered in the Chinese Clinical Trial Registry (https://www.chictr.org.cn/), under the registration number ChiCTR2400088268. Our work has been reported in line with the STROCSS criteria [25].

DNA extraction, PCR amplification, and 5R 16S rRNA-seq

Microbial genomic DNA was extracted from 196 paraffin-embedded OSCC and 132 paraffin-embedded adjacent normal tissues using a TIANamp FFPE DNA Kit (TIANGEN, China) following the manufacturer’s protocol. DNA quality and concentration were assessed using 1% agarose gel electrophoresis and a NanoDrop ND-2000 spectrophotometer (Thermo Scientific Inc., USA) before storage at -80 °C.

16S rRNA-seq was conducted by multiplex amplification of five regions (V2: F1-TGGCGAACGGGTGAGTAA, R1-CCGTGTCTCAGTCCCARTG; V3: F2-ACTCCTACGGGAGGCAGC, R2-GTATTACCGCGGCTGCTG; V5: F3-GTGTAGCGGTGRAATGCG, R3-CCCGTCAATTCMTTTGAGTT; V6: F4-GGAGCATGTGGWTTAATTCGA, R4-CGTTGCGGGACTTAACCC; V8: F5-GGAGGAAGGTGGGGATGAC, R5-AAGGCCCGGGAACGTATT) with a T100 Thermal Cycler (Bio-Rad, USA) [13]. The PCR mix contained 10 µL of 2× Pro Taq, 0.8 µL of each forward and reverse primers (5 µM), 10 ng of template DNA, and ddH₂O to a final volume of 20 µL. PCR amplification was performed with an initial denaturation at 95 °C for 3 min, followed by 32 cycles of denaturation at 95 °C for 30 s, annealing at 53 °C for 30 s, extension at 72 °C for 45 s, and a final extension at 72 °C for 10 min at 4 °C. Negative controls (for sampling, DNA extraction, and no-template PCR) were also included. All samples were amplified in triplicate, followed by extraction and purification of the PCR products from a 2% agarose gel and subsequent quantification using a Synergy HTX system (Biotek, USA). PCR products were detected by 2% agarose gel electrophoresis, with 3 µl loaded for gel imaging. Samples showing visible single bands on the electrophoresis image met the requirements for further sequencing.

After DNA extraction and PCR amplification, the quality of PCR products from 154 tumor samples and 86 adjacent normal samples met the requirements for the 5R 16S rRNA-seq. Equimolar amounts of purified amplicons were pooled and subjected to paired-end sequencing on an Illumina NextSeq 2000 platform (Illumina, San Diego, CA, USA) following the standard protocols of Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China).

5R 16S rRNA-seq data processing and analysis

Reads from the raw sequencing data were demultiplexed according to sample identifiers, filtered, and aligned for each of the five amplified regions using the primer sequences. The short multiple regions framework (SMURF) method was applied to combine read counts from the five regions into a coherent profiling result by solving the maximum likelihood problem. The GreenGenes database (May 2013 version with some improvements) served as a reference [13, 26]. To minimize noise variation at very low abundance levels, samples with fewer than 1000 normalized reads (including negative controls) and species with relative abundances less than 10^{− 4} were excluded from further analysis. To address the presence of common contaminants, any species detected in more than 7.5% of negative DNA/PCR/sequencing controls or more than 7.5% of empty paraffin controls were excluded [27]. The contaminating species identified and subsequently excluded from the analysis are detailed in Table S1.

The sequence data were processed according to the method described above and analyzed as follows. The R package EasyMicroPlot (version 0.5.1.25) was used to calculate alpha diversity indices (Pielou and Shannon index) and the co-occurrence network of genus-level microbiota between the groups. Microorganisms at the genus level with a relative abundance greater than 0.001 and detected in at least 20% of the samples were selected to construct co-occurrence networks. Spearman’s rank correlation coefficient was used to assess relationships among the microorganisms, and those with coefficients greater than 0.3 and p-values < 0.05 were included in the co-occurrence network. Principal coordinate analysis (PCoA) was conducted using the microeco package (version 1.7.0) to assess similarities among microbial communities in samples based on Bray-Curtis dissimilarity. Random forest and differential testing were employed to identify the significant taxa responsible for community differences in groups using the microeco package. Spearman’s rank correlation analysis was conducted to assess the differences in species abundance among samples using the corrplot package (version 0.92).

Survival analysis

Univariate Cox regression analysis was conducted using the survival package (version 3.5-8) based on the patients’ clinicopathological features, species-level microbial abundance, and prognosis data. Variables with univariate Cox regression analysis results indicating a p-value below 0.05 were then chosen for multivariable Cox regression analysis. Subsequently, a nomogram was constructed to assess survival probability based on the Cox regression results, implemented using the rms package (version 6.8-0).

WES

Total DNA was extracted from samples as described above. The quality of the isolated genomic DNA was verified using a combination of two methods: (1) DNA degradation and contamination were assessed by electrophoresis on 1% agarose gels; and (2) DNA concentration was quantified using an ND-2000 spectrophotometer (NanoDrop Technologies). Only high-quality DNA samples, characterized by an OD260/280 ratio of 1.7 to 2.1 and an OD260/230 ratio of 2.0 or higher, were used to construct the sequencing library.

The high-quality genomic DNA isolated from tumor tissues was fragmented using Covaris technology, yielding library fragments with a size range of 180 to 280 bp, to which adaptors were subsequently ligated at both termini. The fragmented DNA was subsequently amplified via ligation-mediated PCR (LM-PCR), purified, and hybridized to the Agilent SureSelect Human Exome V6 for selective enrichment, and nonhybridized fragments were removed through washing. Both uncaptured and captured LM-PCR products underwent real-time PCR to quantify the extent of enrichment. Each captured library was subsequently loaded onto the Illumina NovaSeq platform, and high-throughput sequencing was performed individually for each library to confirm that the desired average fold coverage was achieved.

Raw reads with low quality (mean Phred score < 20), including those containing adapter contamination and unrecognizable nucleotides (N base > 10), were trimmed or discarded using the software Fastp. The trimmed reads were mapped to the GRCh38 reference genome using the BWA-MEME software with default mapping parameters. Following the modified GATK Best Practices, the alignment BAM files were sorted using Samtools and PCR duplicates were marked using MarkDuplicates. After base quality recalibration, germline variant calling, which included SNPs and INDELs in all samples, was performed using the Haplotyper and GVCFtyper programs in Sentieon Genomics Tools. Somatic variants were detected using the tumor-only mode of the Mutect2 module, and germline variants were filtered based on gnomAD variant data (–germline-resource parameter). Variants were filtered using standard hard filtering parameters as per the GATK Best Practices pipeline. Structural variations (SVs) were identified by Manta, which detected variations based on several sources, including split alignments, soft clipping, insert size, paired alignments, and transchromosomal events. Copy number variations (CNVs) were detected by CNVkit, which employed the core window method to identify CNVs. All variants were annotated using the Variant Effect Predictor (VEP). SNPs, INDELs, and structural variants were categorized based on their chromosomal positions (including intergenic regions, exons, introns, splicing sites, untranslated regions, and 1-kb upstream and downstream regions) and their effects (including missense mutations, start codon gain or loss, stop codon gain or loss, and splicing mutations). Mutation annotations were based on data from gnomAD (https://gnomad.broadinstitute.org/), 1000 Genomes Project (1KG) (https://www.internationalgenome.org/), ESP6500 (http://evs.gs.washington.edu/EVS/), dbscSNV (http://www.liulab.science/dbscsnv.html), and dbNSFP (http://database.liulab.science/dbNSFP).

Mutational signatures were extracted and analyzed using the R packages sigminer (version 2.3.1) and Maftools (version 2.20.0). WES HNSCC data were downloaded using the R package TCGAmutations (version 0.4.0). HNSCC driver genes from the IntOGen database were downloaded from https://www.intogen.org/.

Hematoxylin and Eosin (H&E) staining

Commercial tissue microarrays (HN058OC01) comprising 39 primary OSCC tumor samples and 17 adjacent normal samples, were acquired from Zhongke Guanghua Intelligent Biotechnology Co., Ltd. (Xi’an, China). The tissue microarrays were stained using the Hematoxylin-Eosin Stain Kit (G1120, Solarbio) according to the manufacturer’s instructions.

Fluorescence in situ hybridization (FISH)

The tissue microarrays (HN058OC01) were labeled using a Fluorescence in Situ Hybridization Kit for RNA (R0306M, Beyotime) with Cyanine 3-labeled probes targeting bacterial (5′-GCT GCC TCC CGT AGG AGT-3′) and Dialister pneumosintes (5′-CTT GGT GAG CCG TTA CCT CAC CAA CTA CCT-3′) ribosomal 16S rRNA gene sequences. The tissue microarrays were subsequently counterstained with 4′,6-Diamidino-2-phenylindole (DAPI) to visualize the cell nuclei. Fluorescent images were acquired and visualized using a confocal microscope (Pannoramic MIDI, 3DHISTECH) in conjunction with the CaseViewer software (version 2.4). Fluorescence intensities of cells within designated regions of interest were converted into grayscale values and quantified using the ImageJ software.

Data visualization and statistical analysis

The R packages ggplot2 (version 3.5.0) and ggpubr (version 0.6.0) were used for data visualization and statistical analysis. The Chi-squared test was employed to assess whether discrepancies existed in the demographics and clinical characteristics of the included participants. Statistical significance was defined as p < 0.05. When multiple hypothesis tests are conducted on the same dataset, methods such as Benjamini-Hochberg and False Discovery Rate were employed to adjust the p-values.

Results

The microenvironment of OSCC hosts bacteria, and the abundance of the intratumoral bacteria is significantly greater than that in adjacent normal tissue

To confirm the presence of bacteria in OSCC tissues, we designed fluorescently labeled probes to examine the bacterial 16S rRNA gene in tissue microarrays using FISH. The demographics and clinical characteristics of the patients providing the samples for the FISH analysis are shown in Table 1. The results demonstrated the presence of bacteria in both adjacent normal and OSCC tissues, with a markedly elevated total bacterial count in OSCC tissues (Fig. 1B-D). No significant difference was observed in the total bacterial counts between the male and female OSCC tissues (Fig. 1E). While the total bacterial count in OSCC tissues was elevated in both lymph node metastasis-positive and metastasis-negative patients relative to that in adjacent normal tissues, only the disparity between lymph node metastasis-positive patients and adjacent normal tissues was statistically significant (Fig. 1F). In terms of T stage, the total bacterial count in the tumor tissues of T1-T2 stage patients was significantly greater than that in T3-T4 stage patients and adjacent normal tissues; however, no significant difference was found in the total bacterial count between T3-T4 stage patients and adjacent normal tissues (Fig. 1G). The total count of intratumoral bacteria and patient age trend towards a negative correlation (p = 0.058) in OSCC patients (Fig. 1H). These findings suggest that bacteria are present in the microenvironment of OSCC, and that the abundance of total bacteria within tumors is associated with lymph node metastasis and T stage.

Table 1.

Demographics and clinical characteristics of patients who provided samples for FISH and 5R 16S rRNA-seq

	FISH			5R 16S rRNA gene sequencing
	Normal	OSCC	p-values	Normal	OSCC	p-values
	(N=17)	(N=39)		(N=86)	(N=154)
Age (Year)			0.055			0.85
Mean (SD)	48.5 (10.4)	54.5 (10.3)		59.2 (13.7)	59.5 (14.0)
Median [Min, Max]	48.0 [32.0, 73.0]	56.0 [33.0, 78.0]		59.5 [27.0, 85.0]	59.0 [27.0, 85.0]
Gender			0.317			1
Female	3 (17.6%)	2 (5.1%)		26 (30.2%)	46 (29.9%)
Male	14 (82.4%)	37 (94.9%)		60 (69.8%)	108 (70.1%)
Site			0.084			0.867
Tongue	13 (76.5%)	20 (51.3%)		33 (38.4%)	60 (39.0%)
Bucca	2 (11.8%)	7 (17.9%)		26 (30.2%)	40 (26.0%)
Floor of mouth	0 (0%)	3 (7.7%)		12 (14.0%)	19 (12.3%)
Gingiva	0 (0%)	8 (20.5%)		14 (16.3%)	32 (20.8%)
Others	2 (11.8%)	1 (2.6%)		1 (1.2%)	3 (1.9%)
Tumor size (cm)			-			-
Mean (SD)	-	4.06 (1.06)		-	3.06 (1.27)
Median [Min, Max]	-	4.00 [2.00, 7.00]		-	3.00 [0.300, 6.50]
T stage			-			-
T1	-	0 (0%)		-	18 (11.7%)
T2	-	24 (61.5%)		-	50 (32.5%)
T3	-	10 (25.6%)		-	48 (31.2%)
T4	-	4 (10.3%)		-	32 (20.8%)
Unknown		1 (2.6%)		-	0 (0%)
Nodal invasion			-			-
No	-	9 (23.1%)		-	100 (64.9%)
Yes	-	29 (74.4%)		-	54 (35.1%)
Unknown		1 (2.6%)		-	0 (0%)
Differentiation			-			-
Poor	-	-		-	11 (7.1%)
Middle	-	-		-	62 (40.3%)
Well	-	-		-	81 (52.6%)

Abbreviations: FISH, fluorescence in situ hybridization; OSCC, oral squamous cell carcinoma; -: not available

Fig. 1.

The OSCC microenvironment hosts bacteria, and the abundance of the intratumoral bacteria is significantly higher than in normal tissue. (A) Schematic diagram of the overall study design. (B) H&E staining of tissue microarrays. (C) Representative FISH images of OSCC tissues and adjacent normal tissues using a bacteria-specific 16S rRNA gene probe. (D-G) Jitter plots, displaying mean values and error bars, depicting the counts of the intratumoral bacteria detected by FISH in different groups. Each jitter represents an individual sample. p-values were determined using the Wilcoxon rank-sum test (^nsp >0.05, *p < 0.05, **p < 0.01). (H) Spearman’s rank correlation analysis assessing the relationship between total bacterial counts per field and age in OSCC patients

Significant variations exist in the bacterial composition of tumor and adjacent normal tissues in OSCC patients

To elucidate the compositional characteristics of the intratumoral bacteria in OSCC patients, paraffin-embedded tumor and adjacent normal tissues were collected for 5R 16S rRNA-seq. To minimize the impact of environmental microbes on the results, negative and blank controls were included at multiple stages, including sample storage, DNA extraction, and PCR amplification. The study included 11 paraffin control samples, 5 DNA extraction control samples, 5 PCR amplification control samples, 86 normal tissue samples, and 154 tumor samples for 5R 16S rRNA-seq. The demographics and clinical characteristics of the patients providing the samples for the 5R 16S rRNA gene analysis are shown in Table 1. Sequencing data analysis identified 60,466,079 reads, of which 52,539,243 were classified as high-quality reads (Figure S1A).

To investigate the diversity of the intratumoral bacteria, we calculated the α diversity metrics (Pielou and Shannon) at the species level. The results revealed notable differences between the OSCC and normal groups in both Pielou and Shannon metrics, suggesting a pronounced decrease in the intratumoral bacterial diversity (Fig. 2A). We also investigated β diversity in both the OSCC and normal groups using principal co-ordinates analysis (PCoA), which highlighted substantial variations in bacterial composition between the OSCC and normal tissues (Fig. 2B-C). We then used correlation analysis to build co-occurrence networks at the genus level, aiming to clarify competitive or inhibitory relationships among bacteria within each group. Significant differences were noted in the network structure and complexity of the two groups (Fig. 2D, Table S2). These findings revealed notable disparities in the overall bacterial composition between tumor tissues and adjacent normal tissues in OSCC patients.

Fig. 2.

Significant variations in the bacterial composition of tumor tissues and adjacent normal tissues in OSCC patients. (A) Alpha diversity assessed using the Pielou and Shannon indices. Statistical significance was determined using the Wilcoxon test, with p-values adjusted using the Benjamini-Hochberg method (***p < 0.001, ****p < 0.0001). (B) Principal co-ordinates analysis (PCoA) based on Bray-Curtis distance for OSCC and normal samples. (C) Boxplot showing the Bray-Curtis distance in the groups. Statistical significance was determined using the Wilcoxon test (***p < 0.001). (D) Co-occurrence networks of bacteria at the genus-level for normal and OSCC groups. Each point represents a genus, with colors indicating the corresponding phylum. Spearman’s rank correlation coefficient was used to evaluate relationships among microorganisms. Lines connecting points indicate correlations, with red lines representing positive correlations and blue lines representing negative correlations. (E) Pie chart illustrating the composition proportions of microorganisms at the phylum level in tissue groups. (F) Bubble chart depicting the average relative abundance of distinct bacterial taxa across taxonomic levels (phylum, class, order, family, and genus) for normal and OSCC groups. A heatmap displays the taxonomy, abundance changes, and corresponding p-values for the bacteria. p-values were calculated using the Wilcoxon test and adjusted with the Benjamini-Hochberg procedure. (G) Bar chart showing the results of a random forest analysis at the species level, including only bacteria with p-values less than 0.05 (left). Statistical analysis of the relative abundance of these bacteria in different groups is shown on the right. p-values were calculated using the Wilcoxon test and adjusted with the Benjamini-Hochberg procedure. * p < 0.05, ** p < 0.01, *** p < 0.001. (H) Spearman correlation heatmap generated for the top 20 species identified from the random forest analysis and adjusted with the Benjamini-Hochberg procedure. * p < 0.05, ** p < 0.01, *** p < 0.001

Subsequently, we aimed to delineate the specific differences in the bacterial composition of tumor and normal tissues. The bacterial composition within tissues not only differed between tissue groups but also varied among samples for each group (Fig. 2E, Figure S1B). Significant differences in bacterial abundance were identified between the normal and OSCC groups, involving 5 phyla, 10 classes, 12 orders, 12 families, and 13 genera (Fig. 2F, Figure S1C). Subsequently, using random forest analysis, we identified 34 bacterial species that exhibited substantial differences between the normal and OSCC groups. Notably, several anaerobic and periodontal pathogenic bacteria, e.g. Peptostreptococcus stomatis (P. stomatis), Prevotella intermedia (P. intermedia), Capnocytophaga leadbetteri (C. leadbetteri), Treponema denticola (T. denticola), and Dialister pneumosintes (D. pneumosintes), were markedly enriched in OSCC tissues. Conversely, aerobic and facultative anaerobic bacteria, e.g. Pseudomonas fragi (P. fragi), Comamonas terrigena (C. terrigena), and Aeromonas dhakensis (A. dhakensis), were significantly reduced in OSCC tissues (Fig. 2G). Following the random forest analysis results, we selected the top 20 species based on mean DecreaseGini values and conducted Spearman correlation analyses. We observed significant abundance correlations among multiple species. For instance, P. stomatis was positively correlated with Pseudomonas mendocina, P. intermedia, Prevotella tannerae, and Bulleidia moorei. The aforementioned results revealed notable differences in the bacterial composition between tumor and adjacent normal tissues among OSCC patients.

The intratumoral bacterial composition is associated with the short-term disease-free survival (DFS) in OSCC patients

Up to 86% of recurrent OSCC cases recurring within the first 2 years post-surgery and these early recurrences have been reported to have a significantly worse prognosis compared to late relapses [28]. To determine the association between the intratumoral bacteria and the short-term prognosis of OSCC patients, we conducted follow-up assessments. Among the 154 OSCC patients, 53 were lost to follow-up. The clinical and pathological data of the patients who could be followed-up are documented in Table S3. Initially, we analyzed the relationship between intratumoral bacteria at the species level and clinical parameters of DFS among OSCC patients using univariate Cox regression models. The results revealed notable correlations between Methylobacterium zatmanii (M. zatmanii), D. pneumosintes, Brevibacillus borstelensis (B. borstelensis), Gemella morbillorum (G. morbillorum), differentiation grade, tumor site, and DFS in OSCC patients (Fig. 3A). Subsequently, a multivariate COX regression analysis involving these 6 variables demonstrated that only M. zatmanii and D. pneumosintes were the sole variables significantly linked to DFS. The detection of M. zatmanii and D. pneumosintes within tumors was associated with diminished DFS in OSCC patients (Fig. 3A). Based on the presence of D. pneumosintes or M. zatmanii within tumors and clinical variables, we developed two nomograms using a multivariate Cox regression model to predict the probabilities of 1-year and 2-year DFS in OSCC patients (Fig. 3B and Figure S2A). The nomogram developed based on D. pneumosintes (1-year AUC: 0.878, 2-year AUC:0.716) showed higher AUC values for predicting 1-year and 2-year DFS of patients than those developed based on M. zatmanii (1-year AUC: 0.842, 2-year AUC:0.539) (Fig. 3C and Figure S2B). Figure 3D and Figure S2C shows the calibration curves of the models. Collectively, these results have indicated that the nomogram developed based on intratumoral D. pneumosintes could serve as a reliable prognostic model for short-term DFS in OSCC patients, however, the predictive ability of this model for long-term DFS still needs further validation.

Fig. 3.

The intratumoral bacteria have the potential to predict postoperative outcomes in OSCC patients. (A) Forest plot showing hazard ratios (HR) for bacterial and clinicopathological characteristics from both univariate and multivariate Cox regression analyses regarding disease-free survival (DFS) in OSCC patients. (B) Nomogram predicting one-year and two-year DFS probabilities for OSCC patients. (C) Receiver operating characteristic (ROC) curves evaluating the area under the curve (AUC) of the risk score derived from the nomograms for predicting one-year and two-year DFS in OSCC patients. (D) Calibration curves illustrating the prediction accuracy for one-year (left) and two-year (right) DFS outcomes in OSCC patients

We also explored the relationship between the intratumoral bacteria, clinical variables, and overall survival (OS) of OSCC patients using COX regression models. Univariate Cox regression analysis indicated that variables including differentiation degree, Prevotella shahii (P. shahii), Pseudomonas argentinensis (P. argentinensis), M. zatmanii, and Pseudomonas pseudoalcaligenes (P. pseudoalcaligenes) were associated with OS in OSCC patients. However, subsequent multivariate Cox regression analysis of these variables revealed that all aforementioned variables had p-values greater than 0.05, indicating that the intratumoral bacteria in this clinical cohort could not predict the OS of OSCC patients (Table S4). This result may be due to the limited sample size and short follow-up period in our study.

The abundance of the intratumoral D. pneumosintes correlates closely with the clinical pathological features of OSCC patients

The aforementioned results indicate a strong correlation between intratumoral D. pneumosintes and the DFS of OSCC patients. Next, we elucidated the relationship between intratumoral D. pneumosintes and the clinicopathological features of OSCC patients. The results of 5R 16S rRNA-seq revealed a markedly increased relative abundance of the intratumoral D. pneumosintes in OSCC tissues compared to that in adjacent normal tissues, accompanied by a higher occurrence of D. pneumosintes in tumor tissue samples than in adjacent normal tissue samples (Fig. 4A-B). The relative abundance of the intratumoral D. pneumosintes was significantly greater in male OSCC patients than in female patients (Fig. 4C). Furthermore, the relative abundance of intratumoral D. pneumosintes was markedly higher in both T1-T2 and T3-T4 stage OSCC patients than in adjacent normal tissues. Nevertheless, there was no significant difference in the abundance of D. pneumosintes between T1-T2 and T3-T4 stage OSCC patients (Fig. 4D). The relative abundance of intratumoral D. pneumosintes was significantly higher in OSCC patients with or without lymph node metastasis than in the adjacent normal tissues. However, there was no significant difference in the relative abundance of D. pneumosintes within tumors between OSCC patients with or without lymph node metastasis (Fig. 4E). The relative abundance of the intratumoral D. pneumosintes was significantly higher in both poorly differentiated and moderately to highly differentiated OSCC tissues than in the adjacent normal tissues. However, there was no significant difference in the relative abundance of D. pneumosintes within tumors between poorly differentiated and moderate to highly differentiated OSCC patients (Fig. 4F). The relative abundance of the intratumoral D. pneumosintes in OSCC patients showed no significant correlation with age but was significantly positively correlated with tumor size (Fig. 4G-H).

Fig. 4.

The abundance of intratumoral D. pneumosintes is closely correlated with the clinical pathological features of OSCC patients. (A) Jitter plot, displaying mean values and error bars, depicting the relative abundance of intratumoral D. pneumosintes detected by 5R 16S rRNA-seq in the normal and OSCC groups, with each point representing an individual sample. Statistical significance was assessed using the Wilcoxon rank-sum test (***p < 0.001). (B) The stacked bar chart illustrates the proportion of intratumoral D. pneumosintes-positive and -negative patients in the normal and OSCC groups. (C-F) Jitter plots, displaying mean values and error bars, depicting the relative abundance of the intratumoral D. pneumosintes detected by 5R 16S rRNA-seq in different groups, with each point representing an individual sample. Statistical significance was assessed using the Wilcoxon rank-sum test (^nsp > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001). (G) Spearman’s rank correlation analysis evaluating the relationship between the relative abundance of the intratumoral D. pneumosintes and age in OSCC patients. (H) Spearman’s rank correlation analysis evaluating the relationship between the relative abundance of the intratumoral D. pneumosintes and tumor size in OSCC patients. (I) Representative FISH images of OSCC tissues and adjacent normal tissues, using a D. pneumosintes-specific 16S rRNA gene probe. (J-M) Jitter plots, displaying mean values and error bars, depicting the counts of intratumoral D. pneumosintes per field detected by FISH in different groups, with each point representing an individual sample. Statistical significance was assessed using the Wilcoxon rank-sum test (^nsp > 0.05, *p < 0.05, **p < 0.01). (N) Spearman’s rank correlation analysis evaluating the relationship between the counts of the intratumoral D. pneumosintes per field and age in OSCC patients

In addition, specific fluorescence-labeled probes were synthesized to target the 16S rRNA gene of D. pneumosintes in tissue microarrays using FISH. The FISH results demonstrated a markedly higher abundance of D. pneumosintes in OSCC tissues than in adjacent normal tissues (Fig. 4I-J). Although the FISH results also indicated that the abundance of the intratumoral D. pneumosintes was higher in male patients than in female patients, the difference was not statistically significant (Fig. 4K). This might be attributed to the smaller number of female patient samples in the tissue microarray. In subgroup analyses addressing clinical variables such as nodal invasion, T stage, and age, the results of FISH were consistent with those obtained from 5R 16S rRNA-seq (Fig. 4L-N). These findings have confirmed that the abundance of the intratumoral D. pneumosintes correlates closely with the clinical pathological features of OSCC patients, suggesting a crucial role of the intratumoral D. pneumosintes in the progression of OSCC.

Somatic mutational landscape of OSCC patients

Somatic mutational signatures in tumors exhibit variation across ethnic groups and geographic regions. To elucidate the somatic mutational signatures of OSCC patients in Southwest China, WES was performed on 111 OSCC samples (of the 154 OSCC DNA samples used for 5R 16S rRNA-seq, the quality of only 111 OSCC DNA samples met the requirements for WES). The demographics and clinical characteristics of the patients providing the samples for the WES is shown in Table 2 and Fig S3A. Oncoplot analysis revealed that missense mutations were the most prevalent type of mutation, and the 10 most frequently mutated genes were: MUC5B, TCHH, TTN, PRG4, MUC12, PRR36, FAM186A, CYP4F2, ATAD3B, and IFITM3 (Fig. 5A). Among the 30 most mutated genes in our cohort, TP53, TTN, and FLG were also frequently mutated in The Cancer Genome Atlas (TCGA) head and neck squamous cell carcinoma (HNSCC) cohort (Fig. 5B, Figure S3B). In addition, among the 30 most frequently mutated genes in our cohort, only TP53 and KMT2C were documented as HNSCC driver genes in the IntOGen database (Fig. 5C, Figure S4) [29]. Then, based on the WES results, we calculated the TMB for each sample, revealing a median TMB of 10.62 mutations per megabase (muts/Mb) and a range of 6.64–57.58 muts/Mb (Fig. 5A). Employing a threshold of TMB ≥ 10 muts/Mb to define hypermutated samples [30, 31], we noted a significantly higher hypermutation rate of 61% (68 out of 111) in our cohort compared to the TCGA HNSCC cohort (Fig. 5D).

Table 2.

Demographics and clinical characteristics of patients who provided samples for WES

	Hypermutated	Non-hypermutated	Total
	(N=68)	(N=43)	(N=111)
Age (Years)
Mean (SD)	59.9 (14.2)	55.7 (13.9)	58.3 (14.2)
Median [Min, Max]	59.5 [27.0, 85.0]	54.0 [29.0, 83.0]	58.0 [27.0, 85.0]
Gender
Female	20 (29.4%)	15 (34.9%)	35 (31.5%)
Male	48 (70.6%)	28 (65.1%)	76 (68.5%)
Tumor site
Bucca	21 (30.9%)	10 (23.3%)	31 (27.9%)
Floor of mouth	11 (16.2%)	3 (7.0%)	14 (12.6%)
Gingiva	12 (17.6%)	10 (23.3%)	22 (19.8%)
Tongue	24 (35.3%)	20 (46.5%)	44 (39.6%)
Tumor size (cm)
Mean (SD)	3.25 (1.17)	3.20 (1.19)	3.23 (1.18)
Median [Min, Max]	3.00 [1.00, 6.00]	3.00 [1.00, 5.00]	3.00 [1.00, 6.00]
T stage
T1	8 (11.8%)	5 (11.6%)	13 (11.7%)
T2	28 (41.2%)	13 (30.2%)	41 (36.9%)
T3	24 (35.3%)	11 (25.6%)	35 (31.5%)
T4	8 (11.8%)	14 (32.6%)	22 (19.8%)
Nodal invasion
No	34 (50.0%)	30 (69.8%)	64 (57.7%)
Yes	34 (50.0%)	13 (30.2%)	47 (42.3%)
Differentiation
Middle	31 (45.6%)	17 (39.5%)	48 (43.2%)
Poor	7 (10.3%)	3 (7.0%)	10 (9.0%)
Well	30 (44.1%)	23 (53.5%)	53 (47.7%)

Fig. 5.

Somatic mutational landscape of OSCC patients. (A) Heatmap of mutations with multiple tracks presenting the different mutation profiles for mutated genes among OSCC patients. (B) Venn diagram showing the overlap between the 30 most frequently mutated genes in this study and the 30 most frequently mutated genes in the TCGA HNSCC cohort. (C) Venn diagram showing the overlap between the 30 most frequently mutated genes in this study and the genes that drive HNSCC reported in the IntOGen database. (D) Stacked bar chart showing the proportion of hypermutated and non-hypermutated samples in this study and in the TCGA HNSCC dataset. Individuals exhibiting TMB ≥ 10 muts/Mb were classified as hypermutated, whereas those with TMB < 10 muts/Mb were categorized as non-hypermutated

TMB is correlated with the intratumoral bacterial composition in OSCC patients

Certain intratumoral bacteria can induce DNA damage directly, which subsequently escalates mutation rates, potentially culminating in tumorigenesis [15]. To elucidate the association between the intratumoral bacterial composition and mutational signatures in OSCC patients, we integrated the 5R 16S rRNA-seq data with WES data to conduct a comprehensive analysis. As identified above, TP53, TTN, FLG, and KMT2C were the most frequently mutated genes identified in both our cohort and the publicly available cohorts. Based on the presence or absence of mutations in these genes, we categorized OSCC patients into mutated and non-mutated groups. Our analysis revealed that bacterial α diversity indices such as Shannon and Pielou indices did not exhibit significant differences between the mutated and non-mutated groups (Figure S5). Notably, however, we found that both the Shannon and Pielou indices were significantly positively correlated with TMB (Fig. 6A). In addition, the Shannon and Pielou indices were significantly higher for the hypermutated group than for the non-hypermutated group (Fig. 6B). We then compared the overall composition of the intratumoral bacteria in the hypermutated and non-hypermutated groups using β diversity analysis. PCoA using the Bray-Curtis distance revealed distinct clusters for the hypermutated and non-hypermutated groups (Fig. 6C-D). Subsequently, network analysis was used to study microbial co-occurrence patterns. The analysis revealed distinct co-occurrence patterns for the hypermutated and non-hypermutated groups (Fig. 6E). These results suggested a significant difference in the overall composition of the intratumoral bacteria between the hypermutated and non-hypermutated groups.

Fig. 6.

TMB is significantly correlated with the intratumoral bacterial composition. (A) Analysis of the correlation between tumor mutation burden and alpha diversity indices (Pielou and Shannon) calculated using the Spearman method. (B) Alpha diversity assessed for the groups using the Pielou and Shannon indices. Statistical significance was determined using the Wilcoxon test, with P-values adjusted using the Benjamini-Hochberg method (*p < 0.05, **p < 0.01). (C) Principal co-ordinates analysis (PCoA) based on the Bray-Curtis distance. (D) Boxplot showing the Bray-Curtis distance for the groups. Statistical significance was determined using the Wilcoxon test (***p < 0.001). (E) Co-occurrence networks of species-level bacteria for hypermutated and non-hypermutated groups. Each point represents a species, with colors indicating the corresponding phylum. Spearman’s rank correlation coefficient was used to evaluate relationships among bacteria. Lines connecting points indicate correlations, with red lines representing positive correlations and blue lines representing negative correlations. (F) LDA scores for the intratumoral bacteria with significantly differential abundancies between the hypermutated and non-hypermutated groups (left). Statistical analysis of the relative abundance of these bacteria between the groups is shown on the right. P-values were calculated using the Wilcoxon test and adjusted with the Benjamini-Hochberg procedure (**p < 0.01, ***p < 0.001). (G) Cladogram generated by LEfSe showing the differences in the intratumoral bacteria between the hypermutated and non-hypermutated groups. The cladogram was generated with the 200 most abundant taxa and 50 differential features

In subsequent analyses, we investigated the differences in the intratumoral bacteria between the hypermutated and non-hypermutated groups. Figure S6 depicts the 15 most abundant intratumoral bacteria at the phylum, class, order, family, genus, and species levels in both groups. To further identify the bacteria with differential abundance between the hypermutated and non-hypermutated groups, we performed linear discriminant analysis (LDA) of effect size (LEfSe). The LEfSe analysis revealed 42 bacterial classifications that discriminate between the hypermutated and non-hypermutated groups (Fig. 6F). Figure 6G depicts a cladogram of these differentiating bacteria in a taxonomic tree. These results revealed that there were significant differences between the intratumoral bacteria in the hypermutated and non-hypermutated OSCC patients.

Discussion

Recent studies have confirmed the presence of bacteria within the microenvironment of OSCC, with the intratumoral bacteria impacting OSCC progression via mechanisms such as immune modulation and metabolic regulation [28]. Nevertheless, the precise composition of the intratumoral bacteria and its prognostic implications in OSCC patients are still unclear and no studies have yet investigated the association between the composition of intratumoral bacteria and the somatic mutational signatures. This study employed 5R 16S rRNA-seq and FISH on tumor tissues and adjacent normal tissues from OSCC patients to elucidate the compositional features of the intratumoral bacteria and reveal their predictive potential for the short-term prognosis of OSCC patients. In addition, OSCC tissues from the same patients underwent 5R 16S rRNA-seq and WES, followed by integrative analysis of the sequencing data to elucidate the relationship between the composition of intratumoral bacteria and the somatic mutational signatures. Our study provides a crucial theoretical foundation for comprehending the role of intratumoral bacteria in the initiation and progression of OSCC.

Previous studies have detected the presence of bacteria in OSCC tissues using RNAscope-FISH imaging, revealing substantial heterogeneity in the intratumoral bacterial composition between patients [18]. Our study confirmed the presence of bacteria in OSCC tissues, using FISH. Compared with adjacent normal tissues, OSCC tissues displayed a significant increase in total bacterial abundance. Furthermore, the abundance of the intratumoral bacteria was significantly elevated in OSCC patients with lymph node metastasis and those classified as T1-T2 stage. In addition, 5R 16S rRNA-seq was employed to analyze the bacterial composition of tumors and adjacent normal tissues. We observed considerable inter-individual variability in the intratumoral bacterial composition among OSCC patients and a markedly lower bacterial diversity within OSCC tissues than in normal tissues, indicating a pronounced enrichment of specific bacteria in the OSCC microenvironment. A previous study using conventional 16S rRNA-seq also reported a reduction in bacterial diversity within OSCC tumors [21]. Further analysis identified multiple distinct bacteria at different taxonomic levels, including phyla, classes, orders, families, genera, and species that were differentially abundant in the two tissue types. Notably, OSCC tissues showed a significant increase in the relative abundance of bacteria that are obligate or facultative anaerobes (e.g., P. stomatis, D. pneumosintes, and G. morbillorum). Conversely, the numbers of aerobic and facultative anaerobic bacteria (e.g., P. fragi, C. terrigena, and A. dhakensis) in tumors were significantly reduced. Many solid tumors, including OSCC, experience severe hypoxia due to an imbalance caused by accelerated oxygen consumption associated with cancer cell proliferation and inadequate oxygen delivery resulting from a disorganized tumor vasculature [32, 33]. The hypoxic environment in OSCC might promote the colonization and growth of anaerobic bacteria. Therefore, alleviating hypoxia in the tumor microenvironment may offer a potential strategy to restore microbial homeostasis within tumors. Furthermore, we observed that OSCC tissues exhibited significant enrichment of periodontal pathogens [34], including P. intermedia, T. denticola, P. stomatis, D. pneumosintes, and C. gracilis. Among these, positivity for intratumoral D. pneumosintes was associated with the poor short-term DFS in OSCC patients. Periodontitis is independently associated with OSCC, and patients with periodontitis have a 3.7-fold higher likelihood of OSCC than those without [35]. In addition, the oral microbiota associated with periodontitis can facilitate the development of OSCC through γδ T-cell activation [36]. D. pneumosintes bacteria are non-spore-forming, non-motile, non-fermentative, Gram-negative anaerobic rods found in the nasopharynx, oral cavity, intestine, and vagina. Currently recognized as a periodontal pathogen, D. pneumosintes is significantly enriched in the subgingival plaque of periodontitis patients [37]. Moreover, its abundance increases in tissues of proliferative verrucous leukoplakia patients, gastric cancer patients, colorectal cancer patients, the subgingival plaque of Alzheimer’s disease patients, and in non-alcoholic fatty liver disease patients [38–42]. Our findings indicate that D. pneumosintes not only increase in abundance in OSCC tissues, but the abundance also correlates with tumor size, T stage, and differentiation degree. Taken together, our results revealed that anaerobic bacteria and periodontal pathogens are significantly enriched in OSCC tissues, providing new insights into the diagnosis and treatment of OSCC.

Bacteria can induce DNA damage [43]. DNA damage results in heightened mutation rates, ultimately fostering tumorigenesis. In various forms of cancer, notably gastrointestinal malignancies, bacteria play a significant role in promoting DNA damage. Bacteria can induce oxidative/nitrosative DNA damage, thereby promoting tumorigenesis, either directly or indirectly [15]. TMB indicates the quantity of non-synonymous mutations within the tumor genome. Tumors exhibiting a high TMB generally contain a larger number of mutations, producing a greater abundance of neoantigens, which may enhance their susceptibility to immune system recognition and eradication [9]. Our findings revealed a significant positive correlation between TMB and the diversity of the intratumoral bacteria in OSCC patients. In addition, striking differences were observed in the composition of the intratumoral bacteria between patients exhibiting a high frequency of mutations and those with lower rates of mutations. Specifically, Actinobacteria were predominantly enriched in the tumors of hypermutated patients, whereas Proteobacteria were substantially enriched in the tumors of non-hypermutated patients. In 2020, the U.S. FDA sanctioned the anti-PD-1 therapy pembrolizumab for the management of hypermutated cancer patients (TMB ≥ 10 mut/Mb) [9]. Our findings demonstrate a significant positive correlation between TMB and intratumoral bacterial diversity. Future investigations are warranted to determine whether augmenting intratumoral microbiota diversity, e.g., via probiotic supplementation, could potentially enhance the efficacy of immune checkpoint inhibitors in OSCC. Nonetheless, only few studies have documented the correlation between the intratumoral bacteria and the efficacy of immune checkpoint inhibitors. Gao et al. demonstrated that intratumoral Fusobacterium nucleatum (F. nucleatum) could potentiate the efficacy of PD-L1 blockade in colorectal carcinoma [44]. Subsequently, Wang et al. revealed that intratumoral F. nucleatum could facilitate PD-1 blockade in microsatellite-stable colorectal carcinoma [45]. The abundance of intratumoral Streptococcus bacteria might predict prolonged disease-free survival in esophageal squamous cell carcinoma patients undergoing neoadjuvant chemoimmunotherapy, as the abundance of intratumoral Streptococcus positively correlated with GrzB + and CD8 + T cell infiltration [46]. In a mouse model of melanoma, dietary tryptophan metabolites released from intratumoral Lactobacillus reuteri enhanced the efficacy of immune checkpoint inhibitor therapy [47]. These investigations have demonstrated that intratumoral bacteria can influence the efficacy of immune checkpoint inhibitors in various tumor types, including colorectal carcinoma, esophageal squamous cell carcinoma, and melanoma. However, whether the intratumoral bacteria could influence the efficacy of immune checkpoint inhibitors in OSCC requires further investigation.

Conclusion

In summary, our study revealed the intratumoral bacterial composition in OSCC, and found a reduced diversity and significant enrichment of anaerobic and periodontal pathogens. In addition, we identified the intratumoral D. pneumosintes is associated with poorer short-term DFS in OSCC patients. Furthermore, our study systematically elucidated the correlation between the intratumoral bacterial composition and somatic mutational signatures in OSCC patients through the integration of multi-omics data. Notably, we observed a positive correlation between the diversity of the intratumoral bacteria and TMB, implying that enhancing the intratumoral bacterial diversity may augment the efficacy of immune checkpoint inhibitors. Despite the valuable insights our study provides, several limitations must be acknowledged. First, the retrospective design with a limited follow-up period restricts the ability to draw conclusions about long-term outcomes. Second, the reliance on public databases rather than matched normal samples from the same individuals may introduce inaccuracies in excluding germline variants. Lastly, our study is a multi-omics-based descriptive study and the findings offer only preliminary evidence regarding the association between intratumoral bacteria and mutational signatures. Future large-scale, multicenter, prospective studies are needed to validate our findings, and the oncogenic mechanisms require elucidation through molecular experimentation.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

OSCC

Oral squamous cell carcinoma

5R 16S rRNA-seq

Five-region 16S rRNA gene sequencing

WES

Whole exome sequencing

TMB

Tumor mutational burden

FFPE

Formalin fixed paraffin embedded

H&E

Hematoxylin and eosin

FISH

Fluorescence in situ hybridization

DAPI

4′,6-Diamidino-2-phenylindole

PCoA

Principal co-ordinates analysis

DFS

Disease-free survival

OS

Overall survival

TCGA

The Cancer Genome Atlas

HNSCC

Head and neck squamous cell carcinoma

LDA

Linear discriminant analysis

LEfSe

Linear discriminant analysis of effect size

P. stomatis

Peptostreptococcus stomatis

P. intermedia

Prevotella intermedia

C. leadbetteri

Capnocytophaga leadbetteri

T. denticola

Treponema denticola

D. pneumosintes

Dialister pneumosintes

P. fragi

Pseudomonas fragi

C. terrigena

Comamonas terrigena

A. dhakensis

Aeromonas dhakensis

M. zatmanii

Methylobacterium zatmanii

B. borstelensis

Brevibacillus borstelensis

G. morbillorum

Gemella morbillorum

Author contributions

The author contributions for this manuscript are as follows: Y.D. contributed to writing - original draft, funding acquisition, data curation, formal analysis, and visualization. M.Q. contributed to data curation, software, and formal analysis. Y.Z. contributed to methodology and validation. B.Y. contributed to investigation and project administration. H.Z. contributed to data curation. D.X. contributed to data curation and resources. Q.L. contributed to data curation, and formal analysis. R.D. C. contributed to writing - review & editing. P.J. contributed to conceptualization and supervision. X.J. contributed to conceptualization, funding acquisition, writing - review & editing, and supervision.

Funding

This work was supported by grants from the Chongqing medical scientific research project (Joint project of the Chongqing Health Commission and the Science and Technology Bureau) (No. 2025QNXM048), the China Postdoctoral Science Foundation (No. 2024MD754038), the National Natural Science Foundations of China (No. 82370968) and the Natural Science Foundation of Chongqing (No. CSTB2022NSCQ-MSX1148). The funding agencies had no role in the study design, collection, analysis, or interpretation of data, writing of the report, or the decision to submit the article for publication.

Data availability

The raw sequence data of 5R 16S rRNA-seq and WES reported in this paper have been deposited in the Genome Sequence Archive in National Genomics Data Center, China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA-Human: HRA008052 and HRA010901) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human/s/YaQdlb2c and https://ngdc.cncb.ac.cn/gsa-human/s/A92L11WQ.

Declarations

Ethics approval and patient consent statement

Our study conformed to the principles of the Declaration of Helsinki and was approved by the institutional ethical review boards of the Stomatological Hospital of Chongqing Medical University (Ethics No. 2024-087) and the Second Affiliated Hospital of Chongqing Medical University (Ethics No. 2024-064) for anonymous data analysis, and informed consent was waived for this retrospective study.

Consent for publication

Not applicable.

Competing interests

All authors included in this research declare no competing interests.