Microbial dysbiosis and treatment-related gastrointestinal adverse events in locoregionally advanced nasopharyngeal carcinoma: an exploratory study of the oral and gut microbiome

Abstract

Background

Gastrointestinal adverse events are frequent during treatment for locoregionally advanced nasopharyngeal carcinoma (LA-NPC) and can impair nutritional status and treatment tolerance. Radiotherapy and chemotherapy may disrupt microbial communities, but whether oral and gut microbiota are associated with treatment-related gastrointestinal adverse events remains unclear.

Materials and methods

Oral swab and stool samples from 80 patients with LA-NPC were analyzed using 16S rRNA gene sequencing. Associations between treatment-related gastrointestinal adverse events and oral microbial composition were evaluated, together with analyses of microbial functions and gut microbial profiles. Oral microbiota features were then used to develop machine learning models for distinguishing patients with gastrointestinal adverse events.

Results

Among 80 patients with LA-NPC, nausea, vomiting, diarrhea, and constipation occurred in 72.5%, 51.3%, 46.3%, and 67.5%, respectively. Oral microbiota analysis showed enrichment of Alloprevotella, Fusobacterium, and Campylobacter in patients with gastrointestinal adverse events or a high symptom burden, whereas Rothia and Streptococcus were enriched in those without events or with a low burden. Predicted oral microbial functions involved propanoate metabolism, lipoic acid metabolism, retinol metabolism, glycan degradation, sphingolipid metabolism, calcium signaling, and lipopolysaccharide biosynthesis. Alterations in specific gut genera, including Helicobacter, Parasutterella, and Anaerotruncus, were also observed.

Conclusions

This study identified oral and gut microbial features associated with treatment-related gastrointestinal adverse events in patients with LA-NPC. Oral microbial features showed potential value for classifying gastrointestinal adverse event status during treatment.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12885-026-16631-w.

Introduction

Although induction chemotherapy, adjuvant therapy, and immunotherapy have improved disease control in locoregionally advanced nasopharyngeal carcinoma (LA-NPC), treatment-related adverse events continue to complicate management throughout treatment [1–3]. Gastrointestinal adverse events, including nausea, vomiting, diarrhea, and constipation, are common during treatment [4–6]. When persistent, these symptoms may reduce food intake and adherence to oral nutritional supplementation, and increase the likelihood of weight loss, malnutrition, and poor treatment tolerance [7, 8]. Identifying patients with a high gastrointestinal symptom burden and tailoring supportive care accordingly may therefore be important in the comprehensive management of LA-NPC.

Treatment-related gastrointestinal adverse events involve multiple mechanisms. Chemotherapy-induced serotonin release from enterochromaffin cells, neurokinin signaling, mucosal inflammation, barrier injury, motility disturbance, and microbial dysbiosis have been implicated in nausea, vomiting, diarrhea, and constipation [9–11]. Management remains largely symptomatic, relying on antiemetics, hydration, laxatives, antidiarrheal agents, and nutritional support, while tools for identifying patients with a high gastrointestinal symptom burden and tailoring supportive care remain limited [12–15]. Recent studies have linked oral and gut microbiota to cancer treatment-related toxicities [16, 17]. Therefore, microbial markers of gastrointestinal adverse events in LA-NPC may help refine patient stratification and supportive care.

Recent studies have linked oral microbial dysbiosis to several digestive diseases, including upper gastrointestinal disorders and inflammatory bowel disease [18, 19]. Further evidence indicates that oral-derived bacteria can translocate to the gut under conditions of barrier disruption or inflammation, where they may reshape gut microbial communities and affect mucosal barrier function and immune homeostasis [20–22]. These findings provide a plausible biological basis for the development of gastrointestinal symptoms. In cancer treatment, radiotherapy and chemotherapy can disrupt oral and gut microbial communities and may contribute to inflammation and barrier injury [23–25]. In patients with LA-NPC, treatment-related disruption of oral and gut microbiota may be associated with gastrointestinal adverse events. However, the relationship between these microbial communities and gastrointestinal adverse events during LA-NPC treatment remain limited. Whether oral microbiota reflect gastrointestinal adverse event burden, and whether concordant changes occur in the gut microbiota, remains unclear.

Here, we collected oral swab and stool samples from patients with LA-NPC and profiled the microbiota using 16S rRNA gene sequencing. We examined associations between oral and gut microbial features and treatment-related gastrointestinal adverse events, and developed oral microbiota-based machine learning models to classify gastrointestinal adverse event status during treatment.

Materials and methods

Patient eligibility and sample collection

The study protocol was approved by the Ethics Committee of the First Affiliated Hospital of Guangxi Medical University. Eligible patients were 18–70 years of age; were men or non-pregnant women; had pathologically confirmed non-keratinizing nasopharyngeal carcinoma; had stage III–IVA disease according to the 8th edition of the AJCC/UICC staging system; had an ECOG performance status of 0 or 1; and were able to complete the planned treatment and on-treatment assessments. Patients were excluded if they had pre-existing oral or gastrointestinal diseases; acquired immunodeficiency syndrome, active hepatitis B, or active hepatitis C; poorly controlled blood glucose, autoimmune disease, active infection, or a history of other malignancies; major organ dysfunction; or use of antibiotics, glucocorticoids, probiotics, prebiotics, synbiotics, or other microbiota-modulating agents within the previous 3 months. Written informed consent was obtained from all participants or their legally authorized representatives.

Oral swab and stool samples were collected within 2 days before or after the final radiotherapy session. Participants rinsed their mouth with sterile saline 30 min before oral sampling. Sterile oral swabs were used to collect samples from the soft palate and bilateral buccal mucosa, and were then placed in sterile sampling tubes. For stool sampling, the central portion of the stool was collected into sterile stool collection tubes, avoiding urine contamination. Stool samples were aliquoted into 2 mL microcentrifuge tubes, with approximately 0.5 g per tube. All samples were transported to the laboratory immediately after collection and stored at − 80 °C. Six environmental blank controls were processed in parallel and included in downstream analyses to monitor potential contamination.

Treatment and assessment of gastrointestinal adverse events

Patients received induction chemotherapy followed by concurrent chemoradiotherapy. Induction chemotherapy comprised gemcitabine 1000 mg/m² intravenously on days 1 and 8 plus cisplatin 80 mg/m² intravenously on day 1, every 3 weeks for 2–3 cycles. Concurrent chemotherapy comprised cisplatin 100 mg/m² intravenously on day 1, every 3 weeks for 2–3 cycles. All patients received definitive intensity-modulated radiotherapy at a total dose of 66–74 Gy, delivered once daily, 5 days per week. Gastrointestinal adverse events, including nausea, vomiting, diarrhea, and constipation, were assessed during treatment according to the Common Terminology Criteria for Adverse Events version 5.0 [26]. Events were independently evaluated and recorded by two radiation oncologists experienced in the management of head and neck radiotherapy-related toxicities.

16S rRNA sequencing

DNA extraction and PCR amplification

Genomic DNA was extracted using the MagPureSoil DNA LQ Kit (Magen) according to the manufacturer’s instructions. DNA quantity and quality were assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA) and agarose gel electrophoresis. DNA samples were stored at − 20 °C until further use. The V3–V4 region of the bacterial 16 S rRNA gene was amplified using barcoded primers 343 F (5′-TACGGRAGGCAGCAG-3′) and 798R (5′-AGGGTATCTAATCCT-3′), together with Takara ExTaq DNA polymerase.

Library preparation and sequencing

PCR products were checked by agarose gel electrophoresis and purified using AMPure XP beads. The purified products were used as templates for a second round of PCR amplification. Amplicons were purified again using magnetic beads and quantified with Qubit. Concentrations were adjusted prior to sequencing. Paired-end sequencing was conducted on the Illumina NovaSeq 6000 platform by Shanghai OE Biotech. Co., Ltd.

Microbial diversity analysis

Raw sequencing data were obtained in FASTQ format. Cutadapt was first used to process the rawdata and generate clean rawdata. Paired-end rawdata were quality-filtered, denoised, merged, and chimera-filtered using the DADA2 algorithm, with quality control performed in QIIME2 (2020.11) using default parameters. Representative sequences and an ASV abundance table were generated. Representative sequences for each ASV were selected in QIIME2 and aligned against the Silva (version 138) database for taxonomic annotation. Taxonomic assignment was performed using the q2-feature-classifier tool with default parameters. Alpha and beta diversity analyses were performed in QIIME2. The ACE, Chao1, Goods coverage, observed OTUs, Shannon, and Simpson indices were used to evaluate α diversity. β diversity was assessed using Bray-Curtis and Jaccard distance matrices. Principal coordinate analysis (PCoA) and permutational multivariate analysis of variance (PERMANOVA) were then performed to evaluate between-group differences.

Differential microbial analysis

Linear discriminant analysis effect size (LEfSe) was used to identify microbial taxa that contributed significantly to between-group differences. Briefly, features with p < 0.05 were retained for Linear discriminant analysis (LDA), and |LDA score|>2 was set as the threshold for discriminative features. Relative abundances were log10-transformed, and LDA scores were used to estimate the contribution of each differential taxon to group separation.

Microbial functional prediction analysis

Based on the 16S rRNA sequencing data, Phylogenetic investigations of Communities by Reconstruction of Unobserved States II (PICRUSt2) software v2.3.0 was used to predict the potential microbial functions in patients with LA-NPC. KEGG pathway annotation was then performed, and functional pathway differences between groups were further compared.

Machine learning

Microbial taxa associated with gastrointestinal adverse events were used to develop regularized logistic regression (RegLogit) and support vector machine with radial basis function (SVM-RBF) models using the glmnet and e1071 packages, respectively. The final LA-NPC cohort was randomly split 1:1 into training and validation sets. Model performance was evaluated in both sets using the area under the curve (AUC) of the receiver operating characteristic (ROC) curve and confusion matrices.

Oral to gut microbial transfer analysis

Potential oral to gut microbial transfer was assessed using SourceTracker analysis and ASV share analysis. In SourceTracker analysis, oral samples were defined as potential source communities and stool samples as sink communities; the ASV abundance matrix was used to estimate the relative contribution of oral-derived microbiota to each stool sample [27]. ASV share analysis identified identical ASVs shared by paired oral and stool samples within each participant. The transfer ratio of major genera was then calculated. This metric was defined as the proportion of patients with a given genus detected in oral samples who also had identical ASVs of the same genus detected in paired stool samples, reflecting the transfer tendency of different oral taxa to the gut [28].

Statistical analysis

All statistical analyses were performed using R v4.2.2 and SPSS v23.0. All P-values were two-sided, and P-values < 0.05 were considered statistically significant. Continuous variables were summarized as mean ± standard deviation, mean ± standard error of the mean, or median (interquartile range), as appropriate. Between-group comparisons of continuous variables were performed using the independent-samples t test or Wilcoxon rank-sum test, as appropriate. Other statistical methods are described in the corresponding sections above.

Results

Incidence of gastrointestinal adverse events and baseline characteristics of patients with LA-NPC

After exclusions for sample contamination, insufficient sample DNA, failure to provide samples at the prespecified time point, or declined gastrointestinal adverse event assessment during treatment, 80 patients with LA-NPC were included in the final analysis. Oral swab samples were collected for 16S rRNA gene sequencing. During CCRT, nausea, vomiting, diarrhea, and constipation occurred in 58, 41, 37, and 54 patients, respectively. The incidences were 72.5%, 51.3%, 46.3%, and 67.5%, respectively (Table 1). Baseline characteristics, including age, sex, BMI, WHO pathological type, T stage, N stage, clinical stage, plasma EBV DNA, smoking history, and drinking history, are summarized in Table 1.

Table 1.

Baseline information and incidence of adverse gastrointestinal reactions of patients with LA-NPC

Clinical features	LA-NPC(n = 80)
Age, years old, mean ± SD	46.44 ± 11.50
BMI, kg/m², mean ± SD	23.76 ± 4.16
Gender, n (%)
Male	66(82.5%)
Female	14(17.5%)
WHO pathological type, n (%)
Non-keratinizing differentiated type	1(1.2%)
Non-keratinizing undifferentiated type	79(98.8%)
T stage, n (%)
T2	8(10.0%)
T3	24(30.0%)
T4	48(60.0%)
N stage, n (%)
N0	4(5.0%)
N1	17(21.3%)
N2	28(35.0%)
N3	31(38.7%)
Clinical stage, n (%)
Stage III	22(27.5%)
Stage IVa	58(72.5%)
Plasma EBV DNA, n (%)
Positive	29(36.3%)
Negative	51(63.7%)
Smoking, n (%)
Yes	46(57.5%)
No	34(42.5%)
Alcohol consumption, n (%)
Yes	36(45.0%)
No	44(55.0%)
Nausea, n (%)
Yes	58(72.5%)
No	22(27.5%)
Vomiting, n (%)
Yes	41(51.3%)
No	39(48.7%)
Diarrhea, n (%)
Yes	37(46.3%)
No	43(53.7%)
Constipation, n (%)
Yes	54(67.5%)
No	26(32.5%)

Comparison of oral microbial diversity between patients with and without gastrointestinal adverse events

α and β diversity analyses were used to evaluate differences in oral microbial diversity according to gastrointestinal adverse events during CCRT in patients with LA-NPC. α diversity analysis showed no significant between-group differences in ACE, Chao1, Goods coverage, observed ASVs, Shannon, or Simpson indices across nausea, vomiting, diarrhea, or constipation groups (Fig. 1). In the β-diversity analysis, PERMANOVA showed modest, statistically significant differences in oral microbiota composition between patients with and without nausea based on both binary Jaccard distance (p < 0.05) and Bray–Curtis distance (p < 0.05) (Fig. 2A-B). For constipation, a weak difference was detected using binary Jaccard distance (p < 0.05), whereas no significant difference was observed using Bray–Curtis distance (p > 0.05) (Fig. 2A-B). No significant β-diversity differences were observed for vomiting or diarrhea (p > 0.05) (Fig. 2A-B).

Fig. 1.

Oral microbiota α-diversity indices according to gastrointestinal symptoms during CCRT in LA-NPC patients. Boxplots with overlaid black dots show α-diversity distributions in patients with (YES) or without (NO) nausea, vomiting, diarrhea, and constipation. Each dot represents one sample. P values were calculated using the two-sided Wilcoxon rank-sum test

Fig. 2.

Oral microbiota β-diversity in LA-NPC patients with or without gastrointestinal symptoms during CCRT. A Binary Jaccard distance. B Bray-Curtis distance. PCoA plots compare patients with (YES) or without (NO) nausea, vomiting, diarrhea, and constipation. Each point represents one sample. P values were calculated using PERMANOVA

Identification of differential oral microbiota associated with gastrointestinal adverse events

We next compared the dominant microbial taxa between patients with and without each gastrointestinal adverse event. The relative abundances of dominant taxa showed differences across nausea, vomiting, diarrhea, and constipation groups at the genus (Fig. 3A) levels. LEfSe analysis was used to further identify differential oral microbiota associated with gastrointestinal adverse events (Fig. 3B and Figure S1). At the genus level, Prevotella, Alloprevotella, Fusobacterium, and Campylobacter were enriched in patients with nausea, whereas Rothia was enriched in those without nausea. Alloprevotella and Fusobacterium were enriched in patients with vomiting, whereas Rothia was enriched in those without vomiting. Alloprevotella was enriched in patients with diarrhea, whereas Capnocytophaga and Rothia were enriched in those without diarrhea. Haemophilus and Alloprevotella were enriched in patients with constipation, whereas Granulicatella, Cloacibacterium, and Pseudomonas were enriched in those without constipation. Similar patterns were observed in the Wilcoxon rank-sum test comparing the relative abundance of oral genera between patients with and without each gastrointestinal adverse event (Fig. 4).

Fig. 3.

Relative abundance profiles and LEfSe analysis of oral microbial taxa in LA-NPC patients with or without gastrointestinal symptoms during CCRT. A Stacked bar plots show the relative abundance of the dominant oral microbial genera in patients with (YES) or without (NO) nausea, vomiting, diarrhea, and constipation. B LEfSe analysis showing the microbial taxa that differed between patients with (YES) or without (NO) each gastrointestinal symptom. Taxa enriched in patients with the corresponding gastrointestinal symptom are shown in red, whereas taxa enriched in patients without the corresponding symptom are shown in green. The LDA score indicates the effect size of each differentially abundant taxon. Only taxa with an |LDA score| > 2.0 and p < 0.05 were displayed

Fig. 4.

Top 10 differentially abundant oral genera associated with gastrointestinal symptoms during CCRT in LA-NPC patients. The top 10 oral genera were compared between patients with (YES) or without (NO) nausea, vomiting, diarrhea, and constipation using the Wilcoxon rank-sum test. Statistical significance: *p < 0.05, **p < 0.01, ***p < 0.001

Functional prediction analysis of oral microbiota associated with gastrointestinal adverse events

PICRUSt2 analysis was performed to predict the functional potential of the oral microbiota in relation to gastrointestinal adverse events. KEGG enrichment analysis showed differences in predicted oral microbial functional profiles across gastrointestinal adverse event status (Fig. 5). For nausea, predicted functions enriched in patients without nausea included Valine, leucine and isoleucine biosynthesis, Two-component system, Lipoic acid metabolism, Propanoate metabolism, Quorum sensing, and Phosphotransferase system (PTS) (p < 0.05) (Fig. 5). For vomiting, predicted functions enriched in patients without vomiting included Retinol metabolism, Chloroalkane and chloroalkene degradation, Penicillin and cephalosporin biosynthesis, Sesquiterpenoid and triterpenoid biosynthesis, Chlorocyclohexane and chlorobenzene degradation, and Photosynthesis - antenna proteins(p < 0.05) (Fig. 5). Diarrhea-related pathways were mainly enriched in patients with diarrhea and involved Other glycan degradation, Sphingolipid metabolism, Glycosaminoglycan degradation, Lysosome, Zeatin biosynthesis, and Glycosphingolipid biosynthesis - globo and isoglobo series (p < 0.05) (Fig. 5). Constipation-related differential pathways included Calcium signaling pathway, Vibrio cholerae infection, Isoquinoline alkaloid biosynthesis, Lipopolysaccharide biosynthesis, and Monoterpenoid biosynthesis (p < 0.05) (Fig. 5). These findings suggest that gastrointestinal adverse events were accompanied by changes in oral microbial functions.

Fig. 5.

Comparative analysis of predicted KEGG functional pathways of the oral microbiota according to gastrointestinal symptoms during CCRT in LA-NPC patients. Horizontal bar plots show the mean predicted abundance of differentially enriched KEGG pathways in patients with (YES) or without (NO) nausea, vomiting, diarrhea, and constipation, with error bars representing SEM. The x-axis is shown on a log10 scale to improve visualization of low-abundance pathways. Only pathways with p < 0.05 were displayed

Construction of machine learning models based on oral genera to classify gastrointestinal adverse event status

Using oral genera associated with gastrointestinal adverse events, a RegLogit model was developed to classify nausea, vomiting, diarrhea, and constipation status. ROC curve analysis showed AUCs of 0.969, 0.990, 0.950, and 0.849 in the training set, and 0.702, 0.728, 0.763, and 0.698 in the validation set, respectively (Fig. 6A). Confusion matrix analysis showed generally favorable discriminatory performance of the RegLogit model across gastrointestinal adverse events, with variation among individual events (Fig. 6B). SVM-RBF was also used for supplementary modeling, and the results were broadly consistent with those of the RegLogit model (Figure S2).

Fig. 6.

Regularized logistic regression(RegLogit)model based on gastrointestinal adverse event-associated oral genera for distinguishing patients with versus without nausea, vomiting, diarrhea, and constipation during CCRT. A ROC curves of the RegLogit model in the training and validation sets. B Confusion matrices of the RegLogit model in the training and validation sets. The model was developed using selected oral genera and evaluated in the training and validation sets

Comparison of oral microbial features between the high-burden and low-burden groups

Given the shared oral microbial features observed across several gastrointestinal adverse events, particularly involving Alloprevotella and Rothia, the full cohort was further stratified according to gastrointestinal adverse event burden. Patients who experienced 3–4 gastrointestinal adverse events were classified as the high-burden group (n = 39), whereas those who experienced 0–2 events were classified as the low-burden group (n = 41). As shown in Fig. 7A, the relative abundance of Alloprevotella was higher in the high-burden group, whereas Rothia was more abundant in the low-burden group. The relative abundances of Alloprevotella (AUC = 0.745) and Rothia (AUC = 0.705) discriminated between the high-burden and low-burden groups (Fig. 7B). LEfSe analysis was used to identify differential oral microbial features between the two burden groups. In addition to Alloprevotella and Rothia, the high-burden group was enriched in Fusobacterium and Campylobacter, whereas the low-burden group was enriched in Streptococcus, Dorea, Vogesella, and Pelomonas (Fig. 7C). RegLogit and SVM-RBF models were then developed using oral genera associated with gastrointestinal adverse event burden to distinguish high-burden from low-burden status. The RegLogit model yielded an AUC of 0.902 in the training set and 0.867 in the validation set (Fig. 7D). The SVM-RBF model yielded an AUC of 0.895 in the training set and 0.860 in the validation set (Fig. 7E). Both models showed good discriminatory performance, with consistent performance in the validation set. Confusion matrix analysis showed similar classification patterns for the RegLogit and SVM-RBF models (Fig. 7F-G).

Fig. 7.

Oral microbiota features and model performance for distinguishing gastrointestinal adverse event burden status during CCRT in LA-NPC patients. A Relative abundances of Alloprevotella and Rothia in the high-burden and low-burden groups. B ROC curves of Alloprevotella and Rothia for distinguishing the high-burden and low-burden groups. C LDA bar plot from LEfSe analysis between the high-burden and low-burden groups. Only taxa with an absolute log10-transformed LDA score > 2.0 are shown. D-E ROC curves of the regularized logistic regression (RegLogit) model (D) and support vector machine with radial basis function kernel (SVM-RBF) model (E) in the training and validation sets. F-G Confusion matrices of the RegLogit model (F) and SVM-RBF model (G) in the training and validation sets. Gastrointestinal adverse event burden was defined according to the number of gastrointestinal adverse events, including nausea, vomiting, diarrhea, and constipation. Patients with 3–4 events were classified as the high-burden group (n = 39), whereas those with 0–2 events were classified as the low-burden group (n = 41)

Gut microbial features associated with gastrointestinal adverse event burden and potential oral-to-gut microbial transfer

We further examined gut microbial features and potential oral-to-gut microbial transfer patterns in relation to CCRT-related gastrointestinal adverse event burden in patients with LA-NPC. Among the final cohort of 80 patients, 47 had paired stool samples eligible for the present fecal microbiome analysis, including the high-burden group(n = 22) and the low-burden group(n = 25). Samples were eligible for this analysis if they were collected within the predefined sampling window and met sequencing quality control criteria. No significant differences in baseline clinical characteristics or gastrointestinal symptoms were observed between patients included and not included in this analysis (p > 0.05) (Table S1). No significant differences in gut microbial α diversity (Figure S3A) or β diversity (Figure S3B) were observed between the high-burden group and the low-burden group (p > 0.05). LEfSe analysis identified enrichment of gut genera including Helicobacter, Parasutterella, and Anaerotruncus in the high-burden group, whereas Dorea, Lachnospiraceae_UCG_004, and Erysipelotrichaceae_UCG_003 were enriched in the low-burden group (Fig. 8A-B).

Fig. 8.

Gut microbiota features and potential oral-gut microbial transmission associated with gastrointestinal adverse event burden during CCRT in LA-NPC patients. A LDA bar plot from LEfSe analysis of the gut microbiota between the high-burden and low-burden groups. Only taxa with an absolute log10-transformed LDA score > 2.0 are shown. B Cladogram from LEfSe analysis of the gut microbiota between the high-burden and low-burden groups. C Comparison of oral-associated proportions > 1% in the gut microbiota between the high-burden and low-burden groups. D Comparison of oral-associated proportion in the gut microbiota between the high-burden and low-burden groups. Statistical significance: *p < 0.05, **p < 0.01, ***p < 0.001. E Venn diagram showing ASV overlap between the oral and gut microbiota of LA-NPC patients. F Transfer ratio of gastrointestinal adverse event-associated oral microbiota. Gastrointestinal adverse event burden was defined according to the number of gastrointestinal adverse events, including nausea, vomiting, diarrhea, and constipation; patients with 3–4 events were classified as the high-burden group, whereas those with 0–2 events were classified as the low-burden group

SourceTracker analysis indicated detectable oral contribution to stool microbiota, but the estimated oral contribution did not differ significantly between the high-burden and low-burden groups (p > 0.05) (Fig. 8C-D). Although oral and gut microbiota showed distinct diversity profiles (Figure S4A-B), 463 ASVs were shared between the two microbial communities (Fig. 8E). ASV share analysis was then performed in 47 paired oral and stool samples. ASVs belonging to gastrointestinal adverse event burden-associated oral genera, including Alloprevotella, Rothia, and Streptococcus, were detected in paired stool samples, with Streptococcus showing the highest transfer ratio (Fig. 8F).

Discussion

The relationships between oral and gut microbiota and gastrointestinal adverse events during treatment in patients with LA-NPC remain poorly characterized, and their biological basis is not well understood. In this study, we performed 16S rRNA sequencing of paired oral swab and stool samples from patients with LA-NPC to examine the association between gastrointestinal adverse events and oral and gut microbial dysbiosis. These findings extend current understanding of the microbial basis of treatment-related gastrointestinal toxicity in LA-NPC and provide a basis for further studies on the roles of oral and gut microbiota in adverse event development, modifiable microbial markers, and microbiota-related supportive care.

Analysis of oral microbial diversity showed no significant differences in α diversity across groups, whereas βdiversity differed significantly in the nausea and constipation groups. These findings suggest that gastrointestinal adverse events in patients with LA-NPC are not necessarily accompanied by reduced overall oral microbial richness or evenness, but may reflect changes in some microbial components or community structure [29–31]. Because Bray-Curtis distance is more sensitive to differences in relative abundance, whereas binary Jaccard distance mainly captures differences in community membership [29, 32, 33], the modest nausea-associated β-diversity difference may reflect both shifts in taxon abundance and changes in community composition. The constipation-associated difference was weak and was observed only with binary Jaccard distance, suggesting that it may be more closely related to community membership than to changes in dominant taxon abundance. No significant β diversity differences were observed for vomiting or diarrhea, suggesting that these events may be associated with changes in specific taxa rather than broad community level separation [34, 35].These results prompted further LEfSe analysis to identify oral microbiota associated with gastrointestinal adverse events.

Differential analysis of the oral microbiota showed that Alloprevotella tended to be enriched across treatment-related gastrointestinal adverse events, whereas Rothia was more frequently enriched in patients without these events. Previous studies have also reported increased Alloprevotella in oral microbial profiles associated with digestive diseases, suggesting that this genus may be linked to adverse gastrointestinal conditions. Alloprevotella was significantly more abundant in tongue-coating samples from patients with diarrhea-predominant irritable bowel syndrome than in healthy controls, supporting a possible association with diarrhea-dominant gastrointestinal symptom phenotypes [36]. In a mouse model of antibiotic-associated diarrhea, the relative abundance of Alloprevotella increased; after 3′-sialyllactose intervention, Alloprevotella decreased in parallel with reductions in diarrhea rate, fecal water content, and inflammatory levels [37]. Alloprevotella has also been reported to be enriched in the oral microbiota associated with gastritis and gastric cancer, suggesting its potential as a candidate oral marker of upper gastrointestinal lesions [38]。Oral anaerobes can continuously enter the gastrointestinal tract through saliva, and the relative abundance of Alloprevotella has shown consistency between saliva and gastric contents, supporting a potential role for Alloprevotella in the oral and gut microbiota [39]。Consistent with our findings, recent evidence suggests that Rothia may be linked to gastrointestinal homeostasis. Rothia aeria has been shown to efficiently degrade and detoxify gluten, reducing immunogenic peptides implicated in celiac disease and suggesting a potential role in attenuating intestinal inflammation [40]. Another animal study showed that Rothia supplementation upregulated ZO-1 and occludin, reduced TNF-α and IL-1β, and improved intestinal barrier injury, abnormal intestinal permeability, and gut microbial dysbiosis [41]. In addition, Rothia is an important nitrate-reducing genus in the oral cavity and participates in the reduction of salivary nitrate to nitrite [42]. After swallowing, nitrite can be further converted to NO in the acidic gastrointestinal environment, contributing to gastric host defense and acid-dependent killing of swallowed pathogens [42, 43]. In addition, we identified oral microbial features associated with gastrointestinal adverse event burden. Streptococcus was enriched in the low-burden group, whereas Fusobacterium and Campylobacter were more frequently observed in the high-burden group, suggesting that oral microbial patterns may help distinguish patients with different levels of treatment-related gastrointestinal toxicity. Oral Streptococcus, particularly Streptococcus salivarius, is an early colonizer of the oral cavity and gastrointestinal tract. It can reach the gut and has been reported to inhibit NF-κB activation in intestinal epithelial cells and attenuate colitis, suggesting that its enrichment may reflect a relatively stable mucosal microbial state [28, 44]. Clinical studies in head and neck cancer and nasopharyngeal carcinoma radiotherapy have also suggested that Streptococcus salivarius may reduce radiotherapy-related mucosal injury by modulating inflammation and the microbiota [45, 46]. By contrast, Fusobacterium is commonly regarded as an oral opportunistic pathogen. It can ectopically colonize the gut or tumor tissue and has been linked to persistent chronic inflammation, tumor progression, and immune evasion [47, 48]. Previous studies have suggested that ectopic gut colonization by oral bacteria is a feature of intestinal inflammation, and enrichment of Fusobacterium may indicate a more pro-inflammatory gastrointestinal microbial context [49, 50]. This is consistent with the association between Fusobacterium and higher gastrointestinal adverse event burden observed in our cohort. Similarly, Campylobacter, especially oral-derived Campylobacter concisus, has been considered a link between oral colonization and intestinal inflammation and is associated with chronic inflammatory conditions in the gut [51]. Campylobacter concisus can induce IL-8 upregulation in intestinal epithelial cells and, in the presence of immune cells, aggravate epithelial barrier dysfunction, thereby amplifying mucosal immune activation [52, 53]. Enrichment of Campylobacter may therefore mark a more inflammation-prone mucosal state during treatment. Together, these findings suggest that distinct oral microbial features may correspond to different gut microbial and mucosal homeostatic states. Future studies are needed to determine whether microbiota-directed interventions, including probiotic strategies, can reduce gastrointestinal adverse events related to anticancer treatment.

Functional enrichment analysis of the oral microbiota showed that predicted pathways such as propanoate metabolism, lipoic acid metabolism, and retinol metabolism were enriched in patients without nausea or vomiting. α-lipoic acid has been shown to improve intestinal homeostasis in experimental colitis models [54]. Reduced endogenous α-lipoic acid synthesis has also been associated with intestinal epithelial stem cell aging, consistent with a potential role in intestinal epithelial regeneration and gut homeostasis [55]. Propionate is a host-protective short-chain fatty acid involved in mucosal immune regulation and intestinal homeostasis [56], suggesting that enrichment of predicted propanoate metabolism may reflect a functional profile related to mucosal homeostasis. Similarly, microbiota-mediated retinol metabolism has been implicated in the regulation of mucosal immunity and maintain gut homeostasis [57]. These predicted functional pathways may therefore be broadly consistent with mucosal and metabolic homeostasis. By contrast, diarrhea-related predicted functional differences mainly involved other glycan degradation, sphingolipid metabolism, and glycosaminoglycan degradation. Excessive degradation of colonic glycans has been reported to increase susceptibility to infection and inflammatory bowel disease [58]. Altered sphingolipid metabolism is also associated with intestinal inflammation and barrier injury [59]. In the constipation group, predicted enrichment of calcium signaling pathway, lipopolysaccharide biosynthesis, and Vibrio cholerae infection-related functions may suggest potential changes related to enteric nerve–smooth muscle regulation, bacterial product-mediated signaling, intestinal motility, and microbiota-related stimulation. This interpretation is broadly consistent with previous evidence implicating the microbiome in constipation and gastrointestinal motility regulation [60, 61]. Overall, PICRUSt2-based functional inference provides hypothesis-generating clues regarding biological processes potentially associated with treatment-related gastrointestinal adverse events in LA-NPC. However, because these functional profiles were inferred from 16 S rRNA sequencing rather than directly measured using metagenomic, transcriptomic, or metabolomic approaches, further mechanistic validation is required.

RegLogit and SVM-RBF models have been applied in oral microbiome-based disease identification and risk stratification, and have shown potential utility in distinguishing clinical phenotypes such as periodontitis and dental caries [62, 63].Based on this rationale, we further constructed RegLogit and SVM-RBF models using oral microbial features to explore the classification of treatment-related gastrointestinal adverse event status in patients with LA-NPC. The results showed relatively stable classification performance for the overall gastrointestinal adverse event burden. However, for individual symptoms, including nausea, vomiting, diarrhea, and constipation, the validation AUCs of some models were lower than those observed in the training set, indicating that their generalizability should be interpreted with caution. Given the relatively small sample size and the exploratory nature of the machine learning analysis, these findings should be regarded primarily as providing validation of oral microbial features associated with gastrointestinal adverse events, rather than as establishing a clinically ready prediction model. Further validation in larger, multicenter, and externally independent cohorts is required to assess the robustness, generalizability, and potential clinical utility of these models.

In the gut microbiota, LEfSe analysis identified Helicobacter, Parasutterella, and Anaerotruncus as genera associated with a higher burden of treatment-related gastrointestinal adverse events. These genera have also been reported in studies of cancer treatment-related intestinal toxicity. In a model of CTLA-4 blockade-induced intestinal toxicity, enrichment of Anaerotruncus, Parasutterella, and Helicobacter was observed, suggesting that these taxa may reflect changes in the intestinal niche during treatment [64]. These genera have also been implicated in pathological processes related to gastrointestinal inflammation, metabolic disturbance, and mucosal barrier injury in digestive diseases [65–67]. Although no significant baseline or gastrointestinal symptom differences were observed between patients included and not included in the gut microbiome analysis, the exploratory subset design may have constrained the detection of subtle microbial differences, and validation in larger cohorts is warranted. Oral to gut microbial transfer analysis showed no significant differences between burden groups in either the overall oral-derived signal or the transfer ratio of oral genera associated with gastrointestinal adverse events. Oral to gut microbial transfer has been widely reported in several digestive disease settings [68, 69], and its potential contribution to CCRT-related gastrointestinal adverse events in patients with LA-NPC cannot be excluded. However, such a signal may not have been fully captured with the current sample size and 16S rRNA-based analysis. ASV share analysis mainly reflects potential sharing between oral and gut microbiota, but cannot determine the actual abundance, persistent colonization, or functional activity of transferred bacteria. Because 16S rRNA analysis is limited in taxonomic resolution, metagenomic sequencing with strain-level tracking is needed to further clarify these findings [70]. Alternatively, gastrointestinal adverse events may not be driven by large-scale transfer of oral bacteria to the gut. A recent study suggested that increased relative abundance of oral bacteria in stool may result from depletion of the resident gut microbiota [71]. In the present study, gastrointestinal adverse event burden was not accompanied by broad remodeling of the gut microbiota, which may explain the lack of marked differences in oral to gut transfer signals between burden groups. Overall, these findings suggest that gastrointestinal adverse events in this cohort were more closely related to alterations in specific gut genera than to global gut microbiota remodeling.

This study has several limitations. First, this was a single-center study, and external validation is needed to confirm the generalizability of our findings. Second, although we performed 16S rRNA sequencing and integrated analyses of oral swab and stool samples, metagenomic and other omics analyses are needed to improve taxonomic resolution and to better characterize microbiota in treatment-related gastrointestinal adverse events at the levels of microbial genes, metabolites, and host responses. Third, several predicted functional pathways were identified in this study. These findings should be interpreted as association-based clues, and further in vivo and in vitro experiments are needed for mechanistic validation.

Conclusions

This study identified oral microbial genera associated with treatment-related gastrointestinal adverse events in patients with LA-NPC, including Alloprevotella, Rothia, Streptococcus, Fusobacterium, and Campylobacter. Oral microbial features were associated with gastrointestinal adverse event status during treatment and showed potential classification value, which requires further validation in larger independent cohorts. In addition, gastrointestinal adverse events did not appear to be accompanied by global gut microbiota remodeling, but were associated with alterations in specific gut genera, including Helicobacter, Parasutterella, and Anaerotruncus.

Abbreviations

LA-NPC

Locoregionally advanced nasopharyngeal carcinoma

CCRT

Concurrent chemoradiotherapy

PCoA

Principal coordinate analysis

PERMANOVA

Permutational multivariate analysis of variance

LEfSe

Linear discriminant analysis effect size

LDA

Linear discriminant analysis

PICRUSt2

Phylogenetic investigation of communities by reconstruction of unobserved states 2

RegLogit

Regularized logistic regression

SVM-RBF

Support vector machine with radial basis function

AUC

Area under the curve

ROC

Receiver operating characteristic

PTS

Phosphotransferase system

Authors’ contributions

KM and MZ participated in the conception and experimental design of this study, as well as project supervision and management. CY, HCX and QYT were responsible for the methodological design, experimental research, data curation, formal statistical analysis, result visualization, and drafting of the initial manuscript. JYM and LLZ contributed to sample collection and experimental operation. LJL assisted in formal data analysis and result interpretation. KM and MZ jointly revised and edited the manuscript. CY, HCX and QYT made equal contributions to this study. All authors were involved in drafting or critically revising the manuscript, approved the final version for publication, and agreed to the submission of this article to the journal.

Funding

This work was supported by grants from Guangxi Medical University First Affiliated Hospital Specialised and Innovative Teams Project (2020), the Second Batch of Guangxi High-Level Mid-Career and Young Academic Backbone Projects (2020), the Guangxi Medical University High-Level Innovative Team and Xinghu Scholars Programme (2022), Key Research Project of the CSCO(2022), and the Guangxi Medical University Young Leading Talent Programme (2025).

Data availability

The sequencing datasets generated during the current study are available in the Genome Sequence Archive for Human under accession number HRA019707 and can be accessed at https://ngdc.cncb.ac.cn/search/specific? db=hra&q=HRA019707.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Ethics Committee of the First Affiliated Hospital of Guangxi Medical University. All procedures involving human participants and human samples were performed in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants prior to their inclusion in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.