Characteristics of gut microbiota changes in patients with nasopharyngeal carcinoma during radiotherapy Authors: Caihong Yu, Yingying Lin, Xiaonan Cui, Xiaojun Huang, Yanxin Zhang, Min Kang, Jiamei Lu Published: 2026 ABSTRACT BackgroundThis study aimed to characterize the longitudinal and dynamic changes of gut microbiota in patients with nasopharyngeal carcinoma (NPC) during radiotherapy, to inform the development of strategies for maintaining gut microbiota homeostasis.MethodsThirty-one newly diagnosed NPC patients were recruited from the Department of Radiotherapy, First Affiliated Hospital of Guangxi Medical University, and 31 age-matched healthy controls from the same Hospital Physical Examination Center between August and December 2023. Fecal samples were collected from NPC patients at three key time points, including days 1–3 pre-radiotherapy (T0), days 14–18 of mid-radiotherapy (T1), and days 28–32 at end of radiotherapy (T2). Meanwhile, fecal samples from healthy controls were collected on the morning of their physical examination. 16S rDNA sequencing, combined with bioinformatics and statistical analyses, was used to compare the structural characteristics and dynamic changes of gut microbiota in NPC patients during radiotherapy.ResultsAlpha diversity analysis showed significantly lower Shannon and Chao1 indices in NPC patients compared with healthy controls. Beta diversity analysis based on the Bray-Curtis distance revealed distinct gut microbial community structures between NPC patients and healthy controls. In this intergroup comparison, a total of 16 core differentially abundant taxa were consistently identified by two complementary analytical methods (LEfSe and ALDEx2). Linear mixed-effects models demonstrated that the Chao1 index at T0 was significantly lower than at T1 and T2 in NPC patients. Following adjustment for multiple confounding variables, mucositis grade was the only factor significantly associated with gut microbiota alpha diversity during radiotherapy. Radiotherapy time points, treatment regimen, and most clinical and demographic variables showed no such association. Beta diversity analyses based on the Bray-Curtis and unweighted Unifrac distances revealed significant compositional and structural differences linked to radiotherapy time points (T0/T1/T2) in the gut microbiota of NPC patients. This temporal variation was abrogated by stratification according to chemotherapy regimen. Beta diversity based on Bray-Curtis distance showed no significant differentiation across time points in either the concurrent chemotherapy (CCRT) alone cohort or the induction chemotherapy plus CCRT cohort. LEfSe and ALDEx2 concordantly identified seven core differentially abundant taxa across all NPC patients in the longitudinal T0 vs. T1 vs. T2 analysis. Stratification by chemotherapy regimen revealed such core taxa in the CCRT alone cohort, whereas no core differentially abundant taxa were detected by either method in the induction chemotherapy plus CCRT cohort.ConclusionOur findings demonstrate that NPC patients exhibit significant gut microbial dysbiosis compared with healthy controls, characterized by reduced alpha diversity and altered genus-level composition. Longitudinal analyses further revealed that radiotherapy is associated with dynamic alterations in gut microbial diversity, composition, and structure in NPC patients. Notably, these temporal shifts in the gut microbiota are strongly stratified by chemotherapy regimen, with pronounced changes observed in patients receiving CCRT alone but a stable microbial profile in those receiving induction chemotherapy followed by CCRT. Collectively, these results highlight the profound impact of oncological treatment on the gut microbiome in NPC patients and identify treatment regimen as a critical modifier of microbial dynamics during radiotherapy.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12967-026-08265-0. Introduction Nasopharyngeal carcinoma (NPC) is a common head and neck malignancy with a particularly high incidence in South China. The annual incidence in Guangxi Province reaches 10–30 per 100,000 population, ranking among the highest worldwide [1]. Radiotherapy-based comprehensive treatment remains the standard clinical strategy for NPC and has significantly improved patient survival [2]. However, radiotherapy and concurrent chemotherapy (CCRT) trigger systemic immune activation and inflammatory responses that disrupt the gut microbiota, a critical regulator of host metabolism, immune function, and intestinal barrier integrity [3, 4]. These treatments are associated with reductions in beneficial microbes such as Firmicutes, Faecalibacterium, and Bifidobacterium, while potentially pathogenic taxa including Enterobacteriaceae and Bacteroides often increase, ultimately leading to gut dysbiosis [3, 4]. Gut dysbiosis profoundly affects treatment tolerance and prognosis in cancer patients. Loss of beneficial bacteria is linked to chronic intestinal inflammation, impaired nutrient absorption, facilitated pathogenic bacterial proliferation, and damaged intestinal epithelial barrier [5]. Imbalanced gut microbiota is also associated with reduced tolerance to chemoradiotherapy, weakened treatment efficacy, and exacerbated toxicities including oral mucositis and gastrointestinal reactions [6]. Together, these effects worsen clinical outcomes and impose substantial physical and economic burdens on patients and their families [7, 8]. Therefore, exploring dynamic changes in the gut microbiota during NPC treatment and identifying key regulatory taxa are crucial for developing targeted interventions to maintain gut homeostasis, improve treatment tolerance, and optimize clinical outcomes. Current research on the microbiota associated with NPC has focused primarily on the oral mucosal [9], nasopharyngeal microenvironment [10], and gut microbiota in nude mouse models [11]. Clinical studies investigating gut microbiota in NPC patients remain limited, especially longitudinal studies tracking dynamic changes throughout the entire radiotherapy course. Most existing studies are cross-sectional and cannot capture changes linked to radiotherapy or their relationships with treatment-related toxicities. Additionally, core differential taxa between NPC patients and healthy controls, as well as their dynamic alterations during radiotherapy, have not been systematically defined. These gaps limit the development of targeted gut microbiota regulation strategies for NPC patients undergoing radiotherapy. The “gut-nasopharynx axis” has emerged as an important pathway linking gut microbiota to upper respiratory tract diseases. Gut dysbiosis can modulate the nasopharyngeal immune microenvironment through systemic immune signaling, thereby associating with tumor progression and treatment response [12, 13]. Specifically, microbiota can activate CD103(+) dendritic cells (DCs) to regulate γδT17 cell proliferation and activation, contributing to local and systemic immune responses [14]. However, the role of this axis in NPC patients during radiotherapy remains unclear. Short-chain fatty acid (SCFA)-producing bacteria support intestinal barrier function and exert anti-inflammatory effects [15, 16], yet whether radiotherapy is associated with reduced abundance of SCFA-producing bacteria in NPC patients has not been confirmed. This study adopted a combined cross-sectional and longitudinal observational design. Cross-sectionally, we compared the gut microbiota profiles between newly diagnosed NPC patients and healthy controls to identify core differential taxa. Longitudinally, we tracked the dynamic changes linked to radiotherapy in gut microbiota in NPC patients at three key time points, including days 1–3 pre-radiotherapy (T0), days 14–18 of mid-radiotherapy (T1), and days 28–32 at the end of radiotherapy (T2). Using 16S rDNA sequencing and bioinformatics analysis, we characterized compositional and abundance changes in the gut microbiota, with a focus on SCFA-producing bacteria. This study aims to provide a theoretical foundation for formulating personalized strategies to maintain gut microbiota homeostasis, improve treatment tolerance, and optimize prognosis in NPC patients undergoing radiotherapy. Methods Study participants and recruitment From August 2023 to December 2023, we conducted a prospective observational study at the First Affiliated Hospital of Guangxi Medical University (Nanning, Guangxi, China). We enrolled 31 newly diagnosed, treatment-naive patients with NPC (conveniently recruited from the Department of Radiotherapy) and 31 healthy controls (enrolled from the Hospital Physical Examination Center during the same period). To enhance comparability of gut microbiota profiles, we used frequency matching for the healthy controls: age (matched to NPC patients within ± 5 years), and gender (healthy controls: 22 males, 9 females; NPC patients: 24 males, 7 females), resulting in similar male proportions (77.4% vs. 71.0%, respectively) to minimize gender-related confounding in gut microbiota analyses. We additionally collected and verified key exposure factors in controls: (1) dietary pattern: consistent with the general local population (no specialized diets); (2) smoking history: current smokers were identical between groups (8 each), with matching smoking intensity. Of those with a smoking history, 5 controls and 3 NPC patients had abstained for > 2 years. Smoking status (current/former/never) did not differ significantly between groups; (3) alcohol consumption history: participants with a history of alcohol consumption in both groups reported occasional drinking, and all had abstained from alcohol for more than six months; (4) recent antibiotic / probiotic use: confirmed no use within 1 month prior to enrollment. Inclusion and exclusion criteria The inclusion criteria for patients with NPC included the following: (1) histopathologically confirmed nasopharyngeal squamous cell carcinoma; (2) age ≥ 18 years old; (3) scheduled for first-course chemoradiotherapy (treatment-naive); (4) normal communication ability; (5) voluntary ability to provide fecal samples; (6) written informed consent obtained. Patients with the following criteria were excluded: (1) concurrent malignancies in other organs or major organic lesions; (2) alcohol or substance dependence / abuse; (3) history of inflammatory bowel disease (IBD), irritable bowel syndrome (IBS), chronic abdominal pain, diarrhoea, severe constipation, intestinal obstruction, or gastrointestinal surgery; (4) use of antibiotics, probiotics, steroids, or proton pump inhibitors in the prior month; (5) inability to complete fecal sample collection; (6) specialized dietary patterns such as restricted diets or receipt of tube feeding; (7) recent hospitalization (within 3 months prior to enrollment). The inclusion criteria for healthy controls: (1) no disease diagnoses in physical examination reports; (2) age ≥ 18 years old; (3) normal communication ability; (4) able to voluntarily provide fecal samples; (5) written informed consent obtained. Exclusion criteria for healthy controls: (1) alcohol or substance dependence / abuse; (2) use of antibiotics, probiotics, steroids, or proton pump inhibitors in the prior month; (3) pregnant or lactating status; (4) history of IBD, IBS, chronic abdominal pain, diarrhoea, severe constipation, intestinal obstruction, or gastrointestinal surgery; (5) inability to complete fecal sample collection; (6) specialized dietary patterns or receipt of tube feeding; (7) recent hospitalization (within 3 months prior to enrollment). Study design and ethical compliance Our study adopted a mixed observational design integrating cross-sectional and longitudinal approaches. The cross-sectional component compared gut microbiota profiles between NPC patients and healthy controls, while the longitudinal component tracked dynamic changes associated with radiotherapy. Single-blinding was implemented: sample collection, sequencing, and data analysis personnel were masked to group allocation. The study was approved by the Ethics Committee of the First Affiliated Hospital of Guangxi Medical University (Approval NO.: 2023-S919-01). 41 potential NPC patients and 38 potential controls were initially assessed for eligibility, with 10 NPC patients were excluded (1 concurrent extra-organ malignancy, 2 recent antibiotic use, 1 recent probiotic use, 2 incomplete fecal samples, 2 severe constipation, and 2 history of gastrointestinal surgery) and 7 controls excluded (1 alcohol dependence, 2 recent antibiotic use, 2 severe constipation, 1 history of gastrointestinal surgery, and 1 lactating woman). During follow-up, no participants were lost to follow-up, and fecal samples were collected from all 31 enrolled NPC patients and 31 healthy controls at baseline (T0), with additional samples collected from NPC patients at T1 and T2 and no missing samples at any time point. Given the absence of missing data, complete-case analysis was used for all statistical analyses. All participants provided written informed consent, and the protocol was strictly followed. Participant flow is summarized in Fig. 1 (CONSORT 2010-compliant). Fig. 1Flow chart representing the sample population selected in this study (CONSORT) 2010 Flow chart representing the sample population selected in this study (CONSORT) 2010 Treatments for patients with NPC Irradiation was mainly performed using 6–10 MV photons and electrons from linear accelerators. Intensity-modulated radiation therapy (IMRT) was given according to the following details: the total radiation dose ranged from 69.76 to 72.00 Gy (mean ± standard deviation: 70.79 ± 0.63 Gy), with a single-fraction dose of 2.13 to 2.29 Gy (mean ± standard deviation: 2.17 ± 0.05 Gy). Radiation therapy was administered 5 times per week, totaling 31 to 33 sessions (mean ± standard deviation: 32.68 ± 0.48 sessions), and the entire treatment course lasted 7 to 8 weeks. The treatment regimen was formulated with reference to the European Society for Medical Oncology -European Reference Network for Rare Adult Solid Cancers (ESMO-EURACAN) Clinical Practice Guidelines for NPC [2]. Induction chemotherapy regimens: three cycles of GP (Gemcitabine [1000 mg/m2 on day1 and day8 ] and cisplatin [80 mg/m2 on days 2–3] every 3 weeks, for 3 cycles. CCRT: Cisplatin-based concurrent chemotherapy, mainly with 80–100 mg/m2 on days 2 every 3 weeks, for 2 cycles. During radiotherapy, all patients received standardized dietary guidance and oral care protocols, with no prophylactic antibiotics administered. Dietary guidance: Guide patients to maintain a balanced diet: incorporate high-protein, high-calorie, vitamin-rich, and mineral-dense foods. Ensure a combination of refined and coarse grains. Avoid alcohol, tobacco, spicy, fried, stir-fried, and similar irritating foods. Organize patients and their family members to attend mini-lectures on nutrition knowledge. Use methods such as charts, videos, and food model demonstrations to explain food types, intake amounts, food matching skills, and feeding techniques. Help patients and their families understand the principle of food exchange portions, so that patients can easily select alternative foods with equivalent energy content. Monitor patients’ body weight and nutrient intake regularly to ensure adequate energy supply. Advise patients to keep a dietary diary, detailing the type and intake of each daily meal. The primary nurse calculates the patients’ daily total energy needs and actual intake of energy-producing nutrients to meet bodily requirements. Oral care protocols: Instruct patients to pay attention to oral hygiene and strengthen oral care. Emphasize the importance of rinsing the mouth thoroughly after each meal and before bedtime to remove food residues from the mouth and maintain oral cleanliness. Recommend patients to drink water regularly, with a daily intake of 2500–3000 mL. During drinking, patients should hold water in the mouth and swallow slowly to alleviate xerostomia (dry mouth) caused by reduced salivary gland secretion after radiotherapy. Administer nebulized inhalation 1–2 times per day as prescribed by the doctor. Prepare a gargle solution containing 3 vials of lidocaine injection (0.1 g), 2 vials of dexamethasone injection (5 mg), 3 vials of vitamin B12 injection (0.5 mg), and 500 mL of 0.9% sodium chloride solution. Instruct patients to gargle with this solution 3–4 times per day, with each gargling session lasting 3–5 min. Given that cancer therapy is associated with alterations in gut microbiota, treatment parameters were standardized to minimize potential bias from variable treatment regimens. Fecal samples collection Fecal samples were self-collected by participants. For patient with NPC, samples were collected at three time points (T0/T1/T2). On the night before fecal sample collection, researchers provided patients with sterile fecal collection materials and detailed instructions on the process and method of fecal collection. For healthy controls, participants were contacted by phone one day before their physical examination to obtain their consent. On the physical examination day, they signed the informed consent form in the morning. Meanwhile, they were informed of the method for fecal sample collection, provided with sample collection materials, and asked to collect the fecal sample on-site immediately after defecation. The specific procedure as follows: Participants defecated into a disposable sterile bowl in the morning, taking care to avoid contamination by urine, water stains, dust, or other foreign substances. Using a dry sterile spoon, approximately 5–10 g of feces was sampled and transferred into a dry sterile cryovial. Within 10 min of collection, the cryovial containing the fecal sample was placed into a sealed container filled with liquid nitrogen for flash freezing, and transported to the designated biobank within two hours while maintaining cryogenic conditions, where samples were stored at − 80℃ until DNA extraction [17]. All procedures for sample collection, packaging, and storage strictly adhere to the standardized regulations of the Biobank of the First Affiliated Hospital of Guangxi Medical University, and the rationality of the preservation method is supported by relevant studies [17]. DNA extraction and 16S rDNA gene sequencing Prior to DNA extraction, fecal samples were subjected to bead-beating lysis for efficient bacterial cell wall disruption using 0.1 mm glass beads at 60 Hz for 3 min (TissueLyser II, Qiagen, Hilden, Germany). DNA was then isolated from the lysed samples using a fecal genomic DNA extraction kit (Cat. No. AU46111-96, BioTeke Corporation, Beijing, China). To eliminate batch effects, samples from different time points (T0/T1/T2 for NPC patients) and study groups (NPC patients vs. healthy controls) were extracted in a single batch. Subsequently, the quality of the extracted DNA was assessed by agarose gel electrophoresis, and its concentration was quantified using an ultraviolet spectrophotometer. The Variable region 3-Variable region 4 (V3-V4) [18] region of the 16 S rDNA gene was amplified with the universal primers 341 F/805R (341 F: 5′-CCTACGGGNGGCWGCAG-3′; 805R: 5′-GACTACH VGGGTATCTAATC C-3′). The 25 µL reaction system consisted of 12.5 µL of Phusion Hot Start Flex 2X Master Mix (NEB, Ipswich, MA, USA), 2.5 µL of forward primer, 2.5 µL of reverse primer, 50 ng of template DNA, and ddH2O to reach a total volume of 25 µL. The polymerase chain reaction (PCR) cycling conditions were as follows: pre-denaturation at 98 °C for 30 s, followed by 32 cycles of denaturation at 98 °C for 10 s, annealing at 54 °C for 30 s, and extension at 72 °C for 45 s. PCR products were purified using AMPure XT beads (Cat. No. A63880, Beckman Coulter Genomics, Danvers, MA, USA) and quantified with the Qubit dsDNA HS Assay Kit (Cat. No. Q32854, Invitrogen, Life Technologies, Carlsbad, CA, USA). Purified PCR products were evaluated using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and the Illumina Library Quantification Kit (Kapa Biosciences, Woburn, MA, USA), with qualified libraries requiring a concentration of > 2 nM. After gradient dilution of each qualified library (with non-repetitive index sequences), libraries were pooled in appropriate proportions according to the required sequencing output, denatured into single-stranded DNA with NaOH, and subjected to on-instrument sequencing. Sequencing was performed on a NovaSeq 6000 (PE250) sequencer (Hangzhou Lianchuan Biotechnology Co., Ltd, Hangzhou, China) using 2 × 250 bp paired-end sequencing with the NovaSeq 6000 SP Reagent Kit (500 cycles). Data processing and 16S rDNA sequencing analysis For paired-end reads (250 bp per end) generated from the NovaSeq 6000 (PE250) sequencing platform targeting the 16S rDNA V3-V4 region, data processing was performed as follows: first, samples were demultiplexed based on barcode information, followed by removal of sequencing adapters, barcodes, and primer sequences, as well as base sequence balancing using cutadapt (v1.9) with parameters ‘-g R1 -G R2 -n 1 -O 17 -m 100’ (where ‘-n 1’ specifies a maximum of 1 primer mismatch, ‘-O 17’ sets the minimum overlap length for adapter/primer trimming, and ‘-m 100’ retains reads ≥ 100 bp post-trimming). Paired-end reads were assembled into longer tags by merging overlapping regions using FLASH (v1.2.8) with parameters ‘-m 10 -M 100 -x 0.25 -t 1 -z’ (‘-m 10’ = minimum overlap length of 10 bp, ‘-M 100’ = maximum overlap length of 100 bp, ‘-x 0.25’ = maximum overlap mismatch rate of 25%, ‘-t 1’ = 1 thread, ‘-z’ = gzip-compressed output). A window-based quality filter was applied to the merged reads using fqtrim (v0.94) with parameters ‘-P 33 -w 100 -q 20 -l 100 -m 5 -p 1 -V -o trim.fastq.gz’: reads were truncated from the start of any 100 bp window with an average Phred quality score < 20, truncated sequences < 100 bp in length were discarded, and sequences containing > 5% ambiguous bases (N) were discarded (enforced by ‘-m 5’). Chimeric sequences were eliminated via combined de novo and reference-based detection using Vsearch (v2.3.4) with parameters ‘--id 0.97 --threads 1’ (‘--id 0.97’ = 97% sequence similarity threshold for chimera identification). Subsequently, the Divisive Amplicon Denoising Algorithm 2 (DADA2) [19] was implemented through the qiime dada2 denoise-paired function in QIIME2 (v2019.7) with explicit denoising and filtering parameters: --p-trim-left-f 0 (no 5’ truncation for forward reads), --p-trim-left-r 0 (no 5’ truncation for reverse reads), --p-trunc-len-f 240 (truncate forward reads to 240 bp at the 3’ end), --p-trunc-len-r 230 (truncate reverse reads to 230 bp at the 3’ end), --p-max-ee-f 2 (maximum expected errors = 2 for forward reads), --p-max-ee-r 3 (maximum expected errors = 3 for reverse reads), --p-trunc-q 2 (truncate reads at the first base with Phred quality score < 2). Additional length-associated filtering was applied post-denoising: Amplicon Sequence Variants (ASVs) with a mean relative abundance < 0.1% across all samples and a prevalence < 10% within each study group were excluded. This complete denoising and filtering workflow yielded Amplicon Sequence Variant (ASV) feature sequences and an ASV abundance table; as a default quality control step, singleton ASVs (i.e., ASVs with a total sequence count = 1 across all samples) were removed. All Alpha diversity indices were calculated after rarefaction to eliminate the impact of uneven sequencing depth on diversity analysis. Rarefaction was performed to a uniform sequencing depth of 30,000 reads per sample, consistent with the depth at which rarefaction curves reached a plateau for all samples. While rarefaction is primarily implemented for alpha and beta diversity analyses, we retained the rarefied ASV table for downstream differential abundance testing to maintain consistency in the input dataset across all analyses. We acknowledge that compositional methods such as ALDEx2 inherently correct for sequencing depth via centered log-ratio (CLR) transformation, and we confirmed that the core differential taxa identified in this study were consistent between rarefied and non-rarefied input data. Alpha diversity was evaluated using four indices (Shannon, Simpson, Chao1, and Pielou_e) to assess species richness, evenness, and diversity within samples, while beta diversity was analyzed via Principal Coordinates Analysis (PCoA) based on Bray-Curtis and unweighted Unifrac distance metrics to compare intergroup microbial community differences. Species taxonomic annotation was performed using the SILVA database (Release 138, https://www.arb-silva.de/documentation/release 138/) with a confidence threshold of 0.7, and the relative abundance of each taxon in individual samples was calculated from the ASV abundance table. Data collection Study subjects were selected in strict accordance with the inclusion criteria. Before data collection, patients and healthy controls were informed of the purpose, significance, methods, and confidentiality principles of this study, and signed a written informed consent form. Two professionals staff members of the research team were responsible for data collection. They instructed patients to fill out the questionnaires using standardized instructions, and all questionnaires were completed and collected on-site. In addition, basic information (including demographic data, disease-related data, past medical history, etc.) was collected through the hospital information system (HIS). The basic information of healthy controls was extracted from the information system of the physical examination department. The collected data were statistically analyzed and verified by two researchers to ensure the authenticity and accuracy of the data. Statistical methods Based on the obtained species abundance statistics, differential analysis was conducted between comparison groups. Categorical data are presented as counts (percentages), while continuous data following a normal distribution are expressed as mean ± standard deviation (SD) and non-normally distributed data as median (interquartile range) (SPSS 26.0). For intergroup comparisons of alpha diversity indices between NPC patients and healthy controls (primary cross-sectional comparison), the Wilcoxon rank-sum test (R v3.4.4) was used, with post-hoc tests with Benjamini-Hochberg (BH) correction for false discovery rate (FDR) control, with a significance threshold of FDR < 0.1. This threshold was selected given the exploratory nature of this microbiome study, to avoid type II error and the premature exclusion of potentially biologically relevant taxa associated with NPC and radiotherapy exposure, while still accounting for multiple testing. We frame all findings from the FDR < 0.1 threshold as exploratory. Effect sizes were reported as the primary measures of effect magnitude. To account for intra-subject correlation in longitudinal repeated measures data from 31 NPC patients with complete samples at T0, T1 and T2 (n = 31 per time point), linear mixed-effects models (R v3.4.4) were employed to analyze changes in alpha diversity indeices across time points. In these models, alpha diversity indices were included as the dependent variables, time points (T0/T1/T2) as fixed effects, and subject ID as random intercepts. When a significant overall time effect was detected, post-hoc pairwise comparisons were performed using the emmeans package (R v1.10.4), with Benjamini–Hochberg correction for false discovery rate (FDR) control (FDR < 0.1). Effect sizes were reported as estimates of magnitude. In addition, key clinical and demographic covariates were additionally incorporated into the models for adjustment. Beta diversity, reflecting differences in the microbial community composition and structure, was analyzed using Analysis of permutational multivariate analysis of variance (PERMANOVA, adonis2) (R v3.4.4, vegan v2.5.4) stratified by subject ID to control for individual differences. Dispersion testing (betadisper, R v3.4.4, vegan v2.5.4) was performed to confirm that separation is not driven by heterogeneity of variances. Linear Discriminant Analysis Effect Size (LEfSe) (nsegata-lefse, 094f447691f0) was used as an exploratory method to identify candidate differentially abundant taxa, with a threshold of Linear Discriminant Analysis (LDA) score > 3 and BH-corrected P < 0.05, and results from this method are considered preliminary and complementary to the primary analysis. For differential abundance analysis at the phylum, class, order, family, genus, and species levels, Analysis of variance (ANOVA)-like differential expression version 2 (ALDEx2, R v3.4.4) [20, 21] was employed as the primary method to address the compositional nature of microbiome data. This method uses CLR transformation and Monte Carlo sampling to correct for sequencing depth variation and compositional bias. For each taxon, log2 fold change (log2FC) with 95% CI as the primary effect size and BH-corrected P values were reported, with significance defined as P < 0.1 and |log2FC| > 0.5, consistent with the threshold specified above for exploratory analyses, and supported by both Welch’s t-test and Wilcoxon rank-sum test results. Core differential taxa were defined as those that reached significance via both ALDEx2 and LEfSe analyses, to enhance the robustness of our exploratory findings. Two sensitivity analyses were conducted to assess the robustness of findings: (i) stratified analysis by chemotherapy regimen (CCRT vs. induction chemotherapy + CCRT), and (ii) multivariable adjustment for key covariates (clinical stage, sex, age, weight, food intake score, stool form, smoking history, alcohol consumption history, antiemetic use, opioid use, enteral nutritional support, PPI use, enteral feeding, and mucositis grade) in the linear mixed-effects models. Sequencing metrics and quality assessment of gut microbiota in NPC patients and healthy controls A total of 62 fecal samples were collected, including 31 from newly diagnosed NPC patients at days 1–3 pre-radiotherapy (T0, n = 31) and 31 from healthy controls (HC, n = 31). After rigorous quality control, a total of 3,377,490 high-quality and valid sequences were obtained across all samples. The mean number of valid sequences per sample was 54,457 ± 6,715 (mean ± standard deviation), with the minimum, maximum and median values being 38,281, 69,912 and 54,476 reads, respectively. Valid sequences exhibited a read length distribution ranging from 400 to 500 bp, consistent with the expected length of the 16S rDNA V3-V4 amplicon region. Following denoising and error correction with the DADA2 algorithm, a total of 14,485 ASVs were generated for subsequent microbial community analysis. Venn diagram analysis (Fig. 2), based on the ASV feature sequences and corresponding ASV abundance tables, characterized the overlap and uniqueness of ASVs between groups, revealing 4,414 ASVs in the T0 group and 10,071 ASVs in the healthy controls group. Rarefaction curves were constructed from the sequencing dataset to evaluate the sufficiency of sequencing depth (Fig. 3). This plot displays rarefaction curves for two groups: the healthy controls group (light blue curve) and the T0 group (pink curve). When the sequencing depth reached 30,000 reads, the rarefaction curves for both groups tended to reach a plateau (Fig. 3). This indicated that the sampling strategy was rational, and the sequencing depth was sufficient to capture the complete microbial species richness of the study samples. No samples were excluded from further analyses, as all met the sequencing depth criterion corresponding to the plateau region of the curves. Fig. 2Venn diagram of ASV distribution. Note: Venn diagram showing ASVs between NPC patients at T0 (n = 31) and HC (n = 31). Blue circle: ASVs exclusive to T0, dark purple circle: shared ASVs and dark red circle: ASVs exclusive to HC Venn diagram of ASV distribution. Note: Venn diagram showing ASVs between NPC patients at T0 (n = 31) and HC (n = 31). Blue circle: ASVs exclusive to T0, dark purple circle: shared ASVs and dark red circle: ASVs exclusive to HC Fig. 3Rarefaction curves assessing sequencing depth sufficiency. Note: X-axis: Number of sequencing reads, Y-axis: Observed Operational Taxonomic Units (Observed OTUs). Light blue curve: HC, Pink curve: T0 group. When the sequencing depth reached 30,000 reads (the upper limit of the X-axis), the curves tended to plateau, confirming sufficient sequencing depth for capturing microbial species richness Rarefaction curves assessing sequencing depth sufficiency. Note: X-axis: Number of sequencing reads, Y-axis: Observed Operational Taxonomic Units (Observed OTUs). Light blue curve: HC, Pink curve: T0 group. When the sequencing depth reached 30,000 reads (the upper limit of the X-axis), the curves tended to plateau, confirming sufficient sequencing depth for capturing microbial species richness Sequencing metrics, quality assessment and ASV stability of gut microbiota in NPC patients across different radiotherapy stages A total of 93 fecal samples were collected from NPC patients at three time points, including T0, T1 and T2. Following strict quality control, 5,333,681 high-quality and valid sequences were obtained overall, with a mean of 57,351 ± 6,976 valid reads per sample (mean ± standard deviation). The minimum, maximum, and median numbers of valid reads per sample were 37,780, 72,323 and 57,762, respectively. Valid sequences exhibited a read length distribution of 400–500 bp, consistent with the expected amplicon length of the 16S rDNA V3-V4 target region. After denoising and error correction via the DADA2 algorithm, 14,873 ASVs were generated for downstream microbial community analyses. To characterize ASV distribution across time points (Fig. 4), Venn diagram analysis quantified the overlap and uniqueness of ASVs: 1,900 ASVs were exclusive to T0, 1,931 to T1, 2,451 to T2, and 1,635 ASVs were shared across all three time points (with additional pairwise overlaps: 520 between T0 and T1, 356 between T0 and T2, 967 between T1 and T2). Bar plot of paired shared vs. unique ASVs per time point (Fig. 5) showed that the total ASVs per time point increased from 2,511 (T0) to 3,122 (T1) and 2,958 (T2), with unique ASVs accounting for 75.7% (T0), 61.8% (T1), and 82.2% (T2) of the total ASVs at each stage. To evaluate longitudinal stability of the gut microbiota, we calculated Jaccard similarity indices (a measure of community overlap) across time intervals (Figs. 6 and 7). For paired comparisons, the median Jaccard similarity was 0.222 ± 0.074 (T0 vs. T1), 0.184 ± 0.067 (T0 vs. T2), and 0.261 ± 0.059 (T1 vs. T2). Aggregated across adjacent (T0-T1/T1-T2) and non-adjacent (T0-T2) intervals, the overall median Jaccard similarity was 0.242 ± 0.069, indicating low stability of ASV composition in NPC patients during radiotherapy. Rarefaction curves were constructed from the sequencing data set to evaluate sequencing depth sufficiency (Fig. 8), and subsequent analysis was performed to define the optimal sequencing depth for all downstream diversity analyses. This plot displays rarefaction curves for three radiotherapy stages: T0 (pink curve), T1 (dark blue curve) and T2 (light blue curve). When the sequencing depth reached 30,000 reads, the rarefaction curves for all groups tended to reach a plateau, indicating sufficient coverage to capture the microbial community diversity of the samples. As all samples satisfied the sequencing depth criterion corresponding to the plateau region of the rarefaction curves, no samples were excluded from further analyses. Fig. 4Venn diagram of ASV distribution across NPC radiotherapy stages. Note: Venn diagram showing the number of unique and shared ASVs in NPC patients at T0, T1 and T2 (n = 31 per group). Values represent the count of ASVs in each segment, and the color gradient corresponds to ASV count magnitude Venn diagram of ASV distribution across NPC radiotherapy stages. Note: Venn diagram showing the number of unique and shared ASVs in NPC patients at T0, T1 and T2 (n = 31 per group). Values represent the count of ASVs in each segment, and the color gradient corresponds to ASV count magnitude Fig. 5Shared vs. unique ASVs across radiotherapy time points. Note: Stacked bar plot showing the total number of ASVs per time point, stratified into shared (across ≥ 2 time points) and unique (exclusive to one time point) categories. Total ASVs analyzed: 19,989 Shared vs. unique ASVs across radiotherapy time points. Note: Stacked bar plot showing the total number of ASVs per time point, stratified into shared (across ≥ 2 time points) and unique (exclusive to one time point) categories. Total ASVs analyzed: 19,989 Fig. 6Longitudinal core taxa stability analysis (aggregated intervals). Note: Boxplot of Jaccard similarity indices across aggregated time intervals: Interval 0 (adjacent: T0-T1 / T1-T2) and Interval 1 (non-adjacent: T0-T2). The red diamond indicates the median; error bars represent standard deviation (SD) Longitudinal core taxa stability analysis (aggregated intervals). Note: Boxplot of Jaccard similarity indices across aggregated time intervals: Interval 0 (adjacent: T0-T1 / T1-T2) and Interval 1 (non-adjacent: T0-T2). The red diamond indicates the median; error bars represent standard deviation (SD) Fig. 7Jaccard similarity by individual time point comparison. Note: Boxplot of Jaccard similarity indices for pairwise comparisons (T0-T1, T0-T2, T1-T2). Values (mean ± SD) are labeled for each comparison, and the red diamond indicates the median Jaccard similarity by individual time point comparison. Note: Boxplot of Jaccard similarity indices for pairwise comparisons (T0-T1, T0-T2, T1-T2). Values (mean ± SD) are labeled for each comparison, and the red diamond indicates the median Fig. 8Rarefaction curves assessing sequencing depth sufficiency. Note: X-axis: Number of sequencing reads, Y-axis: Observed OTUs. Curves are color-coded: Pink curve (T0), dark blue (T1), light blue (T2). When the sequencing depth reached 30,000 reads (the upper limit of the x-axis), the curves tended to plateau, confirming sufficient sequencing depth for capturing microbial community diversity Rarefaction curves assessing sequencing depth sufficiency. Note: X-axis: Number of sequencing reads, Y-axis: Observed OTUs. Curves are color-coded: Pink curve (T0), dark blue (T1), light blue (T2). When the sequencing depth reached 30,000 reads (the upper limit of the x-axis), the curves tended to plateau, confirming sufficient sequencing depth for capturing microbial community diversity Result Baseline characteristics of NPC patients and healthy controls, and longitudinal changes in patients during radiotherapy A total of 31 patients with nasopharyngeal carcinoma (NPC) were enrolled, with a mean age of 46.10 ± 11.60 years (range: 28–69 years). According to clinical staging, 3 patients had stage II or lower disease, 11 had stage III, and 17 had stage IVa. Treatment regimens included CCRT alone (n = 16) and induction chemotherapy followed by CCRT (n = 15), with all patients receiving definitive intensity-modulated radiotherapy (IMRT). The 31 age-matched healthy controls had a mean age of 42.50 ± 9.60 years (range: 31–62 years). Demographic and clinical characteristics, including ethnicity, lifestyle factors, and clinical variables, were comparable between NPC patients and healthy controls, with no statistically significant differences observed (all P > 0.05; Table 1). Among NPC patients, longitudinal assessments across radiotherapy time points (T0, T1, T2) revealed no significant changes in body weight, food intake score, stool form, opioid use and enteral feeding (all P > 0.05; Table 2). By contrast, significant temporal changes were observed for mucositis grade, antiemetic use, PPI use, and enteral nutritional support. Table 1Baseline demographic and clinical characteristics of NPC patients and healthy controlsProjectsClassificationT0HCAbsolute differences (T0 - HC)95% CIsX2/Z/t P Age (years)≥ 1846.10 ± 11.6042.50 ± 9.603.60(-1.81, 9.01)1.3310.188GenderMale24 (77.42%)22 (70.97%)6.45%(-15.27%, 28.17%)0.3370.561Female7 (22.58%)9 (29.03%)-6.45%(-28.17%, 15.27%)Ethnic groupThe Han people18 (58.06%)14 (45.16%)12.90%(-8.72%, 34.52%)1.0330.309The Zhuang people13 (41.94%)17 (54.84%)-12.90%(-34.52%, 8.72%)Educational levelJunior high school and below18 (58.06%)15 (48.39%)9.67%(-12.05%, 31.39%)0.5860.746Senior high school / Secondary Technical school8 (25.81%)10 (32.26%)-6.45%(-28.17%, 15.27%)Junior College / Bachelor’s Degree and above5 (16.13%)6 (19.35%)-3.22%(-20.75%, 14.31%)SpouseYes24 (77.42%)26 (83.87%)-6.45%(-28.17%, 15.27%)0.4130.520No7 (22.58%)5 (16.13%)6.45%(-15.27%, 28.17%)Place of residenceThe urban areas11 (35.48%)15 (48.39%)-12.91%(-34.53%, 8.71%)1.0600.303The rural areas20 (64.52%)16 (51.61%)12.91%(-8.71%, 34.53%)Monthly per capita household income (yuan)< 30008 (25.81%)6 (19.36%)6.45%(-15.27%, 28.17%)0.4400.8033000–600011 (35.48%)11 (35.48%)0.00%(-21.63%, 21.63%)> 600012 (38.71%)14 (45.16%)-6.45%(-28.17%, 15.27%)Medical insurance payment typeUrban and Rural Residents’ Basic Medical Insurance21 (67.74%)20 (65.52%)2.22%(-19.40%, 23.84%)0.0720.788Urban Employees’ Basic Medical Insurance10 (32.26%)11 (35.48%)-3.22%(-20.75%, 14.31%)Smoking historyYes11 (35.48%)13 (41.94%)-6.46%(-28.18%, 15.26%)0.2720.602No20 (64.52%)18 (58.06%)6.46%(-15.26%, 28.18%)Drinking historyYes15 (48.39%)12 (38.71%)9.68%(-12.04%, 31.40%)0.5910.442No16 (51.61%)19 (61.29%)-9.68%(-31.40%, 12.04%)Food intake (score)3 ~ 54.58 ± 0.504.45 ± 0.620.13(-0.16, 0.42)0.8970.373Stool formType II6 (29.03%)8 (25.81%)3.22%(-20.75%, 14.31%)0.9240.356Type III15 (38.71%)11 (35.48%)3.23%(-20.74%, 14.32%)Type IVa10 (32.26%)12 (38.71%)-6.45%(-28.17%, 15.27%)Body weight45.1 ~ 81.163.01 ± 9.2363.31 ± 9.87-0.30(-5.15, 4.55)0.1660.685EBV statusNo infection0 (100%)31 (100%)-100.00%(-100.00%, -100.00%)--Previous infection31 (100%)0100.00%(100.00%, 100.00%)Recent antibiotics useYes00----No31(100%)31(100%)----Recent probiotics useYes00----No31(100%)31(100%)----Diet typeSpecialized dietary00----Tube feeding00----Regular diet31(100%)31(100%)----Gastrointestinal (GI) symptomsYes00----No31(100%)31(100%)----Recent Proton Pump Inhibitors (PPI) useYes00----No31(100%)31(100%)----Recent Metformin useYes00----No31(100%)31(100%)----Note: This table shows baseline demographic and clinical characteristics of NPC patients and healthy controls. Continuous variables (e.g., age, food intake score) are mean ± SD and analyzed via Student’s t-test, while categorical variables are n (%) and analyzed via chi-square tests. Group comparisons include absolute differences and 95% CIs Table 1Baseline demographic and clinical characteristics of NPC patients and healthy controlsProjectsClassificationT0HCAbsolute differences (T0 - HC)95% CIsX2/Z/t P Age (years)≥ 1846.10 ± 11.6042.50 ± 9.603.60(-1.81, 9.01)1.3310.188GenderMale24 (77.42%)22 (70.97%)6.45%(-15.27%, 28.17%)0.3370.561Female7 (22.58%)9 (29.03%)-6.45%(-28.17%, 15.27%)Ethnic groupThe Han people18 (58.06%)14 (45.16%)12.90%(-8.72%, 34.52%)1.0330.309The Zhuang people13 (41.94%)17 (54.84%)-12.90%(-34.52%, 8.72%)Educational levelJunior high school and below18 (58.06%)15 (48.39%)9.67%(-12.05%, 31.39%)0.5860.746Senior high school / Secondary Technical school8 (25.81%)10 (32.26%)-6.45%(-28.17%, 15.27%)Junior College / Bachelor’s Degree and above5 (16.13%)6 (19.35%)-3.22%(-20.75%, 14.31%)SpouseYes24 (77.42%)26 (83.87%)-6.45%(-28.17%, 15.27%)0.4130.520No7 (22.58%)5 (16.13%)6.45%(-15.27%, 28.17%)Place of residenceThe urban areas11 (35.48%)15 (48.39%)-12.91%(-34.53%, 8.71%)1.0600.303The rural areas20 (64.52%)16 (51.61%)12.91%(-8.71%, 34.53%)Monthly per capita household income (yuan)< 30008 (25.81%)6 (19.36%)6.45%(-15.27%, 28.17%)0.4400.8033000–600011 (35.48%)11 (35.48%)0.00%(-21.63%, 21.63%)> 600012 (38.71%)14 (45.16%)-6.45%(-28.17%, 15.27%)Medical insurance payment typeUrban and Rural Residents’ Basic Medical Insurance21 (67.74%)20 (65.52%)2.22%(-19.40%, 23.84%)0.0720.788Urban Employees’ Basic Medical Insurance10 (32.26%)11 (35.48%)-3.22%(-20.75%, 14.31%)Smoking historyYes11 (35.48%)13 (41.94%)-6.46%(-28.18%, 15.26%)0.2720.602No20 (64.52%)18 (58.06%)6.46%(-15.26%, 28.18%)Drinking historyYes15 (48.39%)12 (38.71%)9.68%(-12.04%, 31.40%)0.5910.442No16 (51.61%)19 (61.29%)-9.68%(-31.40%, 12.04%)Food intake (score)3 ~ 54.58 ± 0.504.45 ± 0.620.13(-0.16, 0.42)0.8970.373Stool formType II6 (29.03%)8 (25.81%)3.22%(-20.75%, 14.31%)0.9240.356Type III15 (38.71%)11 (35.48%)3.23%(-20.74%, 14.32%)Type IVa10 (32.26%)12 (38.71%)-6.45%(-28.17%, 15.27%)Body weight45.1 ~ 81.163.01 ± 9.2363.31 ± 9.87-0.30(-5.15, 4.55)0.1660.685EBV statusNo infection0 (100%)31 (100%)-100.00%(-100.00%, -100.00%)--Previous infection31 (100%)0100.00%(100.00%, 100.00%)Recent antibiotics useYes00----No31(100%)31(100%)----Recent probiotics useYes00----No31(100%)31(100%)----Diet typeSpecialized dietary00----Tube feeding00----Regular diet31(100%)31(100%)----Gastrointestinal (GI) symptomsYes00----No31(100%)31(100%)----Recent Proton Pump Inhibitors (PPI) useYes00----No31(100%)31(100%)----Recent Metformin useYes00----No31(100%)31(100%)----Note: This table shows baseline demographic and clinical characteristics of NPC patients and healthy controls. Continuous variables (e.g., age, food intake score) are mean ± SD and analyzed via Student’s t-test, while categorical variables are n (%) and analyzed via chi-square tests. Group comparisons include absolute differences and 95% CIs Baseline demographic and clinical characteristics of NPC patients and healthy controls Note: This table shows baseline demographic and clinical characteristics of NPC patients and healthy controls. Continuous variables (e.g., age, food intake score) are mean ± SD and analyzed via Student’s t-test, while categorical variables are n (%) and analyzed via chi-square tests. Group comparisons include absolute differences and 95% CIs Table 2Clinical and demographic characteristics of npc patients across radiotherapy time pointsProjectsClassificationT0T1T2Absolute differences between groups (T0 vs. T1, T0 vs. T2, T1 vs. T2)95% CIs(T0 vs. T1, T0 vs. T2, T1 vs. T2)F/H/χ² P Body weight41.9 ~ 81.163.31 ± 9.8661.49 ± 9.5958.73 ± 8.921.82; 4.58; 2.76(-2.81, 7.07); (-0.20, 9.36); (-2.25, 7.15)1.9200.060Food intake (score)1 ~ 54.45 ± 0.623.58 ± 0.962.87 ± 0.670.87; 1.58; 0.71(0.46, 1.28); (1.25, 1.91); (0.29, 1.13)1.2480.268Stool formType II6 (29.03%)12 (38.71%)10 (32.26%)-9.68%; -3.23%; 6.45%(-31.40%, 12.04%); (-20.74%, 14.32%); (-15.27%, 28.17%)2.7720.250Type III15 (38.71%)11 (35.48%)12 (38.71%)3.23%; 0.00%; -3.23%(-20.74%, 14.32%); (-21.63%, 21.63%); (-20.74%, 14.32%)Type IV10 (32.26%)8 (25.81%)9 (29.03%)6.45%; 3.23%; -3.22%(-15.27%, 28.17%); (-20.74%, 14.32%); (-20.75%, 14.31%)Mucositis gradeGrade 0015 (48.39%)7 (22.58%)48.39%; 22.58%; 25.81%(31.20%, 65.58%); (5.37%, 39.79%); (8.10%, 43.52%)18.365< 0.001***Grade I015 (48.39%)18 (58.06%)48.39%; 58.06%; -9.67%(31.20%, 65.58%); (40.87%, 75.25%); (-26.88%, 7.54%)Grade II01 (3.22%)6 (19.36%)3.22%; 19.36%; -16.14%(-3.01%, 9.45%); (2.15%, 36.57%); (-33.35%, 1.07%)AntiemeticsYes014 (45.16%)12 (38.71%)45.16%; 38.71%; 6.45%(28.05%, 62.27%); (21.60%, 55.82%); (-10.76%, 23.66%)6.7530.034*No31 (100%)17 (54.84%)19 (61.29%)-45.16%; -38.71%; -6.45%(-62.27%, -28.05%); (-55.82%, -21.60%); (-23.66%, 10.76%)OpioidsYes01 (3.23%)1 (3.23%)3.23%; 3.23%; 0.00%(-3.01%, 9.45%); (-3.01%, 9.45%); (-6.02%, 6.02%)1.0220.600No31 (100%)30 (96.77%)30 (96.77%)-3.23%; -3.23%; 0.00%(-9.46%, 3.00%); (-9.46%, 3.00%); (-6.02%, 6.02%)PPIYes014 (45.16%)12 (38.71%)45.16%; 38.71%; 6.45%(27.64%, 62.68%); (21.56%, 55.86%); (-18.07%, 31.07%)18.365< 0.001***No31 (100%)17 (54.84%)19 (61.29%)-45.16%; -38.71%; -6.45%(-62.68%, -27.64%); (-55.86%, -21.56%); (-31.07%, 18.07%)Enteral nutritional supportYes05 (16.13%)6 (19.35%)16.13%; 19.35%; -3.22%(3.19%, 29.07%); (5.43%, 33.27%); (-22.23%, 15.79%)6.3920.041*No31 (100%)26 (83.87%)25 (80.65%)-16.13%; -19.35%; 3.22%(-29.07%, -3.19%); (-33.27%, -5.43%); (-15.79%, 22.23%)Enteral feedingYes000----No31 (100%)31 (100%)31 (100%)--Note: This table summarizes changes in key clinical variables (weight, medication use, nutritional support) for NPC patients at T0, T1, and T2. Continuous variables (e.g., weight) are presented as mean ± SD and analyzed via repeated-measures ANOVA, while categorical variables (e.g., medication use) are n (%) and analyzed via chi-square tests. Pairwise comparisons include absolute differences and 95% CIs. Significance: *P < 0.05, **P < 0.01, ***P < 0.001 Table 2Clinical and demographic characteristics of npc patients across radiotherapy time pointsProjectsClassificationT0T1T2Absolute differences between groups (T0 vs. T1, T0 vs. T2, T1 vs. T2)95% CIs(T0 vs. T1, T0 vs. T2, T1 vs. T2)F/H/χ² P Body weight41.9 ~ 81.163.31 ± 9.8661.49 ± 9.5958.73 ± 8.921.82; 4.58; 2.76(-2.81, 7.07); (-0.20, 9.36); (-2.25, 7.15)1.9200.060Food intake (score)1 ~ 54.45 ± 0.623.58 ± 0.962.87 ± 0.670.87; 1.58; 0.71(0.46, 1.28); (1.25, 1.91); (0.29, 1.13)1.2480.268Stool formType II6 (29.03%)12 (38.71%)10 (32.26%)-9.68%; -3.23%; 6.45%(-31.40%, 12.04%); (-20.74%, 14.32%); (-15.27%, 28.17%)2.7720.250Type III15 (38.71%)11 (35.48%)12 (38.71%)3.23%; 0.00%; -3.23%(-20.74%, 14.32%); (-21.63%, 21.63%); (-20.74%, 14.32%)Type IV10 (32.26%)8 (25.81%)9 (29.03%)6.45%; 3.23%; -3.22%(-15.27%, 28.17%); (-20.74%, 14.32%); (-20.75%, 14.31%)Mucositis gradeGrade 0015 (48.39%)7 (22.58%)48.39%; 22.58%; 25.81%(31.20%, 65.58%); (5.37%, 39.79%); (8.10%, 43.52%)18.365< 0.001***Grade I015 (48.39%)18 (58.06%)48.39%; 58.06%; -9.67%(31.20%, 65.58%); (40.87%, 75.25%); (-26.88%, 7.54%)Grade II01 (3.22%)6 (19.36%)3.22%; 19.36%; -16.14%(-3.01%, 9.45%); (2.15%, 36.57%); (-33.35%, 1.07%)AntiemeticsYes014 (45.16%)12 (38.71%)45.16%; 38.71%; 6.45%(28.05%, 62.27%); (21.60%, 55.82%); (-10.76%, 23.66%)6.7530.034*No31 (100%)17 (54.84%)19 (61.29%)-45.16%; -38.71%; -6.45%(-62.27%, -28.05%); (-55.82%, -21.60%); (-23.66%, 10.76%)OpioidsYes01 (3.23%)1 (3.23%)3.23%; 3.23%; 0.00%(-3.01%, 9.45%); (-3.01%, 9.45%); (-6.02%, 6.02%)1.0220.600No31 (100%)30 (96.77%)30 (96.77%)-3.23%; -3.23%; 0.00%(-9.46%, 3.00%); (-9.46%, 3.00%); (-6.02%, 6.02%)PPIYes014 (45.16%)12 (38.71%)45.16%; 38.71%; 6.45%(27.64%, 62.68%); (21.56%, 55.86%); (-18.07%, 31.07%)18.365< 0.001***No31 (100%)17 (54.84%)19 (61.29%)-45.16%; -38.71%; -6.45%(-62.68%, -27.64%); (-55.86%, -21.56%); (-31.07%, 18.07%)Enteral nutritional supportYes05 (16.13%)6 (19.35%)16.13%; 19.35%; -3.22%(3.19%, 29.07%); (5.43%, 33.27%); (-22.23%, 15.79%)6.3920.041*No31 (100%)26 (83.87%)25 (80.65%)-16.13%; -19.35%; 3.22%(-29.07%, -3.19%); (-33.27%, -5.43%); (-15.79%, 22.23%)Enteral feedingYes000----No31 (100%)31 (100%)31 (100%)--Note: This table summarizes changes in key clinical variables (weight, medication use, nutritional support) for NPC patients at T0, T1, and T2. Continuous variables (e.g., weight) are presented as mean ± SD and analyzed via repeated-measures ANOVA, while categorical variables (e.g., medication use) are n (%) and analyzed via chi-square tests. Pairwise comparisons include absolute differences and 95% CIs. Significance: *P < 0.05, **P < 0.01, ***P < 0.001 Clinical and demographic characteristics of npc patients across radiotherapy time points Note: This table summarizes changes in key clinical variables (weight, medication use, nutritional support) for NPC patients at T0, T1, and T2. Continuous variables (e.g., weight) are presented as mean ± SD and analyzed via repeated-measures ANOVA, while categorical variables (e.g., medication use) are n (%) and analyzed via chi-square tests. Pairwise comparisons include absolute differences and 95% CIs. Significance: *P < 0.05, **P < 0.01, ***P < 0.001 Structural characteristics of gut microbiota in NPC patients Gut microbiota diversity between NPC patients and healthy controls In this study, Alpha diversity indices, namely the Chao1 index, Pielou-E index, Shannon index, and Simpson index, were used to evaluate microbial community characteristics, including the species richness, evenness, and diversity of microbial communities. All Alpha diversity indices were calculated after rarefaction to eliminate the impact of uneven sequencing depth. For the primary cross-sectional comparison (T0 vs. HC), the Wilcoxon rank-sum test (R v3.4.4) was used to compare the Alpha diversity indices. Effect sizes were calculated to quantify the magnitude of differences, and post-hoc tests with BH correction were performed to control for FDR. Violin plots visualized the results of the Alpha diversity index analysis (Fig. 9a, b, c, d). The results indicated that the Shannon index of T0 was lower than that of HC (effect size = 0.317, FDR-adjusted P = 0.075), the Chao1 index of T0 was also lower (effect size = 0.565, FDR-adjusted P < 0.001). There were no statistically significant differences in the Simpson index (effect size = 0.208, FDR-adjusted P = 0.613) and Pielou-e index (effect size = 0.162, FDR-adjusted P = 1.000) between the two groups. Beta diversity reflects differences in microbial community composition and structure between samples. In this study, ANOSIM was used to compare similarities in gut microbiota composition, while PERMANOVA was applied to test the statistical significance of group differences, and Betadisper was used for homogeneity of dispersion test. For the primary cross-sectional comparison (T0 vs. HC), we analyzed beta diversity using Bray-Curtis and weighted Unifrac distance metrics, and the generated principal co-ordinate analysis (PCoA) plots (Fig. 10a, b) visualized the differences in microbial community comparison and structure. Each sample is represented by a single point, and the closer the distance between the points, the higher the similarity in the composition and structure of the microbiota between the respective samples. For the Bray-Curtis distance, there was a statistically significant difference in the microbial composition and structure between T0 and HC groups (PERMANOVA R2 = 0.037, F = 2.301, P = 0.001; ANOSIM R = 0.147, P = 0.001; 999 permutations, no overdispersion / constraints). In contrast, no statistically significant difference based on weighted Unifrac distance was detected (PERMANOVA R2 = 0.027, F = 1.693, P = 0.072; ANOSIM R = 0.021, P = 0.118; 999 permutations, no overdispersion / constraints). Additionally, betadisper tests confirmed homogeneous variances (Bray-Curtis: F = 1.211, P = 0.283; weighted_Unifrac: F = 0.216, P = 0.640). Fig. 9Violin plot of Alpha diversity index analysis. Note: Violin plot showed analysis of differences in Alpha diversity indices between patients (T0, light purple) and healthy controls (HC, blue). *P < 0.1; ***P < 0.001 Violin plot of Alpha diversity index analysis. Note: Violin plot showed analysis of differences in Alpha diversity indices between patients (T0, light purple) and healthy controls (HC, blue). *P < 0.1; ***P < 0.001 Fig. 10Principal co-ordinate analysis plot. Note: PCoA plot showed the differences in the composition and structure of microbial communities across samples between patients (T0, light purple points) and healthy controls (HC, light blue points) Principal co-ordinate analysis plot. Note: PCoA plot showed the differences in the composition and structure of microbial communities across samples between patients (T0, light purple points) and healthy controls (HC, light blue points) Differentially abundant gut microbiota taxa between NPC patients and healthy controls LEfSe analysis and ALDEx2 were used to identify specific differentially abundant taxa in the gut microbiota between patients with NPC and healthy controls. The BH method was applied for FDR correction, and volcano plots were used to visualize the observed differential abundance profiles. For identifying differentially abundant taxa in the primary cross-sectional comparison (T0 vs. HC), LEfSe was used as an exploratory method and identified 28 genus-level taxa with differentially relative abundance (LDA > 3, BH-corrected P < 0.05; Fig. 11). ALDEx2 was used as the primary method and detected 33 genus-level taxa with differentially relative abundance (FDR < 0.1, |log2 FC| > 0.5; Fig. 12). Notably, 16 core differentially abundant taxa overlapped between the two analytical approaches (Figs. 13a, b; Table 3), with ALDEx2-derived log2FC (and 95% CIs) as the primary effect measure. Specifically, these taxa included the genera Lactiplantibacillus, Lactobacillus, Akkermansia, Enterococcus, Veillonella, Escherichia-Shigella, and Megasphaera, all of which were significantly enriched in the T0 group relative to the Healthy group. In contrast, the genera Fusicatenibacter, Alistipes, Ligilactobacillus, Lachnospira, Ruminococcus, Monoglobus, Subdoligranulum, NK4A214_group, and Clostridia_UCG-014_ unclassified were significantly enriched in the healthy controls group compared with the T0 group. Fig. 11LEfSe LDA score bar chart. Note: LEfSe (Kruskal-Wallis rank sum test and Wilcoxon rank sum test) identified genus-level differentially abundant taxa between T0 and HC groups, using threshold of LDA > 3 and BH-corrected P < 0.05 LEfSe LDA score bar chart. Note: LEfSe (Kruskal-Wallis rank sum test and Wilcoxon rank sum test) identified genus-level differentially abundant taxa between T0 and HC groups, using threshold of LDA > 3 and BH-corrected P < 0.05 Fig. 12ALDEx2 volcano plot. Note: Volcano plot visualizes genus-level differential abundance between T0 and HC groups. X-axis: log2FC (effect size), Y-axis: -log10 (FDR-adjusted P value). Red dots denote significant taxa (FDR < 0.1, |log2FC| > 0.5): upper right = enriched in T0, upper left = enriched in Health, and gray dots = non-significant taxa ALDEx2 volcano plot. Note: Volcano plot visualizes genus-level differential abundance between T0 and HC groups. X-axis: log2FC (effect size), Y-axis: -log10 (FDR-adjusted P value). Red dots denote significant taxa (FDR < 0.1, |log2FC| > 0.5): upper right = enriched in T0, upper left = enriched in Health, and gray dots = non-significant taxa Fig. 13Venn diagram and heatmap of core differential taxa. Note: (a) Venn diagram shows overlap of genus-level differential taxa identified by LEfSe (28 taxa) and ALDEx2 (33 taxa), and 16 taxa are shared, supporting result consistency. (b) Heatmap displays relative abundance of the 16 shared taxa across T0 / HC samples. Horizontal axis: samples; vertical axis: taxa (genus level); color intensity corresponds to relative abundance (darker = higher abundance) Venn diagram and heatmap of core differential taxa. Note: (a) Venn diagram shows overlap of genus-level differential taxa identified by LEfSe (28 taxa) and ALDEx2 (33 taxa), and 16 taxa are shared, supporting result consistency. (b) Heatmap displays relative abundance of the 16 shared taxa across T0 / HC samples. Horizontal axis: samples; vertical axis: taxa (genus level); color intensity corresponds to relative abundance (darker = higher abundance) Table 3Differentially abundant gut microbiota genera between NPC patients and healthy controlsFeature (genus level)Group comparisonlog2FC95%CI (lower)95% CI (upper)P-adj (Welch)P-adj (Wilcoxon)−log₁₀ (FDR-adjusted P)LactiplantibacillusT02.8302.8082.8541.739 × 10 − 9 ***1.918 × 10 − 15 ***8.760LactobacillusT00.9360.9130.9603.276 × 10 − 5 ***1.462 × 10 − 5 ***4.485FusicatenibacterHC−0.777−0.801−0.7544.627 × 10 − 4 ***6.533 × 10 − 4 ***3.335AkkermansiaT00.7500.7260.7730.002 **2.599 × 10 − 4 ***2.634AlistipesHC−0.631−0.655−0.6080.009 **0.003 **2.066EnterococcusT00.8080.7840.8310.013 **0.006 **1.888VeillonellaT00.5330.5090.5560.015 **0.019 **1.817LigilactobacillusHC−0.603−0.626−0.5790.020 **0.026 **1.697LachnospiraHC−0.549−0.573−0.5260.021 **0.080 *1.683Escherichia-ShigellaT00.5620.5390.5860.021 **0.026 **1.682MegasphaeraT00.6170.5940.6410.022 **0.020 **1.653RuminococcusHC−0.532−0.555−0.5080.033 **0.027 **1.481MonoglobusHC−0.503−0.527−0.4800.034 **0.018 **1.472SubdoligranulumHC−0.559−0.583−0.5360.035 **0.053 *1.457NK4A214_groupHC−0.538−0.562−0.5150.080 *0.088 *1.096Clostridia_UCG−014_unclassifiedHC−0.515−0.538−0.4910.082 *0.079 *1.087Note: ***P < 0.001, **P < 0.05, *P < 0.1 Table 3Differentially abundant gut microbiota genera between NPC patients and healthy controlsFeature (genus level)Group comparisonlog2FC95%CI (lower)95% CI (upper)P-adj (Welch)P-adj (Wilcoxon)−log₁₀ (FDR-adjusted P)LactiplantibacillusT02.8302.8082.8541.739 × 10 − 9 ***1.918 × 10 − 15 ***8.760LactobacillusT00.9360.9130.9603.276 × 10 − 5 ***1.462 × 10 − 5 ***4.485FusicatenibacterHC−0.777−0.801−0.7544.627 × 10 − 4 ***6.533 × 10 − 4 ***3.335AkkermansiaT00.7500.7260.7730.002 **2.599 × 10 − 4 ***2.634AlistipesHC−0.631−0.655−0.6080.009 **0.003 **2.066EnterococcusT00.8080.7840.8310.013 **0.006 **1.888VeillonellaT00.5330.5090.5560.015 **0.019 **1.817LigilactobacillusHC−0.603−0.626−0.5790.020 **0.026 **1.697LachnospiraHC−0.549−0.573−0.5260.021 **0.080 *1.683Escherichia-ShigellaT00.5620.5390.5860.021 **0.026 **1.682MegasphaeraT00.6170.5940.6410.022 **0.020 **1.653RuminococcusHC−0.532−0.555−0.5080.033 **0.027 **1.481MonoglobusHC−0.503−0.527−0.4800.034 **0.018 **1.472SubdoligranulumHC−0.559−0.583−0.5360.035 **0.053 *1.457NK4A214_groupHC−0.538−0.562−0.5150.080 *0.088 *1.096Clostridia_UCG−014_unclassifiedHC−0.515−0.538−0.4910.082 *0.079 *1.087Note: ***P < 0.001, **P < 0.05, *P < 0.1 Differentially abundant gut microbiota genera between NPC patients and healthy controls Note: ***P < 0.001, **P < 0.05, *P < 0.1 Changes in gut microbiota of NPC patients during radiotherapy Changes in gut microbiota diversity of NPC patients during radiotherapy In this study, Alpha diversity indices, namely the Chao1 index, Pielou-E index, Shannon index, and Simpson index, were used to evaluate the species richness, evenness, and diversity of gut microbiota in NPC patients. All Alpha diversity indices were calculated after rarefaction to eliminate the impact of uneven sequencing depth on diversity analysis. Violin plots were generated to visualize the results of the Alpha diversity index analysis, as shown in Figs. 14a, b, c, d. Linear mixed-effects models were employed to assess the impact of radiotherapy time points (T0/T1/T2) on gut microbiota alpha diversity indices, with ANOVA-type F tests evaluating overall time-point effects and post-hoc pairwise contrasts examining specific group differences. For the Chao1 index, which reflects microbial richness, the overall effect of radiotherapy time point was statistically significant (fixed effect, F = 13.572, df = 2 / 60, P = 1.372 × 10− 5). Post-hoc contrasts revealed that the Chao1 index at T0 was significantly lower than at both T1 (estimate = -112.913, SE = 28.770, P_adj = 3.395 × 10− 4) and T2 (estimate = − 141.828, SE = 28.770, P_adj = 2.050 × 10− 5), whereas no significant difference was observed between T1 and T2 (estimate = − 28.915, SE = 28.770, P_adj = 0.319). In contrast, for the Shannon (diversity), Simpson (dominance), and Pielou-e (evenness) indices, no statistically significant overall effects of radiotherapy time point were detected (Shannon: F = 1.601, P = 0.210; Simpson: F = 0.841, P = 0.436; Pielou-e: F = 0.314, P = 0.732), and post-hoc pairwise comparisons further confirmed that none of the time-point contrasts reached statistical significance (all P > 0.05). Beta diversity quantifies compositional differences in gut microbiota across distinct samples, and to characterize these differences among NPC patients at T0, T1, and T2, we used principal coordinates analysis (PCoA) to visualize community structure via two widely used distance metrics, paired with statistical tests to validate group separation (Figs. 15a, b). For the Bray-Curtis distance (Fig. 17a), statistical analyses (999 permutations) revealed no group difference, with PERMANOVA (R2 = 0.028, F = 1.298, P = 0.051). For the unweighted UniFrac distance (Fig. 17b), there was stronger evidence of group differentiation, with PERMANOVA (R2 = 0.069, F = 3.351, P = 0.001). Critically, Betadisper tests confirmed homogeneous within-group variances for both distance metrics (Bray-Curtis: P = 0.436; unweighted UniFrac: P = 0.612), which ensures that the observed compositional differences across time points were not driven by heterogeneity in within-group dispersion. Fig. 14Violin plot of alpha diversity index in NPC patients at three time points. Note: Violin plots showed the differences of alpha diversity indices for NPC patients at T0 (purple), T1 (light blue), and T2 (dark blue). ***P < 0.001 Violin plot of alpha diversity index in NPC patients at three time points. Note: Violin plots showed the differences of alpha diversity indices for NPC patients at T0 (purple), T1 (light blue), and T2 (dark blue). ***P < 0.001 Fig. 15Principal coordinate analysis plots of gut microbiota composition. Note: PCoA plot shows the differences in the composition and structure of microbial communities across samples among NPC patients at different time points (T0/T1/T2). Each point represents a single sample, with points of different colors corresponding to samples from different time points: purple (T0), light blue (T1), and dark blue (T2). *P < 0.1; **P < 0.05 Principal coordinate analysis plots of gut microbiota composition. Note: PCoA plot shows the differences in the composition and structure of microbial communities across samples among NPC patients at different time points (T0/T1/T2). Each point represents a single sample, with points of different colors corresponding to samples from different time points: purple (T0), light blue (T1), and dark blue (T2). *P < 0.1; **P < 0.05 Identification of differentially abundant gut microbiota taxa in NPC patients during radiotherapy via LEfSe and ALDEx2 LEfSe and ALDEx2 were used to identify specific differentially abundant taxa in the gut microbiota among patients with NPC at different time points during radiotherapy. The BH method was applied for FDR correction, and volcano plots were used to visualize the observed differential abundance profiles. At the genus levels, LEfSe analysis identified 21 gut microbial taxa with differentially relative abundance among patients with NPC across the three time points (T0, T1, and T2), with 4 taxa significantly enriched in the T1 group and 17 in the T2 group (Fig. 16). At the same taxonomic levels (genus), ALDEx2 detected 3 gut microbial taxa with differential relative abundances when comparing patients with NPC at T0 versus T1 (T0 vs. T1), with 1 taxon wassignificantly enriched in the T0 group and 2 in the T1 group (Fig. 17a; Table 4). Additionally, ALDEx2 detected 18 gut microbial taxa with differential relative abundances in the T0 versus T2 comparison for patients with NPC (T0 vs. T2), with 3 taxa significantly enriched in the T0 group and 15 in the T2 group (Fig. 17b; Table 4). In contrast, ALDEx2 failed to identify any gut microbial taxa with differential relative abundances betweenpatients with NPC at T1 and T2 (T1 vs. T2; Fig. 17c; Table 4). For the intragroup comparison of patients with NPC across the three time points (T0 vs. T1 vs. T2), ALDEx2 identified 12 gut microbial taxa with differential relative abundances. Notably, 7 core differentially abundant taxa overlapped between the two analytical approaches (LEfSe and ALDEx2) (Figs. 18a, b; Table 4). Specifically, these taxa included the genera Hungatella, Pantoea, UBA1819, Alistipes, Desulfovibrio, Lachnospiraceae_NK4A136_group, and Pseudomonas. Fig. 16LEfSe LDA Score Bar Chart. Note: LEfSe (Kruskal-Wallis rank sum test and Wilcoxon rank sum test) identified genus-level differentially abundant taxa among NPC patient across the three time points (T0, T1, and T2), using threshold of LDA > 3 and BH-corrected P < 0.05 LEfSe LDA Score Bar Chart. Note: LEfSe (Kruskal-Wallis rank sum test and Wilcoxon rank sum test) identified genus-level differentially abundant taxa among NPC patient across the three time points (T0, T1, and T2), using threshold of LDA > 3 and BH-corrected P < 0.05 Fig. 17ALDEx2 volcano plot. Note: Volcano plot visualizes genus-level differential abundance among NPC patient across the three time points (T0, T1, and T2). X-axis: log2FC (effect size), Y-axis: -log10 (FDR-adjusted P value). Red dots denote significant taxa (FDR < 0.1, |log2FC| > 0.5): upper right = enriched in T1 / T2, upper left = enriched in T0, and gray dots = non-significant taxa ALDEx2 volcano plot. Note: Volcano plot visualizes genus-level differential abundance among NPC patient across the three time points (T0, T1, and T2). X-axis: log2FC (effect size), Y-axis: -log10 (FDR-adjusted P value). Red dots denote significant taxa (FDR < 0.1, |log2FC| > 0.5): upper right = enriched in T1 / T2, upper left = enriched in T0, and gray dots = non-significant taxa Fig. 18Venn diagram and heatmap of core differential taxa. Note: (a) Venn diagram (genus level) shows overlap of differentially abundant taxa identified by ALDEx2 (12 taxa) and LEfSe (21 taxa), and 7 taxa are shared, supporting result consistency. (b) Heatmap displays relative abundance of the 7 shared taxa across NPC patient samples (T0 / T1 / T2 groups). Horizontal axis: samples; vertical axis: genus-level taxa; color intensity corresponds to relative abundance (darker = higher abundance) Venn diagram and heatmap of core differential taxa. Note: (a) Venn diagram (genus level) shows overlap of differentially abundant taxa identified by ALDEx2 (12 taxa) and LEfSe (21 taxa), and 7 taxa are shared, supporting result consistency. (b) Heatmap displays relative abundance of the 7 shared taxa across NPC patient samples (T0 / T1 / T2 groups). Horizontal axis: samples; vertical axis: genus-level taxa; color intensity corresponds to relative abundance (darker = higher abundance) Table 4Differentially abundant gut microbiota genera in NPC patients across radiotherapy time pointsFeature (genus level)Group comparisonlog2FC95% CI (lower)95% CI (upper)P_adj (Welch)P_adj (Wilcoxon)−log₁₀ (FDR-adjusted P)UBA1819T1 (T0 vs. T1)0.5940.5800.6070.031 **0.023 **1.512PhenylobacteriumT1 (T0 vs. T1)0.5920.5790.6060.045 **0.037 **1.346ActinomycesT0 (T0 vs. T1)−0.500−0.514−0.4870.077 *0.038 **1.112UBA1819T2 (T0 vs. T2)0.7600.7440.7760.005 **0.001 **2.350PantoeaT2 (T0 vs. T2)0.7060.6900.7220.011 **0.004 **1.970AnaerotruncusT2 (T0 vs. T2)0.6900.6740.7060.013 **0.002 **1.870EggerthellaT2 (T0 vs. T2)0.6800.6640.6960.026 **0.003 **1.582ActinomycesT0 (T0 vs. T2)−0.555−0.571−0.5390.031 **0.019 **1.511BrevundimonasT2(T0 vs. T2)0.5980.5820.6140.040 **0.029 *1.399PseudomonasT2 (T0 vs. T2)0.6580.6420.6740.043 **0.045 **1.368CarnobacteriumT2 (T0 vs. T2)0.6400.6240.6560.061 *0.081 **1.214Lachnospiraceae_NK4A136_groupT2 (T0 vs. T2)0.5720.5560.5880.063 *0.024 **1.197DesulfovibrioT2 (T0 vs. T2)0.6060.5900.6220.064 *0.054 *1.194Desulfovibrionaceae_unclassifiedT2 (T0 vs. T2)0.6090.5930.6250.076 *0.026 **1.119DubosiellaT2 (T0 vs. T2)0.6380.6220.6530.076 *0.031 **1.117FaecalibacteriumT0 (T0 vs. T2)−0.531−0.547−0.5150.080 *0.095 *1.098ParamuribaculumT2 (T0 vs. T2)0.5860.5700.6020.083 *0.030 **1.083HelicobacterT2 (T0 vs. T2)0.6110.5950.6270.084 *0.041 **1.075RoseburiaT0 (T0 vs. T2)−0.526−0.542−0.5100.096 *0.1661.019MuribaculumT2 (T0 vs. T2)0.5480.5320.5640.097 *0.043 *1.013SphingomonasT2 (T0 vs. T2)0.5510.5350.5670.099 *0.037 **1.005UBA1819T0 vs. T1 vs. T2///3.924 × 10 − 4***3.924 × 10 − 4***3.406PantoeaT0 vs. T1 vs. T2///0.007 **0.007 **2.140HungatellaT0 vs. T1 vs. T2///0.032 **0.032 **1.501DesulfovibrioT0 vs. T1 vs. T2///0.065 *0.065 *1.187Lachnospiraceae_NK4A136_groupT0 vs. T1 vs. T2///0.069 *0.069 *1.158PseudomonasT0 vs. T1 vs. T2///0.076 *0.076 *1.118AlistipesT0 vs. T1 vs. T2///0.081 *0.081 *1.094Note: Significance of difference (***P < 0.001, **P < 0.05, *P < 0.1) Table 4Differentially abundant gut microbiota genera in NPC patients across radiotherapy time pointsFeature (genus level)Group comparisonlog2FC95% CI (lower)95% CI (upper)P_adj (Welch)P_adj (Wilcoxon)−log₁₀ (FDR-adjusted P)UBA1819T1 (T0 vs. T1)0.5940.5800.6070.031 **0.023 **1.512PhenylobacteriumT1 (T0 vs. T1)0.5920.5790.6060.045 **0.037 **1.346ActinomycesT0 (T0 vs. T1)−0.500−0.514−0.4870.077 *0.038 **1.112UBA1819T2 (T0 vs. T2)0.7600.7440.7760.005 **0.001 **2.350PantoeaT2 (T0 vs. T2)0.7060.6900.7220.011 **0.004 **1.970AnaerotruncusT2 (T0 vs. T2)0.6900.6740.7060.013 **0.002 **1.870EggerthellaT2 (T0 vs. T2)0.6800.6640.6960.026 **0.003 **1.582ActinomycesT0 (T0 vs. T2)−0.555−0.571−0.5390.031 **0.019 **1.511BrevundimonasT2(T0 vs. T2)0.5980.5820.6140.040 **0.029 *1.399PseudomonasT2 (T0 vs. T2)0.6580.6420.6740.043 **0.045 **1.368CarnobacteriumT2 (T0 vs. T2)0.6400.6240.6560.061 *0.081 **1.214Lachnospiraceae_NK4A136_groupT2 (T0 vs. T2)0.5720.5560.5880.063 *0.024 **1.197DesulfovibrioT2 (T0 vs. T2)0.6060.5900.6220.064 *0.054 *1.194Desulfovibrionaceae_unclassifiedT2 (T0 vs. T2)0.6090.5930.6250.076 *0.026 **1.119DubosiellaT2 (T0 vs. T2)0.6380.6220.6530.076 *0.031 **1.117FaecalibacteriumT0 (T0 vs. T2)−0.531−0.547−0.5150.080 *0.095 *1.098ParamuribaculumT2 (T0 vs. T2)0.5860.5700.6020.083 *0.030 **1.083HelicobacterT2 (T0 vs. T2)0.6110.5950.6270.084 *0.041 **1.075RoseburiaT0 (T0 vs. T2)−0.526−0.542−0.5100.096 *0.1661.019MuribaculumT2 (T0 vs. T2)0.5480.5320.5640.097 *0.043 *1.013SphingomonasT2 (T0 vs. T2)0.5510.5350.5670.099 *0.037 **1.005UBA1819T0 vs. T1 vs. T2///3.924 × 10 − 4***3.924 × 10 − 4***3.406PantoeaT0 vs. T1 vs. T2///0.007 **0.007 **2.140HungatellaT0 vs. T1 vs. T2///0.032 **0.032 **1.501DesulfovibrioT0 vs. T1 vs. T2///0.065 *0.065 *1.187Lachnospiraceae_NK4A136_groupT0 vs. T1 vs. T2///0.069 *0.069 *1.158PseudomonasT0 vs. T1 vs. T2///0.076 *0.076 *1.118AlistipesT0 vs. T1 vs. T2///0.081 *0.081 *1.094Note: Significance of difference (***P < 0.001, **P < 0.05, *P < 0.1) Differentially abundant gut microbiota genera in NPC patients across radiotherapy time points Note: Significance of difference (***P < 0.001, **P < 0.05, *P < 0.1) Microbial diversity and taxonomic enrichment stratified by chemotherapy regimen Alpha diversity changes by chemotherapy regimen For patients treated with CCRT alone, alpha diversity analyses (Pielou-e for evenness, Shannon for diversity, Simpson for dominance, Chao1 for richness) revealed no statistically significant differences in the Shannon, Simpson, or Pielou-e indices across the three radiotherapy time points (T0/T1/T2) (Figs. 19a, b, c). However, the Chao1 index (microbial richness) at T0 was significantly lower than that at both T1 and T2 (Fig. 19d). For patients treated with induction chemotherapy followed by CCRT, analysis of the same alpha diversity indices showed no statistically significant differences in the Shannon, Simpson, Pielou-e, or Chao1 indices across T0, T1, and T2 (Figs. 20a, b, c, d). Fig. 19Alpha diversity in NPC patients receiving CCRT during radiotherapy.Note: Violin plots show the distribution of (a) Shannon index, (b) Simpson index, (c) Pielou-e index, and (d) Chao1 index. Samples are colored by radiotherapy time point: T0 (blue), T1 (orange), and T2 (light yellow). **P < 0.01, ***P < 0.001 Alpha diversity in NPC patients receiving CCRT during radiotherapy.Note: Violin plots show the distribution of (a) Shannon index, (b) Simpson index, (c) Pielou-e index, and (d) Chao1 index. Samples are colored by radiotherapy time point: T0 (blue), T1 (orange), and T2 (light yellow). **P < 0.01, ***P < 0.001 Fig. 20Alpha diversity in NPC patients receiving induction chemotherapy followed by CCRT during radiotherapy. Note: Violin plots show the distribution of (a) Shannon index, (b) Simpson index, (c) Pielou-e index, and (d) Chao1 index. Samples are colored by radiotherapy time point: T0 (blue), T1 (orange), and T2 (light yellow) Alpha diversity in NPC patients receiving induction chemotherapy followed by CCRT during radiotherapy. Note: Violin plots show the distribution of (a) Shannon index, (b) Simpson index, (c) Pielou-e index, and (d) Chao1 index. Samples are colored by radiotherapy time point: T0 (blue), T1 (orange), and T2 (light yellow) Linear mixed-effects models of gut microbiota alpha diversity Linear mixed‑effects models with likelihood ratio tests were used to evaluate the associations between radiotherapy time points, treatment regimen, clinical and demographic covariates, and gut microbiota alpha diversity indices (Shannon, Simpson, Chao1, and Pielou‑e). Patient ID was included as a random intercept to account for repeated measurements within individuals, and all models were adjusted for the following covariates: clinical stage, mucositis grade, sex, age, smoking status, alcohol consumption, antiemetic use, opioid use, PPI use, enteral nutritional support, body weight, food intake score, and stool form. For the four alpha diversity indices assessed, the only statistically significant fixed effect was the intercept term, corresponding to the baseline level at T0. Mucositis grade was significantly associated with the Chao1, Shannon, and Pielou‑e indices. Otherwise, no significant associations were observed for radiotherapy time points, treatment regimen, or any other clinical or demographic covariates (all P > 0.05; Table 5). These results indicate that gut microbiota alpha diversity (including diversity, dominance, richness, and evenness) remained stable during radiotherapy. After adjustment for all assessed confounders, alpha diversity was not significantly associated with treatment regimen or most clinical and demographic variables, except for mucositis grade. Table 5Likelihood ratio test results for linear mixed-effects models of alpha diversity indicesTermShannon(χ², P-value)Simpson(χ², P-value)Chao1(χ², P-value)Pielou-e(χ², P-value)(Intercept)(7.661, 0.006**)(23.499, 1.250 × 10 − 6***)(2.578, 0.108)(9.627, 0.002**)Radiotherapy time point(1.134, 0.567)(2.296, 0.317)(5.177, 0.075)(0.979, 0.613)Treatment regimen(3.251, 0.071)(0.765, 0.382)(2.631, 0.105)(2.315, 0.128)Gender(0.087, 0.768)(0.044, 0.834)(0.761, 0.383)(0.008, 0.928)Age(0.463, 0.496)(0.001, 0.975)(1.963, 0.161)(0.088, 0.767)Clinical stage(0.159, 0.690)(0.938, 0.333)(0.020, 0.888)(0.198, 0.656)Mucositis grade(5.968, 0.015*)(3.273, 0.070)(5.436, 0.020*)(4.434, 0.035*)Smoking status(0.049, 0.825)(0.148, 0.700)(1.161, 0.281)(0.035, 0.851)Alcohol consumption(0.012, 0.911)(0.254, 0.614)(0.194, 0.659)(0.036, 0.849)Antiemetic use(0.165, 0.685)(0.008, 0.928)(0.703, 0.402)(0.093, 0.760)Opioid use(2.765, 0.096)(0.369, 0.543)(1.227, 0.268)(2.411, 0.121)PPI use(0.451, 0.502)(0.013, 0.909)(0.092, 0.761)(0.778, 0.378)Enteral nutritional support(0.019, 0.889)(1.230, 0.267)(0.301, 0.583)(0.100, 0.751)Body weight(0.156, 0.693)(0.001, 0.984)(1.029, 0.310)(0.033, 0.855)Food intake score(0.384, 0.535)(0.016, 0.898)(0.081, 0.775)(0.488, 0.485)Stool form(0.069, 0.793)(0.120, 0.729)(0.739, 0.389)(0.003, 0.960)Note: ***P < 0.001, **P < 0.01, *P < 0.05 Table 5Likelihood ratio test results for linear mixed-effects models of alpha diversity indicesTermShannon(χ², P-value)Simpson(χ², P-value)Chao1(χ², P-value)Pielou-e(χ², P-value)(Intercept)(7.661, 0.006**)(23.499, 1.250 × 10 − 6***)(2.578, 0.108)(9.627, 0.002**)Radiotherapy time point(1.134, 0.567)(2.296, 0.317)(5.177, 0.075)(0.979, 0.613)Treatment regimen(3.251, 0.071)(0.765, 0.382)(2.631, 0.105)(2.315, 0.128)Gender(0.087, 0.768)(0.044, 0.834)(0.761, 0.383)(0.008, 0.928)Age(0.463, 0.496)(0.001, 0.975)(1.963, 0.161)(0.088, 0.767)Clinical stage(0.159, 0.690)(0.938, 0.333)(0.020, 0.888)(0.198, 0.656)Mucositis grade(5.968, 0.015*)(3.273, 0.070)(5.436, 0.020*)(4.434, 0.035*)Smoking status(0.049, 0.825)(0.148, 0.700)(1.161, 0.281)(0.035, 0.851)Alcohol consumption(0.012, 0.911)(0.254, 0.614)(0.194, 0.659)(0.036, 0.849)Antiemetic use(0.165, 0.685)(0.008, 0.928)(0.703, 0.402)(0.093, 0.760)Opioid use(2.765, 0.096)(0.369, 0.543)(1.227, 0.268)(2.411, 0.121)PPI use(0.451, 0.502)(0.013, 0.909)(0.092, 0.761)(0.778, 0.378)Enteral nutritional support(0.019, 0.889)(1.230, 0.267)(0.301, 0.583)(0.100, 0.751)Body weight(0.156, 0.693)(0.001, 0.984)(1.029, 0.310)(0.033, 0.855)Food intake score(0.384, 0.535)(0.016, 0.898)(0.081, 0.775)(0.488, 0.485)Stool form(0.069, 0.793)(0.120, 0.729)(0.739, 0.389)(0.003, 0.960)Note: ***P < 0.001, **P < 0.01, *P < 0.05 Likelihood ratio test results for linear mixed-effects models of alpha diversity indices Note: ***P < 0.001, **P < 0.01, *P < 0.05 Beta diversity analysis by chemotherapy regimen Stratified by chemotherapy regimen, beta diversity analysis based on the Bray-Curtis distance metric revealed no evidence of significant inter-group differentiation across the three radiotherapy time points (T0/T1/T2) in either treatment cohort, with both patients treated with CCRT alone (PERMANOVA R² = 0.049, F = 1.159 and P = 0.153) and patients receiving induction chemotherapy followed by CCRT (PERMANOVA R² = 0.030, F = 0.654 and P = 0.955). Betadisper tests further indicated no significant differences in group dispersion between T0, T1, and T2 in either cohort (P = 0.732 and P = 0.271, respectively), suggesting that the observed PERMANOVA results were not confounded by unequal group variances. Visualization of PCoA plots (Figs. 21a, b) showed substantial overlap between the three time points, consistent with the statistical findings of minimal beta diversity changes during radiotherapy. Fig. 21PCoA of beta diversity by chemotherapy regimen during radiotherapy. Note: (a) PCoA based on Bray-Curtis distance in patients treated with CCRT alone. PCoA1 explained 11.8% of variance, and PCoA2 explained 6.95%. (b) PCoA based on Bray-Curtis distance in patients receiving induction chemotherapy followed by CCRT. PCoA1 explained 11.01% of variance, and PCoA2 explained 8.53%. Samples are colored by radiotherapy time point: T0 (red), T1 (green), T2 (blue). Ellipses represent 95% confidence intervals PCoA of beta diversity by chemotherapy regimen during radiotherapy. Note: (a) PCoA based on Bray-Curtis distance in patients treated with CCRT alone. PCoA1 explained 11.8% of variance, and PCoA2 explained 6.95%. (b) PCoA based on Bray-Curtis distance in patients receiving induction chemotherapy followed by CCRT. PCoA1 explained 11.01% of variance, and PCoA2 explained 8.53%. Samples are colored by radiotherapy time point: T0 (red), T1 (green), T2 (blue). Ellipses represent 95% confidence intervals LEfSe-based taxonomic enrichment by chemotherapy regimen To characterize regimen-specific taxonomic shifts in gut microbiota across radiotherapy time points, we performed LEfSe analysis with an LDA score greater than 3 and an FDR-adjusted P value less than 0.05, stratified by chemotherapy regimen and focusing on genus-level taxa. For patients treated with CCRT alone (Fig. 22a), distinct stage-specific enrichment patterns were observed across T0, T1, and T2. Specifically, UBA1819, Hungatella, and Morganella were uniquely enriched at T1, while a broader set of genera including Ruminococcus_gnavus_group, Pantoea, Lactiplantibacillus, Brevundimonas, Muribaculaceae_unclassified, Alloprevotella, Desulfovibrio, Lactobacillus, Ligilactobacillus, Lachnospiraceae_NK4A136_group, Pseudomonas and Enterobacter exhibited significant enrichment at T2. All these taxa displayed LDA scores exceeding 3, which indicates strong discriminatory power for distinguishing between time points. In contrast, for patients receiving induction chemotherapy followed by CCRT (Fig. 22b), the differential enrichment profile was more constrained. Specifically, Rothia was the sole genus enriched at the pre-radiotherapy baseline (T0), Hungatella and Dialister were enriched at T1, and UBA1819 dominated the taxa enriched at T2. Notably, the overlap in enriched genera such as Hungatella and UBA1819 between the two regimens was limited. The breadth of taxa enriched at T2 was markedly narrower in the induction plus CCRT cohort, and this aligns with our prior observation of regimen-specific microbiota dynamics. These findings highlight that chemotherapy regimen shapes the temporal trajectory of gut microbiota alterations during radiotherapy, with distinct sets of taxa driving stage-specific community shifts in each treatment group. Fig. 22LEfSe of differentially abundant gut microbiota genera in NPC patients by chemotherapy regimen during radiotherapy. Note: (a) LEfSe genus-level identified differentially abundant taxa among NPC patient treated with CCRT alone across the three time points (T0, T1, and T2). Genera with LDA scores (green) are enriched at T1, while those with scores (red) are enriched at T2. (b) LEfSe analysis comparing T0, T1, and T2 time points for patients receiving induction chemotherapy followed by CCRT. Genera are colored by the time point at which they are most abundant: T0 (yellow), T1 (green), and T2 (red). Only taxa with an LDA score > 3.0 are shown LEfSe of differentially abundant gut microbiota genera in NPC patients by chemotherapy regimen during radiotherapy. Note: (a) LEfSe genus-level identified differentially abundant taxa among NPC patient treated with CCRT alone across the three time points (T0, T1, and T2). Genera with LDA scores (green) are enriched at T1, while those with scores (red) are enriched at T2. (b) LEfSe analysis comparing T0, T1, and T2 time points for patients receiving induction chemotherapy followed by CCRT. Genera are colored by the time point at which they are most abundant: T0 (yellow), T1 (green), and T2 (red). Only taxa with an LDA score > 3.0 are shown Genus-level differential abundance of gut microbiota during radiotherapy in NPC patients by chemotherapy regimen: ALDEx2 analysis Differential abundance analysis using ALDEx2 at the genus level identified a broad range of taxa with significant changes in relative abundance across radiotherapy time points (T0, T1, and T2) in patients treated with CCRT alone (Fig. 23; Table 6). The genus UBA1819 exhibited the most pronounced differential abundance with the highest statistical significance, followed by the Lactobacillus, Phenylobacterium and Clostridium_ innocuum_group (all FDR-adjusted P < 0.05). Additional genera with significant temporal changes in relative abundance included Hydrogenophaga, Pantoea, Desulfovibrionaceae_unclassified, Dubosiella, Lachnospiraceae_ NK4A136_group, Brevundimonas, Helicobacter, Pseudomonas, Sphingomonas, Eggerthella, Aeromonas, Caulobacter, Escherichia, Muribaculum, Anaerotruncus, and Carnobacterium (all FDR-adjusted P < 0.05). These findings suggest that radiotherapy is associated with marked and widespread shifts in the relative abundance of specific gut microbiota genera in NPC patients receiving CCRT alone, indicating distinct temporal remodeling of the gut microbial community over the course of radiotherapy. In contrast, differential abundance analysis using ALDEx2 at the genus level revealed no taxa with significant changes in relative abundance across radiotherapy time points (T0, T1, and T2) in patients receiving induction chemotherapy followed by CCRT. This observation indicates that the temporal dynamics of gut microbiota at the genus level during radiotherapy are markedly influenced by the chemotherapy regimen, with pronounced remodeling occurring in patients treated with CCRT alone, while the gut microbiota of patients receiving induction chemotherapy followed by CCRT remains relatively stable throughout the treatment course. Fig. 23Differential abundance analysis of gut microbiota at genus level across radiotherapy time points (T0 vs. T1 vs. T2) in NPC patients. The y-axis shows -log10 (FDR-adjusted P value), and the dashed line indicates the significance threshold (FDR_adjusted P = 0.05) Differential abundance analysis of gut microbiota at genus level across radiotherapy time points (T0 vs. T1 vs. T2) in NPC patients. The y-axis shows -log10 (FDR-adjusted P value), and the dashed line indicates the significance threshold (FDR_adjusted P = 0.05) Table 6Genus-level differential abundance of gut microbiota during radiotherapy in NPC patients receiving CCRT alone (ALDEx2 analysis)Feature (genus level)Group comparisonlog2FC95% CI (lower)95% CI (upper)P_adj(Welch)P_adj(Wilcoxon)−log₁₀(FDR-adjusted P)UBA1819T0 vs. T1 vs. T25.5234.7416.3050.003 **0.003 **2.527LactobacillusT0 vs. T1 vs. T22.6722.2553.0900.015 **0.015 **1.811PhenylobacteriumT0 vs. T1 vs. T22.8682.2563.4800.015 *0.015 *1.811Clostridium]_innocuum_groupT0 vs. T1 vs. T24.2773.5624.9920.019 **0.019 **1.718PantoeaT0 vs. T1 vs. T24.0052.9295.0820.020 **0.020 **1.699HydrogenophagaT0 vs. T1 vs. T22.9432.3923.4930.020 **0.020 **1.699Lachnospiraceae_NK4A136_groupT0 vs. T1 vs. T24.1163.0655.1670.025 **0.025 **1.596DubosiellaT0 vs. T1 vs. T23.5892.8884.2900.025 **0.025 **1.596Desulfovibrionaceae_unclassifiedT0 vs. T1 vs. T23.4592.8514.0680.025 **0.025 **1.596BrevundimonasT0 vs. T1 vs. T24.1523.2105.0940.028 **0.028 **1.548PseudomonasT0 vs. T1 vs. T23.4922.4444.5400.028 **0.028 **1.548SphingomonasT0 vs. T1 vs. T24.0793.3664.7920.028 **0.028 **1.548HelicobacterT0 vs. T1 vs. T24.1633.4544.8720.028 **0.028 **1.548EggerthellaT0 vs. T1 vs. T23.8153.0364.5930.044 **0.044 **1.354CaulobacterT0 vs. T1 vs. T22.9962.4043.5880.044 **0.044 **1.349MuribaculumT0 vs. T1 vs. T24.6913.9865.3960.044 **0.044 **1.349EscherichiaT0 vs. T1 vs. T2/−0.6920.6920.044 **0.044 **1.349AeromonasT0 vs. T1 vs. T23.4752.8414.1080.044 **0.044 **1.349AnaerotruncusT0 vs. T1 vs. T22.7131.9993.4270.052 *0.052 *1.285CarnobacteriumT0 vs. T1 vs. T23.5452.5634.5270.052 *0.052 *1.285ParamuribaculumT0 vs. T1 vs. T23.5952.9374.2530.052 *0.052 *1.285BradyrhizobiumT0 vs. T1 vs. T22.6712.1503.1930.052 *0.052 *1.285PhreatobacterT0 vs. T1 vs. T22.6462.0923.1990.052 *0.052 *1.283LeuconostocT0 vs. T1 vs. T22.8422.1673.5170.056*0.056*1.248HT002T0 vs. T1 vs. T24.0523.3874.7180.056*0.056*1.248Methylobacterium-MethylorubrumT0 vs. T1 vs. T2/−0.4460.4460.060 *0.060 *1.219EnterorhabdusT0 vs. T1 vs. T23.2972.7053.8890.060 *0.060 *1.219LigilactobacillusT0 vs. T1 vs. T23.2482.1534.3420.064 *0.064 *1.195MassiliaT0 vs. T1 vs. T23.7812.9854.5780.064 *0.064 *1.195Allorhizobium-Neorhizobium-Pararhizobium-RhizobiumT0 vs. T1 vs. T2/−0.4940.4940.064 *0.064 *1.195Clostridia_UCG−014_unclassifiedT0 vs. T1 vs. T20.617−0.3161.5510.065 *0.065 *1.187Ruminococcaceae_unclassifiedT0 vs. T1 vs. T22.4841.7583.2100.069 *0.069 *1.162Rahnella1T0 vs. T1 vs. T24.0443.3664.7230.069 *0.069 *1.162CoprobacillusT0 vs. T1 vs. T24.9444.3455.5430.069 *0.069 *1.162AlloprevotellaT0 vs. T1 vs. T21.7780.8402.7150.070 *0.070 *1.156PelomonasT0 vs. T1 vs. T2/−0.2950.2950.076*0.076*1.120VibrioT0 vs. T1 vs. T22.6291.9043.3530.090 *0.090 *1.046AllobaculumT0 vs. T1 vs. T23.1002.5233.6760.093 *0.093 *1.030MorganellaT0 vs. T1 vs. T28.0367.3468.7270.094 *0.094 *1.029Oscillospiraceae_unclassifiedT0 vs. T1 vs. T21.9281.2232.6320.094 *0.094 *1.029Prevotellaceae_UCG−001T0 vs. T1 vs. T23.5542.8854.2220.094 *0.094 *1.029Note: Significance of difference (***P < 0.001, **P < 0.05, *P < 0.1) Table 6Genus-level differential abundance of gut microbiota during radiotherapy in NPC patients receiving CCRT alone (ALDEx2 analysis)Feature (genus level)Group comparisonlog2FC95% CI (lower)95% CI (upper)P_adj(Welch)P_adj(Wilcoxon)−log₁₀(FDR-adjusted P)UBA1819T0 vs. T1 vs. T25.5234.7416.3050.003 **0.003 **2.527LactobacillusT0 vs. T1 vs. T22.6722.2553.0900.015 **0.015 **1.811PhenylobacteriumT0 vs. T1 vs. T22.8682.2563.4800.015 *0.015 *1.811Clostridium]_innocuum_groupT0 vs. T1 vs. T24.2773.5624.9920.019 **0.019 **1.718PantoeaT0 vs. T1 vs. T24.0052.9295.0820.020 **0.020 **1.699HydrogenophagaT0 vs. T1 vs. T22.9432.3923.4930.020 **0.020 **1.699Lachnospiraceae_NK4A136_groupT0 vs. T1 vs. T24.1163.0655.1670.025 **0.025 **1.596DubosiellaT0 vs. T1 vs. T23.5892.8884.2900.025 **0.025 **1.596Desulfovibrionaceae_unclassifiedT0 vs. T1 vs. T23.4592.8514.0680.025 **0.025 **1.596BrevundimonasT0 vs. T1 vs. T24.1523.2105.0940.028 **0.028 **1.548PseudomonasT0 vs. T1 vs. T23.4922.4444.5400.028 **0.028 **1.548SphingomonasT0 vs. T1 vs. T24.0793.3664.7920.028 **0.028 **1.548HelicobacterT0 vs. T1 vs. T24.1633.4544.8720.028 **0.028 **1.548EggerthellaT0 vs. T1 vs. T23.8153.0364.5930.044 **0.044 **1.354CaulobacterT0 vs. T1 vs. T22.9962.4043.5880.044 **0.044 **1.349MuribaculumT0 vs. T1 vs. T24.6913.9865.3960.044 **0.044 **1.349EscherichiaT0 vs. T1 vs. T2/−0.6920.6920.044 **0.044 **1.349AeromonasT0 vs. T1 vs. T23.4752.8414.1080.044 **0.044 **1.349AnaerotruncusT0 vs. T1 vs. T22.7131.9993.4270.052 *0.052 *1.285CarnobacteriumT0 vs. T1 vs. T23.5452.5634.5270.052 *0.052 *1.285ParamuribaculumT0 vs. T1 vs. T23.5952.9374.2530.052 *0.052 *1.285BradyrhizobiumT0 vs. T1 vs. T22.6712.1503.1930.052 *0.052 *1.285PhreatobacterT0 vs. T1 vs. T22.6462.0923.1990.052 *0.052 *1.283LeuconostocT0 vs. T1 vs. T22.8422.1673.5170.056*0.056*1.248HT002T0 vs. T1 vs. T24.0523.3874.7180.056*0.056*1.248Methylobacterium-MethylorubrumT0 vs. T1 vs. T2/−0.4460.4460.060 *0.060 *1.219EnterorhabdusT0 vs. T1 vs. T23.2972.7053.8890.060 *0.060 *1.219LigilactobacillusT0 vs. T1 vs. T23.2482.1534.3420.064 *0.064 *1.195MassiliaT0 vs. T1 vs. T23.7812.9854.5780.064 *0.064 *1.195Allorhizobium-Neorhizobium-Pararhizobium-RhizobiumT0 vs. T1 vs. T2/−0.4940.4940.064 *0.064 *1.195Clostridia_UCG−014_unclassifiedT0 vs. T1 vs. T20.617−0.3161.5510.065 *0.065 *1.187Ruminococcaceae_unclassifiedT0 vs. T1 vs. T22.4841.7583.2100.069 *0.069 *1.162Rahnella1T0 vs. T1 vs. T24.0443.3664.7230.069 *0.069 *1.162CoprobacillusT0 vs. T1 vs. T24.9444.3455.5430.069 *0.069 *1.162AlloprevotellaT0 vs. T1 vs. T21.7780.8402.7150.070 *0.070 *1.156PelomonasT0 vs. T1 vs. T2/−0.2950.2950.076*0.076*1.120VibrioT0 vs. T1 vs. T22.6291.9043.3530.090 *0.090 *1.046AllobaculumT0 vs. T1 vs. T23.1002.5233.6760.093 *0.093 *1.030MorganellaT0 vs. T1 vs. T28.0367.3468.7270.094 *0.094 *1.029Oscillospiraceae_unclassifiedT0 vs. T1 vs. T21.9281.2232.6320.094 *0.094 *1.029Prevotellaceae_UCG−001T0 vs. T1 vs. T23.5542.8854.2220.094 *0.094 *1.029Note: Significance of difference (***P < 0.001, **P < 0.05, *P < 0.1) Genus-level differential abundance of gut microbiota during radiotherapy in NPC patients receiving CCRT alone (ALDEx2 analysis) Note: Significance of difference (***P < 0.001, **P < 0.05, *P < 0.1) Discussion This study systematically characterized gut microbiota profiles in newly diagnosed NPC patients and their dynamic changes linked to radiotherapy, while exploring the impact of chemotherapy regimen on microbial dynamics. Our findings confirm that NPC patients exhibit distinct gut microbial dysbiosis compared to healthy controls, and radiotherapy is associated with temporal shifts in gut microbial diversity and composition. These effects are strongly stratified by chemotherapy regimen. These observations fill critical gaps in the current understanding of NPC-related gut microbiota and provide novel insights for translational interventions targeting the gut microbiome to improve NPC treatment outcomes. Gut microbiota characteristics in newly diagnosed NPC patients Alpha diversity analysis revealed that newly diagnosed NPC patients had significantly lower Shannon (species diversity) and Chao1 (species richness) indices compared with healthy controls, consistent with prior work on NPC gut microbiota [13] and supporting the presence of reduced gut microbial diversity in this patient population. Beta diversity analysis using the Bray-Curtis distance metric further identified distinct microbial community structures between NPC patients and healthy controls, aligning with Jiang et al.’s [13] observation of intergroup clustering. However, genus-level compositional differences between our study and theirs may reflect confounding from induction chemotherapy, which was administered to 15 of our NPC patients, as treatment-related toxicities, dietary shifts, and psychological stress are known to be associated with altered gut microbiota composition [22–24]. Notably, we identified 16 core differential taxa between NPC patients and healthy controls using complementary LEfSe and ALDEx2 analyses, enhancing the robustness of our findings. Among these, Lactiplantibacillus and Lactobacillus, which are typically considered beneficial due to their ability to produce lactic acid and induce regulatory T (Treg) cells [25, 26], were enriched in NPC patients. While their overabundance may seem counterintuitive, it is plausible that dysregulated growth of these taxa disrupts overall microbial balance, potentially contributing to a pro-inflammatory microenvironment. disrupt microbial balance. This raises the hypothesis that dysregulated Lactobacillaceae could modulate the nasopharyngeal immune microenvironment via the “gut-nasopharynx axis” [12, 13]. This pathway warrants further validation in future interventional studies, as direct evidence for this mechanism was not captured in our observational cohort. Enriched genera including Enterococcus, Veillonella, Escherichia-Shigella, and Megasphaera in NPC patients align with known links to intestinal inflammation [27–30], while elevated Akkermansia may reflect compensatory mucosal repair through mucin degradation [31]. However, excessive Akkermansia growth could deplete the intestinal mucus layer and exacerbate barrier damage. This relationship remains speculative, as we did not measure intestinal barrier integrity such as fecal calprotectin or zonulin levels in the current study. Reduced abundances of beneficial taxa such as Fusicatenibacter and Ligilactobacillus further support the presence of gut dysbiosis in NPC patients, which is likely driven by a combination of disease-related systemic inflammation and pre-treatment perturbations [32, 33]. Collectively, these findings highlight that NPC is associated with specific gut microbial signatures that may serve as potential biomarkers or therapeutic targets. Dynamic changes of gut microbiota in NPC patients during radiotherapy Longitudinal analysis revealed significant increases in the Chao1 index (species richness) from T0 to T1 / T2, while beta diversity analyses demonstrated time-dependent shifts in microbial community structure across the radiotherapy course, especially based on unweighted UniFrac distance. The initial reduction in diversity at T0 may be linked to EBV-induced systemic inflammation, which inhibits commensal bacterial growth [5, 34], and tumor-derived pro-inflammatory factors such as IL-6 and TNF-α [35, 36]. We did not measure these inflammatory mediators in the current study, which represents a key limitation. At the phylum level, declining Bacteroidota abundance during radiotherapy aligns with the hypothesis that radiation-induced systemic inflammation, which has been reported in prior studies [37], alters the gut microenvironment to suppress inflammation-sensitive Bacteroidota [38]. This trend may also be exacerbated by radiotherapy-related dietary changes such as reduced fiber intake, altered macronutrient composition despite standardized dietary guidance. Rising Verrucomicrobiota, largely composed of Akkermansia, could reflect mucin-driven compensatory repair in response to radiation-induced intestinal damage [31], while the transient increase followed by a decline in Firmicutes may reflect early tolerance of Lactobacillaceae to radiotherapy, followed by energy metabolism stress as treatment progresses. Rebounding Proteobacteria abundance in late radiotherapy (T2) raises concern for pathogenic translocation, a process hypothesized to stem from intestinal barrier impairment [39, 40] which remains unvalidated in our cohort but merits further investigation given its potential link to treatment related infections. At the genus level, key trends included declining Faecalibacterium and Roseburia (both SCFA-producing taxa) alongside rising Klebsiella, further highlighting persistent dysbiosis during radiotherapy. Reduced SCFA producing taxa such as Faecalibacterium may weaken anti-inflammatory signaling via the GPR43 / 109 A pathway [15, 16], which could exacerbate radiotherapy induced intestinal inflammation and mucositis. We did not quantify SCFAs to confirm this relationship. Enriched Klebsiella and Escherichia-Shigella in late radiotherapy link to the hypothesized release of pro-inflammatory lipopolysaccharides (LPS) [41], which may contribute to treatment-related toxicities. Additionally, we identified 7 core differential taxa (Hungatella, Pantoea, UBA1819, Alistipes, Desulfovibrio, Lachnospiraceae_NK4A136_group, Pseudomonas) overlapping between LEfSe and ALDEx2 analyses. Among these, Desulfovibrio vulgaris has been shown to induce intestinal inflammation and exacerbate colitis in animal models [42], while Alistipes, as a genus closely associated with inflammation and cancer, may participate in the progression of NPC and radiotherapy-induced intestinal damage [43]. These taxa have been previously associated with intestinal inflammation [44–46], although their specific functional roles in NPC radiotherapy remain speculative and require further validation. Impact of chemotherapy regimen on gut microbiota dynamics A key novel finding of this study is that chemotherapy regimen strongly stratifies gut microbial dynamics during radiotherapy. Specifically, patients treated with CCRT alone exhibited significant increases in Chao1 index and pronounced genus-level differential abundance such as UBA1819, Lactobacillus and Pantoea during radiotherapy, while those receiving induction chemotherapy followed by CCRT showed no significant changes in alpha / beta diversity or genus-level composition. This suggests that induction chemotherapy may pre-condition the gut microbiota to be more resilient to subsequent radiotherapy, potentially via early modulation of microbial communities or suppression of radiotherapy-sensitive taxa. LEfSe analysis further confirmed regimen-specific taxonomic enrichment patterns. Patients treated with CCRT alone exhibited a wider range of enriched taxa at T2, while those receiving induction chemotherapy followed by CCRT presented a more restricted differential abundance profile. These findings are clinically relevant, as they indicate that chemotherapy regimens should be considered when designing interventions targeted at gut microbiota for NPC patients. Limitations and strengths This study has several limitations that should be acknowledged. First, while the longitudinal Chao1 index showed significant increases, we did not explore its correlation with inflammatory markers or clinical outcomes such as tumor regression, which limits our ability to link microbial changes to clinical relevance. Second, we adopted 16S rDNA sequencing targeting only the V3–V4 variable regions, which limits the resolution of low-abundance taxa and prevents functional characterization of microbial communities. In subsequent studies, metagenomic sequencing will be employed to obtain more comprehensive genetic information, facilitating the discovery of novel functional genes and pathways. Third, the absence of extraction blanks and negative controls in the 16S rDNA sequencing workflow represents an important methodological limitation. This may introduce uncertainty in the interpretation of low-abundance ASVs, as potential exogenous contamination from reagent residues, laboratory consumables, or environmental microbes during experimental procedures cannot be fully ruled out. Critically, this limitation may affect the interpretation of rare taxa identified in our analyses, including some of the low-abundance core differential taxa reported in this study. While we mitigated this risk by excluding ASVs with a mean relative abundance < 0.1% across all samples and a prevalence < 10% within each study group to filter out potential sporadic contaminant sequences, we cannot definitively exclude the possibility of contamination contributing to the observed rare taxa profiles. Future studies should incorporate strict negative controls including extraction blanks and no-template PCR controls to more accurately distinguish true low-abundance taxa from contaminant sequences. Fourth, differential abundance analyses were conducted using rarefied data to ensure consistency with diversity analyses. Although rarefaction is not typically recommended for compositional approaches such as ALDEx2, which corrects for sequencing depth bias via CLR transformation. We verified that core differential taxa remained consistent when using rarefied versus non-rarefied data. Future analyses will prioritize the use of non-rarefied data for compositional differential abundance testing to align with standard best practices. Fifth, we employed a FDR threshold of < 0.1 for primary analyses, which is more lenient than the conventional 0.05 threshold. This choice was adopted to reduce type II error and retain biologically relevant taxa related to NPC and radiotherapy, given the exploratory design. However, this lenient threshold may increase type I error risk. Accordingly, all results at FDR < 0.1 are considered exploratory and require validation in larger independent cohorts. Six, while we collected dietary intake scores, detailed records of food structure such as fiber and fat intake are lacking, which limits our ability to disentangle the association of diet vs. radiotherapy exposure with gut microbiota changes. Finally, as an observational study, we cannot establish causal relationships between gut microbiota alterations and NPC or radiotherapy outcomes. Interventional studies are therefore needed to validate these exploratory associations. Despite these limitations, this study has several key strengths. It employs a mixed cross-sectional/longitudinal design to systematically track gut microbiota changes throughout NPC radiotherapy. The use of complementary LEfSe and ALDEx2 analyses enhances the robustness of differential taxa identification, while stratification by chemotherapy regimen provides novel insights into treatment-specific microbial dynamics. Additionally, strict frequency matching of healthy controls (age, gender, smoking status) minimizes confounding, and standardized treatment / dietary guidance reduces variability in microbial profiles. Finally, complete follow-up with no lost-to-follow-up ensures the reliability of longitudinal results. Translational implications and future directions The findings of this study have important translational implications for improving NPC treatment outcomes via gut microbiota modulation. First, the core differential taxa identified between NPC patients and healthy controls (e.g., Lactiplantibacillus, Fusicatenibacter) and those dynamically altered during radiotherapy (e.g., Akkermansia, Faecalibacterium) represent potential biomarkers for monitoring treatment response or predicting treatment-related toxicities. Second, targeted interventions to restore gut homeostasis, including dietary fiber supplementation to support SCFA-producing taxa or strain-specific probiotics such as Clostridium butyricum, may mitigate radiotherapy-induced dysbiosis and reduce toxicities such as oral mucositis. Previous studies have confirmed that dietary intervention can regulate intestinal microbiota and improve disease prognosis through multiple molecular mechanisms [47], which provides a theoretical basis for our proposed intervention strategy. A prospective randomized controlled trial is warranted to evaluate the efficacy of such interventions. Future directions should focus on three key areas. Validating the functional links between gut microbial changes and clinical outcomes such as tumor regression using correlational analyses. Employing integrated multi-omics approaches including 16S rDNA sequencing, metabolomics for SCFA quantification, inflammatory marker detection to clarify the mechanisms underlying the gut-nasopharynx axis in NPC. Conducting interventional studies to test the efficacy of gut microbiota-targeted strategies in improving treatment tolerance and prognosis. Additionally, larger independent cohorts are needed to confirm the microbial signatures identified in this study and evaluate their generalizability. Conclusion This study demonstrates that newly diagnosed NPC patients exhibit significant gut microbial dysbiosis compared with healthy controls, characterized by reduced species diversity and richness, distinct community structures, and specific genus-level signatures. Over the course of radiotherapy, NPC patients display dynamic gut microbiota alterations, including increased species richness and modified phylum and genus level compositions. These temporal changes are significantly stratified by chemotherapy regimen, with marked microbial remodeling observed in patients receiving CCRT alone but relative stability in those undergoing induction chemotherapy followed by CCRT. Several candidate taxa were identified as exploratory signatures, although their functional roles and relevance to the gut-nasopharynx axis remain to be validated. Incomplete collection of potential confounding factors limits causal inference in this observational study, and all observed associations should be interpreted as non-causal. Nevertheless, our findings provide novel insights into gut microbial dynamics associated with NPC and its treatment, supporting the potential value of gut microbiota-targeted strategies to improve clinical outcomes. Future studies in independent cohorts with rigorous control of confounding factors are warranted to confirm these observations. Interventional trials are also needed to validate strategies targeting the gut microbiota to alleviate dysbiosis associated with radiotherapy and improve treatment tolerance in NPC patients. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1