Participants in this study were recruited from an ongoing study supported by the Natural Science Foundation of Liaoning Province, China (Grant No. 2022JH2/101500014), titled “Pathogenesis of intestinal flora dysbiosis in allergic rhinitis and the application of washed microbiota transplantation in AR treatment” (2022–2025). A total of 23 patients met the criteria of sample collection, and agreed to join in this study (Group A). In addition, 15 age- and sex-matched healthy controls were recruited from the community (Group B). All participants signed the written informed consent. The following demographic and clinical data were collected via in-person interviews: age, sex, weight, height, body mass index (BMI), educational attainment, smoking history, alcohol consumption history, marital status, family medical history, disease duration, medication and dietary supplement usage, Total Nasal Symptom Score (TNSS), and Rhinoconjunctivitis Quality of Life Questionnaire (RQLQ) outcomes. This study protocol was conducted in accordance with the Declaration of Helsinki and approved by the Ethical Committee of General Hospital of Northern Theater Command of PLA (Approval number: Y(2025)092).

The diagnosis of AR was based on ARIA guidelines (2016 revision), Chinese Society of Allergy Guidelines for Diagnosis and Treatment of Allergic Rhinitis (2018) and Chinese guideline for diagnosis and treatment of allergic rhinitis (2022 revision) (Brożek et al., 2017; Cheng et al., 2018; Subspecialty Group of Rhinology, Editorial Board of Chinese Journal of Otorhinolaryngology Head and Neck Surgery, and Subspecialty Group of Rhinology, Society of Otorhinolaryngology Head and Neck Surgery, Chinese Medical Association, 2022). The inclusion criteria of Group A were as follows: (1) paroxysmal sneezing, clear water-like runny nose, itch, sneezing, and other symptoms appear 2 or more, and the daily symptoms persist or accumulate more than 1 h; (2) the presence of allergen-specific IgE antibody (sIgE) test ≥0.35 IU/mL and/or positive skin prick tests (SPT); (3) a history of a reaction in the past year; (4) age 18–65 years; (5) voluntary participation in this study. In addition, 15 age- and sex-matched healthy controls were recruited from the community. The inclusion criteria of Group B were as follows: (1) there is no history of allergies or family allergies;(2) there are no allergy-related symptoms;(3) voluntary participation in this study. Exclusion criteria for both groups included: (1) receipt of systemic or topical antibiotics, immunomodulatory agents (including glucocorticoids), antihistamines, probiotics, prebiotics, or synbiotics within 3 months prior to enrollment; (2) use of laxatives or antidiarrheal medications, or experience of constipation, diarrhea, or respiratory tract infection within the preceding 4 weeks; (3) comorbid respiratory conditions including chronic obstructive pulmonary disease, asthma, bronchiectasis, tuberculosis, pneumonia, pulmonary heart disease, or pulmonary malignancies; (4) history of hypertension, coronary heart disease, hyperthyroidism, hypothyroidism, hepatic or renal dysfunction, or hematologic disorders; (5) history of psychiatric or neurological conditions; (6) presence of clinically significant abnormalities upon pre-trial assessment deemed likely to confound study outcomes, as determined by the investigators (Liu et al., 2020; Zhou M. S. et al., 2021; Zhu et al., 2020). All study participants shared comparable ethnic/geographic and dietary backgrounds.

To comprehensively evaluate the symptom severity and quality of life impact of AR, we utilized two validated clinical tools: TNSS and RQLQ. The TNSS assesses the severity of four key nasal symptoms: nasal obstruction, rhinorrhea, sneezing, and nasal itching. Each symptom was scored by patients on a 4-point Likert scale: 0 = no symptoms; 1 = mild symptoms (present but not bothersome); 2 = moderate symptoms (noticeable and occasionally bothersome); 3 = Severe symptoms (frequent and significantly bothersome). The total TNSS ranged from 0 to 12, with higher scores indicating greater symptom severity. The RQLQ is a disease-specific instrument designed to measure the impact of AR on patients’ quality of life. The questionnaire comprises 28 items across seven domains: sleep disturbances, nasal symptoms, ocular symptoms, practical problems, emotional function, activity limitations, and general well-being. Each item was rated on a 7-point scale (0–6), where: 0 = No impairment; 6 = Severe impairment. The overall RQLQ score was calculated as the mean of all item scores, yielding a total range of 0 to 6, with higher scores reflecting worse quality of life (QoL). Both TNSS and RQLQ have demonstrated high reliability and validity in prior studies of AR (Bousquet et al., 2025; Sánchez and Castro, 2019), making them suitable for capturing symptom burden and its functional consequences in our cohort.

Fecal samples were collected from all participants by retaining the mid-to-late portion of bowel movements. Using sterilized spoons, the inner layer of fecal material was carefully sampled, and all specimens were transferred into sterile plastic tubes under aseptic conditions, with precautions taken to avoid contamination from urine or contact with toilet surfaces. Additionally, all female participants provided stool samples exclusively during their non-menstrual phase. Within 2 h post-collection, samples were transported to the laboratory in ice-packed coolers to maintain a cold chain, followed by immediate storage at −80°C until subsequent analyzes.

Fecal samples from AR patients and healthy controls were collected using sterile protocols, stored at −80°C, and processed for 16S rRNA gene sequencing. Genomic DNA was extracted (QIAamp DNA Stool Mini Kit), and the V3-V4 region was amplified with primers 357F/806R using a two-step PCR protocol (Phusion polymerase). Libraries were sequenced on Illumina NovaSeq (250-bp paired-end). Bioinformatic analysis included quality filtering [Trimmomatic (Bolger et al., 2014)], amplicon sequence variant (ASV) clustering using the DADA2 (Callahan et al., 2016) pipeline, chimeric removal (integrated within DADA2), and taxonomic annotation (SILVA 138). Alpha diversity indices (Observed species, Chao1, ACE, Shannon, Simpson, and Phylogenetic Diversity whole tree) and beta diversity metrics (Bray-Curtis, Jaccard, unweighted and weighted UniFrac) and LEfSe (Segata et al., 2011) (LDA > 2, P < 0.05) were calculated using QIIME 2 (Bolyen et al., 2019).

Fecal metabolites were profiled via UHPLC-QTOF-MS (Agilent 6545). Samples were homogenized in 80% methanol, centrifuged, and filtered. Raw data were processed with XCMS for peak alignment, normalization, and QC-based filtering (RSD < 30%). Multivariate analysis included PCA and OPLS-DA (SIMCA-P, validated by permutation tests). The OPLS-DA model was used strictly as an exploratory tool for variable selection and not for predictive purposes. Differential metabolites were identified (VIP > 1.0, FC ≥ 1.5/ ≤ 0.667, P < 0.05) and mapped to KEGG pathways [MetaboAnalyst 5.0 (Pang et al., 2021)] using hypergeometric tests. For the receiver operating characteristic (ROC) analysis of candidate metabolites, internal validation was performed using a 10-fold cross-validation scheme.

Statistical analysis was performed using R software (version 4.0.2). For baseline characteristics of study participants, continuous variables were first assessed for normality using the Kolmogorov-Smirnov test. Normally distributed variables were compared between groups using Independent t-test, while non-normally distributed variables were analyzed with Mann-Whitney U test to assess median differences. Categorical variables were evaluated using Fisher’s exact test considering the limited sample size. Difference of community structure of gut microbiome among groups was analyzed using the method of permutational multivariate ANOVA (PERMANOVA). Metabolomic profiles were processed with log-transformation prior to statistical analysis. Differentially expressed metabolites were identified using MS/MS spectral data and verified through non-parametric statistical testing (Mann-Whitney U). Spearman’s rank correlation analysis was conducted to explore associations between gut microbiota and metabolites, with false discovery rate (FDR) correction applied using Benjamini-Hochberg procedure to adjust for multiple comparisons. All reported p-values were two-tailed and P < 0.05 was considered significant.

The study cohort comprised 38 participants, including 23 AR patient (Group A) and 15 healthy controls (Group B). Demographic and anthropometric characteristics were comparable between groups (Table 1). The AR group (median age: 37 years, IQR: 31–42) and controls (median age: 35 years, IQR: 26–43) showed no significant differences in age (P = 0.58, Mann-Whitney U test) or gender distribution (47.8% vs. 40.0% males, P = 0.65, χ² test). BMI was marginally higher in AR patients (mean ± SD: 23.8 ± 3.1 kg/m²) than in controls (21.8 ± 3.5 kg/m²), though this difference was not statistically significant (P = 0.12, independent t-test). Ethnic homogeneity was observed, with Han Chinese constituting 82.6% of AR patients and 93.3% of controls (P = 0.35, Fisher’s exact test).

Lifestyle factors and clinical parameters further defined the cohorts. Smoking and alcohol consumption were infrequent in both groups: 8.7% of AR patients (median smoking duration: 3 years, IQR: 3–12) and 13.3% of controls (median: 5 years, IQR: 5–5) reported smoking (P = 0.62), while alcohol use was reported by 13.0% of AR patients (median: 10 years, IQR: 7–15) and 13.3% of controls (median: 7 years, IQR: 7–7, P = 1.00). Symptom severity and quality of life impact were quantified by the Total Nasal Symptom Score (TNSS: 5, IQR: 4–8) and Rhinoconjunctivitis Quality of Life Questionnaire (RQLQ: 42, IQR: 32–61), respectively, reflecting clinically relevant allergic burden.

Quality control of sequencing data revealed a unimodal distribution of effective sequence lengths, with a median of 420 bp (IQR: 414.5–419.5 bp), indicating high consistency across samples (Supplementary Figure 1). Rarefaction curves approached asymptote at a sequencing depth of 40,000 reads per sample (Supplementary Figure 2), indicating that the majority of microbial diversity within each sample was effectively captured. This confirms that the sequencing effort was sufficient for robust downstream analyses of alpha and beta diversity. Although the species accumulation curve did not reach a complete plateau, the rate of new species discovery markedly decreased after approximately 20 samples (Supplementary Figure 3), indicating that our cohort size was sufficient to capture the majority of the microbial diversity.

The gut microbiota composition of AR patients (Group A, n = 23) and healthy controls (Group B, n = 15) exhibited significant differences at both phylum and genus levels. At the phylum level (Figure 1A), Firmicutes and Bacteroidetes were dominant in both groups. However, AR patients exhibited a marked increase in the relative abundance of the pro-inflammatory phylum Fusobacteriota, alongside a significant reduction in the SCFA-producing phylum Verrucomicrobiota.

Analysis at the genus level (Figure 1B) revealed that Faecalibacterium, a keystone SCFA-producing genus enriched in healthy individuals, was significantly depleted in AR patients. Other beneficial SCFA producers, including Ruminococcus and Roseburia, also showed a tendency toward reduced abundance in the AR group. Conversely, genera with potential pro-inflammatory associations demonstrated an opposite trend: Fusobacterium, which was nearly absent in HC, was present at high abundance in AR patients. Genus-level heatmap analysis (Figure 1C) and a star plot of the top 10 most abundant genera (Figure 1D) further confirmed distinct clustering patterns between AR patients and HC, reflecting clear differences in microbial community structure.

Linear discriminant analysis (Figure 1E) identified Clostridia, Ruminococcaceae, and Faecalibacterium as key discriminators for HC (LDA score > 3.0), while Synergistetes and Bacillales were enriched in AR. Phylogenetic cladogram analysis (Figure 1F) further substantiated these systematic differences from an evolutionary perspective. HC exhibited enrichment of SCFA-producing families such as Ruminococcaceae and Lachnospiraceae, whereas AR patients showed overrepresentation of potentially pathogenic bacteria, including Collinsella. Collectively, these findings indicate that AR is associated with gut microbial dysbiosis, characterized by a decline in SCFA-producing bacteria and an expansion of pro-inflammatory taxa, which may contribute to disease pathogenesis.

Alpha diversity analysis revealed no statistically significant differences in gut microbial richness and diversity between AR patients and HC (Supplementary Figure 4). Indices reflecting microbial richness—including Observed species (P = 0.174), Chao1 (P = 0.174), and ACE (P = 0.181)—and indices representing microbial diversity, such as Shannon (P = 0.191) and Simpson (P = 0.145), all showed P-values greater than 0.05. Notably, the phylogenetic diversity index phylogenetic diversity whole tree (PD whole tree) approached statistical significance (P = 0.051), suggesting a potential tendency toward reduced phylogenetic complexity of the gut microbiota in AR patients compared with healthy individuals, though this trend did not reach the conventional significance threshold (P < 0.05). Collectively, these findings indicate that the overall taxonomic breadth of the gut microbial community does not differ substantially between AR patients and healthy controls, while the evolutionary composition of the microbiota may have a subtle difference that requires further verification with a larger sample size.

Beta diversity analysis highlighted differences in microbial community structures between the AR patients and HC. Using four common distance metrics (Bray-Curtis, Jaccard, weighted UniFrac, and unweighted UniFrac) and PERMANOVA analysis (with effect size represented by R²), significant differences were observed in Bray-Curtis distance (P = 0.041, R² = 0.043; Figure 2A), Jaccard distance (P = 0.025, R² = 0.031; Figure 2B), and unweighted UniFrac distance (P = 0.047, R² = 0.039; Figure 2D), while weighted UniFrac distance showed no significant difference (P = 0.299, R² = 0.040; Figure 2C). Principal Coordinate Analysis (PCOA) visualization consistently showed that samples based on Bray-Curtis (Figure 2E), Jaccard (Figure 2F), and unweighted UniFrac (Figure 2H) distances exhibited a tendency of separation between the AR patients and HC, whereas the PCOA plot based on weighted UniFrac (Figure 2G) distance showed no obvious group separation. Principal Coordinate 1 (PC1) accounted for 8.6% (Bray-Curtis), 5.29% (Jaccard), 24.97% (weighted UniFrac) and 10.76% (unweighted UniFrac) of the total variance, while Principal Coordinate 2 (PC2) explained 6.78% (Bray-Curtis), 3.98% (Jaccard), 18.95% (weighted UniFrac) and 5.54% (unweighted UniFrac) of the total variance, respectively. These results collectively indicate that AR is associated with alterations in gut microbial community structure, and the compositional dissimilarity between AR patients and HC is mainly driven by the presence/absence of rare taxa rather than the relative abundance of dominant taxa.

The orthogonal partial least squares-discriminant analysis (OPLS-DA) revealed distinct clustering between AR patients and HC in both positive (POS) and negative (NEG) ionization modes (Figures 3A, B). The model demonstrated robust validity, with high explanatory power (R² = 0.97 for POS, R² = 0.94 for NEG) based on permutation testing (Figures 3C, D).

We further identified differentially abundant metabolites between the two groups. In the POS mode (Figure 3E), metabolites such as Epinephrine, Normeperidine, Quercetin, and Maltol exhibited distinct z-scores between AR patients and healthy controls. For instance, Maltol showed a notably higher abundance in AR patients, which was validated by its ROC curve with an AUC of 0.841 [95% CI: 0.706–0.976], 100% sensitivity, and 73.9% specificity (Figure 3G). In the NEG mode (Figure 3F), metabolites including Indolepyruvate, trans-Cinnamic acid, and 4-Coumaric acid were differentially expressed. 4-Coumaric acid had an AUC of 0.852 [95% CI: 0.731–0.973], 86.7% sensitivity, and 78.3% specificity (Figure 3H), indicating its potential as a discriminatory metabolite.

Pathway enrichment analysis highlighted significant perturbations in metabolic networks associated with AR. In the POS mode (Figure 3I), pathways such as biosynthesis of unsaturated fatty acids, steroid hormone biosynthesis, and nicotinate and nicotinamide metabolism were significantly impacted. In the NEG mode (Figure 3J), key pathways included pantothenate and CoA biosynthesis, glycolysis/gluconeogenesis, and pyruvate metabolism, suggesting alterations in energy metabolism and vitamin biosynthesis in AR patients.

To comprehensively assess the interactions between gut microbiota and metabolites in AR patients and HC, we performed Spearman correlation analysis across multiple metabolite classes, including amino acids, lipids, nucleotides, peptides and other metabolites (e.g., carbohydrates, energy-related compounds, vitamins and cofactors). All p-values were adjusted for multiple comparisons using the false discovery rate (FDR) method, and only FDR-corrected p-values (q-values) are reported. Correlations with q < 0.05 were considered statistically significant, while those approaching significance (q < 0.10) are discussed as suggestive trends.

Significant correlations were observed primarily for peptides and other metabolites. Among peptides (Figure 4D), the dipeptide Asp-Glu showed a strong positive correlation with Ruminococcus (ρ = 0.481, q = 0.031). Among other metabolites (Figure 4E), significant associations included positive correlations of D-phenyllactic acid with Faecalibacterium (ρ = 0.515, q = 0.046), 2-methoxy-4-vinylphenol with Ruminococcus (ρ = 0.565, q = 0.021), 4-hydroxybenzaldehyde with Faecalibacterium (ρ = 0.514, q = 0.046), and epinephrine with Olsenella (ρ = 0.537, q = 0.037). A significant negative correlation was observed between 2-methoxy-4-vinylphenol and Ruminococcus torques (ρ = −0.526, q = 0.040). No significant correlations were found for amino acids, lipids, or nucleotides after FDR correction.

Although amino acids, lipids and nucleotides did not yield statistically significant results, several trends were noted. Amino acid (Figure 4A), 17α-Ethynylestradiol was positively associated with Olsenella (ρ = 0.405, q = 0.149). Among lipids (Figure 4B), 2-Hydroxycaproic acid showed positive correlations with Ruminococcus (ρ = 0.440, q = 0.096), and 12-Oxo phytodienoic acid was positively associated with Faecalibacterium (ρ = 0.503, q = 0.084). Nucleotides (Figure 4C), such as flavin mononucleotide, showed positive trends with multiple genera including Ruminococcus (ρ = 0.340, q = 0.061).

In summary, this integrative analysis revealed significant microbe-metabolite interactions primarily involving peptides and other metabolites, particularly those related to microbial metabolism of aromatic compounds, neurotransmitters, and phenolic acids. These findings suggest a specific role for certain bacterial taxa in modulating metabolic pathways relevant to AR, warranting further validation in larger cohorts.