Multi-omics analysis of gut microbiota and metabolome reveals potential biomarkers in pediatric snoring

Abstract

Pediatric snoring, a common manifestation of obstructive sleep apnea (OSA), can significantly impact children’s development. This study aimed to characterize the alterations in gut microbiota and metabolome associated with pediatric snoring. Fecal samples were collected from 30 snoring children and 30 matched healthy children, and analyzed using 16 S rRNA gene sequencing and untargeted metabolomics. Analysis of the gut microbiota revealed distinct community structures between the two groups. Key genera such as Faecalibacterium and Bacteroides were enriched in snoring children, whereas Bifidobacterium and Akkermansia were more abundant in healthy children. Functional prediction indicated significant perturbations in microbial metabolic pathways, including amino acid and lipid metabolism. Metabolomic profiling identified 214 significantly altered metabolites, with 101 upregulated and 113 downregulated in the snoring group. Notable changes were observed in metabolites such as L-Arginine, Guanosine, and N1-Acetylspermidine. Pathway enrichment analysis highlighted dysregulation in purine metabolism and bile secretion. A panel of the top differential metabolites demonstrated high diagnostic accuracy for distinguishing snoring children from controls. In conclusion, this multi-omics study reveals significant and coordinated disruptions in the gut microbiota and metabolome of children who snore. These findings provide a foundation for understanding the role of the gut-microbiota-metabolite axis in pediatric snoring and identify potential non-invasive biomarkers for early detection and future mechanistic investigations.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12866-026-04952-6.

Introduction

Pediatric snoring, a common clinical symptom of sleep-related breathing disorders (SRBDs) in children, is closely linked to obstructive sleep apnea (OSA), a condition characterized by recurrent upper airway collapse, intermittent hypoxia, and sleep fragmentation [1–4]. OSA poses significant multisystem risks to children’s health, including growth impairment, neurocognitive deficits, and cardiovascular complications [5–8]. Epidemiological studies indicate a high disease burden, with habitual snoring affecting 1.5% to 27.6% of children globally and OSA prevalence ranging from 1.2% to 5.7% [9–11]. Despite its clinical significance, pediatric snoring remains underdiagnosed and inadequately managed. Current treatments such as adenotonsillectomy and continuous positive airway pressure (CPAP) are limited by efficacy and compliance issues, while parental misconceptions often delay intervention [2, 9, 12, 13]. Thus, there is an urgent need for non-invasive biomarkers to facilitate early detection and guide personalized management.

In recent years, the bidirectional crosstalk between the gut and systemic health, mediated by the “gut-brain axis” and “gut-lung axis”, has emerged as a central framework for unraveling the pathogenesis of pediatric snoring and its severe clinical correlate, OSA [14, 15]. The gut microbiota, a complex microbial community dominated by phyla such as Firmicutes and Bacteroidetes, functions far beyond aiding digestion [16–18]. It regulates host metabolic homeostasis, modulates immune responses, and even synchronizes circadian rhythms, all of which are disrupted in children with snoring and OSA [18]. Mounting evidence links gut microbial dysbiosis to the development of pediatric snoring. In animal models of intermittent hypoxia (IH), researchers observed significant reductions in gut microbial diversity, depletion of beneficial short-chain fatty acid (SCFA)-producing bacteria, and overgrowth of pro-inflammatory taxa. Notably, restoring normal oxygen levels partially reversed these dysbiotic changes, suggesting a potential causal relationship between IH-induced hypoxia and gut microbial imbalance [19, 20]. Despite these valuable preclinical insights, critical gaps remain in translating this knowledge to pediatric snoring. Animal models of IH do not fully recapitulate the clinical features of pediatric snoring, and the extent to which hypoxia-driven gut dysbiosis contributes to snoring in children has not been systematically characterized. Furthermore, most existing research on pediatric snoring and gut health focuses solely on microbial composition, with little attention given to how microbial shifts translate into metabolic alterations. Understanding these metabolic changes is a crucial step in elucidating how the gut influences systemic health in this population.

To address these gaps, we conducted an integrated multi-omics study comparing gut microbiota and metabolome profiles between snoring children and healthy children. Using 16 S rRNA sequencing and untargeted metabolomics, we aimed to characterize the associated microbial and metabolic alterations, identify potential non-invasive biomarkers, and explore the functional pathways that may link gut dysbiosis to pediatric snoring. Our findings reveal distinct microbiota composition and metabolite profiles in snoring children, providing novel insights into the gut-metabolite axis and offering potential non-invasive biomarkers for early clinical assessment.

Materials and methods

Clinical sample collection

This study was conducted at Anhui Provincial Children’s Hospital from October 2024 to December 2024, with a total of 60 pediatric participants enrolled, including 30 snoring children (Group A, disease group) and 30 age-matched healthy children (Group B, control group). The study protocol was approved by the Institutional Review Board of Anhui provincial children’s hospital (Approval No.: EYLL-2025-073), and written informed consent was obtained from the legal guardians of all participants prior to any sample collection or clinical assessment. All work was conducted in accordance with the Declaration of Helsinki (1964). For Group A (snoring children), participants were aged 2–9 years and confirmed to have snoring with concurrent rhinitis via otolaryngological examination; all had experienced sleep-related snoring for at least 1 month (with or without mouth breathing) and underwent surgical treatment (nasal endoscopic adenoid ablation combined with bilateral partial tonsillectomy) during the study period. Exclusion criteria for Group A included a history of antibiotics, probiotics, or anti-inflammatory drugs use within 2 weeks before sample collection, as well as chronic diseases (e.g., diabetes, hypertension, autoimmune disorders) or gastrointestinal abnormalities. For Group B (healthy children), participants were aged 2–9 years (median age: 6 years) and frequency-matched with Group A by age and gender; they had no snoring, sleep disorders, or chronic diseases confirmed by routine physical examination, no recent use of interfering medications (same as Group A), and no gastrointestinal symptoms (e.g., diarrhea, constipation) or recent infections.

Fecal samples were collected from all participants in the morning within 1 h of defecation using sterile stool collection tubes. For Group A, samples were collected 1 day before surgery to avoid potential interference from surgical procedures or anesthetics; for Group B, samples were collected during routine physical examinations. Immediately after collection, all samples were snap-frozen in liquid nitrogen and transferred to a -80 °C freezer within 2 h to maintain the stability of gut microbiota and metabolites, preventing degradation before subsequent experiments.

Participant characteristics

Comprehensive demographic, anthropometric, and clinical laboratory data were collected for all participants (Table 1). The snoring group (Group A) and the healthy control group (Group B) were well-matched in terms of age, sex, height, weight, body mass index, residential area, and key laboratory parameters including white blood cell count, C-reactive protein, and albumin levels (all P > 0.05). As anticipated, and consistent with the concomitant diagnosis of rhinitis in the snoring cohort, children in Group A exhibited significantly higher levels of total immunoglobulin E (IgE) and a greater prevalence of sensitization to at least one common allergen compared to the healthy controls (both P < 0.05).

Table 1.

Baseline characteristics of snoring children and healthy children

Characteristic	Snoring Children (Group A, n = 30)	Healthy Children (Group B, n = 30)	P-value
Age (years)	5.1 ± 1.7	5.0 ± 2.2	0.8445
Gender (Male/Female)	18/12	16/14	0.6054
Height (cm)	117.1 ± 13.4	114.7 ± 15.3	0.5206
Weight (kg)	22.9 ± 8.3	21.3 ± 7.2	0.4284
BMI (kg/m2)	16.2 ± 2.8	15.8 ± 2.0	0.5268
WBC (×109/L)	8.2 ± 2.4	7.5 ± 2.1	0.2341
CRP (mg/L)	1.2 ± 0.7	1.8 ± 2.8	0.2595
Albumin (g/L)	46.5 ± 3.0	46.3 ± 2.3	0.7730
Total IgE (IU/mL)	161.0 ± 245.8	59.1 ± 85.4	0.0362
Any allergen positive (Yes/No)	16/14	8/22	0.0366
Residential area (Urban/Rural)	16/14	19/11	0.4360

16 S rRNA gene sequencing

Reagents and consumables

Fecal genomic DNA was extracted using the Magnetic Bead-Based Fecal Genomic DNA Extraction Kit (Prepackaged, catalog No. AU46111-96) purchased from BioTeke (China). DNA concentration and purity were quantified using the Qubit dsDNA HS Assay Kit (catalog No. Q32854, Invitrogen, USA) and a NanoDrop ND-1000 Spectrophotometer (Thermo Fisher Scientific, USA), respectively. PCR amplification was performed with Phusion Hot Start Flex 2X Master Mix (catalog No. M0536L, NEB, USA), and custom-synthesized primers targeting the V3-V4 hypervariable region (341 F: 5′-CCTACGGGNGGCWGCAG-3′; 805R: 5′-GACTACHVGGGTATCTAATCC-3′) were obtained from Sangon Biotech (China). PCR products were purified using AMPure XT Beads (catalog No. A63880, Beckman Coulter, USA), and library construction was performed using the Nextera XT Index Kit (catalog No. FC-131-1001, Illumina, USA). Sequencing was conducted with the NovaSeq 6000 SP Reagent Kit (500 cycles, catalog No. 20029137, Illumina, USA) to generate paired-end reads.

16 S rRNA gene sequencing and analysis

DNA extraction and quality control

Fecal genomic DNA was extracted using the Magnetic Bead-Based Fecal Genomic DNA Extraction Kit (catalog No. AU46111-96, BioTeke, China) in strict accordance with the manufacturer’s protocol. After extraction, DNA concentration was quantified using a Qubit fluorometer (Invitrogen, USA), and only samples with a DNA concentration of ≥ 50 ng/µL were selected for subsequent experiments to ensure the reliability of PCR amplification.

PCR amplification and library construction

The V3-V4 hypervariable region of the 16 S rRNA gene was amplified in a 25 µL reaction system, which included 12.5 µL Phusion Hot Start Flex 2X Master Mix (catalog No. M0536L, NEB, USA), 2.5 µL forward primers (341 F: 5′-CCTACGGGNGGCWGCAG-3′), 2.5 µL reverse primers (805R: 5′-GACTACHVGGGTATCTAATCC-3′), 50 ng template DNA, and ddH₂O to make up the total volume. The PCR program was set as follows: pre-denaturation at 98℃ for 30 s; 25 cycles of denaturation at 98℃ for 10 s, annealing at 54℃ for 30 s, and extension at 72℃ for 30 s; and a final extension at 72℃ for 5 min. PCR products were purified using AMPure XT Beads (Beckman Coulter, USA) to remove primer dimers and non-specific amplification products, were then ligated with dual-index adapters using the Nextera XT Index Kit (Illumina, USA) to construct sequencing libraries. Qualified libraries (concentration ≥ 2 nM) were sequenced on the Illumina NovaSeq 6000 platform with 2 × 250 bp paired-end reads, using the NovaSeq 6000 SP Reagent Kit (500 cycles) as the sequencing reagent.

Bioinformatic analysis

Raw sequencing reads were processed using QIIME2 and related bioinformatic tools: primer and adapter sequences were removed using Cutadapt (v1.9), paired-end reads were assembled into full-length sequences using FLASH (v1.2.8) with an overlap threshold of ≥ 10 bp and mismatch rate ≤ 2%, low-quality reads (average quality score < 20, read length < 100 bp) were filtered out using fqtrim (v0.94), and chimeric sequences were eliminated using Vsearch (v2.3.4) against the SILVA database. Denoising was performed with DADA2 to generate amplicon sequence variants (ASVs), and singletons (ASVs with a total read count of 1) were excluded to reduce noise. The ASV abundance table was normalized to relative abundance for downstream ecological analyses. Alpha diversity indices (Chao1, observed_species, Shannon, Simpson, goods_coverage, Pielou_E, ACE) were calculated using QIIME2. Due to the non-normal distribution of most diversity indices (assessed by Shapiro–Wilk test), inter-group differences were compared using the Wilcoxon rank-sum test. Beta diversity analysis was performed to evaluate between-sample differences in microbial community structure, using Principal Component Analysis (PCA), Principal Coordinates Analysis (PCoA), and Non-metric Multidimensional Scaling (NMDS) based on Bray-Curtis, Jaccard, and unweighted-Unifrac distances; the significance of inter-group differences was verified via Anosim test. Species composition analysis at the genus level included clustering analysis of the top 30 genera to visualize similarities in community composition among samples, and species difference analysis was conducted to screen for taxa with significant abundance differences between groups. Linear Discriminant Analysis Effect Size (LEfSe) (with thresholds of LDA > 3.0 and p < 0.05) and Indicator analysis were used to identify potential microbial biomarkers. For genus-level abundance comparison between groups, the Wilcoxon rank-sum test was applied after confirming the non-normal distribution of microbial abundance data via Shapiro-Wilk test. Functional prediction of microbial communities was performed using PICRUSt2 to predict KEGG orthologs and metabolic pathways, thereby exploring differences in functional profiles between groups.

Untargeted metabolomics detection and analysis

Reagents and instrumentation

HPLC-grade methanol (4 L/bottle, catalog No. A-456-4, Fisher Scientific, USA), HPLC-grade acetonitrile (4 L/bottle, catalog No. 955-4, Fisher Scientific, USA), and LC-MS-grade formic acid (50 mL/bottle, catalog No. A17-50, Fisher Scientific, USA) were used for metabolite extraction and chromatographic separation. The analytical system consisted of a Thermo Vanquish Flex ultra-high-performance liquid chromatography (UHPLC) system coupled with a Q-Exactive high-resolution mass spectrometer (Thermo Scientific, USA), equipped with an ACQUITY UPLC T3 column (100 mm × 2.1 mm, 1.8 μm, Waters, USA).

Sample pretreatment

Frozen fecal samples (100 mg) were thawed on ice and mixed with 1 mL pre-cooled 50% methanol (v/v). The mixture was vortexed for 1 min, incubated at room temperature for 10 min, and stored at -20℃ overnight to maximize metabolite extraction. After incubation, samples were centrifuged at 4000 × g for 20 min at 4℃, and the supernatant was transferred to a new 96-well plate. A pooled quality control (QC) sample was prepared by mixing 10 µL of supernatant from each participant sample to monitor instrument stability during analysis. All samples (including QC samples) were stored at − 80℃ until UHPLC-MS/MS analysis.

UHPLC-MS/MS conditions

Chromatographic separation was performed with the ACQUITY UPLC T3 column maintained at 40℃, with a flow rate of 0.3 mL/min. The mobile phase consisted of solvent A (5 mM ammonium acetate + 5 mM acetic acid) and solvent B (acetonitrile), using the following gradient elution program: 0 ~ 0.8 min (2% B), 0.8 ~ 2.8 min (2%~70% B), 2.8 ~ 5.6 min (70%~90% B), 5.6 ~ 6.4 min (90%~100% B), 6.4 ~ 8.0 min (100% B), and 8.1 ~ 10 min (2% B). Mass spectrometric detection was conducted in both positive and negative ionization modes: full-scan precursor spectra (m/z 70 ~ 1050) were collected at a resolution of 70,000 with an automatic gain control (AGC) target of 3e6 and maximum injection time of 100 ms; in data-dependent acquisition (DDA) mode, the top 3 most abundant ions were selected for MS/MS fragmentation, with fragment spectra collected at a resolution of 17,500, AGC target of 1e5, and maximum injection time of 80 ms. A QC sample was analyzed every 10 test samples to ensure instrument stability, with a relative standard deviation (RSD) of peak intensity < 30% considered acceptable.

Data processing and analysis

Raw mass spectrometry data were converted from .raw format to mzXML format using MSConvert (ProteoWizard, v3.0.23193). Peak picking, retention time correction, and peak alignment were performed using XCMS (v3.18.1), and subsequent data normalization (median normalization) to eliminate systematic errors, removal of low-quality features (present in < 50% of samples), and missing value imputation (Minimum method) to complete the metabolite matrix were conducted with the metaX package (v1.4.0) in R (v4.1.3). Metabolite annotation was achieved by matching the accurate mass (m/z) of samples with the Human Metabolome Database (HMDB, v5.3) and Kyoto Encyclopedia of Genes and Genomes (KEGG, v109) (mass tolerance < 10 ppm), and validated using an in-house MS/MS library (matching score > 75%) to ensure the reliability of secondary metabolite identification; classification annotations of identified metabolites in HMDB (SuperClass level) and KEGG (level 1, 2, 3 pathways) were also conducted.

Multivariate analysis included Principal Component Analysis (PCA) to visualize overall differences in metabolic profiles between groups and Partial Least Squares Discriminant Analysis (PLS-DA) to enhance group separation. Differential metabolite screening was performed using the following thresholds: fold change (FC) ≥ 1.2 or ≤ 1/1.2, p < 0.05 (t-test), and variable importance in projection (VIP) ≥ 1 (from PLS-DA); the number of differential metabolic ions and differential secondary metabolites was statistically summarized, and a volcano plot was used to visualize the distribution of differential metabolites. The top 30 differential metabolites were displayed using a heatmap (Z-score normalization, allowing only horizontal comparison of the abundance of the same metabolite across different samples, with vertical comparison being meaningless) and boxplots (log₂-transformed abundance values to normalize the metabolite abundance distribution and reduce data skewness, showing inter-group differences). Receiver Operating Characteristic (ROC) curves were plotted to evaluate the diagnostic potential of differential metabolites, with the top 5 metabolites and an overall combined metabolite panel being analyzed. For inter-group comparison of differential metabolite abundances, the Wilcoxon rank-sum test was used when the metabolite abundance data were confirmed to be non-normally distributed via Shapiro-Wilk test. Functional analysis of differential metabolites included KEGG pathway enrichment analysis (hypergeometric test, P < 0.05 after Benjamini-Hochberg correction) to identify perturbed metabolic pathways, and a network diagram was constructed to visualize the regulatory relationships between the top 30 differential metabolites and their associated KEGG pathways (triangles for metabolites, circles for pathways).

Statistical analysis

All statistical analyses were performed using SPSS 26.0 (IBM, USA) and R 4.1.3. The Shapiro-Wilk test was used to assess the normality of continuous variables: normally distributed data were presented as mean ± standard deviation (SD), and comparisons between groups were performed using the independent two-sample Student’s t-test; non-normally distributed data were presented as median (interquartile range, IQR), and the Wilcoxon rank-sum test was applied for inter-group comparison to ensure robust statistical inference for non-normal data, which was the case for most alpha diversity indices, microbial genus abundance data, and metabolite abundance data in this study. Inter-group comparisons of clinical characteristics, alpha diversity indices, and differential metabolite abundances were conducted using the appropriate test based on the normality assessment. Correlations between differential microbial taxa and metabolites were analyzed using Spearman’s rank correlation coefficient. A two-tailed P < 0.05 was considered statistically significant, and the Benjamini-Hochberg method was used to correct for multiple comparisons, controlling the false discovery rate (FDR) to reduce the risk of type I errors.

Results

Gut microbiota diversity and community structure differ between snoring and healthy children

Analysis of gut microbial alpha diversity revealed no significant differences between snoring (Group A) and healthy (Group B) children in species richness or evenness indices (Chao1, Shannon, Simpson, etc.; p.adj > 0.05), except for a lower sequencing coverage (goods_coverage, p.adj = 0.04) in the snoring group (Fig. 1). Rarefaction curves confirmed sufficient sequencing depth (Fig. 1B–E). However, beta diversity analysis demonstrated a distinct separation of gut microbial communities between the two groups. Principal coordinate analysis (PCoA) and non‑metric multidimensional scaling (NMDS) based on Bray‑Curtis, Jaccard, unweighted‑UniFrac and weighted‑UniFrac distances showed clear clustering by group (Fig. 2). This separation was statistically validated by ANOSIM tests, which confirmed that inter‑group differences were significantly greater than intra‑group differences (Bray‑Curtis: R = 0.0708, P = 0.006; Jaccard: R = 0.1271, P = 0.001; unweighted‑UniFrac: R = 0.1621, P = 0.001) (Fig. 3). For weighted‑UniFrac, although a trend toward separation was observed, the difference did not reach statistical significance (R = 0.035, P = 0.063). These results indicate that while overall within‑sample diversity was similar, the composition and structure of the gut microbiota were significantly altered in children who snore.

Fig. 1.

Alpha diversity and amplicon sequence variant (ASV) profiles of gut microbiota in snoring children and healthy children. A Venn diagram depicting shared (1290) and group-unique ASVs (2167 in Group A (snoring children), 2592 in Group B (healthy children)). B–E Rarefaction curves for Chao1, observed_species, Shannon, and Simpson indices confirm sufficient sequencing depth. F–L Violin-box plots of Chao1, observed_species, Shannon, Simpson, pielou_e and ace indices show no significant inter-group differences; only goods_coverage (reflecting sequencing completeness) differs significantly between groups (p.adj = 0.04, data not visualized here). M Rank-abundance curves illustrating species richness and evenness, with overlapping distribution patterns across groups. *P < 0.05

Fig. 2.

Beta diversity analysis of gut microbiota between snoring children and healthy children. A Principal Components Analysis (PCA) plot, where PC1 and PC2 account for 30.9% and 9.2% of total variance, respectively, showing clustering trends of samples by group. B–E Principal Co-ordinates Analysis (PCoA) plots using Bray-Curtis, Jaccard, unweighted UniFrac, and weighted UniFrac distance metrics, respectively, all illustrating separation tendencies between Group A (snoring children, pink) and Group B (healthy children, blue). F–I Non-metric Multidimensional Scaling (NMDS) plots with stress values ranging from 0.1348 to 0.2206, further demonstrating distinct microbial community clustering between the two groups

Fig. 3.

Analysis of similarities (Anosim) for gut microbial communities between snoring children and healthy children. A Anosim plot based on Bray-Curtis distance, demonstrating significantly greater dissimilarity between groups than within groups (P = 0.006, R = 0.0708). B Anosim plot based on Jaccard distance, with significant inter-group differences (P = 0.001, R = 0.1271). C Anosim plot based on unweighted-UniFrac distance, confirming robust inter-group dissimilarity (P = 0.001, R = 0.1621). D Anosim plot based on weighted-UniFrac distance, where differences did not reach statistical significance (P = 0.063, R = 0.035). Collectively, Bray-Curtis, Jaccard, and unweighted UniFrac metrics confirm significant differences in gut microbial community structure between snoring children and healthy children. *P < 0.05, **P < 0.01

Altered gut microbiota composition and predicted function in snoring children

Genus-level composition differed significantly between groups (Fig. 4). Snoring children exhibited higher relative abundances of Faecalibacterium (≈ 18%) and Bacteroides (≈ 15%), whereas healthy children were enriched in Bifidobacterium (≈ 20%). Unsupervised clustering confirmed that samples were grouped by health status (Fig. 4B, C).

Fig. 4.

Composition and abundance patterns of gut microbiota at the genus level between snoring children and healthy children. A Stacked bar plot displaying the relative abundance of the top 30 genera, illustrating inter-group compositional differences. B Heatmap of the top 30 genera, with hierarchical clustering (rows and columns) and z-score normalization to visualize relative abundance variations; color coding denotes phylum affiliation and group-specific abundance trends. C Cluster stacked bar plot of the top 30 genera, further emphasizing group-wise compositional clustering. D Venn diagram showing shared (348) and group-unique genera (119 in Group A (snoring children), 99 in Group B (healthy children)). E Bubble plot of taxonomy for key genera, where bubble size represents relative abundance and color indicates phylum, highlighting differential genus distribution across groups. These analyses collectively reveal distinct genus-level gut microbial signatures in snoring children compared to healthy children

Differential and biomarker analyses identified specific taxa associated with snoring (Fig. 5). LEfSe analysis identified Faecalibacterium (LDA score = 4.5) as a key biomarker for the snoring group, while indicator species analysis highlighted Akkermansia as strongly associated with health (Fig. 5B, C).

Fig. 5.

Taxonomic and functional characterization of gut microbiota in snoring children and healthy children. A Histogram of logarithmic relative abundance for differentially abundant genera, showing genus-level abundance differences between groups. B Linear discriminant analysis effect size (LEfSe) plot (LDA score > 3.0, P < 0.05), identifying genera enriched in Group A (snoring children, gold) and Group B (healthy children, green). C Taxonomy indicator bubble plot, where bubble color represents p-value range, bubble size denotes sqrtIV (indicator value), and position indicates group-specific genus enrichment. D, E Functional pathway prediction via PICRUSt2 at KEGG Level 2 (D) and Level 3 (E), displaying differences in mean proportions of functional pathways between groups; orange bars=Group A (snoring children), blue bars=Group B (healthy children), with 95% confidence intervals and p-values for inter-group comparisons. Collectively, these analyses reveal distinct taxonomic biomarkers and functional perturbations in the gut microbiota of snoring children. *P < 0.05

Functional prediction (PICRUSt2) revealed significant perturbations in microbial metabolic potential. At KEGG Level 2, pathways including Cellular Processes and Signaling (P = 0.0358) and Lipid Metabolism were more abundant in snoring children, while Amino Acid Metabolism (P = 0.0191) and Translation were enriched in controls (Fig. 5D). Level 3 analysis further identified dysregulation in specific sub-pathways such as Fatty acid biosynthesis (P = 0.0136) and RNA polymerase (P = 0.0142) (Fig. 5E).

Distinct fecal metabolomic profiles in snoring children and their association with gut microbiota

Metabolite profiling revealed extensive alterations in the fecal metabolome of snoring children (Fig. 6). Annotation identified metabolites primarily belonging to Lipids and lipid-like molecules (548 metabolites), Organoheterocyclic compounds (451 metabolites), and Organic acids and derivatives (354 metabolites) (Fig. 6A). KEGG pathway analysis showed enrichment in core Metabolic pathways, with key sub-pathways including Carbon metabolism, Purine metabolism, and Fatty acid metabolism (Fig. 6B). Multivariate analysis confirmed clear separation between the metabolic profiles of the two groups. Partial Least Squares-Discriminant Analysis (PLS-DA) showed distinct clustering of snoring (Group A) and healthy (Group B) children (Fig. 6C). Permutation testing validated the robustness of the model without overfitting (Fig. 6D). Collectively, these findings demonstrate a disrupted fecal metabolome in snoring children, characterized by perturbations in lipid/organic acid metabolism and core metabolic pathways. These metabolic shifts align with the observed gut microbiota alterations, supporting a link between microbial dysbiosis and metabolic dysfunction in pediatric snoring.

Fig. 6.

Metabolomic profiling of fecal samples from snoring children and healthy children. A Histogram of metabolite counts classified by HMDB SuperClass, showing the distribution of identified metabolite classes. B Histogram of KEGG Level 3 pathway counts, color-coded by KEGG Level 1 categories (Environmental Information Processing, Human Diseases, Metabolism, Organismal Systems), illustrating pathway coverage in metabolomic annotation. C Principal component analysis (PCA) plot, where PC1 (8.79%) and PC2 (7.18%) separate quality control (QC, red), Group B (healthy children, yellow), and Group A (snoring children, blue) samples, revealing distinct metabolic clustering between groups. D Correlation plot for model evaluation: red dots (QC samples) show high correlation (indicating analytical stability), while blue dots (study samples) display model fitting with intercepts of R2 (0.774) and Q2 (-0.275) reflecting model performance. These results demonstrate significant metabolic differences between snoring children and healthy children

Identification and characterization of differentially abundant metabolites

Differential metabolite screening identified 214 significantly altered metabolites in snoring children compared to healthy controls (101 upregulated, 113 downregulated) (Fig. 7A, B, F). Multivariate analyses confirmed distinct global metabolic profiles between groups. Principal component analysis (PCA) showed clear separation along the first principal component (PC1, 71.17% variance) (Fig. 7C). Supervised PLS-DA further enhanced group discrimination, with a robust model validated by permutation testing (R2 intercept 0.859, Q2 intercept − 0.209) (Fig. 7D, E). Key differentially abundant metabolites included L-Arginine, PG (16:0/18:3(6Z,9Z,12Z)), gamma-Glutamylleucine, Guanosine, and N1-Acetylspermidine (upregulated in snoring), and 1-Methyladenosine, all-trans-5,6-Epoxyretinoic acid, and 2-Hydroxyhexadecanoic acid (downregulated) (Fig. 7G, Fig. S1). These metabolites are implicated in pathways such as purine metabolism and bile secretion, reinforcing the connection between metabolic dysregulation and pediatric snoring.

Fig. 7.

Differential metabolomic analysis between snoring children and healthy children. A Bar plot showing the total number of differential metabolites identified (1473 upregulated, 1809 downregulated). B Bar plot of significantly differential metabolites after strict screening (101 upregulated, 113 downregulated). C Principal Components Analysis (PCA) plot illustrating separation of quality control (QC, purple) and study samples, with PC1 accounting for 71.17% of variance. D PCA plot differentiating Group A (snoring children, blue) and Group B (healthy children, red) samples, with PC1 explaining 11.20% of variance. E Correlation plot for model evaluation: red dots (QC samples) indicate analytical stability, while blue dots (study samples) show model fitting with intercepts of R2 (0.859) and Q2 (-0.206). F Volcano plot of differential metabolites (AVSB comparison): red dots=significantly upregulated (101), blue dots=significantly downregulated (113), gray dots = non-differential metabolites (1341); x-axis = log2(fold change), y-axis=-log10(p-value). G Heatmap of top differential metabolites, with color gradient representing relative abundance (blue = low, red=high), highlighting distinct metabolic signatures between groups. These analyses confirm substantial metabolomic differences between snoring children and healthy children

Differential metabolites show diagnostic potential and converge on core metabolic pathways

ROC analysis indicated strong diagnostic value for the differential metabolites. The top five individual metabolites showed good performance, with AUC values ranging from 0.77 to 0.81. A combined logistic regression panel achieved superior diagnostic accuracy (AUC = 0.94) (Fig. 8A, B). KEGG enrichment analysis revealed that the altered metabolites were significantly enriched in core metabolic pathways, most notably Purine metabolism and Lipid metabolism (Fig. 8C, D). A metabolite-pathway interaction network highlighted purine metabolism as a central hub, connecting to 11 differential metabolites (e.g., guanosine, adenosine) (Fig. 8E). This visualizes the functional convergence of metabolic disturbances in snoring.

Fig. 8.

Diagnostic potential and metabolic pathway enrichment of differential metabolites. A Receiver operating characteristic (ROC) curves for the top 5 individual differential metabolites, with area under the curve (AUC) values ranging from 0.7733 to 0.8133. B ROC curve for the combined logistic regression model of differential metabolites, showing an AUC of 0.9389, indicating robust diagnostic performance. C KEGG enrichment bar plot, where bar length represents the number of metabolites annotated to each pathway, and color denotes KEGG functional category. D KEGG enrichment scatter plot, where bubble size reflects the number of metabolites in a pathway, color indicates p-value, and the x-axis (Rich Factor) represents the ratio of differential metabolites to total metabolites in a pathway. E KEGG metabolic pathway network, visualizing connections between differential metabolites (nodes) and enriched pathways (edges), with node properties indicating metabolite significance and pathway affiliation. These analyses underscore the diagnostic utility of differential metabolites and perturbed metabolic pathways in snoring children

Discussion

This integrated multi-omics study, utilizing 16 S rRNA gene sequencing and untargeted metabolomics, characterized the gut microbiome and metabolome in 30 snoring children compared to age-matched healthy controls. Our aim was to elucidate perturbations along the gut-brain/lung axis in pediatric snoring, a common precursor to obstructive sleep apnea (OSA) with significant developmental implications [21, 22]. By situating our results within the context of pediatric OSA and gut homeostasis literature, we identify key patterns that advance the understanding of pathophysiological links between gut dysbiosis, metabolic dysfunction, and early sleep-disordered breathing.

Our gut microbiota findings reflect both confirmatory and novel aspects when contextualized within the existing pediatric OSA literature, potentially highlighting differences related to disease stage between habitual snoring and established OSA. In line with studies reporting reduced Bifidobacterium in children with OSA [14], we observed lower levels of this beneficial, SCFA-producing genus in snoring children, suggesting an early loss of microbes crucial for intestinal barrier integrity and anti-inflammatory responses [20, 23, 24]. However, unlike the marked reductions in α-diversity often reported in full-blown OSA [14, 19], we found no significant differences in richness or evenness indices, except for sequencing coverage. This may indicate that a loss of microbial diversity progresses with increasing severity along the OSA spectrum. Furthermore, while the increase in Bacteroides aligns with reports of pro-inflammatory taxon expansion in OSA [20], the elevation of Faecalibacterium in our cohort contrasts with its reported decrease in adult OSA [19], possibly due to age-related microbial dynamics or the milder intermittent hypoxia characteristic of pediatric snoring.

Most notably, our integrated metabolomic analysis uncovered significant and novel perturbations. We identified 214 significantly changed metabolites, with enrichment in pathways including purine metabolism and bile secretion—the latter finding is consistent with prior evidence linking OSA to bile acid homeostasis disturbances [25, 26]. Specifically, the prominent dysregulation of purine metabolism, to our knowledge, has not been a focal point in prior gut-focused studies of pediatric snoring or early OSA. This finding, along with the distinct co-variation patterns we observed between specific microbial taxa and fecal metabolites, provides a novel multi-omics perspective on the gut-environment in the early phases of sleep-disordered breathing. Dysregulated bile acid profiles may contribute to pathogenesis by impairing intestinal barrier function and promoting systemic inflammation [27, 28], suggesting a plausible mechanism through which snoring influences systemic health. The observed fecal metabolic and microbial signatures further suggest that gut dysbiosis may remotely influence upper airway physiology. Gut-derived metabolites such as short-chain fatty acids, bile acids, and purine-related compounds can enter circulation and act as signaling molecules. Their alterations could weaken anti-inflammatory responses, disrupt circadian and immune homeostasis, and potentially affect pharyngeal mucosal immunity or neural signaling via the gut-brain axis, thereby linking fecal metabolic perturbations to snoring pathophysiology. Notably, a combined panel of differential metabolites exhibited high diagnostic accuracy (AUC = 0.9389), outperforming individual markers. This supports the potential of multi-metabolite signatures as non-invasive biomarkers for pediatric snoring and underscores the clinical relevance of the observed metabolic disturbances.

These fecal-based biomarkers are promising for clinical translation, given their non-invasive nature particularly suitable for pediatric screening. Beyond diagnosis, specific metabolites such as guanosine (within the purine metabolism pathway) hold biological relevance. Purine metabolites are involved in immune modulation, cellular energy sensing, and neural signaling. Their dysregulation may thus reflect or contribute to the systemic inflammation and altered neuromodulatory tone seen in sleep-disordered breathing, suggesting both mechanistic insights and potential targets for future intervention.

This study has several limitations. First, the relatively small sample size (n = 60) may limit the generalizability of our findings. Second, the cross-sectional design and lack of long-term follow-up prevent us from establishing the temporal relationship between gut perturbations and the progression of snoring. Third, we did not measure critical functional metabolites such as SCFAs [29] and trimethylamine N-oxide (TMAO) [30], which are known to mediate gut-host interactions in OSA. Fourth, while we excluded participants with antibiotic use within 2 weeks of sample collection, residual effects of prior medications cannot be fully ruled out. Finally, the study does not establish causal relationships between gut microbiota alterations and metabolic changes.

In conclusion, this multi‑omics study delineates unique gut microbial and metabolic signatures in pediatric snoring. Altered abundance of Bifidobacterium corroborates previous findings, while novel perturbations in purine metabolism offer new insights into early sleep‑disordered breathing. These differential metabolites and pathways may serve as non‑invasive biomarkers, supporting early diagnosis and targeted management of this common condition [9, 14, 31].

Authors’ contributions

Guarantor of integrity of the entire study, study concepts, study design, definition of intellectual content: Lihua Zhou; study concepts: Xiaoyan Chen, Lihua Huang, Lihua Zhou; study design: Xiaoyan Chen, Lihua Huang, Zhao Cheng, Lihua Zhou; definition of intellectual content: Xiaoyan Chen, Lihua Huang, Lihua Zhou; literature research: Xiangjun Wan, Ying Yang; clinical studies: Zhao Cheng, Ying Yang, Xiangjun Wan; experimental studies: Yan Xu, Shuaitong Chen; data acquisition: Xiaoyan Chen, Lihua Huang, Yan Xu, Shuaitong Chen; data analysis: Lihua Huang, Yan Xu, Shuaitong Chen; statistical analysis: Zhao Cheng, Yan Xu; manuscript preparation: Xiaoyan Chen, Lihua Huang; manuscript editing: Xiaoyan Chen, Lihua Huang; manuscript review: Lihua Zhou. All authors reviewed and approved the manuscript.

Funding

This study was supported by the 2024 Anhui Provincial Quality Engineering Project: Medical Communication and Cultural Competence (Grant No. 2024jcjs052) and 2020 Anhui Provincial Demonstration Grassroots Teaching Organization (Obstetrics and Gynecology Nursing Teaching and Research Office, Grant No. 215).

Data availability

The datasets generated and/or analysed during the current study are available in the Sequence Read Archive (SRA) repository, BioProject: PRJNA1384292 (16S rRNA); and in the OMIX repository (hosted by the National Genomics Data Center, China National Center for Bioinformation), BioProject: PRJCA053754 (metabolomic data).

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Institutional Review Board of Anhui provincial children’s hospital (Approval No.: EYLL-2025-073), and written informed consent was obtained from the legal guardians of all participants prior to any sample collection or clinical assessment. All work was conducted in accordance with the Declaration of Helsinki (1964).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.