A total of 25 fecal samples were obtained from children in the Changchun region of Northeast China, including 12 samples from children diagnosed with norovirus (NoV) infection and 13 samples from healthy children as controls. The NoV-positive children were confirmed through clinical diagnosis and laboratory testing, including RT-PCR detection of NoV RNA in fecal samples. Healthy children were selected based on the absence of gastrointestinal symptoms and no history of NoV infection in the past six months. Study cohort stratification systematically accounted for pediatric age, dietary patterns, antibiotic exposure history, and comorbid conditions to ensure group allocation independence from confounding variables. Written informed consent was obtained from the parents or legal guardians of all participants. The study protocol was reviewed and approved by the Ethics Committee of the China-Japan Union Hospital of Jilin University (approval number: 2025012117). Fresh fecal samples were processed in sterile containers, immediately frozen at -80°C, and subsequently transported to the laboratory for further processing and analysis.

Genomic DNA was extracted from fecal samples using either the MagPure Stool DNA KF Kit B or the Magnetic Bead Fecal and Soil Genome Extraction Kit (MAGEN, Guangzhou, China), according to the manufacturer’s instructions. In brief, 100–200 mg of fecal sample was placed into a centrifuge tube preloaded with grinding beads, followed by the addition of 1 mL ATL/PVP-10 buffer. Mechanical homogenization was performed using a tissue grinder (Shanghai Jingxin Tech) followed by thermal incubation at 65°C for 20 minutes. Following centrifugation at 14,000 × g for 5 minutes (Eppendorf, Germany), the clarified supernatant was aspirated into fresh reaction vessels. Phase separation was achieved by adding 0.6 mL PCI buffer with 15-second vortex mixing prior to high-speed centrifugation (18,213 × g, 10 minutes). The aqueous phase was combined with magnetic bead binding buffer (600 µL), Proteinase K (20 µL), and RNase A (5 µL) in deep-well plates. Automated nucleic acid isolation was conducted on the KingFisher platform (Thermo Fisher Scientific, USA) using sequential 700 µL aliquots of three distinct wash buffers, culminating in target molecule elution with 100 µL elution buffer. Purified DNA aliquots were cryopreserved at -80°C in 1.5 mL microcentrifuge tubes for downstream analytical processes.

DNA libraries were prepared using the BGI Optimal DNA Library Prep Kit (BGI, Shenzhen, China). Genomic DNA was sheared and size-selected using magnetic bead purification. Fragmented DNA underwent end repair to produce blunt ends, followed by the addition of a 3’-adenine overhang. Sequencing adapters were ligated to the repaired DNA fragments, and the ligated products were PCR-amplified to construct the final library. Rigorous quality control was performed to assess the integrity and uniformity of the final libraries. The double-stranded DNA libraries were denatured to single-stranded DNA and circularized to generate single-stranded circular DNA molecules. Residual linear DNA molecules were eliminated by enzymatic digestion. Each DNB contains approximately 300 tandem copies of the original library molecule. DNBs were loaded onto a patterned nanoarray and sequenced on the DNBSEQ-T7 platform (BGI, Shenzhen, China), generating 150 bp paired-end reads.

Raw sequencing reads were quality-filtered and trimmed using SOAPnuke v1.5.2 (Chen et al., 2018), with low-quality reads and adapter sequences removed. Host-derived reads were identified and removed by alignment against the host genome using SOAP2 (Li et al., 2009). High-quality reads were assembled de novo using MEGAHIT v1.1.4 (Li et al., 2015), and contigs shorter than 200 bp were excluded from further analysis. Open reading frames (ORFs) were predicted from the assembled contigs using MetaGeneMark (Zhu et al., 2010). Redundant genes were clustered and removed using CD-HIT v4.6.6 (Fu et al., 2012) with identity and alignment coverage thresholds set at 95% and 90%, respectively.

Gene abundance matrices were generated and quantified using Salmon (v1.5.2) (Patro et al., 2017). Predicted protein sequences were aligned to functional databases, including BacMet, CARD, KEGG, EggNOG, COG, Swiss-Prot, and CAZy, using DIAMOND v0.9.24 (Buchfink et al., 2015) with an E-value threshold of 1e−5. Taxonomic classification was performed using Kraken2 (Wood et al., 2019) based on the lowest common ancestor (LCA) algorithm. For human gut samples, taxonomic annotation was performed against the Unified Human Gastrointestinal Genome (UHGG) v1.0 database (Almeida et al., 2021), which includes more than 200,000 gut microbial genomes. For non-human samples, taxonomic classification was based on the NCBI Nt database (release 202011). Taxonomic and functional abundance profiles were computed using Bracken (https://github.com/jenniferlu717/Bracken) with default settings. Differential abundance analyses of genera, phyla, and KEGG orthologs (KOs) were performed using the Wilcoxon rank-sum test (Matsouaka et al., 2018). Reporter score analysis (Patil and Nielsen, 2005) was applied to identify significantly enriched KEGG pathways, with a significance threshold set at an absolute score of 1.65. Alpha diversity indices, including Shannon, Chao1, and Simpson indices, were calculated at the gene, genus, and KO levels using the vegan R package. Beta diversity was evaluated based on Bray-Curtis dissimilarity (JRBJT, 1957) and Jensen-Shannon divergence (Majtey et al., 2005). All statistical analyses, including Wilcoxon rank-sum and Kruskal-Wallis H tests, were performed in the R environment (v4.0.2).

This longitudinal study prospectively acquired 25 stool specimens from pediatric subjects in Jilin province during a 24-month surveillance period, stratified into NoV-infected (n=12) and asymptomatic control (n=13) cohorts. These samples were sent to BGI for DNA extraction and metagenomic sequencing using the DEBSEQ-T7 platform ( Figure 1A ). To evaluate and predict the increase in species richness with sample size, species accumulation curves were generated for both healthy and NoV-infected groups ( Figures 1B, C ). The results indicated that the number of species in both groups was approximately 6,000, with a slightly higher species richness observed in the NoV-infected group compared to the healthy group. Quantitative analysis of alpha diversity metrics (Chao1, Shannon, and Simpson indices) enabled systematic comparison of microbial community richness and structural evenness across experimental cohorts. ( Figure 1D , Supplementary Table S1 ). Consistent with the species accumulation curves, the NoV-infected group exhibited higher alpha diversity than the healthy group, although the difference was not statistically significant. Principal Component Analysis (PLS-DA) was implemented to discern discriminative features in microbiota structural profiles between the two groups ( Figure 1E ). The results revealed that while the two groups were closely positioned on the PLS-DA plot, indicating similar species composition, they were distinctly separated in direction, suggesting notable differences between the groups.

Species annotation was performed against the NCBI-NR database, and the relative abundance of gut microbiota was calculated at six taxonomic hierarchies (phylum to species) for each individual sample. The top 10 most abundant species were selected to construct stacked bar plots, providing a visual representation of the composition and relative proportions of the major gut microbiota ( Figures 2A–F ). Although individual variation in microbial composition was observed, Bacteroidaceae and Lachnospiraceae were identified as the predominant families in both groups. Notably, sample con3 exhibited a distinct microbial profile dominated by Bifidobacteriaceae, which markedly differed from the other samples.

To further investigate the average distribution of gut microbiota in healthy and NoV-infected children, the taxonomic abundance at different levels (phylum, class, order, family, genus, and species) was calculated as the group mean ( Figures 3A–F ). Notably, undefined taxa showed upregulation following norovirus infection ( Figures 3E, F ), suggesting potential novel virus-associated microbiota. Cross-phylogenetic analysis revealed that healthy controls exhibited elevated Bacteroidetes abundance at phylum, class and order taxonomic ranks compared to NoV-infected subjects. Comparative analysis at the ordinal taxonomic rank revealed significant enrichment of Eubacterium populations in healthy controls compared to NoV-infected individuals. Notably, the proportion of unclassified microbial taxa increased at the family, genus, and species levels, especially in the NoV-infected group. Heatmaps were generated to visualize the relative abundance of gut microbiota across different taxonomic levels ( Figures 3G–L ). The results demonstrated that Bacteroidetes was the most dominant phylum, while Tectiliviricetes exhibited a relatively low abundance. At the family, genus, and species levels, both Intestiviridae and Carjivirus were identified as low-abundance taxa.

A GraPhlAn phylogenetic tree was constructed to illustrate the evolutionary relationships of microbiota between the healthy and NoV-infected groups ( Figure 4A , Supplementary Table S2 ). The tree revealed that Actinomycetota, Bacillota, Bacteroidota, and Pseudomonadota were the core microbiota with clear evolutionary relationships. LEfSe was employed to detect differentially abundant species across groups (LDA >3), revealing Bacteroides uniformis as the most discriminative taxon (LDA >4) in NoV-infected individuals ( Figure 4B ).This suggests that changes in Bacteroides uniformis may be closely associated with NoV infection.

Taxa with significant inter-group abundance variations were identified and quantified, revealing a >3-fold increase in Bacteroides uniformis in NoV-infected individuals ( Figure 5A , Supplementary Table S2 ). Other taxa, including Veillonella sp. S12025-13, Carjivirus communis, and Veillonella nakazawae, also exhibited significant differences, suggesting their potential roles in NoV infection. Additionally, the mean values and confidence intervals of differential microbiota were calculated ( Figure 5B ). The results confirmed that Bacteroides uniformis had the highest confidence interval and degree of difference, while Veillonella sp. S12025-13, Carjivirus communis, Veillonella nakazawae, and Limosilactobacillus mucosae also showed significant upregulation in the NoV-infected group.

KEGG pathway analysis demonstrated a broader functional gene repertoire in the NoV-infected cohort, with significant enrichment in carbohydrate, lipid, and cofactor/vitamin metabolism compared to healthy controls ( Figure 6A , Supplementary Table S3 ). This metabolic reprogramming suggests a potential adaptation to support viral replication or host immune evasion. Detailed metabolic profiling ( Figure 6B , Supplementary Table S4 ) revealed upregulated pathways in the NoV-infected group, including galactose utilization, C5-branched dibasic acid processing, and branched-chain amino acid (valine, leucine, isoleucine) biosynthesis. In contrast, the healthy group exhibited enrichment in pathways such as porphyrin metabolism, fluorobenzoate degradation, and atrazine degradation. Swiss-Prot pathway analysis ( Figure 6C ) revealed downregulation of bacterial proteins such as P7351, O05515, Q8G509, and P55620 in the NoV-infected group, while proteins like A6KYJ6, Q9Z4V8, P0C939, P76071, and Q5LIM4 were upregulated. Finally, quantitative scoring of KEGG, COG, and NOG enrichment ( Figures 6D–F ) showed that the NoV-infected group had significantly higher KEGG TPM values but lower COG and NOG TPM values compared to the healthy group, indicating that NoV infection primarily affects KEGG pathways while reducing COG and NOG pathway enrichment.