Integrated fecal macrogenomic and metabolomic analyses reveal celiac disease flora and metabolic profiles associated with Chinese populations

Abstract

Background and aims

Genes and gluten intake are necessary but not sufficient to cause celiac disease (CeD), and altered intestinal flora is an additional risk factor for the development of CeD. The present study was conducted to investigate the intestinal flora and metabolic characteristics of CeD among the Chinese population, with the use of CeD patients from Xinjiang, China.

Methods

Macrogenomic sequencing was performed to analyze the composition and differences of the intestinal flora of 40 CeD patients and 40 healthy subjects. Non-targeted metabolomics analysis was carried out using LC–MS metabolomics technology in 30 CeD patients and 30 control subjects. A model for CeD diagnosis was constructed based on differential flora and metabolites. Blood was collected from all subjects for HLA typing assay.

Results

CeD-associated alterations were identified in the gut microbiome and metabolome. 15 differential bacterial strains (AUC = 0.85) and 8 differential metabolites (AUC = 0.9799) constructed a diagnostic panel that was effective in differentiating CeD patients from healthy subjects. Compared with non-CeD patients carrying HLA-DQ^±, the abundances of Agathobacter_rectalis, Bifidobacterium_pseudocatenulatum, Clostridia_bacterium, Coprococcus_comes, and Fusicatenibacter_saccharivorans in CeD patients were significantly lower (P < 0.05). Metabolomics analysis showed that Leoheteronin D, Pc (34:2), and GPEtn (18:1/16:0) were the major metabolites involved in multiple metabolic pathways in CeD patients.

Conclusion

Our study revealed specific alterations in the gut microbiome and metabolome of Chinese CeD patients through a multi-omics integration strategy. We found that CeD individuals carrying CeD risk genes may possess a unique intestinal flora composition, and this intestinal flora may, to some extent, explain the pathogenesis of CeD beyond the contributions of genes and gluten intake.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-06991-5.

Introduction

Celiac disease (CeD) is a unique intestinal autoimmune disorder caused by gluten ingestion in genetically susceptible individuals [1]. The prevalence of CeD in the general population ranges from 0.5 to 2%, with an average prevalence of approximately 1% [2, 3]. The pathogenesis of CeD is complex and is the result of a combination of genetic factors, environmental factors, and the immune system. Genetic factors are the main pathogenetic prerequisites for patients with CeD, and nearly all individuals with CeD carry the human leukocyte antigen (HLA)-DQ2/DQ8 genes [4]. Gluten is considered to be a major external trigger for CeD. Gluten is a protein composed of soluble gliadin and insoluble gluten components, which is rich in proline residues and glutamine residues; it is found mainly in wheat, but also in barley, rye, and oats [2, 5]. When gluten is ingested by individuals carrying genetic susceptibility genes, proline levels are elevated in the body, and gluten proteins resist the catabolic action of digestive proteases to produce proline and glutamine-rich peptides (33-and 13-mer gliadin peptides) [6]. These peptides cross the epithelial barrier of the digestive tract and enter the lamina propria of the intestine, where they are amidated by the enzyme transglutaminase 2 (TG2) in local tissues. The amidated proteins are recognized by the APC via HLA-DQ2/DQ8, which ultimately triggers a specific CD4 + T cell-mediated immune response. This series of immune responses leads to systemic inflammation and intestinal tissue damage [7–11]. A strict gluten-free diet is currently the only effective treatment for CeD [3].

To date, the pathogenesis of CeD is unclear. Although the HLA-DQ2/8 gene-mediated gluten-specific immune response has been well-studied, neither gluten intake nor individual genetic susceptibility adequately explains the pathogenesis of CeD. The carrier rate of HLA-DQ2/8 in most gluten-consuming populations is approximately 30–40%, but only about 1% of these individuals will develop CeD [12]. Therefore, there may be other factors that play a prominent role in triggering the manifestations of CeD and promoting the progression of CeD, such as the mode of infant delivery, infections, and alterations in the quantity and composition of the gut microbiome [13–15]. The role of infections in the development of CeD has been confirmed in previous observational studies. These studies point to the potential role of viral infections, especially adenovirus, enterovirus, rotavirus, and eutherovirus infections, in the pathogenesis of CeD. Viral infections are directly or indirectly involved in the development of CeD through inflammatory responses to dietary antigens [16–18]. The intestinal flora play an important role in host nutrient metabolism, the maintenance of intestinal mucosal barrier integrity, immunomodulation, and resistance to pathogens [19]. Earlier studies have identified a variety of gut microorganisms associated with the risk of developing CeD. For example, opportunistic pathogenic bacterial proteases (e.g., elastase), which are a type of bacterial protease, increase the antigenicity of gluten, thereby enhancing gluten-specific T-cell immune responses in CeD patients. In contrast, intestinal bacteria such as Lactobacillus contribute to the complete digestion of gluten peptides, decreasing their immunogenicity [20]. The essential amino acid tryptophan is difficult to metabolize by microorganisms in the duodenum of CeD patients, leading to impaired activation of aryl hydrocarbon receptors in the intestinal mucosa and subsequent inflammation [21]. In addition, microbiota-driven mechanisms include pathways independent of bacterial elastase, such as the cross-reactivity of T-cell receptors between gliadin and Pseudomonas aeruginosa peptides in HLA-DQ 2.5-positive individuals [22]. This suggests an important role for gut microbes in the development of CeD. Gut microbial changes usually lead to metabolite changes; previous studies have found differences in short-chain fatty acid metabolism and one-carbon metabolism between CeD patients and healthy individuals [23, 24]. Serine and glycine may be specific biomarkers for CeD [25]. However, the unique gut microbiome profile and metabolic profile of CeD have not yet been clarified, given the combination of factors such as the clinical trials involved, the sample source (duodenal and fecal), and the differences in sampling locations and study populations [26].

In the present study, we performed metagenomic sequencing and untargeted metabolomics analysis of fecal samples from patients diagnosed with CeD and their matched healthy controls in Xinjiang, China, aiming to characterize the altered intestinal flora and metabolic profiles specific to CeD. On this basis, we further analyzed the differences in intestinal flora between CeD patients carrying the CeD risk gene (HLA-DQ2^±) and healthy subjects. Based on the metagenomic profiling and metabolic profiling of CeD patients, we screened potential biomarkers for the diagnosis of CeD. In addition, we extracted viral group data from metagenomic sequencing data to assess whether the viruses detected in feces were associated with CeD and to determine the mechanism of viral infection in the development of CeD.

Materials and methods

Patient enrollment and sample collection

In this cohort study, we recruited 40 patients with CeD who were first diagnosed in the Department of Gastroenterology of the People’s Hospital of Xinjiang Uygur Autonomous Region, China, between June 2023 and August 2024, and 40 healthy volunteers matched for ethnicity, gender, and age (with an age difference of no more than 5 years). All included CeD patients were screened for serum tTG-IgA, and the diagnosis was confirmed following endoscopic biopsy of the small bowel in combination with histopathology (Marsh ≥ grade II). The Marsh scores were evaluated by specialized pathologists. All included CeD patients met the diagnostic criteria of the World Gastroenterology Organization Global Guidelines for Celiac Disease (2017) [27]. Patients with concomitant inflammatory bowel disease or other autoimmune diseases were excluded.

The controls included in this study were healthy volunteers who came to the People’s Hospital of the Autonomous Region for a physical examination. All control subjects were screened negative for serum anti-tTG (tTG < 20 IU/mL). Exclusion of participants who had used antibiotics or probiotics within one month prior to enrollment was an exclusion criterion for all participants in this study. For all CeD patients and healthy control subjects included in the study, each of them signed an informed consent form after being informed of the study details. Their blood samples and fecal samples were retained with their consent. The stool samples were collected using stool sample collection tubes and stored at − 80 °C until further processing.

The study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the Ethics Committee of the Xinjiang Uygur Autonomous Region People’s Hospital (Ethics No. KY2023060173). All participants were informed about the study, voluntarily participated in the study, and signed an informed consent form.

Patient genotyping of HLA-DQ2/DQ8

After obtaining the patient’s consent, take their whole blood for DNA extraction and HLA genotyping. The detailed methods and HLA typing results are shown in Supplementary Table 1.

DNA extraction, library construction, and metagenomic sequencing and analysis

0.2 g of stool material was used to extract total genomic DNA with the FastPure Stool DNA Isolation Kit (Magnetic bead) (MJYH, shanghai, China) according to manufacturer’s instructions. Concentration and purity of extracted DNA was determined with SynergyHTX and NanoDrop2000, respectively. DNA quality was checked on 1% agarose gel. DNA extract was fragmented to an average size of about 350 bp using Covaris M220 (Gene Company Limited, China) for paired-end library construction. Paired-end library was constructed using NEXTFLEX Rapid DNA-Seq (Bioo Scientific, Austin, TX, USA). Paired-end sequencing was performed on Illumina NovaSeq™ X Plus (Illumina Inc., San Diego, CA, USA) at Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China) using NovaSeq X Series 25B Reagent Kit according to the manufacturer’s instructions (www.illumina.com).

The software Fastp (Version:v0.20.0; https://github.com/OpenGene/fastp) was used to control the quality of the original data, the software BWA (Version:v0.7.17; http://bio-bwa.sourceforge.net) was used to align the reads to the host DNA sequence, and the contaminated reads with high similarity were removed. The quality-filtered data were assembled using MEGAHIT (Version:v1.1.2; https://github.com/voutcn/megahit). Contigs with a length ≥ 300 bp were selected as the final assembling result. Open reading frames (ORFs) from each assembled contigs were predicted using Prodigal (Version:v2.6.3; https://github.com/hyattpd/Prodigal) and a length ≥ 100 bp ORFs were retrieved.A non-redundant gene catalog was constructed using CD-HIT [28] (Version:v4.6.1; http://weizhongli-lab.org/cd-hit/) with 90% sequence identity and 90% coverage. Gene abundance for a certain sample was eatimated by SOAPaligner [29] (Version:soap2.21release; https://github.com/ShujiaHuang/SOAPaligne) with 95% identity.

The best-hit taxonomy of non-redundant genes was obtained by aligning them against the NCBI NR database by DIAMOND (Version:v2.0.13; http://ab.inf.uni-tuebingen.de/Software/diamond/) [30] with an e-value cutoff of 1 × 10^–5. Similarly, the functional annotation (GO, KEGG, COG) of non-redundant genes was obtained. Based on the taxonomic and functional annotation and the abundance profile of non-redundant genes, the differential analysis was carried out at each taxonomic, functional, or gene-wise level by Kruskal–Wallis test.

α-Diversity analysis including Chao, Shannon, and Simpson indexes was conducted and visualized using the vegan and fossil packages in R, respectively. PCoA was used to visually evaluate the overall difference and similarity of bacterial and viral communities between the CeD and HC groups. The abundance calculation method between samples is mainly based on RPKM (Relative Path LengthMetric). The comparison between groups was conducted using Wilcoxon rank sum test, and after FDR multiple test correction, P value < 0.05, Consider the difference significant.

Metabolite extraction and UHPLC-MS/MS analysis

50 mg stool sample was added to a 2 mL centrifuge tube and a 6 mm diameter grinding bead was added. 400μL of extraction solution (methanol: water = 4:1 (v:v)) containing 0.02 mg/mL of internal standard (L-2-chlorophenylalanine) was used for metabolite extraction. Samples were ground by the Wonbio-96c (Shanghai wanbo biotechnology co., LTD) frozen tissue grinder for 6 min (− 10C, 50 Hz), followed by low-temperature ultrasonic extraction for 30 min (5C, 40 kHz). The samples were left at − 20 °C for 30 min, centrifuged for 15 min (4C, 13000 g), and the supernatant was transferred to the injection vial for LC–MS/MS analysis. The LC–MS/MS analysis of sample was conducted on a Thermo UHPLC-Q Exactive HF-X system equipped with an ACQUITY HSS T3 column (100 mm × 2.1 mm i.d., 1.8 μm; Waters, USA) at Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China).

The pretreatment of LC/MS raw data was performed by Progenesis QI (Waters Corporation, Milford, USA) software, and a three-dimensional data matrix in CSV format was exported. The information in this three-dimensional matrix included: sample information, metabolite name and mass spectral response intensity. Internal standard peaks, as well as any known false positive peaks (including noise, column bleed, and derivatized reagent peaks), were removed from the data matrix, deredundant and peak pooled. At the same time, the metabolites were identified by searching database, and the main databases were the HMDB (http://www.hmdb.ca/), Metlin (https://metlin.scripps.edu/) and Majorbio Database.The data matrix obtained by searching database was uploaded to the Majorbio cloud platform (https://cloud.majorbio.com) for data analysis [31].

Then, the R package “ropls” (Version 1.6.2) was used to perform principal component analysis (PCA) and orthogonal least partial squares discriminant analysis (OPLS-DA), and 7-cycle interactive validation evaluating the stability of the model. The metabolites with VIP > 1, p < 0.05 were determined as significantly different metabolites based on the Variable importance in the projeciton (VIP) obtained by the OPLS-DA model and the p-value generated by student’s t test. Differential metabolites among two groups were mapped into their biochemical pathways through metabolic enrichment and pathway analysis based on KEGG database (http://www.genome.jp/kegg/). These metabolites could be classified according to the pathways they involved or the functions they performed.

Statistical analysis

The clinical profile data were statistically analyzed using IBM SPSS Statistics 26.0 software (IBM Corporation, Armonk, NY, USA) and visualized using GraphPad Prism 8.0 software (La Jolla, CA, USA). Each value was expressed as the mean ± standard error of the mean (SEM). If the data conformed to the assumptions of normal variance distribution and passed the chi-square test, one-way analysis of variance (One-way ANOVA) followed by least significant difference (LSD) t-tests were used to analyze the differences between groups. If the data did not conform to the above assumptions, the Wilcoxon rank-sum test was used. A p-value less than 0.05 was considered to indicate a statistically significant difference. One-way or two-way analysis of variance (ANOVA) with Duncan’s multiple range test at a significance level of p < 0.05 was used to assess component differences. For pairwise comparisons between two groups, the Wilcoxon rank-sum test was employed, followed by the false discovery rate (FDR) correction for multiple testing.

Results

Design of experiment

In order to analyze the gut microbiome composition and metabolic profiles of CeD patients, in this study, we collected fecal samples from 80 individuals in a cross-sectional cohort and analyzed them for their microbiota taxonomic composition. Subsequently, we conducted liquid chromatography-mass spectrometry (LC–MS) metabolomics analysis on 60 of these samples, with 30 samples from CeD patients and 30 from healthy controls (HCs). We conducted correlation analyses of metagenomic and metabolomic data to identify disease-specific changes in the microbial and metabolic profiles of individual patients. We explored the relationships between disease-enriched microbiota and metabolites, and investigated the potential diagnostic value of these specific microbial species and metabolites for CeD.

General information and clinical characteristics of the study cohort

A total of 40 CeD patients and 40 healthy volunteers were enrolled in this study, and their clinical characteristics are presented in Table 1. The mean age of the enrolled CeD patients was 44 ± 12.0 years. Among them, there were 29 females (72.5%) and 11 males (27.5%), and the male-to-female ratio was approximately 1:3.79. There was no significant difference in BMI values between the two groups (P = 0.691). The ethnic composition of the 40 CeD patients included 6 Han Chinese (15%), 15 Kazakhs (37.5%), and 19 Uyghurs (47.5%). Thyroid disease, osteoporosis, anemia, and vitamin D deficiency were the most common comorbidities among CeD patients. Specifically, 12 out of 40 CeD patients had comorbid thyroid disease, 12 had comorbid osteoporosis, and 14 had comorbid anemia. The incidence of these comorbidities was significantly lower in the included healthy control subjects than in the CeD group (P < 0.05). During sample recruitment, the control group was matched to the CeD group for ethnicity, gender, and age (with an age difference of ± 5 years), resulting in no significant differences in ethnicity, gender, or age between the two groups.

Table 1.

Clinical parameters of the two groups enrolled in the study

Parameters	Cases (n = 40) CeD patients	Controls (n = 40) Healthy volunteers	Statistical value	P value
Min–max age	19–65	19–66
Mean age ± SD	44 ± 12.0	42.6 ± 11.9	t = 0.513	0.994
Sex F/M (%F)	29F/11 M (72.5%)	29F/11 M (72.5%)	–	–
BMI	20.16 ± 4.3	20.33 ± 3.93	t = − 0.185	0.691
Nationality (%)			–	–
Han	6/40 (15%)	6/40 (15%)
Kazakh	15/40 (37.5%)	15/40 (37.5%)
Uygur	19/40 (47.5%)	19/40 (47.5%)
Disease comorbidities (%)
Thyroid diseases	12/40 (30%)	4/40 (7.5%)	χ2 = 5.000	0.025
Osteoporosis	12/40 (30%)	4/40 (10%)	χ2 = 5.000	0.025
Anemic	14/40 (35%)	2/40 (5%)	χ2 = 11.250	0.001
Vitamin D deficiency	17/40 (42.5%)	7/40 (22.5%)	χ2 = 5.952	0.015
Anxiety and depression	2/40 (5%)	0/40 (0%)	χ2 = 2.051	0.152
Mean tTG-IgA (% +)	2596.2 40/40 (+) (100%)	0/40 (−) (0%)
Duodenal biopsy  (Marsh index) (%)	5/40 M3a (12.5%) 16/40 M3b (12.5%) 19/40 M3c (47.5%)	40/40 M0 (100%)
Clinical symptoms (%)  (Classical/non-classical)	18/40 Cla (45%) 22/40 N-Cla (55%)	–

CeD celiac disease, BMI body mass index, F female, M male, tTG -IgA anti tissue-transglutaminase 2 IgA, Marsh Index M3a: Marsh 3A, M3b: Marsh 3B, M3c: Marsh 3C

Gut microbiota of patients with CeD is different from that of the health controls

After quality control, the filtering process removed Homo sapiens sequences and low-quality reads, resulting in more than 5 million clean reads per sample. The applied sequencing depth was sufficient to fully cover the biodiversity of the samples. Alpha diversity reflects the abundance and evenness of microbial species within a population, and it encompasses the Shannon index, Chao index, and Simpson index. The results of the Shannon index showed a significant decrease in biodiversity and abundance in the CeD group compared to the control group (P < 0.001) (Fig. 1A). The Chao and Simpson indices also indicated significant differences (P < 0.05) in diversity and richness between the two groups (Fig. 1B, C). Beta diversity reflects the differences in community structure and microbial species composition between the two sample groups. Principal Coordinate Analysis (PCoA) based on the Bray–Curtis distance algorithm showed that patients with CeD were distinctly separated from the control group, with significant differences in distance and degree of dispersion (P = 0.001). This indicated that the microbial compositions of the two groups differed significantly from each other (Fig. 1D). Additionally, there was some overlap in the microbial species composition between the two groups, yet the degree of dispersion was greater in the CeD group. The results indicated that the diversity and abundance of intestinal flora in CeD patients were lower than those in healthy controls, and that there was a deviation in the composition of intestinal microorganisms in CeD patients, accompanied by a significant increase in the heterogeneity of intestinal microorganisms among these patients.

Fig. 1.

Metagenomic features of celiac disease (A) Shannon index (P < 0.0001). (B) Chao index (P = 0.0136). (C) Simpson index. (P = 0.0002) (α-diversity). (D) Principal coordinate analysis (PCoA) of Bray–Curtis analysis demonstrated that individuals with CeD patients were significantly different from HCs. Component proportions of bacterial phylum (E), Genus (F) in each group. (G) Top 10 differential bacterial strains at the species level between CeD patients and HCs

Through gene annotation, we identified a total of 20,729 taxa (5 domains, 15 phyla, 169 classes, 308 orders, 531 families, 1004 genera, 3086 species, and 15,611 strains) between the two groups. The major phyla composition of the CeD and control groups was as follows: Bacillota (Firmicutes) (50.82% vs. 67.11%), Bacteroidota (Bacteroidetes) (22.85% vs. 12.84%), Pseudomonadota (Proteobacteria) (11.81% vs. 2.80%), Actinomycetota (6.41% vs. 8.51%), Uroviricota (1.72% vs. 2.63%), Verrucomicrobiota (2.71% vs. 1.33%) (Fig. 1E). Compared with healthy controls, the ratio of Firmicutes/Bacteroidetes(F/B) was decreased in the CeD group. A change (either an increase or a decrease) in the F/B ratio is generally considered a marker of intestinal dysbiosis [32]. These findings suggest that patients in the CeD group may experience intestinal flora dysbiosis.

To analyze the differences in microbiota composition between CeD patients and the control group, we analyzed the microbiota composition at different taxonomic levels (phylum, order, family, genus, and species) using the Wilcoxon rank-sum test. At the genus level, a total of 3086 genera were identified. Among them, 2814 genera were identified in the CeD group and 2612 genera in the control group. Of these, 474 genera were unique to the control group and 272 genera were unique to the CeD group (Fig. 1F). Through differential analysis, a total of 332 genera showed significant differences between the two groups (P < 0.05, adjusted P-value < 0.05) (Supplementary Table 2). At the species level, a total of 15,611 species were identified. Among them, 3635 species showed significant differences between the two groups (P < 0.05), and 1555 species remained significantly different after multiple testing correction (adjusted P-value < 0.05). Among these species, 28 species were absent in CeD patients and 5 species were more abundant in CeD patients (Supplementary Table 3). Based on species abundance, the top 10 differential species were screened, which were Escherichia coli, Faecalibacterium prausnitzii, Bacteriophage sp., Caudoviricetes sp., Agathobacter rectalis, Oscillospiraceae bacterium, Clostridiales bacterium, Blautia wexlerae, Anaerobutyricum hallii, and Lachnospiraceae bacterium. Among them, except for Escherichia coli, which was significantly up-regulated in the CeD group, the remaining species were significantly down-regulated (P < 0.05) in the CeD group (Fig. 1G).

Construction of a diagnostic model based on CeD macrogenomics

We constructed gene sets representing significant differences between the two groups at the species level. The screening criteria for significantly different strains were set as Corrected P-value < 0.05. After screening, the number of significantly different strains was 1555. Random forest models were constructed based on the gene sets of differentially abundant strains. The AUC-random forests algorithm was employed to identify the optimal random forest model that maximized the AUC value of the ROC curve (Fig. 2A). After feature selection with tenfold cross-validation, we finally identified 15 differentially abundant bacterial species, including Roseburia_intestinalis, Ruminococcus_sp, Bacilli_bacterium and Gemmiger_sp, for the construction of the CeD diagnostic model. All of these species belong to the phylum Firmicutes (Supplementary Table 3). Through principal component analysis (PCA), we further investigated whether the 15 differentially abundant strains could discriminate between CeD patients and healthy controls. The results showed that the 15 differentially abundant strains could effectively distinguish CeD patients from healthy controls (Fig. 2B). Based on receiver operating characteristic (ROC) analysis, the diagnostic performance of the 15 differentially abundant species for CeD was characterized by an AUC of 0.85 (95% CI 0.76–0.93), indicating that the combination of these 15 differentially abundant strains has good diagnostic performance for CeD (Fig. 2C). These results confirm the high potential of classifiers based on these species as promising non-invasive tools for fecal-based detection in CeD patient populations.

Fig. 2.

Prediction model and functional analysis based on differential bacterial strains (A) Random forest model based on species level differences in bacterial strains. (B) Principal Component Analysis (PCA) of 15 different bacterial species. (C) The best predictive model with the highest AUC value (AUC = 0.85) was constructed. Functional classification of 15 different bacterial strains by the (D) Clusters of Orthologous Groups of proteins (COG), and (E) Kyoto Encyclopedia of Genes and Genomes (KEGG)

Different microbiota have different molecular functions and are involved in different biological processes

To further analyze the molecular functions of the 15 differentially abundant species, we performed functional annotation of the homologous protein groups using the Clusters of Orthologous Groups (COG), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Gene Ontology (GO) databases. COG analysis showed that the differentially abundant species were mainly associated with functions such as DNA-binding response regulator, transcriptional regulator containing XRE-family helix-turn-helix (HTH) structural domain, site-specific tyrosine recombinase XerD, signal transduction histidine kinase, and glycosyltransferase involved in cell wall biosynthesis (Fig. 2D).

KEGG analysis showed that the enrichment of differentially abundant microbes was mainly related to the biosynthesis of secondary metabolites, amino acid biosynthesis, ABC transporters, and cofactor biosynthesis (Fig. 2E).

GO analysis showed that the biological processes of the 15 differentially abundant species were mainly involved in transmembrane transport, response to other organisms, cell communication, and secondary metabolic processes (Fig. S1) (Supplementary Table 4).

Viral composition and differential viruses between CeD patients and healthy controls

To further investigate the mechanism by which viral infection influences CeD, we extracted viral genomic sequences within the viral domain. Among the genomic sequences detected in this metagenome sequencing, the proportion of sequences related to the viral domain was 3.392%. Based on the metagenome sequencing results and annotations from the NR database, within the viral domain, we detected a total of 8 viral kingdoms, 9 viral phyla, 10 viral orders, 27 viral families, 161 viral genera, and 1,956 viral species (Supplementary Table 5). Through the analysis of the differences in viral composition between CeD patients and healthy controls, both PCA and PCoA revealed significant differences in viral composition between the two groups (P < 0.05) (Fig. 3A, B). At the phylum level, there were three viral phyla that showed significant differences between the two groups. Among them, the most prominent viral phylum with a significant difference was Uroviricota, which had a significantly higher abundance in the CeD group compared to the healthy control group (CeD: 69.77%, Con: 62.46%; P = 0.020) (Fig. 3C). At the genus level, a total of 11 virus genera showed significant differences. The top three genera with significant differences in terms of species abundance were Junduvirus (CeD: 0.073%, Con: 0.295%), Traversvirus (CeD: 0.296%, Con: 0.004%), and Lederbergvirus (CeD: 0.254%, Con: 0.009%) (Fig. 3D, E).

Fig. 3.

Viral analysis between Celiac disease patients and HCs (A) Principal component analysis shows differences in viral composition between CeD patients and HCs. (B) Principal coordinate analysis (PCoA). Component proportions of virus phylum (C), Genus (D) in each group. (E) Differences in viruses between CeD patients and HCs at the genus level

Differences in gut microbiota under different genetic risks of CeD

It has been demonstrated that the genotype of infants with a familial risk of CeD who carry the HLA-DQ2 haplotype influences the composition of the early gut microbiota, implying that individuals carrying the CeD risk gene possess a distinctive gut microbiota [33]. We further analyzed the differences in gut microbiota between CeD patients with a genetic risk and non-CeD patients with a genetic risk to explore the characteristic gut microbiota of CeD within the same genetic background. In the present study, we examined the HLA-DQ genotypes of all patients enrolled in the study, and the results are presented in Supplementary Table 1. We defined patients carrying the HLA-DQ2^+/+ or DQ2^+/- genotypes as individuals at a higher risk of developing CeD. According to this criterion, we categorized the patients enrolled in this study into four groups: CeD patients (32, CeD-HLA1) and non-CeD patients (11, NCeD-HLA1) who carried the CeD risk gene, CeD patients (8, CeD-HLA2) and non-CeD patients (29, NCeD-HLA2) who did not carry the CeD risk gene. Alpha-diversity and beta-diversity revealed differences in the composition and abundance of microbial communities among the four groups (Fig. 4A, B). Through the analysis of differences in microbial communities, we found that the abundances of multiple microbial species differed among the four groups (Fig. S2).

Fig. 4.

Microbial differences between CeD patients carrying HLA-DQ2^+/- and HCs (A) Chao index (P = 0.2155) (α-diversity). (B) Principal Component Analysis (β-diversity). (C) Component proportions of bacterial genus. (D) The branching diagram generated by LEfSe analysis shows the dominant microbial communities between CeD patients and HCs. (E) Differences in bacterial between CeD patients and HCs at the species level

We analyzed the differences in gut microbiota between CeD patients carrying risk genes and healthy subjects carrying risk genes. At the phylum level, the abundances of Bacillota (49.6% vs. 61.1%), Bacteroidota (22.7% vs. 13.2%), and Pseudomonadota (13.9% vs. 4.3%) were lower in the CeD-HLA1 group compared to the NCeD-HLA1 group, while the abundance of Actinomycetota (5.5% vs. 11.4%) was higher (Fig. 4C). At the species level, the abundances of Agathobacter rectalis, Bifidobacterium pseudocatenulatum, Clostridia bacterium, Coprococcus comes, and Fusicatenibacter saccharivorans were significantly lower in the CeD-HLA1 group (P < 0.05) (Fig. 4E) (Supplementary Table 6). Linear discriminant analysis effect size (LEfSe) showed that the dominant bacterial groups in the CeD-HLA1 group included c__Gammaproteobacteria, o__Enterobacterales, f__Enterobacteriaceae, g__Escherichia, and s__Escherichia_coli, among others (Fig. 4D).

Metabolic differences in fecal samples between CeD and healthy controls

Alterations in the gut microbiota are indicative of changes in fecal metabolites, and a correlation exists between these two aspects. A total of 5107 metabolites were identified through metabolite profiling of fecal samples (Fig. 5A). Based on the partial least squares discriminant analysis (PLS-DA) scores of metabolites between the two groups, the results indicated that the clustering of samples from the two groups was satisfactory and suitable for subsequent analysis (Fig. 5B). Through principal component analysis (PCA) of the metabolites between the two groups (Fig. 5C), a significant difference was observed in the overall metabolic profiles of CeD patients and healthy controls (ANOSIM test, P = 1 × 10^–3), indicating that fecal metabolites were modified in CeD patients.

Fig. 5.

Fecal metabolomics for quantification of metabolites in both the HCs and CeD patients (A) The number of metabolites between two groups. (B) Partial least squares discriminant analysis (PLS-DA) of CeD patients and HCs showed significant classification performance between the two groups. (C) Principal component analysis (PCA) of differential metabolites between two groups. (D) Volcano plot of untargeted metabolomic data. Fold change indicates the changes in metabolites in the CeD group compared with those in the HCs group. (E) Pie chart: HMDB compound classification corresponding to the two groups of differential metabolites

Orthogonal projections to latent structures discriminant analysis (OPLS-DA) revealed differences in fecal metabolic profiles between the two groups, and permutation testing demonstrated no evidence of overfitting and validated the OPLS-DA model (Fig. S3). A total of 1,476 differential metabolites were identified between the two groups based on differential analysis, including 716 positive ion metabolites (ES+) and 760 negative ion metabolites (ES−) (P-value < 0.05, VIP_pred_OPLS-DA > 1) (Supplementary Table 7). Analysis of 1476 differential metabolites revealed that 1359 metabolites were down-regulated and 117 metabolites were up-regulated in the CeD group compared to healthy controls (Fig. 5D). Annotation of the 1476 differentially abundant metabolites using the Human Metabolome Database (HMDB) version 4.0 was carried out. The metabolites were mainly categorized into Lipids and lipid-like molecules (n = 439), Organic acids and derivatives (n = 119), Organic heterocyclic compounds (n = 116), Benzenoids (n = 77), Phenylpropanoids and polyketides (n = 65), Organoheterocyclic compounds (n = 62), Organic oxygen compounds (n = 62) and other markers (Fig. 5E) (Supplementary Table 8).

Annotation of the 1476 differentially abundant metabolites between the two groups using the KEGG database revealed that these differentially abundant metabolites were mainly enriched in functional pathways such as the Digestive and Nervous system, Lipid metabolism, Biosynthesis of other secondary metabolite, Amino acid metabolism, Cancer: overview, Infectious disease: bacterial, Signal transduction (Environmental Information Processing) (Fig. 6A). Steroid hormone biosynthesis (P adjust = 2.38 × 10^–7), Arachidonic acid metabolism (P adjust = 1.69 × 10^–6), Primary bile acid biosynthesis (P adjust = 5.3 × 10^–4), Retinol metabolism (P adjust = 2.3 × 10^–3), Neuroactive ligand-receptor interactions (P adjust = 2.5 × 10^–4) were among the significantly enriched signaling pathways (Fig. 6B) (Supplementary Table 9).

Fig. 6.

Functional pathways of differential metabolites and construction of diagnostic models based on differential metabolites (A) KEGG functional enrichment analysis based on differential metabolites. (B) Enrichment analysis of KEGG pathway based on differential metabolites. (C) AUC values of 8 differential metabolites in distinguishing between CeD patients and healthy subjects.Tyr-Gly:AUC = 0.9011; (E)-3,10-Dihydroxy-4,9-dimethyldodec-6-enedioic acid: AUC = 0.9211; N-(2-hydroxy-3-(1-piperidinyl)propoxy)-3-Pyridinecarboximidamide: AUC = 0.9022; 6-gingerol: AUC = 0.9156; Solasodiene: AUC = 0.9044; GPEtn(18:1/16:0): AUC = 0.9067; Pc(34:2): AUC = 0.9267; Leoheteronin D: AUC = 0.9289). (D) RF method was used with train function of R’s caret package. Combined ROC analysis showed higher levels of AUC = 0.9799

Construction of CeD diagnostic model based on differential metabolites

Differences in metabolites between CeD patients and HCs prompted us to investigate whether gut flora has the potential to differentiate these two groups. To construct a diagnostic model for distinguishing CeD patients from HCs using a random forest approach, we included the top 50 differentially abundant metabolites in a disease classifier and refined the features via Recursive Feature Elimination with Cross-Validation (RFECV). This process ultimately identified 8 differential metabolites, including Pc (34:2), GPEtn (18:1/16:0), Leoheteronin D and Solasodiene. All 8 metabolites exhibited good diagnostic efficacy in distinguishing CeD patients from HCs (Fig. 6C), among which Pc(34:2) (AUC = 0.9267, 95% CI 0.8653, 0.988), GPEtn(18:1/16:0) (AUC = 0.9067, 95% CI 0.8271, 0.9863), and 6-gingerol (AUC = 0.9156, 95% CI 0.8388, 0.9923) were significantly enriched in CeD patients. We further developed a diagnostic model based on these eight differential metabolites. Results showed that the combined ROC analysis of the eight metabolites yielded excellent diagnostic efficacy (AUC = 0.9799, 95% CI 0.9737–0.9861), thereby exhibiting strong discriminatory ability (Fig. 6D). These findings suggest that the metabolite-based classification model could serve as a diagnostic tool for CeD.

Correlation of gut microbes and fecal metabolites in CeD

Using Platts analysis, we examined the concordance between alterations in the microbiota and changes in metabolite profiles. The results demonstrated that changes in the abundance of the microbiome were consistent with changes in metabolite expression levels in both the CeD group and the control group (M2 = 0.674, P = 0.001) (Fig. 7A). We further investigated the correlation between the differentially abundant flora and differentially abundant metabolites in the two groups (Fig. 7B).The results indicated that most of the differentially abundant flora were negatively correlated with Solasodiene, GPEtn(18:1/16:0), Pc(34:2), and 6-gingerol. They were positively correlated with Tyr-Gly and Leoheteronin D (P < 0.05). Specifically, Gemmiger_sp was significantly negatively correlated with GPEtn(18:1/16:0) and significantly positively correlated with (E)-3,10-Dihydroxy-4,9-dimethyldodec-6-enedioic acid and N-(2-hydroxy-3-(1-piperidinyl)propoxy)-3-Pyridinecarboximidamide.Ruminococcus_sp was significantly negatively correlated with Solasodiene. Bacilli bacterium was significantly positively correlated with Tyr-Gly(P < 0.05).

Fig. 7.

Correlation analysis of gut microbiota characteristics and metabolic characteristics, and construction of a microbial metabolic joint diagnostic model. (A) Procrustes analysis showed consistency between metabolite changes and microbial changes (P = 0.001). (B) Spearman correlation heatmap between 15 differential microorganisms and 8 differential metabolites. (C) Ranking 15 differential bacterial species and 8 differential metabolites based on the average descent accuracy of the random forest model. (D) Based on the Random Forest model. The ROC analysis showed AUC = 0.644 for differential strains, AUC = 0.894 for differential metabolites, and the combined ROC analysis showed a higher level of AUC = 0.9647

Construction of a combined diagnostic model based on fecal microorganisms and metabolites in CeD

We further investigated whether the diagnostic performance for CeD could be enhanced by a diagnostic panel consisting of fecal microbiota and metabolites. We merged differential strains with differential metabolites and constructed a random forest model based on 15 differential strains (AUC = 0.85) and 8 differential metabolites (AUC = 0.9799) after screening. The 15 differential strains and 8 differential metabolites merged were modeled separately by Random Forest analysis, and the features were calculated using RFECV to obtain the best model for each of the strains, metabolites, and strain metabolite merged. Ranking based on mean descent accuracy of random forest models, we found that three metabolites, 6-gingerol, Pc (34:2), and GPEtn (18:1/16:0), had higher Mean Decrease Accuracy in distinguishing CeD from healthy populations (Fig. 7C), suggesting that they may be the key features of CeD. By integrating the microbial and metabolic features, we developed a composite diagnostic model with an AUC of 0.939, indicating that the model has good diagnostic utility for CeD (Fig. 7D). However, its diagnostic efficacy was lower than that of the diagnostic model constructed using the differentially abundant metabolites. The main reason for this might be that the inclusion of microbiome-related factors could have affected the overall diagnostic efficacy. The integration of metabolite and microbiota data into a single model significantly increases the complexity of the model, which in turn leads to a reduction in diagnostic performance [34].

Discussion

The gut microecosystem, which is composed of bacteria, archaea, protists, fungi, and viruses, influences multiple aspects of development, digestion, and immunity in mammals [13]. CeD, a typical autoimmune intestinal disease, is closely associated with alterations in the gut microecology, specifically dysbiosis of the intestinal flora, during its development [13]. However, owing to differences in the geographical dietary environment, genetic background, sampling sites, and sampling methods, the gut microbial characteristics of CeD vary to some extent across different studies [26]. In the present study, we uncovered extensive alterations in the gut microbiome and serum metabolome of CeD patients from the Xinjiang cohort. Through fecal metagenomic analysis and metabolic profiling, we screened and obtained differentially abundant flora and metabolites between CeD patients and healthy controls. Based on metagenomic sequencing, we identified differences in the viral composition between CeD patients and healthy subjects.In addition, our study also identified differences in the intestinal flora between CeD patients carrying the CeD risk gene and healthy subjects. Specifically, Escherichia coli may be associated with the CeD risk gene HLA-DQ^±. Based on the differentially abundant microorganisms and metabolites, we developed a set of CeD diagnostic panels with good diagnostic performance.

Based on metagenomic sequencing, we identified significant differences in the gut microbial composition between CeD patients and healthy subjects. Both alpha-diversity and beta-diversity analyses indicated significant differences in microbial composition and abundance between the two groups (P < 0.05). The diversity and abundance of the intestinal flora in CeD patients were lower than those in healthy controls, consistent with previous reports [35]. In the present study, a total of 169 phyla were identified in the samples from the two groups, of which 45 showed significant differences (P < 0.05). The most predominant phyla in both groups were Bacillota, Bacteroidota, Pseudomonadota, and Actinomycetota. Among these phyla, the relative abundances of Firmicutes and Actinobacteria were significantly downregulated in the CeD group, while the relative abundances of Proteobacteria and Uroviricota (1.72% vs. 2.63% in the virus domain) were significantly upregulated. Previous studies have also reported that fecal levels of Firmicutes and Actinobacteria were lower in active CeD patients than in healthy controls, and this finding may be related to gluten metabolism in individuals [36]. In addition, the Firmicutes/Bacteroidetes (F/B) ratio was lower in the CeD group compared to healthy controls. Changes in the F/B ratio, whether an increase or decrease, are typically regarded as markers of intestinal dysbiosis [37]. These results suggest that dysbiosis of the intestinal flora exists in CeD patients, which is consistent with previous research findings. Intestinal flora dysbiosis is evident in CeD patients, as demonstrated by alterations in the abundance of specific flora and a reduction in microbial diversity [38–40].

There were significant differences in the flora composition between CeD patients and healthy controls at the phylum, order, family, genus, and species levels. At the genus level, a total of 3086 genera were identified in the samples from the two groups, and 332 of these genera showed significant differences (Corrected P-value < 0.05). At the species level, a total of 15,611 species were identified, and 1555 of these species showed significant differences after multiple corrections (Corrected P-value < 0.05). Based on species abundance, we found that the species with the most significant difference between the two groups was Escherichia coli. Escherichia coli was significantly upregulated in the CeD group. In a recent study, Escherichia coli was found to be significantly more abundant in the duodenum at site D1 in CeD patients, but not in the feces or other sites of the duodenum [26]. In the present study, we found that Escherichia coli was similarly more abundant in the fecal samples of CeD patients. Indeed, Escherichia coli strains containing virulence genes are associated with CeD, and these strains may contribute to the development of CeD by exacerbating the intestinal inflammatory response [41, 42]. For instance, Adherent-invasive E. coli (AIEC) has been reported to be associated with inflammatory bowel disease (IBD), and IBD shares certain immunological features with CeD. AIEC can invade intestinal epithelial cells and induce the release of inflammatory factors, thereby exacerbating intestinal damage [43]. Notably, E. coli strains isolated from the fecal samples of patients with active CeD exhibited higher invasion coefficients than those isolated from healthy subjects and patients with inactive CeD [44]. These findings suggest that Escherichia coli may be one of the major causative agents of CeD.

We constructed a random forest model based on the species-level differentially abundant bacterial gene sets and found that 15 bacterial species, including Agathobacter_rectalis, Blautia_wexlerae, Roseburia_intestinalis, Ruminococcus sp., Bacilli_bacterium, Clostridioides_difficile, and Gemmiger sp., could effectively distinguish CeD patients from healthy volunteers (Supplementary Table 1). All of these bacteria were downregulated in CeD patients compared with healthy volunteers. Agathobacter_rectalis and Roseburia_intestinalis are both butyrate-producing bacteria [45, 46]. Butyrate, a short-chain fatty acid, promotes the function of the intestinal epithelial barrier and plays a crucial role in maintaining the intestinal mucosal barrier and regulating intestinal immunity [47]. Agathobacter_rectalis and Roseburia _intestinalis may play a protective role in CeD patients, and the reduction in their abundance may be associated with disease development. Blautia_wexlerae belongs to the genus Blautia, a beneficial intestinal bacterium that produces short-chain fatty acids, such as acetic acid [48]. It has been determined to be negatively associated with the development of obesity and type 2 diabetes [49]. Oral administration of Blautia_wexlerae was found to induce metabolic changes and anti-inflammatory effects in mice, thus reducing high-fat diet-induced obesity and diabetes. Blautia_wexlerae has been less well-studied in CeD. Some studies have reported significant increases in the abundance of Blautia_wexlerae in CeD patients after a gluten-free diet [50]. It is hypothesized that the increased abundance of this strain may contribute to the symptomatic relief of CeD through its anti-inflammatory effects. However, it has also been proposed that the genus Blautia is a risk factor for CeD [47]. Ruminococcus sp. and Gemmiger sp. are common species in the gut. Previous studies have also found that the abundances of Ruminococcus and Gemmiger are downregulated in CeD patients [51]. Ruminococcus and Gemmiger are considered beneficial bacteria in most studies, as they contribute to the balance of the intestinal microecology and inhibit the growth of harmful bacteria [52]. Bacilli_bacterium and Clostridioides_difficile both belong to Firmicutes. The genus Bacilli has been implicated as a protective factor against CeD in some studies [47]. A study by De Angelis et al. found that Bacilli can improve gluten digestion in gluten-sensitive patients by hydrolyzing immunogenic peptides during gastrointestinal digestion [53]. A study reported that Bacillus subtilis LZU-GM attenuated the adverse effects of gluten-containing foods and balanced the gut microbiota in mice [54]. In addition, Bacillus_cereus strains, also belonging to the genus Bacilli, have been reported to degrade gluten and CeD immunotoxic peptides [55]. Therefore, the genus Bacilli may have beneficial effects in enhancing gluten digestion and balancing the intestinal microbiota, thereby attenuating the adverse effects associated with gluten consumption and alleviating the symptoms of CeD. Clostridioides_difficile has been associated with a variety of intestinal disorders, such as IBD, IBS, and CeD [56]. One study reported that CeD patients were at a higher risk of contracting Clostridioides_difficile (P < 0.0001), which may be attributed to nutrient malabsorption and the immune characteristics of the disease [57]. In addition, we identified several viruses associated with CeD, including three major viruses: Junduvirus, Traversvirus, and Lederbergvirus. This is a novel finding; however, due to the scarcity of reports on these viruses, we will not delve into their specific roles in CeD in this study. Viral infections play a crucial role in disease progression. Similar to bacteria, viruses are implicated in the pathophysiology of numerous diseases, including CeD [58]. In the TEDDY cohort study, the cumulative number of fecal enterovirus exposures from 1 to 2 years of age was associated with an elevated risk of CeD autoimmunity [17]. Adenoviruses, enteroviruses, rotaviruses, echoviruses, and Epstei-Barr virus (EBV) infections have all been reported to potentially contribute to the pathogenesis of CeD [13, 59].

We further investigated the differences in the composition of the intestinal flora between CeD patients carrying the HLA-DQ^± risk gene and non-CeD patients. To some extent, we minimized the influence of genetic factors on CeD to investigate the central role of the intestinal flora in the pathogenesis of CeD. Our results showed that the abundances of Agathobacter_rectalis, Bifidobacterium_pseudocatenulatum, Clostridia_bacterium, Coprococcus_comes and Fusicatenibacter_saccharivorans were significantly lower in the CeD-HLA1 group compared to the NCeD-HLA1 group (P < 0.05). The genus Bifidobacterium is a common probiotic genus [60]. A case–control study has found that there are fewer lactobacilli and bifidobacteria in CeD children compared to healthy controls. Some bacterial species within the genus Bifidobacterium confer a protective effect on epithelial cells against damage induced by maltolysin [61]. In addition, Escherichia coli is the dominant bacterial group in the CeD-HLA1 group. We found that the abundance of Escherichia coli was higher in individuals harboring the HLA-DQ^± genotype than in those without it, regardless of whether they had CeD or not, suggesting an important role for this bacterium. Collectively, our findings suggest that alterations in the intestinal flora may be one of the primary factors contributing to the development of CeD in some patients who do not carry the typical risk genes for CeD pathogenesis. These alterations may contribute to CeD development by disrupting the function of the intestinal mucosal barrier [62].

Through functional enrichment analysis, we found that the functions of the differentially abundant species were mainly enriched in the pathways of amino acid biosynthesis, ABC transporter-related pathways and cofactor biosynthesis. Various amino acids play crucial roles in the pathogenesis of CeD. A case–control analysis of the levels of 23 amino acids in children with CeD showed that the levels of 7 amino acids (tryptophan, taurine, glutamic acid, proline, ornithine, alanine, and methionine) were elevated [63]. Glutamate and proline are typically the products of gluten digestion and are closely associated with the progression of CeD [64]. Tryptophan may contribute to the pathogenesis of CeD by modulating the immune system [65]. ABC transporter proteins play vital roles in cellular physiological functions, including nutrient absorption and metabolic waste elimination [66]. The essence of CeD pathogenesis is a CD4⁺-mediated immune response triggered by the antigenic processing of peptide epitopes presented by MHC class I molecules and the presentation of toxic gluten peptides by MHC class II molecules. In this process, various transporter proteins, such as CFTR, PSF1 and PSF2, play crucial roles [67, 68]. Previous studies have also reported impaired solute transporter proteins and water channel proteins in CeD patients [69]. These results suggest that the ABC transporter protein family plays a crucial role in the pathogenesis of CeD.

Our current fecal metabolomics analyses revealed excessive accumulation of lipids and lipid-like molecules in CeD patients. One of the major lipid molecules involved in multiple metabolic pathways related to CeD is Pc(34:2). Pc(34:2) is a phosphatidylcholine (PCs), and KEGG enrichment analysis has demonstrated that it is involved in pathways such as arachidonic acid metabolism, alpha-linolenic acid metabolism, and other related pathways. PCs have been reported to participate in the progression of various diseases. Specifically, PC(34:1) is closely associated with UC and can exert anti-UC effects by inhibiting the conversion of glutamate to N-acetylglutamate [70]. Alterations in lipid metabolism may represent a major mechanism in the pathogenesis of CeD. Lipid metabolism and its dysregulation can lead to acute and chronic inflammation under various conditions. The intake of whey soluble proteins may contribute to the development of CeD by affecting lipid metabolic pathways [71]. Previous studies have also reported that lipid metabolism is associated with refractory environmental intestinal dysfunction, among which CeD is included [72]. Our correlation analysis also indicated that Pc(34:2) was negatively correlated with Ruminococcus sp. and Fusicatenibacter saccharivorans. In addition, the levels of GPEtn(18:1/16:0), 6-gingerol, and Solasodiene were elevated in CeD patients, while the level of retinal was decreased along with several amino acids (Tyr-Gly, Lasalocid A). Vitamin A is absorbed by intestinal epithelial cells and converted to retinal, a product essential for intestinal adaptive immunity. Retinal directs the development of B and T cells and facilitates their recruitment into the gut. Thus, a deficiency of retinol leads to an increased susceptibility to intestinal infectious diseases [73, 74].

We constructed a diagnostic model for CeD based on eight differential metabolites, such as Pc(34:2), and the results demonstrated that these eight differential metabolites exhibited excellent diagnostic value for CeD (AUC = 0.9799, 95% CI 0.9737–0.9861). Subsequently, we conducted further investigations to determine whether the diagnostic efficacy for CeD could be enhanced through a combined diagnostic panel consisting of fecal microbes and metabolites. The diagnostic efficacy of the combined diagnostic model, which was generated based on microbiome and metabolic profiles for CeD, reached an AUC of 0.939. Although the diagnostic efficacy of this combined model was lower than that of the diagnostic model constructed based on differential metabolites, it remained capable of effectively distinguishing CeD patients from healthy subjects. This study provides a foundation for the development of noninvasive detection kits for CeD.

In addition, the network diagrams of the correlations indicate multiple associations between gut microbes and metabolites, as well as the potential impact of these distinct gut microbiota and metabolic substances on the pathogenesis and progression of CeD. However, the present study has several limitations. Firstly, the disease cohort included in this study was relatively small, which might thus limit the generalizability of the study’s findings to a certain degree. Whether the distinctive dietary structure (high gluten intake) of the population in the Xinjiang region of China could potentially influence the results remains to be further verified. Moreover, the limitations in the annotation of viral groups and the inadequate coverage of viral databases might lead to an underestimation of viral diversity and its functions. The CeD-associated viruses identified in this study have not undergone functional characterization, and we can only ascertain that these viruses predominantly belong to the phage group. The theory of “Phage-Bacteria-Host” interactions may be beneficial for understanding the specific mechanism by which these viruses are involved in CeD [75]. However, the specific mechanism warrants further exploration.

In conclusion, our study unveiled specific modifications in the gut microbiome and metabolome of Chinese patients with CeD via a multi-omics integration approach. Based on viral macrogenomic data, we discovered that three viruses, namely Junduvirus, Traversvirus, and Lederbergvirus, were correlated with CeD. Individuals with CeD who carry CeD risk genes might possess a distinctive composition of gut flora, which could, to some extent, account for the pathogenesis of CeD beyond the factors of genes and gluten consumption. Furthermore, we developed diagnostic models for CeD. This research may contribute to the comprehension of the pathogenesis of CeD and additionally offer a scientific foundation for non-invasive diagnosis.

Author contributions

Conceptualization, S.X., W.L., and F.G.; Data curation, S.X., T.S., N.L., and W.K.; Formal analysis, S.X., and H.L.; Funding acquisition, W.L., and F.G.; Investigation, S.X., S.Y., J.X., F.G.; Methodology, S.X.; and W.L.; Project administration, T.S., and F.G.; Resources, T.S., H.L., and F.G.; Software, S.X., J.X., and S.T.; Supervision, T.S., N.L., and F.G.; Visualization, S.X., T.S., J.X., and F.G.; Writing—original draft, S.X.; Writing—review & editing, S.X., T.S., and F.G.

Funding

This research was funded by Natural Science Foundation of Xinjiang Uygur Autonomous Region, Grant Number: 2022D01C831. National Natural Science Foundation of China, Grant Number: 82260116; National Natural Science Foundation of China, Grant Number: 82460117.

Data availability

All raw data for this study are included in supplementary documents. Further inquiries can be directed to the corresponding author.

Declarations

Ethics approval and consent to participate

This study was approved by the ethics Committee of Xinjiang Uygur Autonomous Region People’s Hospital (Xinjiang, China), (protocol n. KY2023060173; approval date: June 2023). All participants were informed about the study, voluntarily participated in the study, and signed an informed consent form.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no conflict of interest.