Introduction

Irritable bowel syndrome (IBS) is a common functional gastrointestinal disorder with an estimated local prevalence of 6.6% in Hong Kong1. The symptoms of IBS can seriously affect a person’s quality of life. Under the Rome IV guidelines, IBS is divided into four key subtypes based on stool patterns: IBS-D (diarrhea-predominant), IBS-C (constipation-predominant), IBS-M (mixed), and unclassified IBS. Despite the gastrointestinal symptoms, IBS patients suffer a very high frequency of psychological problems, such as anxiety and depression2. With advances in gut microbiota analyses, abnormalities in intestinal microflora, known as gut dysbiosis, have been implicated in IBS3,4. Gut dysbiosis is involved in the pathogenesis of IBS including gastrointestinal and psychiatric symptoms5. While much of the research focuses on the bacterial components, it is crucial to recognize that the human gut microbiome is an ecosystem of diverse microorganisms. Among these, viruses have recently garnered increasing attention due to their significant role in shaping the microbiota and their potential therapeutic implications6. Advancements in sequencing technologies and bioinformatics tools have particularly driven this heightened interest.

The human gut virome is thought to significantly impact the microbiome and human health. The viral sequences detected in the human gut are dominated by bacteriophages, which are viruses that specifically infect and replicate within bacteria. Despite most of the phage sequences corresponding to “dark matter” remaining to be characterized, phages have gained increasing attention in recent years as potential modulating tools against bacterial infections, including drug-resistant infections7,8, and those targeting the gastrointestinal tract9. Several studies have highlighted the relationship between bacteriophages and psychiatric disorders such as depression and anxiety10,11. Shotgun metagenomics captures genomic DNA from all gut organisms, not only prokaryotes, making it an optimal tool for studying gut multi-kingdom profiles12,13. The well-benchmarked assembly-based approaches overcome the fundamental limitation of the lack of reference genomes for the majority of phages, enabling the discovery of tens of thousands of viral species14,15,16 and their functional capacities15,17, and providing virome researchers with great insights into the structure and composition of the human gut virome. We hypothesize that gut virome dysbiosis may drive IBS heterogeneity and psychiatric comorbidities through bacteriophage-mediated microbial community shifts. Distinct virome signatures could serve as diagnostic biomarkers for IBS stratification and associated mental disorders.

In this study, we aim to elucidate the intricate relationships between gut virome–bacteriome interactions and IBS. Furthermore, we focus on a subgroup of IBS patients who exhibit comorbid psychiatric conditions, offering detailed insights into how these mental health issues may correlate with gut viral dynamics. By examining these interactions, we thus expect to uncover novel insights that could pave the way for innovative intervention strategies targeting the gut virome of IBS, particularly in those with concurrent mental health disorders.

Results

The distribution of gut viruses drives the differentiation of IBS from HC

A total of 444 stool samples were obtained from IBS (n = 360) and HC (n = 84) controls and sequenced using shotgun metagenomics (Fig. 1a). The IBS patients were further categorized according to the Rome IV criteria based on their predominant stool pattern(Supplemental Fig. 1a). While IBS is recognized as a multifactorial and heterogeneous disorder, its gut microbiome characteristics remain elusive due to this inherent variability, highlighting the need for more nuanced approaches to accurately capture IBS-associated microbial signatures. To assess IBS microbiome heterogeneity, we performed Hopkins clustering analysis on multi-kingdom microbial profiles, suggesting significant clustering tendency (Hopkins statistic = 0.97), indicating substantial heterogeneity among IBS patients. The quality of clustering, as represented with a silhouette plot (Supplementary Fig. 1b), was highest (0.93) with k = 2, suggesting that was the optimal number of clusters. To further resolve this heterogeneity, we integrated multi-kingdom microbial data (205 bacteria, 91 fungi, and 242 bacteriophages) using weighted similarity network fusion (WSNF), which identified two robust clusters (Cluster 1 [C1]: n = 185; Cluster 2 [C2]: n = 132; Fig. 1a). Comparative analysis of clinical characteristics revealed a significantly higher Bristol stool score in C2 (mean ± SD: 4.66 ± 1.59) compared to C1 (5.06 ± 1.30; p = 0.012, Fig. 1b), suggesting distinct phenotypic profiles between clusters.

Fig. 1: Gut viruses drive the differentiation of IBS from HC.
figure 1

a Integration of gut multi-biome data through weighted similarity network fusion (WSNF) approach. Heatmap of similarity scores derived from WSNF analysis in IBS patients, with spectral clustering identifying two distinct patient subgroups (labeled Cluster 1 and Cluster 2). Color gradient represents pairwise similarity strength (scale bar at right). b Comparison of Bristol score of patients between two identified patient clusters. c Principal coordinate analysis (PCoA) of gut multi-biome of based on Bray–Curtis dissimilarity illustrates two patient clusters. d Random forest classifier model trained on multi-biome can predict the clustering of IBS patients. e Features from gut viruses contribute most to differentiating clusters in the random forest models. f, g Network visualization of key taxa in two clusters in HC (f) and IBS clusters (g). Each node represents a microbial taxon (e.g., a bacterial, viral or fungal species) included in the co-occurrence network. The shape of a node is a visual attribute used to distinguish between different types of microbes (e.g., bacteria vs. viruses vs. fungi). Circles, triangle, and inverted triangle represent microbes and lines represent their associated interactions. Node size (degree) reflects the number of direct interactions for a given microbe. Border thickness represents calculated stress centrality for each microbe, while color depth reflects the positive or negative of the interactions.

Full size image

The two WSNF-derived clusters exhibited significant differences in their multi-kingdom microbiome profiles, as demonstrated by β-diversity analysis (PERMANOVA: R² = 0.07, p < 0.01; Fig. 1c, Supplementary Fig. 1d–f). The comparison of alpha-diversity (Shannon index and richness) between the two clusters revealed no significant differences in the multi-kingdom profile. However, when examining bacteria, fungi, and viruses separately, we observed compensatory dynamics within the multi-kingdom communities (Supplementary Fig. 1g–j). Specifically, while the alpha-diversity of bacteria and fungi decreased, viral diversity increased, suggesting potential coevolutionary dynamics among these microbial groups.

To evaluate the potential of microbial clusters for IBS stratification, we constructed multi-kingdom random forest models, achieving discriminative AUCs of 0.73–0.93 across IBS clusters (IBS-C1: 0.897–0.898; IBS-C2: 0.930–0.931; HC: 0.726–0.731), demonstrating their utility in subtyping (Fig.1d). We ranked microbial contributions across multiple taxonomic levels by assessing feature importance in the random forest model. We found gut viruses had a much more important effect (54%) than the other taxa in the analysis (bacteria 26%, fungi 20%) (Fig.1e). These results pointed to a crucial role of gut virome in shaping the division between IBS two clusters and healthy controls. To understand the interaction among gut microbiota, the co-occurrence of multi-kingdom taxa was measured and depicted with relative abundances in both IBS and healthy controls (Fig. 1f and g). The strong interactions among bacteriophages were observed in healthy controls (Fig. 1f). Interestingly, Escherichia coli, Escherichia phage HK639, and Escherichia phage TL-2011b had an inverse relationship with Bacteroides uniformis, which is known as indicators of healthy intestines18, only in healthy controls. Nevertheless, gut multi-kingdom taxa interactions were more noticeable in healthy controls than in IBS subjects (HC 423 vs. C1-IBS 168, C2-IBS 228) (Fig. 1g, Supplementary Fig. 2).

Viral signatures associated with anxiety and depression in IBS

Previous epidemiological studies showed that the incidence of anxiety and depression in IBS patients is much higher than that in the general population. The frequent co-occurrence of mental disorders and IBS is well established, more than a quarter of individuals with IBS had depressive symptoms, and over a third had anxiety symptoms, which was considerably higher than in healthy individuals2,19. A recent meta-analysis revealed that the prevalence of depressive and anxiety symptoms in IBS patients was 28.8% and 39.1%, respectively2. Growing evidence suggests that the gut microbiota is closely linked to mental health20,21, and microbial imbalance may contribute to psychiatric conditions. However, most gut microbiota studies in IBS have focused exclusively on bacteria, neglecting the potential roles of other kingdoms (e.g. viruses) within the microbiome ecosystem. Here, we investigated the associations between gut microbiota, especially viruses, and psychiatric comorbidities in IBS.

To investigate the potential association between microbiome signatures and psychological symptoms, we classified samples into five groups: healthy controls (HC), regular IBS patients without depression (rIBS, SAS < 50 and SDS < 53), IBS with depression (dIBS, SDS ≥ 53 and SAS < 50), IBS with anxiety (aIBS, SAS ≥ 50 and SDS < 53), and IBS with both depression and anxiety (adIBS, SDS ≥ 53 and SAS ≥ 50). We tested the differential abundance of viral species using microbiome multivariable association with linear models (MaAsLin2) with the individual participant as a random factor. We observed differential abundance patterns of gut viruses in IBS with depression and anxiety when compared with the HC (Fig. 2a). Notably, the changes observed in dIBS and adIBS were more pronounced compared to aIBS and rIBS. These alterations may signify the strong correlations of the gut viral population with depression. We found increased Enterobacteria phages and Escherichia phages in dIBS and adIBS, including Enterobacteria_phage_cdtI, Enterobacteria_phage_HK225, and Escherichia_phage_Cartapus. We found shared decreased levels of Salmonella phage 118970_sal3, Clostridium phage phiCD211, and Mannheimia phage vB_MhS_587AP2 in aIBS and adIBS. MaAsLin2 analysis revealed distinct bacterial signatures of IBS with psychiatric co-morbidity (Supplementary Fig. 3a). Compared to rIBS, adIBS showed the most pronounced dysbiosis, with significant depletion of butyrate producers Eubacterium rectale22 and next-generation beneficial microbe Akkermansia muciniphila (FDR < 0.05)23, alongside enrichment in pathobionts (Klebsiella pneumoniae, Escherichia coli, FDR < 0.05). Notably, both adIBS and dIBS groups exhibited depletion of Eubacterium eligens, a putative gut–brain axis modulator implicated in energy homeostasis, colonic motility, and anti-inflammatory immunoregulation. These findings suggest psychiatric comorbidities in IBS are associated with taxon-specific ecological imbalances, affecting short chain fatty acid producers and immunomodulatory species24. However, the causal direction of these associations requires further validation through longitudinal studies and mechanistic experiments. Distinct phage–bacteria interaction patterns are illustrated across IBS subtypes in the Supplementary Fig. 3b–e, with the strongest negative correlations (e.g., bacteroides–phage pairs) being observed in adIBS, while minimal associations are displayed in rIBS. These differential networks suggest that psychiatric comorbidities may exacerbate phage-mediated dysbiosis, particularly affecting key bacterial taxa involved in gut–brain axis regulation.

Fig. 2: Multi-omics based machine learning for mental disorder diagnosis in IBS.
figure 2

a MaAslin2 analysis of IBS with mental disorder illustrating discriminant taxa compared to HC. b The area under the receiver operating characteristic curve (AUROC, center for the error bands is median). c Feature importance summary of top 30 features from random forest classifier.

Full size image

Next, we explored whether the gut microbiome could predict psychiatric comorbidities in IBS. Given that anxiety and depression (adIBS, 42%, 134 out of 317) are the most common subgroups, we conducted machine learning training and evaluated the prediction performance for this group. We constructed a machine learning classifier with ten repeats of fivefold cross-validation to differentiate adIBS from regular IBS. Combining this with patient stratification based on the microbiome could offer a focused and precision-based diagnostic strategy. Using gut multi-kingdom microbiome and metabolites data, we trained a random forest machine learning model to differentiate adIBS from rIBS. The random forest model achieved a mean AUROC of 0.78 (Fig. 2b), suggesting that multi-class disease classification based on the fecal microbiome was feasible. To further test whether a simplified panel of biomarkers could be used as biomarkers to distinguish the mental disorder in IBS according to the multi-omics data, we performed predictions using top-ranked features contributing to the random forest classifier (Fig. 2b). We tested how many of the features representatives of the mental disorder in IBS were necessary to achieve the comparable predictive performance by training the classifier model with different number of top-ranking features that were chosen based on the mean decrease in GINI from the classifier trained with the full set of features (Supplementary Fig. 4a). The results showed that using as few as 100 features (Supplementary Fig. 4b) achieved best average AUC 0.84 for IBS with mental disorder. Applying the variable selection strategy based on random forest, important features were identified based on the importance score. Particularly, important features of random forests in cluster 2 are distinct with healthy control (Fig. 2c). Viruses have the highest importance (36.2%) among the top 100 features, suggesting the model relies heavily on gut viruses, with fungi (31.1%), bacteria (21.5%), and metabolites (11.2%) playing supporting roles in the model’s predictions (Supplementary Fig. 4c). Biomarker discovery in the context of mental disorders in IBS has the potential to revolutionize patient care by offering more precise diagnostics, personalized treatment strategies, and a deeper understanding of the complex interplay between gut health and mental well-being. This knowledge can inform preventive measures and interventions, potentially reducing the overall burden of mental health issues within the population. Gut viruses, particularly bacteriophages can influence the composition and function of the gut microbiome, which in turn affect host metabolism. A recent study from our group found that impaired bile acid synthesis regulation and excessive bile acid excretion, driven by a Clostridia-rich microbiota, are linked to the severity of diarrheal symptoms in IBS-D patients25. Given the contribution of gut viruses to prediction of the mental disorders in IBS, we test the relationship between the gut virus composition and the bile acid concentrations. We performed MaAsLin2 analyses of 37 bile acid metabolites in HC, rIBS, and adIBS, and found 9 bile acid metabolites that were significantly decreased in IBS patients (Supplementary Fig. 4d). The correlation between bile acids and gut viruses varied across the HC, rIBS, and adIBS groups (Supplementary Fig. 4d). In general, the adIBS group exhibited fewer positive and negative correlations, indicating a potential disruption or weakening of the interplay between bile acids and gut viruses in these patients (Supplementary Fig. 4e).

Construction of phage genome catalog in the Hong Kong IBS cohort

Improvements in high-throughput sequencing and bioinformatic technologies have allowed the virome to be analyzed in unprecedented detail. Using state-of-the-art bioinformatics tools, we characterized the global virome in IBS by analyzing 1.28 TB of sequences derived from 444 fecal metagenomes. Rarefaction analysis (Supplementary Fig. 5a) revealed that more gut viruses remain to be identified. The average read count per sample was 20,436,661.5 and per-sample alignment of bacteria, fungi and viruses has been shown in Supplementary Fig.5c. Following recent viromics benchmarking approaches and stringent criteria, we identified 626,559 putative viral contigs. Assessing with CheckV revealed 1302 complete viral genomes, 4843 high quality and 7026 medium quality viral genomes. We were able to classify 29,254 viral genomes (4.67% of the total) using PhaGCN as belonging to existing families. We have collected intrinsic and extrinsic factors (e.g. age, sex, body mass index [BMI], lifestyle, dietary habits, diseases, and medications) through questionnaires, face-to-face interviews, and medical records. The influences of these factors on the composition of gut microbiota from IBS patients were determined by the Adonis test. We assessed the confounding effects of these factors on the IBS gut microbiome. The results indicated that although none of the trends were statistically significant (p > 0.05), the IBS-SSS score was the primary factor influencing gut microbiome composition across all IBS samples (Supplementary Fig. 5b).

To examine the taxonomic spread covered by metagenome-assembled viral genomes, we compared them against the RefSeq prokaryotic virus database using vConTACT2, a network-based method to classify viral contigs, with all viral sequences from bulk metagenomes as input, a network-based method to classify viral contigs. Clustering of the viral sequences with 95% sequence similarity generated 105,365 viral genomes (corresponding to the species level) (Fig. 3a). The majority of the phages were predicted to belong to the Caudoviricetes, among which Sepvirinae (n = 2388), Peduoviridae (n = 2312) were most abundant (Fig. 3b). Viral families Tectiviridae, Hendrixvirinae, Adenoviridae, Phycodnaviridae, Fernvirus in HC were more abundant than that in IBS (Fig. 3b). Virus clustering patterns generated by vContact2, revealing distinct virome compositions across the IBS with depression and anxiety (Fig. 3c). In addition, dominant families such as Kostyavirus, Eucampyvirinae, and Boydwoodruffvirinae exhibited differential distributions, with rIBS showing unique viral signatures compared to psychologically distressed subgroups (aIBS, dIBS, adIBS) (Fig. 3d). These findings suggest a potential link between gut virome profiles and IBS patients with anxiety or depression.

Fig. 3: Taxonomic diversity of gut viruses in IBS.
figure 3

a Gene-sharing network of the metagenome-assembled viral genomes in IBS and healthy controls obtained from vConTACT2 and visualized using Cytoscape, with the viral genomes being colored according to the family assignment. Nodes represent viral genomes and edges indicate similarity based on shared protein clusters. b Comparison of the viral family distribution of viral genomes in HC and IBS. c Gene-sharing network of the metagenome-assembled viral genomes in IBS with mental disorders. d Significantly altered viral families of viral genomes in IBS with mental disorders (Wilcoxon test, p < 0.05).

Full size image

Applying the integrated phage–host prediction, we were able to identify 8517 (8.08%) DNA vOTUs associated with a host genus or family (Supplementary Fig. 5d). Among the most abundant host families, a higher proportion of vOTU mapped to Bacteroidaceae and Ruminococcaceae in HC, whereas higher number of viral population with host families Lachnospiraceae, Acutalibacteraceae, and Enterobacteriaceae were observed in IBS (Supplementary Fig. 5e).

IBS viruses have a broader range of bacteria host

The host range is one of the central traits to understand in phages, which is governed by intricate molecular interactions between phages and host throughout the infection cycle. Host prediction is important to understand the potential roles of gut viruses in an ecosystem. While many well studied model phages seem to display a narrow host specificity, recent ecological and metagenomic evidence suggests that phage host ranges can vary considerably26,27, from narrow to broad. Recent investigations have elucidated key molecular mechanisms underpinning phage multiple-host infectivity28,29, a phenomenon with profound implications for phage therapy, microbial community dynamics, and biotechnological applications. Here, we systematically characterize the host ranges of gut bacteriophages in IBS, revealing distinct patterns of phage–bacteria interactions. By predicting the probable hosts of the gut viruses, we identified 52,690 bacteriophages from the MAG. Host prediction of the vOTUs revealed that the most common phylum of the predicted hosts was Firmicutes (IBS:67.2%, HC: 66.5%), followed by Bacteroidota (IBS:17.11%, HC: 19.77%), and Pseudomonadota (IBS:4.48%, HC: 6.46%) (Fig. 4a, b). Phage–host interactions differed notably between IBS patients and healthy controls (Fig. 4c). Comparative assessment of phage abundance per host phylum further highlighted elevated phage predation on Bacillota (e.g., Bacillota I, mean ± SD 2.43 ± 7.81 vs. 7.32 ± 18.27, p < 0.001) in IBS, whereas diminished predation on Bacteroidota was observed relative to HC (mean ± SD 20.52 ± 17.51 vs. 24.99 ± 18.82, p = 0.049) (Fig. 4d). These findings suggest phage–bacterial interactions may contribute to microbial dysbiosis associated with IBS. At the genus level, the commonly assigned hosts were Parabacteroides, Bilophila, and Faecalibacillus. Notably, bacteriophages were predicted to infect a broader range of host species in IBS compared to HC. We calculated the average number of phages per host genus, and the average number of phages per individual host genome for bacteriophage (Fig. 4a, b, e). The number of phages per host genome varied, with Bilophia having the most abundant prophages in IBS (Fig. 4e). A Unique host genome was defined as a MAG that served as the sole predicted bacterial host for a given phage based on our metagenomic analyses. 2237 phages were identified in the genus Bilophia (77 unique host genome) (Fig. 4e). In comparison, only 394 gut phages could infect Bilophia (13 unique host genome). The unique host genome of prophages varied among host genera, and IBS has a greater number of unique genomes than HC (Fig. 4c). Following the viral-host pattern via matched prophages in viral families, the host genera of IBS and HC virus were mainly from 7 and 5 bacteria phyla, respectively (Fig. 4f). We found number of bacteria host for Peduoviridae (p = 0.02) was significantly enriched in IBS, reinforcing the unique host-specific associations in IBS (Supplementary Fig. 5f). IBS virus comprised of more abundant host genera in the bacteria community. For example, the viral contigs belonging to Peduoviridae spanned 10 bacteria genera in IBS while only 3 host genera linked in HC (Fig. 4f). Viruses in HC seemed to be specific with viral contig family linked to one specific bacterial genus, whereas a considerable fraction of viruses in IBS have broad host ranges with phage contigs predicted to infect multiple genera.

Fig. 4: Bacterial host range of the gut viruses in IBS.
figure 4

a and b Genome-based phylogenetic trees of bacterial genera contained the predicted hosts of viral genomes. c Percentage of phage infections (host phylum) in IBS and healthy controls. d Numbers of phages per phylum of host in IBS and healthy controls. e Comparison of lysogeny rates, numbers of phages per genus of host in IBS and HC. f Sankey plot showing the virus–host linkages in IBS and HC.

Full size image

Gut viral auxiliary metabolic genes (AMGs) link viral infection to host metabolism and psychiatric comorbidities in IBS

Gut viral infection can affect host metabolism via the expression of viral AMGs. In addition to their physical impact on microbial communities, phages carry host-derived AMGs, which are used to manipulate the host cell metabolism during infection. To better understand the ecological effects of viruses in IBS, we searched the AMGs by VIBRANT30 in IBS viral genomes and calculated the relative abundances. A total of 31,819 genes were annotated to be AMGs (Fig. 5a). According to the KEGG annotation, the identified AMGs of gut viruses were involved in a variety of metabolic pathways, including those related to amino acid metabolism (29.9%), cofactors and vitamins metabolism (20.3%), carbohydrate metabolism (14.6%) (Fig. 5a). The most frequently annotated AMGs across all vOTUs were related to sulfur metabolic pathways (dcm, cysH, the most widespread AMGs) and could be found in 37.81% of AMGs encoding vOTUs (Supplementary Fig. 5). Notably, the most prevalent AMG was DNMT1, a DNA (cytosine-5)-methyltransferase that protects viruses from their hosts’ antiviral restriction-modification systems, which was detected in 27.5% of all AMGs. The high proportion of such an AMG probably represents a defense mechanism of the gut virome. Several previous studies have suggested a link between gut viruses and depression11,31.

Fig. 5: Viral auxiliary metabolic genes (AMG) on host metabolism.
figure 5

a Auxiliary metabolic categories identified in virome bacteriophage contigs, with amino acid metabolism representing the highest proportion of overall AMGs in the assembled bacteriophage dataset. b Diversity and richness of AMGs in IBS with and without psychiatric comorbidities. c Volcano plot of AMG showing differential abundance between HC and IBS patients with psychiatric comorbidities.

Full size image

Previous work has shown more altered gut viruses than gut bacteria in depression, suggesting that gut virome may play a role at least equivalent to that of the gut bacteriome in the pathology of depression10. Gut viruses may influence host behaviors by regulating their host bacteria and their metabolism. We next compared the AMG levels between IBS individuals with and without psychiatric comorbidities to explore the potential link between gut viral functions and mental disorders in IBS. We found higher AMG diversity in IBS with depression and anxiety (Fig. 5b), indicating more metabolic functions on the viral genome. The volcano plot (Fig. 5c) illustrated the comparison of AMG levels of IBS with or without mental disorders, showing a significant decrease in ubiG and ahcY in the IBS with depression, with ubiG also exhibiting a consistent trend of decrease in individuals who have both anxiety and depression. Notably, the ubiG gene encodes an S-adenosyl-L-methionine (SAM-e)-dependent methyltransferase enzyme involved in ubiquinone biosynthesis, a process reported to be crucial in various neurological diseases. SAMe is one of the most extensively studied supplements for treating depression, with low levels reported in individuals with depressive disorders32,33. Interestingly, the IBS with anxiety exhibits a markedly different AMG pattern, characterized by an increase in the expression of pcaC (xenobiotic metabolism), UGDH (carbohydrate), asnB (amino acid metabolism), while coaD (cofactors and vitamins), acpP (secondary metabolites), gala(glycan), EARS (cofactors and vitamins), and guaA (nucleotide) showed decreased levels than the HC. This contrasting AMG profile underscored the distinct viral function underpinning associated with IBS when comorbid with either depression or anxiety.

Discussion

In the present study, we have performed a large-scale analysis for viral profiles of deeply phenotyped IBS individuals (n = 317) and shown extensive virome variation and its association with the bacteriome and mental disorders. This study is, to the best of our knowledge, the largest single cohort analysis for the human gut virome of IBS population. The analysis uncovered novel viral clades, interactions between the virome and bacterial viral functional genes, and clinical features strongly associated with the virome, expanding our knowledge of the gut virome structure and variation and offering potential biomarkers for diagnosis and new insights into the pathogenesis of the disorder.

In addition to delineating the general virome features of IBS, we have conducted a focused investigation into the virome of IBS patients who also suffer from depression and anxiety. Our findings reveal distinct viral compositions and interactions within this subgroup, suggesting that the gut virome may play a significant role in the gut–brain axis and the manifestation of psychiatric comorbidities in IBS patients. The observed increase in viral-host interactions in IBS patients with depression or anxiety may be driven by several interconnected pathways. Depression and anxiety are associated with low-grade systemic inflammation (e.g., elevated IL-6, TNF-α), which may amplify host susceptibility to viral activity or persistence34,35. The mental disorder may increase intestinal permeability (“leaky gut”) could facilitate viral particle translocation or immune activation11,36. Dysbiosis of the gut microbiota—a hallmark of both IBS and mood disorders—might indirectly promote viral replication by disrupting colonization resistance or altering mucosal immunity37,38. While further research is needed to validate these mechanisms, their convergence in comorbid IBS and psychiatric conditions provides a plausible framework for our findings. The identification of specific viruses and their interactions in IBS patients with mental health issues underscores the complexity of IBS as a multifactorial disorder and highlights the need for tailored therapeutic approaches that address both gastrointestinal and mental health symptoms.

Our analysis indicates that the gut virome in IBS patients exhibits a broader host range compared to healthy controls, suggesting a higher prevalence of phage–bacteria interactions in IBS. This expanded host range may facilitate more dynamic and extensive interactions within the gut microbiome, potentially exacerbating the dysbiotic state observed in IBS. Moreover, we have identified distinct AMGs in the IBS virome, which are not only involved in the modulation of bacterial metabolism but also in pathways potentially linked to mental disorders. These findings suggest that gut viruses may contribute to the metabolic and neuropsychiatric dimensions of IBS, providing new avenues for research into the viral contributions to IBS pathophysiology and its psychiatric comorbidities.

In summary, our study provides a comprehensive analysis of the gut virome in IBS, uncovering novel insights into its complexity and potential implications for both gastrointestinal and mental health. The identification of unique viral signatures, expanded host ranges, and specific AMGs highlights the intricate interplay between gut viruses and the broader microbiome, paving the way for future studies aimed at elucidating the role of the virome in IBS and its associated comorbidities.

Methods

Sample collection

Individuals meeting the Rome IV criteria for IBS12,13 were prospectively recruited from two Chinese medicine clinics affiliated with the School of Chinese Medicine at Hong Kong Baptist University. A total of 360 IBS patients (meeting Rome IV diagnostic criteria, including 317 with complete psychiatric comorbidity assessments), and 84 healthy controls (HC) were recruited for fecal metagenomic sequencing. Among the 360 IBS patients, subtype distribution was as follows: 67.22% IBS-D (diarrhea-predominant), 8.05% IBS-C (constipation-predominant), 8.33% IBS-M (mixed), and 16.39% IBS-U (undefined). Specifically, IBS patients were included if they were 18–65 years, and met Rome IV criteria; had an IBS symptom severity scale (IBS-SSS) score over 75 points at baseline; had a normal colonic evaluation within the past 5 years via colonoscopy or barium enema. Bowel movement frequency was recorded as number per day and consistency was scored on a 7-point scale and stool consistency was evaluated using the Bristol stool scale39. Subjects were excluded if they were pregnant; had a medical history of IBD; surgical histories involving gallbladder removal, the gastrointestinal (GI) tract, or cerebral cranium; had a medications know to influenc GI function, antidepressants and anxiolytics. Individuals without the medical history of GI diseases, neurodegenerative diseases, cardiovascular diseases, metabolic disorders were also recruited as control. All participants were required to stop using microbiota-related supplements. Probiotics, antibiotics, prebiotics, at least three weeks before stool sampling. Specimens (stool) were transported to the laboratory on dry ice and stored at −80 °C until DNA extraction. Details of patients’ diagnoses, subject recruitment, and sampling are described in the previous work25,40. The study was conducted in accordance with the declaration of Helsinki. This study was approved by the Ethics Committee of Hong Kong Baptist University (HASC/15-16/0300 and HASC/16-17/0027). Written informed consent was signed and obtained from all participants.

DNA extraction and metagenomic sequencing

Phenol/chloroform/isoamyl alcohol method was applied to to extract microbial DNA from stool samples (200 mg) of included subjects. The DNA that passed quality control was then subjected to library construction using the TruSeq DNA HT Sample Prep Kit. Paired-end sequencing (2 × 150 bp) was carried out using Illumina Hi-Seq platform.

Depression and anxiety

IBS patients with depression and anxiety were identified using a combination of the Zung self-rating depression scale (SDS), the 17-item Hamilton depression rating scale (HAMD-17), and the Zung self-rating anxiety scale (SAS). Firstly, all subjects were requested to complete the SDS and SAS, with a cut-off index score of 50. Patients scoring above 53 on the SDS were subsequently evaluated with the HAMD-17 for a professional diagnosis40.

Taxonomic classification

We first performed short-read taxonomic classification of metagenomic sequencing using Metaphlan341 and Kraken v1.1.1 with default settings was performed as previously described (database, NCBI RefSeq, release 202)42,43. The relative abundance were estimated with Bracken44. Abundance profiles across samples with species with medium abundance ≤0.01% were filtered out.

Integration and clustering analysis of multi-biome data

Integration of bacterial, fungal, and viral profiles was achieved through weighted SNF (online tool (WSNF, https://integrative-microbiomics.ntu.edu.sg)), with biome-specific weighting based on the observed richness of each biome. WSNF integrates multiple data types by constructing and fusing similarity networks, assigning adaptive weights to each data source to optimize the combined representation. This approach offers advantages over traditional clustering methods by effectively capturing complementary information across omics layers while reducing noise and bias from individual datasets. Using the merged dataset of microbiome profile, the tool generates a corresponding patient similarity network using a clustering algorithm (Bray–Curtis metric, default settings), outputting the cluster assignments for each patient. The optimal clustering configuration (n = 2) was established through eigengap method and silhouette-optimized k-nearest neighbors (KNN) parameters.

Genome assembly, binning, and analysis for the species-level representative MAGs

Metagenomes (metagenome-assembled genomes, or MAGs) were assembled by using MEGAHIT v1.2.945 to investigate the genomic diversity of bacteria species46. Sequence coverage profiles were then generated by aligning quality-filtered reads to their respective assemblies by BWA v0.7.1747. Contigs were binned with MetaBAT248 (v2.15, default parameters). The quality of assembly bins were evaluated with CheckM49, retaining those with over 90% completeness and <5% contamination. The taxonomy of MAGs were inferre to SGBs by applying ‘phylophlan_metagenomic’, a subroutine of PhyloPhlAn v3.0.6750. PhyloPhlan was employed to establish taxonomic classifications of MAGs, which were crucial for subsequent host–phage association analysis.

Viral sequence recognition and clustering

Viral sequences were identified from metagenomic assemblies using VirSorter2 (v2.2.4) with the parameters --exclude-lt2gene --db 2. Scaffolds longer than 1 kbp were considered putative viral sequences if they met at least one of the following criteria51,52: (1) VirSorter-positive (including all six categories), (2) BLASTn alignments to NCBI viral dereplicated sequences (e ≤ 10−10, >90% query coverage, >50% ANI), (3) being circular53, (4) longer than 3 kbp with no hits (e value of 10−10, 90% ANI, alignments >100 nucleotides) to the nt database (release 249). A total of 10,458,078 scaffolds were initially classified as viral. Putative viral scaffolds were subjected to quality control using CheckV54 to removed potential contaminations. Taxonomic classification at the family level was performed using PhaGCN v2.155. The viral contigs were next analyzed based on the predominant assignment of their open reading frames (ORFs). The ORFs were predicted with MetaProdigal v2.6.356. To annotate the predicted ORFs, their amino acid sequences against the viral RefSeq protein database (v84) were queried using Diamond57 (e-value threshold of <10−5 and a bitscore >50). Contigs were taxonomically annotated according to the predominant assignment of their constituent ORFs to a taxon. Viral abundance was estimated by mapping reads to viral contigs using BWA, and the resulting alignments were processed using custom scripts. Abundances were normalized as reads per kilobase per million mapped reads (RPKM). Viral contigs were clustered into viral clusters (VCs) using vContact2 (v0.11.3)58.

Host prediction and construction of phylogenetic tree

Host prediction were achieved by CRISPR spacers and prophage sequence alignment to MAGs. We manually constructed a database of MAG and viral sequences for each independent metagenomic sample. For CRISPR spacer hit, we used BLASTn v2.12.0+ to compare the CRISPR spacers59 identified from the MAG to the database built with the viral genomes with parameters: -task blastn-short -perc_identity 100 -qcov_hsp_perc 100. For prophage sequence alignment, we used BLASTn v2.12.0+ to align all viral genomes into MAG with parameters: -task megablast -perc_identity 90 -qcov_hsp_perc 75, and results with MAG length>2500 bp were retained. To reveal the infection patterns of IBS and HC gut phage, we established genome-level phylogenetic trees on MAGs reconstructed from the IBS and HC gut metagenomes, respectively. For each genus, we used a script to randomly select a MAG and count the infection status of all MAGs in the genus (the number of times each genus was infected, the average number of times the MAGs in the genus were infected). Afterwards, phylogenetic trees were constructed using IQ-TREE v2.3.360 (-bb 1000 -m MFP -nt 128) and then visualized using the R package ggtree v3.8.2.

Identification of auxiliary metabolic genes (AMGs)

AMGs were predicted for all metagenome-assembled viral genomes using VIBRANT30, which employs a rigorous pipeline including viral contig identification, Prodigal-based gene prediction, and HMM searches against KEGG, Pfam and VOG databases with stringent filters (E-value, viral-like score). The relative abundance of AMGs is determined by calculating the relative abundance of viral genomes carrying those AMGs in each sample. Differential abundance analysis of AMGs between groups was performed using DESeq261 with default parameters.

Microbiome co-abundance analysis

To determine the interaction among gut microbiota in IBS and HC, a multi-kingdom interaction network were constructed. Co-occurrence analysis were performed with statistical significance testing using SparCC network analysis62. SparCC analysis were performed using the R package “SpiecEasi v1.1.1” with 20 iterations in the outer loop and 10 iterations in the inner loop63. The correlation strength exclusion threshold is 0.1, using SparCC default settings. Correlations with an absolute value <0.1 were considered zero by the inner SparCC loop, and p-values <0.05 were considered significant. Network metrics such as node degree (busy), stress centrality (critical), and betweenness centrality (influential) were calculated, and visualized in cytoscape63.

Multi-class classification by machine learning

All classifier models are implemented using Python 3.6 as described previouly64. They are trained and evaluated on the microbiome datasets using the cross-validation (CV) method. The machine learning-based classifiers are implemented using the python library Sklearn65. We randomly split the dataset into 70% for training and 30% for validation. To diagnose different phenotypes, random forests (RF) was employed using taxonomic profiles of the fecal microbiome with to the default SciKit-learn settings (n_estimators = 2000, class_weight = balanced). The optimal models, determined through cross-validation, were then evaluated on a separate evaluation dataset to assess their final performance in predicting incident disease. Microbial features that were consistently highly ranked and frequently selected were identified as predictive signatures for further analysis. The prediction performance was derived using the same training datasets.

Statistics and reproducibility

Calculations of diversity (Shannon’s index), richness (Chao1 index) and rarefaction calculation were performed using the vegan package. Compositional data were analyzed and visualized via principal coordinates analysis (PCoA) based on on Bray–Curtis dissimilarities. We used two-sided Wilcoxon signed-rank tests to assess differences between groups. The effect size of host factors were explored to identify covariated of gut microbiome compositional variation by using permutational multivariate analysis of variance (PERMANOVA; Adonis)66 (Adonis) in the vegan R package (999 permutations, FDR < 0.05).Taxonomic differences were calculated using Multivariate Association with Linear Models (MaAsLin2)67.