Disrupted Gut Viral‐Bacterial Ecology of Patients With Liver Cirrhosis

ABSTRACT

Background

The gut microbiota contributes to liver cirrhosis (LC), yet the gut virome and its cross‐kingdom ecology with bacteria are less well defined.

Methods

To characterize LC‐associated virome alterations and assess their clinical relevance, we reanalyzed publicly available faecal metagenomes from patients with LC and healthy controls. After quality control and removal of human reads, sequences were mapped to the Chinese Gut Viral Catalogue at 95% nucleotide similarity, viral operational taxonomic units (vOTUs) were annotated using the latest ICTV framework, and viral functions were inferred by KEGG annotation. Differential vOTUs and bacterial species, virus–bacteria networks and random forest classifiers were constructed with internal and external validation.

Results

LC showed reduced viral richness and Shannon diversity, and a distinct Bray‐Curtis separation from controls. Ten viral families and 473 vOTUs differed between groups (59 LC‐enriched). KEGG‐based profiling highlighted functional shifts in LC‐enriched viruses, including increased K01185 (lysozyme) and K02172 (blaR1). Virus‐bacteria networks were markedly sparser in LC than in controls (130 vs. 509 significant correlations). A virome‐based random forest model distinguished patients from controls with high accuracy in internal (optimal AUC = 0.911) and external (optimal AUC = 0.773) validation cohorts, and the model combining viral and bacterial features achieved similarly robust performance.

Conclusions

LC is associated with disrupted gut viral–bacterial ecology, and virome features show promise as non‐invasive biomarkers, warranting longitudinal and mechanistic follow‐up.

Key Points

• In cirrhotic patients, gut viral diversity is reduced, and the symbiotic relationship between enteric viruses and bacteria is disrupted.

• Faecal viral features can effectively distinguish patients with liver cirrhosis from healthy individuals.

• Gut viral‐bacterial ecological abnormalities have the potential to be developed into a simple, non‐invasive diagnostic biomarker for liver cirrhosis.

Lay Summary

Patients with liver cirrhosis have significant imbalances in their gut viruses, including lower viral diversity and a shift towards viruses that attack harmful bacteria. This study found that these viral changes, combined with bacterial imbalances, could help distinguish cirrhosis patients from healthy individuals with high accuracy, suggesting that gut viruses may serve as a non‐invasive diagnostic tool for liver cirrhosis in the future.

Abbreviations

cnGVC

Chinese Gut Viral Catalogue

FDR

false discovery rate

ICTV

International Committee on Taxonomy of Viruses

KEGG

Kyoto Encyclopedia of Genes and Genomes

KO

KEGG Orthology

LC

liver cirrhosis

PCoA

principal coordinates analysis

PERMANOVA

permutational multivariate analysis of variance

SCFAs

short‐chain fatty acids

TLRs

toll‐like receptors

vOTUs

viral operational taxonomic units

1. Introduction

Liver cirrhosis (LC), an end‐stage manifestation of chronic liver disease, is pathologically defined by hepatocyte loss and irreversible fibrotic scarring, wherein collagen‐rich tissue progressively replaces functional hepatic parenchyma [1]. Global epidemiological patterns reveal HBV accounts for approximately 29% of LC‐related mortality, predominantly in Asia‐Pacific and African regions, whereas HCV and alcohol‐associated liver disease prevail in the United States, Europe and China [2, 3]. Although etiological diversity exists (HBV, HCV and alcohol), shared pathobiological mechanisms involve chronic inflammatory signalling, dysregulated extracellular matrix remodelling and sustained hepatic stellate cell activation [4, 5, 6]. If not properly managed, these pathological processes can trigger severe complications such as infections, hepatic encephalopathy and ascites [7], or even progress to hepatocellular carcinoma [8].

In recent years, the role of the gut microbiota in LC has attracted increasing attention. Growing evidence suggests that LC is associated with gut microbial dysbiosis. Patients with LC exhibit reduced gut microbial diversity, enrichment of potentially pathogenic bacteria such as Enterobacteriaceae and Veillonella [9], and depletion of beneficial short‐chain fatty acid (SCFA)‐producing bacteria, including members of Bacteroidetes and Firmicutes [10]. These microbial alterations may contribute to impaired intestinal barrier function and activation of hepatic Toll‐like receptor (TLR) signalling through the gut‐liver axis, thereby promoting systemic inflammation and liver injury [11]. However, most studies have focused primarily on the bacteriome, whereas the gut virome, particularly bacteriophages, remains less well characterized in LC [12]. Gut viruses can influence host physiology through direct and indirect mechanisms [13], and bacteriophages may shape bacterial community structure and functional potential in several disease contexts, including hepatitis, nasopharyngeal cancer, cervical cancer and gastrointestinal malignancies [14, 15, 16] but the cross‐kingdom co‐abundance patterns between gut viruses and commensal bacteria remain insufficiently explored [17], but the cross‐kingdom co‐abundance patterns between gut viruses and commensal bacteria remain insufficiently explored [18].

In this study, we reanalyzed 231 publicly available faecal metagenomes from the study PRJEB6337, comprising 96 LC patients and 83 healthy controls (HCs) in the discovery cohort and 21 LC patients and 31 controls in the validation cohort, and further included an additional independent validation set of 31 LC patients and 48 healthy controls from the study PRJEB65440. We characterized gut virome diversity, identified LC‐associated viral taxa, inferred predicted functional potential, explored virus‐bacteria co‐abundance networks and developed exploratory predictive models to evaluate whether viral and bacterial signatures may serve as candidate non‐invasive biomarkers for LC.

2. Material and Methods

2.1. Participants, Sequencing Protocol and Data Preprocessing

In this study, we reanalyzed 231 publicly available faecal bulk shotgun metagenomic sequencing samples downloaded from the European Bioinformatics Institute database under the accession code PRJEB6337 [19]. Based on the original two‐stage design of PRJEB6337, the discovery cohort comprised 96 LC patients and 83 healthy controls after quality control, whereas the validation cohort comprised 21 LC patients and 31 healthy controls. In addition, we downloaded and reanalyzed faecal metagenomic data from another project under the accession code PRJEB65440 [20], including 31 LC patients and 48 healthy controls, as an additional independent validation set.

In the present reanalysis, raw reads were preprocessed using fastp software (v0.20.1) [21] with the following parameters: ‐l90 ‐q20 ‐u30 ‐y –trim_poly_g to remove low‐quality sequences. Quality‐filtered reads were aligned to the human reference genome GRCh38 using bowtie2 (v2.4.1) [22], and host‐derived contaminants were subsequently removed using default parameters. Therefore, our virome analysis was based on viral sequences recovered from bulk faecal metagenomes rather than purified viral‐like particle (VLP) fractions. Because VLP‐enriched virome sequencing and bulk shotgun metagenomic sequencing can capture different viral fractions and produce distinct viral community profiles, this methodological distinction was considered when interpreting the virome results and is further discussed as a limitation [23, 24].

2.2. Gut Virome Profiling and Analyses

A comprehensive gut virus catalogue, termed the Chinese Gut Viral Catalogue (cnGVC) [25], was constructed from over 10 000 publicly available faecal metagenomes and includes more than 90 000 nonredundant viral operational taxonomic units (vOTUs). High‐quality reads from all samples were mapped to the cnGVC database using Bowtie2 with a nucleotide similarity threshold of 95% to define viral ‘species‐level’ distinctions [22]. To generate the abundance profile of vOTUs in each faecal sample, we aggregated the reads mapped to each vOTU and normalized the relative abundance by the total number of mapped reads in each sample. Subsequently, the relative abundance at the viral family level was determined by aggregating the relative abundances of vOTUs assigned to the same family. This study implemented the latest taxonomic framework released by the International Committee on Taxonomy of Viruses (ICTV, https://ictv.global/) [26] for systematic annotation of viral sequences.

For assessing the gut virome diversity, we analysed the profiles at the vOTU level. The total number of observed vOTUs was determined by counting the number of vOTUs with a relative abundance greater than zero. To further quantify the diversity of the gut virome, Shannon diversity indices were calculated using the diversity function of the R vegan package.

2.3. Functional Annotation of the Viral Genomes

We performed functional annotation of the vOTUs using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database [27]. To achieve this, the viral protein‐coding genes for each vOTU were aligned against the KEGG database using DIAMOND [28] with the following options ‘‐query‐cover 50 ‐subject‐cover 50‐e 1e‐5 ‐min‐score 50 ‐max‐target‐seqs 50’. Then, each protein was assigned a KEGG ortholog (KO) based on the best‐hit protein in the database. These annotations were interpreted as predicted functional potential rather than experimentally verified viral activity.

2.4. Gut Bacteriome Profiling

The prokaryotic composition of the gut microbiome in faecal metagenomes from all samples was profiled using the MetaPhlAn4 algorithm [29]. Species‐level relative abundances were generated by MetaPhlAn4 and used for downstream bacterial community analyses. Family‐level profiles were obtained by aggregating the relative abundances of species assigned to the same family.

2.5. Construction of Cross‐Kingdom Co‐Occurrence Network

Cross‐kingdom co‐abundance networks between bacterial species and vOTUs were constructed as exploratory association analyses. Bacterial‐vOTU co‐abundance patterns were analysed using Spearman's rank correlation with a stringent threshold (|r| > 0.6). Only taxa with a prevalence greater than 10% were included in the correlation analysis. The false discovery rate (FDR) method was utilized for p‐value correction to improve the reliability of the results. Significant associations were visualized as interaction networks through ggraph (v2.2.1) and igraph (v2.0.3). Because metagenomic relative‐abundance data are compositional and sparse, these networks were interpreted as exploratory co‐abundance patterns rather than direct virus‐bacteria interactions. To assess the robustness of the Spearman‐based results, we additionally performed a FastSpar‐based sensitivity analysis using prevalence‐filtered viral and bacterial abundance profiles.

2.6. Statistical Analysis

All statistical analyses in this study were conducted on the R (v4.2.1) platform. Principal coordinates analysis (PCoA) of the Bray‐Curtis distance was carried out and visualized using the vegan package [30]. Permutational multivariate analysis of variance (PERMANOVA) was performed using the adonis function in the vegan package, with statistical significance assessed using 1000 permutations. For comparisons between two groups, Fisher's exact test was used for categorical variables, and the Wilcoxon rank‐sum test was used for continuous variables or microbial abundance comparisons, as appropriate. p‐Values were adjusted for multiple testing using the Benjamini‐Hochberg FDR method where applicable, and adjusted q‐values < 0.05 were considered statistically significant unless otherwise specified.

Random Forest classifiers were constructed using the randomForest package with 1000 trees to distinguish LC patients from HCs based on microbial abundance profiles. Candidate features included LC‐associated vOTUs and bacterial species identified in the discovery cohort. Feature importance was estimated within the discovery cohort, and models were retrained using increasing numbers of top‐ranked features to identify the feature set with the highest cross‐validated area under the receiver operating characteristic curve (AUC). This importance‐ranking and incremental feature‐search strategy was used to select the optimal vOTU‐only model and the optimal combined virome‐bacteriome model. Model performance was evaluated using AUC, sensitivity, specificity and accuracy where applicable.

3. Results

3.1. Structural Variations of the Gut Virome in LC Patients

In total, 860 Gb of high‐quality sequence data were obtained from a previous study involving 117 LC patients and 114 HC subjects [19]. The raw metagenomic dataset was quality‐filtered and decontaminated to remove human sequences. High‐quality reads were then mapped to the cnGVC catalogue, a comprehensive gut viral reference constructed from Chinese populations (comprising 93 462 vOTUs; see ‘Materials and Methods’). This allowed for the profiling of gut viral composition across 179 faecal samples (96 LC patients and 83 healthy controls) from the discovery cohort (Table S1). In total, 40 768 vOTUs were quantified from the dataset. Taxonomic annotation was performed using the latest framework released by the ICTV. Among the identified viral sequences, 17.5% (7135/40768) were classified into 38 established viral families. These taxonomically resolved vOTUs were used for subsequent in‐depth analyses.

We first assessed within‐sample viral richness and diversity in patients with LC compared to HC. At the vOTU level, LC patients exhibited a significant reduction in viral richness, as indicated by the number of observed vOTUs (Wilcoxon test, p < 0.001; Figure 1A). A comparable decrease in viral diversity was observed based on the Shannon index (Figure 1B). To evaluate inter‐sample differences in viral composition, beta diversity was calculated using Bray–Curtis dissimilarity. This analysis revealed a distinct separation between LC and HC samples (PERMANOVA R² = 0.0215, p = 0.001; Figure 1C), indicating an LC‐associated shift in the overall viral community structure. These results collectively highlight significant alterations in gut vOTU composition among LC patients. Furthermore, Adonis multivariate analysis was conducted to quantify the contributions of nine clinical indicators to virome variation between the LC and HC groups (Figure 1D). Among these factors, ‘Cirrhotic’ status accounted for the largest proportion of variance (R ² = 0.016), followed by albumin concentration (Alb, R ² = 0.0099) and hepatitis B virus‐related variables (HBV‐related, R ² = 0.0097), all showing strong statistical significance (p < 0.01). These findings collectively suggest that gut virome alterations are associated with LC status, although causal involvement in LC pathogenesis remains to be determined.

FIGURE 1.

Comparative analysis of gut viral communities in liver cirrhosis (LC) patients and healthy controls (HC). (A, B) Boxplots comparing vOTU richness (A) and Shannon index (B) of gut virome between LC and HCs. (C) PCoA analysis based on Bray‐Curtis distances shows the distribution of gut viruses at the vOTU level. Ellipsoids represent the 80% confidence interval for each group. (D) Contributions of clinical parameters to the explanatory power of gut virome differences between the two groups. (E) Stacked bar plot displaying relative abundance distribution of dominant viral families (Top10). (F) Boxplots illustrating difference of representative gut viome families between groups. Asterisks denote statistical significance levels (*p < 0.05, **p < 0.01, ***p < 0.001).

Comparative analysis of virome composition at the family level revealed dominant gut virus families including Microviridae, Winoviridae, Herelleviridae and Aliceevansviridae (Figure 1E). Ten viral families exhibited significant abundance differences between LC and HC groups, with Aliceevansviridae, Peduoviridae and Drexlerviridae being LC‐enriched, while Herelleviridae, Intestiviridae, Suoliviridae, Salasmaviridae, Guelinviridae, Crevaviridae and Phycodnaviridae were predominantly enriched in the HC group (Figure 1F, Table S2).

3.2. LC‐Associated Viral Signatures Identification and Function Analysis

Given the significant differences observed at the viral family level, we further identified distinct viral signals at the vOTU level. Applying stringent selection criteria (q < 0.05, mean relative abundance > 0.01, fold change > 1.2), a total of 473 vOTUs were found to differ significantly between LC patients and HC (q < 0.01, Figure 2A, Table S3). Among these, 59 vOTUs were enriched in the LC group, while 414 were enriched in the HC group.

FIGURE 2.

Gut viral signatures and functions and host associations. (A) The volcano plot displays the fold change versus q‐value of vOTUs (viral operational taxonomic units) with an average abundance greater than 1% across all samples. The X‐axis represents the ratio (log2 transformed) of vOTU abundance between LC and HC, while the Y‐axis shows the q‐values (−log10 transformed) of vOTUs. vOTUs enriched in LD are marked in red, and those enriched in HC are marked in blue. (B) Differences in the occurrence rate of KOs between LC‐enriched and HC‐enriched vOTUs. (C) Host distribution difference between the two groups.

To explore the predicted functional potential of these LC‐associated viral signals, we annotated the protein‐coding genes within all LC‐related vOTUs using the KEGG database [19]. This analysis identified 2393 protein‐coding genes with homologous KEGG entries, which were further categorized into 834 KEGG Orthology terms (KOs). Comparative analysis revealed nine KOs with significantly different occurrence rates between LC‐enriched and HC‐enriched vOTUs (Fisher's exact test, q < 0.05; Figure 2B, Table S4). We identified six differentially represented KOs between LC‐enriched and HC‐enriched viruses, including significantly lower occurrence rates of K00986 (RNA‐directed DNA polymerase) and K01358 (ATP‐dependent Clp protease) in LC‐enriched viruses, along with higher occurrence rates of K01185 (lysozyme), K02172 (bla regulator protein BlaR1), K00656 (formate C‐acetyltransferase) and K21528 (serine recombinase). Additionally, we observed distinct expression patterns for several genes in LC‐associated viruses, although these differences did not reach statistical significance in Fisher's test. Notably, K00558 (DNA (cytosine‐5)‐methyltransferase 1), K01356 (repressor lexA) and K01448 (N‐acetylmuramoyl‐L‐alanine amidase) showed notable enrichment in HC‐enriched viruses, while K08600 (sortase B), K07258 (serine‐type D‐Ala‐D‐Ala carboxypeptidase) and K12132 (eukaryotic‐like serine/threonine‐protein kinase) were preferentially enriched in LC‐enriched viruses.

Next, to explore putative phage‐host associations, we predicted prokaryotic host associations for the differential vOTUs using genome homology and CRISPR spacer matching against the UHGG database [31] (Figure 2C, Table S5). LC‐enriched vOTUs predominantly infected hosts from the genera Streptococcus, Bacteroides, Faecalicatena and Escherichia, whereas HC‐enriched vOTUs primarily targeted Bacteroides, Faecalibacterium, Blautia_A and UBA9502. Notably, Streptococcus and Escherichia hosts were exclusively found in LC‐enriched vOTUs, whereas the majority of Faecalibacterium served as hosts for HC‐enriched vOTUs. These findings suggest distinct predicted phage‐host association patterns between LC‐enriched and HC‐enriched vOTUs.

3.3. Composition of Gut Bacteria and LC‐Associated Bacterial Signatures

Comparative analysis of bacterial alpha diversity between the LC and HC groups demonstrated significantly lower values in both richness and Shannon index (p < 0.05; Figure 3A,B), reflecting reduced bacterial diversity in LC patients. PCoA of gut bacterial communities showed distinct clustering between the LC and HC groups (PERMANOVA R ² = 0.0337, p = 0.001, Figure 3C) consistent with the virome patterns and indicating substantial bacterial compositional differences associated with LC. Adonis analysis identified nine clinical parameters contributing to intergroup bacterial community differences (Figure 3D), with ‘Cirrhotic’ status showing the strongest association (R ² = 0.028, p = 0.001), followed by albumin levels (Alb, R ² = 0.0208, p = 0.001) and HBV‐related factors (R ² = 0.0175, p = 0.001). Secondary contributors included age and alcohol‐related parameters (p ≤ 0.01).

FIGURE 3.

Differences in gut bacterial communities between LC patients and HC group. (A, B) Boxplots showing the richness index (A) and Shannon diversity index (B) of gut bacterial communities at the species level between the two groups. (C) PCoA analysis based on Bray‐Curtis distances of gut bacterial composition. (D) Contributions of clinical parameters to the explanatory power of gut bacterial communities. (E) Stacked bar plot displaying relative abundance distribution of dominant bacteria families (Top10). (F) Boxplots displaying gut bacterial family‐level differences between the two groups. (G) Volcano plot illustrating the fold change versus q‐value of bacterial species with mean relative abundance greater than 1% across all samples.

At the family level, a total of 148 bacterial families were identified, with dominant families including Bacteroidaceae, Lachnospiraceae and unclassified Bacteroidales (Figure 3E). Among them, seven bacterial families showed differential relative abundance between LC and HC groups (q < 0.05, mean relative abundance > 0.1, fold change > 5). Specifically, unclassified Bacilli and unclassified Alphaproteobacteria were significantly enriched in the HC group, while five families, including Veillonellaceae, Enterobacteriaceae and Streptococcaceae, were enriched in LC patients (Figure 3F, Table S6). Species‐level comparative analysis identified 169 differentially abundant bacterial signatures associated with LC, including 46 species significantly enriched in LC patients and 123 species predominantly abundant in HC (q < 0.01, Figure 3G, Table S7).

3.4. Cross‐Kingdom Network Between Gut Viral and Bacterial Signatures

We performed an exploratory co‐abundance analysis between 473 LC‐associated vOTUs and 169 LC‐associated bacterial species. Abundance‐based networks were constructed in both the HC and LC groups using Spearman correlation analysis (|r| > 0.6 and q < 0.05) to explore virus‐bacteria co‐abundance patterns rather than direct interactions. In the HC group, the network contained 509 significant correlations (Figure 4A, Table S8), whereas the LC patients exhibited only 130 significant correlations (Figure 4B, Table S9). Certain bacterial species, such as Alistipes putredinis , Bacteroides uniformis and Oscillibacter sp. ER4, showed high connectivity in the HC network (Figure 4C). In contrast, four bacterial species, including Alistipes putredinis , Bacteroides uniformis , Ruminococcus bromii and Eubacterium rectale , showed relatively high connectivity in the LC network (Figure 4D). Notably, species such as Eubacterium rectale and Veillonella dispar showed increased network connectivity within the LC‐associated network, suggesting that they may represent candidate taxa involved in altered co‐abundance patterns in LC.

FIGURE 4.

Correlation analysis between gut bacteria and virus signatures in LC patient and HC. (A, B) Significant correlation network between gut virus and bacteria signatures in LC patients (A) and HC (B). (C, D) The bar chart displays the top 10 species in terms of connectivity in the HC correlation network (C) and LC correlation network (D).

Because metagenomic relative‐abundance data are compositional and sparse, we further performed a FastSpar‐based sensitivity analysis to evaluate the robustness of the Spearman‐based network. In this sensitivity analysis, only 18 viral taxa showed significant associations with Bacteroides uniformis in the HC network, whereas no virus‐bacteria associations remained statistically significant in the LC network (all p > 0.05). These results suggest that the Spearman‐based networks should be interpreted as exploratory co‐abundance patterns rather than robust evidence of direct virus‐bacteria interactions.

3.5. Diagnostic Potential of the Gut Viral and Bacterial Signatures

Finally, to evaluate whether gut virome features could distinguish between LC and HC groups, we constructed exploratory Random Forest classifiers based on 473 significantly associated LC‐related vOTUs identified in the discovery cohort. The vOTU‐based model achieved a cross‐validation area under the curve (AUC) of 0.909 (95% CI: 0.866–0.953) for differentiating LC patients from HC in the discovery cohort (Figure 5A). Incorporating gut bacterial features into the model modestly improved classification performance (AUC = 0.925). In the vOTU model, the top 10 most important features were LC‐enriched vOTUs (e.g., v16960, v69725, v43147). In the vOTU‐bacteria combined model, the highest‐ranking features were bacterial species, including Veillonella sp., Streptococcus sp. and Haemophilus parainfluenzae (Figure 5B,C). To generate an optimal set of gut microbial features for LC classification, we retrained the models using vOTU and bacterial features ranked by importance within the discovery cohort. The optimal vOTU model achieved an AUC of 0.923 with the top 71 vOTUs, while the vOTU‐bacteria combined model reached a maximum AUC of 0.938 with the top 98 vOTU/bacterial features (Figure 5D). These results suggest that gut virome features, alone or in combination with bacterial taxa, may have potential as candidate microbial signatures for distinguishing LC patients from HC.

FIGURE 5.

Classification and prediction of LC and HC groups based on gut virome and bacteriome features. (A) Random Forest models distinguishing LC patients from HC individuals using gut virome features (vOTU model) and combined virome–bacteriome features (vOTU‐bacteria combined model). (B, C) Top 20 most important features in the vOTU model (B) and vOTU‐bacteria combined model. (D) Model performance based on the number of microbial features, ranked by importance. Red lines represent the vOTU model, and green lines represent the vOTU‐bacteria combined model. Data points indicate the mean AUC from 10 replicate models, with error bars representing variance. (E) Predictive performance of the vOTU model and the vOTU‐bacteria combined model based on all features in the validation cohorts. (F) Predictive performance of the vOTU model and the vOTU‐bacteria combined model at their optimal feature counts on the validation cohorts.

To further evaluate model performance, we applied the trained models to the validation cohort from the original two‐stage PRJEB6337 study (comprising 21 LC patients and 31 HCs), as well as to an additional independent validation set from PRJEB65440 (comprising 31 LC patients and 48 HCs, see ‘Materials and Methods’). Predictions were performed using both the full‐feature vOTU model and the vOTU‐bacteria combined model, along with their respective optimal feature versions. In the PRJEB6337 and PRJEB65440 validation cohorts, the full‐feature vOTU model achieved AUCs of 0.911 (95% CI: 0.833–0.989) and 0.773 (95% CI: 0.669–0.878), respectively, while the full‐feature vOTU‐bacteria combined model reached AUCs of 0.946 (95% CI: 0.890–1.000) and 0.751 (95% CI: 0.637–0.865), respectively (Figure 5E). The optimal vOTU model (71 features) maintained comparable performance (AUC = 0.901; 95% CI: 0.833–0.909) and 0.758 (95% CI: 0.651–0.865) in the validation cohorts of PRJEB6337 and PRJEB65440, respectively, whereas the optimal vOTU‐bacteria combined model (98 features) achieved improved predictive accuracy, with AUCs of 0.965 (95% CI: 0.924–1.000) and 0.755 (95% CI: 0.646–0.865), respectively (Figure 5F). Overall, these findings suggest that both gut virome and combined virome‐bacteriome features exhibit promising yet preliminary discriminative performance across internal and external validation cohorts.

4. Discussion

Although the gut microbiota's hepatic impacts are well characterized through the gut‐liver axis, the contribution of the gut virome to LC‐associated microbial dysbiosis remains insufficiently explored, particularly regarding its co‐abundance patterns and predicted functional relationships with commensal bacteria [17]. Moreover, recent advancements in viral taxonomy and the substantial expansion of reference databases, including the updated cnGVC [25], provide an opportunity to revisit the LC‐associated gut virome with improved resolution. To address this gap, we comprehensively profiled the gut virome in LC patients, analysing its diversity, taxonomic composition, LC‐associated viral signatures, predicted functional potential and exploratory cross‐kingdom co‐abundance networks. Additionally, we developed exploratory predictive models based on viral and bacterial features to evaluate whether these microbial signatures may serve as candidate non‐invasive biomarkers for LC.

Our study revealed LC‐associated alterations in the gut virome of LC patients, characterized by a significant reduction in viral α‐diversity and notable shifts in community composition. In contrast to previous reports where no significant virome changes were observed [17], our findings may be partly attributable to the improved resolution provided by updated viral reference databases and differences in analytical strategies. However, because this study is based on bulk faecal metagenomic data, these results should be interpreted as viral signals recovered from shotgun metagenomes rather than as a complete profile of purified viral‐like particle fractions. The loss of viral diversity likely reflects a disruption of the phage–bacteria balance, where reduced phage diversity limits regulatory control over bacterial populations, contributing to microbial dysbiosis.

In parallel with virome alterations, we observed significant disruptions in the gut bacteriome, including decreased bacterial diversity and expansion of pathobionts such as Streptococcus, and Veillonella. These bacterial changes are consistent with previous findings in cirrhosis, where the overgrowth of opportunistic and pro‐inflammatory taxa has been linked to disease progression and complications [19]. Notably, our phage–host association analysis predicted shifts in predominant phage hosts among LC patients, particularly involving genera such as Streptococcus, and Prevotella. This reveals a strong virus‐bacteria dysbiosis interconnection, with viral community alterations demonstrating high concordance with corresponding bacterial compositional changes in this study and previous studies [19], but they should not be interpreted as direct evidence of virus‐bacteria interactions. The study suggests that Streptococcus spp. and Veillonella spp. are typical pro‐inflammatory microbes, while Prevotella species, well known for their involvement in plant polysaccharide metabolism [32, 33, 34]. These findings demonstrate that cirrhosis‐related gut dysbiosis involves both bacterial and viral disturbances, underscoring the need to study these microbial components together in liver disease development.

The coexistence of phages and bacteria in the intestine is dynamic and interdependent, and our findings suggest that viruses, particularly bacteriophages, may actively shape the dysbiotic ecosystem in LC patients [18]. In this study, decreased expression of RNA‐directed DNA polymerase (K00986) in LC‐enriched viruses was observed, consistent with previous studies [17]. Conversely, the upregulation of lysozyme (K01185), blaR1 (K02172), formate C‐acetyltransferase (K00656) and serine recombinase (K21528) in LC‐enriched viruses reflects enhanced bacterial lysis, antibiotic resistance and genetic rearrangement. The increased lytic activity likely contributes to bacterial turnover and community restructuring, potentially depleting commensal bacteria and releasing microbial products. One consequence of such enhanced bacterial lysis is the release of endotoxins and other pathogen‐associated molecules from bacterial cells. In the context of cirrhosis, this release can exacerbate inflammation and immune activation through the gut‐liver axis [35]. Collectively, these functional differences highlight distinct viral ecological strategies between LC patients and healthy individuals. In LC, the viruses seem to adopt a more aggressive, lytic lifestyle, potentially synergizing with bacterial hosts to drive disease pathogenesis.

Moreover, the discovery of a distinct virome signature in LC opens new avenues for clinical translation. Microbiome‐based diagnostics have already shown promise in liver diseases. Qin et al. [19] identified gene‐based microbial biomarkers that distinguished cirrhosis patients from healthy controls with high accuracy. In fact, as few as 15 microbial genes were sufficient to discriminate cirrhosis in their cohort, underscoring the diagnostic power of microbiota profiles [19]. Our findings suggest that incorporating viral markers could further enhance such diagnostic tools. The gut virome may provide additional, non‐redundant information beyond bacterial composition alone. For instance, recent work in non‐alcoholic fatty liver disease demonstrated that decreased faecal viral diversity and specific shifts in phage taxa correlate with advanced fibrosis, and a predictive model combining virome features with clinical data outperformed models based on clinical or bacterial data alone [36]. Advances in metagenomic technologies have rendered virome analysis clinically feasible, though its implementation still necessitates validation in large‐scale cohorts and demonstration of added value over existing clinical indicators. Future development may focus on gut virome profiling or combined bacterial‐viral biomarker panels as novel tools for non‐invasive diagnosis and treatment response monitoring.

Despite the promising findings, this study has several limitations. First, although we accounted for individual factors such as gender, age, body mass index and selected clinical parameters, other potential confounders—such as diet, medication use, lifestyle and environmental exposures—were not included due to limited data availability. Numerous studies have demonstrated that these variables, especially dietary patterns and physical activity, significantly influence both the gut microbiota and virome [37, 38, 39]. Second, the cross‐sectional design of this study prevents us from determining whether the observed virome alterations are a cause or a consequence of liver cirrhosis. To resolve this, future studies should incorporate prospective longitudinal designs and mechanistic experiments in animal models to help clarify causal relationships between virome changes and LC development.

5. Conclusion

This study emphasizes the crucial role of the gut virome in LC pathogenesis and demonstrates its potential as a novel diagnostic marker. We observed significant changes in the gut virome of LC patients, including reduced viral α‐diversity and shifts in community structure, which were associated with specific functional gene alterations. The virome in LC patients was marked by the enrichment of phages targeting pro‐inflammatory bacteria and the upregulation of genes involved in bacterial lysis and antibiotic resistance. Our findings suggest that a combined approach targeting both the virome and bacteriome could offer new strategies for diagnosing and managing cirrhosis. The development of a predictive model based on these microbial features, with strong performance in both internal and external validation, underscores the potential for incorporating the gut virome in clinical practice. However, further longitudinal studies and mechanistic investigations are essential to establish causal relationships and explore therapeutic applications, ultimately enhancing our understanding and treatment of cirrhosis.

Author Contributions

Huanhuan Chang: Data curation, formal analysis, investigation, methodology, validation, visualization, writing – original draft, writing – review and editing. Yan Yang: Data curation, formal analysis, investigation, methodology, validation, visualization, writing – original draft, writing – review and editing. Pan Zhang: Data curation, funding acquisition, formal analysis, investigation, methodology, validation, visualization, writing – original draft, writing – review and editing. Zhixin Lei: Data curation, formal analysis, investigation, methodology, validation, visualization, writing – original draft, writing – review and editing. Yuetong Zhang: Data curation, resources, writing – review and editing. Shenghui Li: Data curation, resources, writing – review and editing. Lumin Wang: Data curation, writing – review and editing. Ying Wang: Data curation, writing – review and editing. Jue Jiang: Data curation, writing – review and editing. Lu Li: Conceptualization, project administration, supervision, writing – review and editing. Haitao Shi: Conceptualization, project administration, supervision, writing – review and editing. Ameng Shi: Conceptualization, funding acquisition, project administration, supervision, writing – review and editing. Huanhuan Chang, Yan Yang, Pan Zhang and Zhixin Lei contributed equally to this work and share first authorship. Lu Li, Haitao Shi and Ameng Shi contributed equally to this work and share corresponding authorship. All authors meet the ICMJE criteria for authorship. All authors have read and approved the final version of the manuscript and agree to be accountable for all aspects of the work.

Funding

This work was supported by the Natural Science Basic Research Program of Shaanxi Province (Grant No. 2025JC‐YBQN‐1179) and the Scientific Research Supporting Fund of the Second Affiliated Hospital of Xi'an Jiaotong University (No. 2025SCIPT‐63).

Disclosure

Use of large language models, AI and machine learning tools: None declared.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Sample information from 96 LC patients and 83 healthy individuals.

Table S2: Detailed information of the 10 differential families identified by Wilcoxon rank‐sum test (FC > 1.2, mean relative abundance > 0.01, pvalue < 0.05).

Table S3: Detailed information of the 473 differential vOTUs identified by Wilcoxon rank‐sum test (FC > 1.2, mean relative abundance > 0.01, pvalue < 0.05).

Table S4: Detailed information of the 834 KOs.

Table S5: Host Information for differential vOTUs.

Table S6: Detailed information of the seven differential bacteria families identified by Wilcoxon rank‐sum test (FC > 5, mean relative abundance > 0.1, qvalue < 0.05).

Table S7: Detailed information of the 169 differential species identified by Wilcoxon rank‐sum test (FC > 1.2, mean relative abundance > 0.01, qvalue < 0.05).

Table S8: Details of the HC group cross‐kingdom network identified by the Spearman method (|r| > 0.6, qvalue < 0.05).

Table S9: Details of the LC patients cross‐kingdom network identified by the Spearman method (|r| > 0.6, qvalue < 0.05).

Chang H., Yang Y., Zhang P., et al., “Disrupted Gut Viral‐Bacterial Ecology of Patients With Liver Cirrhosis,” Liver International 46, no. 9 (2026): e70836, 10.1111/liv.70836.

Data Availability Statement

The data that support the findings of this study are openly available in European Bioinformatics Institute database at https://www.ebi.ac.uk/, reference number PRJEB6337 and PRJEB65440.