Characterization of the intrahepatic microbiome in patients with HBV-related end-stage liver disease

ABSTRACT

Background

Dysregulated microbiota is a hallmark of end-stage liver disease (ESLD). This study aimed to elucidate the intrahepatic microbiome in hepatitis B virus (HBV)-related ESLD.

Methods

We collected liver tissue samples from patients undergoing liver transplantation due to decompensated cirrhosis (DC) (n = 20) or acute-on-chronic liver failure (ACLF) (n = 24), as well as 18 samples from donors. Metatranscriptomic sequencing was performed to profile liver microbiome and transcriptome.

Results

2208 bacterial species were detected across 13 phyla and 165 genera. Metatranscriptomic profiling revealed that Proteobacteria and Actinobacteria dominated the intrahepatic microbiome, with Escherichia coli and Pseudomonas most prevalent across groups. Principal coordinate analysis showed distinct microbial community structures among donors, DC, and ACLF patients. Compared with donors, both groups exhibited increased abundance of Bacteroides heparinolyticus, Moraxella osloensis, and Gardnerella vaginalis, while ACLF patients were further enriched with Alcaligenes faecalis and Burkholderia insecticola, and DC patients had higher B. heparinolyticus. Most taxa originated from the gut, with additional oral- and respiratory-derived species. Despite similar abundance between groups, E. coli in ESLD displayed marked functional activation, including nutrient acquisition systems and virulence factors linked to adhesion, invasion, and toxin production. Integrated host – microbiome analysis revealed taxa-specific associations with impaired hepatic metabolic, immune, and structural integrity.

Conclusion

This study delineates the compositional and functional reprogramming of the intrahepatic microbiome in patients with ESLD and its coupling with liver metabolic, immune, and structural pathways. These findings suggest the intrahepatic microbiome as a promising therapeutic target for ESLD.

HIGHLIGHTS

• Metatranscriptomic profiling revealed marked compositional and functional dysregulation of the intrahepatic microbiome in end-stage liver disease (ESLD).

• Opportunistic taxa, including including Escherichia coli with increased transcriptional activity were enriched and showed associations with virulence-associated functions, immune signatures, and metabolic disruption.

• Integrated host – microbiome analysis revealed taxa-specific associations with impaired hepatic metabolic, immune, and structural integrity.

Introduction

Decompensated cirrhosis (DC) and acute-on-chronic liver failure (ACLF) represent advanced stages of chronic liver disease, characterized by severe impairment of liver function, systemic inflammation, multi-organ injuries/failure, and poor outcome [1,2]. The pivotal role of the gut-liver axis in the pathogenesis of DC and ACLF has been well recognized [3,4]. Patients with DC and ACLF present intestinal dysbiosis, which is characterized by reduced microbial diversity, overgrowth of pathogenic bacteria such as Enterobacteriaceae, Escherichia coli (E. coli) and Klebsiella pneumoniae (K. pneumoniae), and depletion of beneficial commensal species such as Lactobacillus, Bifidobacterium, and Faecalibacterium prausnitzii [5–8]. Such dysbiosis is linked to increased intestinal permeability, translocation of bacteria and their products such as LPS and microbial metabolites [9,10], which contribute to systemic inflammation, hepatic complications such as hepatic encephalopathy, and organ injuries or failure [11].

However, while the gut microbiome has been extensively studied, the intra-hepatic microbiome remains poorly understood [12]. Emerging evidence suggests that the liver is not a sterile environment but harbors a unique microbial population that may contribute to liver disease pathogenesis. These bacteria were believed to be largely derived from intestinal translocation due to intrinsic anatomical linkage to the liver via the portal vein [13–15], and interact with parenchymal and non-parenchymal cells, contributing to hepatic inflammation and hepatocarcinogenesis. For example, it was reported that the proteobacteria was enriched in the liver of obese patients and associated with lobular and portal inflammation [16]. And intratumoural bacteria, such as K. pneumoniae, Cutibacterium, Stenotrophomonas maltophilia, Fusobacterium, promoted the development of HCC by impairing immune surveillance and antitumour capacity of local immune cells, and contributing to a chronic inflammatory local environment [17–20].

Given that microbial translocation is frequent in patients with DC and ACLF, understanding the intrahepatic microbiome in these end-stage liver diseases (ESLD) may be crucial to provide new insights into the pathophysiological mechanisms and explore the potential for microbiome-based interventions. This study aims to investigate microbial diversity, compositional shifts, functional alterations of intrahepatic microbiome in patients with decompensated cirrhosis and ACLF by metatranscriptomic sequencing, as well as potential associations with changes of host transcriptome.

Materials and methods

Patients

We enrolled 20 patients with DC and 24 patients with ACLF, all on a background of cirrhosis due to Hepatitis B virus (HBV) infection, who underwent liver transplantation at the First Affiliated Hospital of Zhejiang University School of Medicine between May 2023 and December 2023. ACLF was diagnosed by EASL-CLIF criteria [21]. For the control group (Ctrl), liver tissues were collected from 18 donors who underwent living-related liver transplantation. We excluded patients with confirmed or suspected malignancies, prior organ transplants, pregnancy, and severe extra-hepatic diseases. The study was conducted following the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the First Affiliated Hospital of Zhejiang University School of Medicine (2022–1105), with informed consent obtained from all participants or their legal representatives. Detailed demographic, clinical, and laboratory data of the enrolled subjects are provided in Table 1. The Severity score was calculated as previously described [22–24]:

Table 1.

Baseline characteristics of Ctrl, DC, and ACLF patients.

	Ctrl N = 18	DC N = 20	ACLF N = 24	P value
Age, median (IQR)	48 (42–55)	53 (48–60)	51 (22-57)	0.181
Gender, Female no. (%)	4 (22.22%)	3 (15.00%)	4 (16.67%)	0.877
Complications at admission – no. (%)
Ascites	–	10(50.00%)	18(75.00%)	0.120
Hepatic encephalopathy	–	3(15.00%)	21(87.50%)	<0.001
Gastrointestinal bleeding	–	4(20%)	3(12.50%)	0.684
Bacterial infection	–	3(15%)	10(41.67%)	0.096
Laboratory data, median (IQR)
TB	0.55 (0.44–0.79)	1.26 (0.83–2.49)	23.95 (16.12–31.64)	<0.001
INR	1.01 (0.97–10.03)	1.29 (1.17–1.73)	3.14 (2.26–3.64)	<0.001
Creatinine	0.68 (0.61–0.83)	0.84 (0.70–0.92)	0.58 (0.47–0.75)	0.05
BUN	4.15 (3.40–4.97)	4.54 (3.84–7.12)	3.95 (1.48–6.99)	0.340
ALT	12.50 (9.00–19.75)	24.50 (13.75–36.50)	81.00 (34.50–135.25)	<0.001
AST	15.00 (12.75–17.25)	29.00 (21.25–41.00)	75.50 (51.75–134.00)	<0.001
Albumin	46.05 (43.50–46.68)	34.70 (29.40–40.58)	34.20 (31.00–37.88)	<0.001
WBC count	5.90 (5.48–7.18)	2.56 (1.92–3.59)	11.06 (5.10–11.52)	<0.001
NLR	1.69 (1.34–2.11)	3.52 (2.41–5.32)	10.60 (5.96–14.83)	<0.001
Platelet count	271.50 (212.25–307.25)	48.0 (29.25–84.50)	62.0 (41.00–86.50)	<0.001
Hemoglobin	137 ± 17.14	101 ± 28.57	93.38 ± 25.26	<0.001
Serum sodium	142.00 (140.00–143.25)	142.00 (137.75–143.00)	137.00 (135.00–140.00)	<0.001
HBsAg	–	180.11 (0.02–193.96)	2374.87 (133.64–11865.33)	<0.001
Treatments – no. (%)
Antibiotics	–	20 (100.00%)	23 (95.83%)	1.000
Vasopressors	–	0 (0.00%)	11 (45.83%)	<0.001
Diuretics	–	1(5.00%)	16 (66.67%)	<0.001
Albumin	–	18 (90.00%)	24 (100.00%)	0.201
PPIs	–	20 (100.00%)	23 (95.83%)	1.000
ALSS	–	0 (0.00%)	19 (79.17%)	<0.001
RRT	–	0 (0.00%)	12 (50.00%)	<0.001
Mechanical ventilation	–	0 (0.00%)	12 (50.00%)	<0.001
Organ Failure——no. (%)
Liver	–	–	23 (95.83%)	–
Coagulation	–	–	17 (70.83%)	–
Kidney	–	–	12 (50.00%)	–
Cerebral	–	–	15 (62.50%)	–
Lung	–	–	12 (50.00%)	–
Circulation	–	–	11(45.83%)	–
ACLF Grade——no. (%)
1	–	–	3 (12.5%)	–
2	–	–	5 (20.83%)	–
3	–	–	16 (66.67%)	–
Severity Score——median (IQR)
MELD	–	10.84 (8.75–15.46)	31.90 (26.74–34.83)	<0.001
MELD-Na	–	10.84 (8.75–15.55)	31.90 (27.90–35.73)	<0.001
ClIF-C ACLF	–	–	59.83 (48.18–65.34)	–

ACLF, acute-on-chronic liver failure; ALT, alanine aminotransferase; ALSS, Artificial liver support system; AST, aspartate aminotransferase; BUN, blood urea nitrogen; CLIF-C ACLF, Chronic Liver Failure- Consortium ACLF; Ctrl, control; DC, decompensated cirrhosis; HBsAg, Hepatitis B surface antigen; INR, international normalized ratio; MELD(−Na), model for end-stage liver disease(−sodium); NLR, neutrophil-to-lymphocyte ratio; PLT, platelet count; PPIs, proton pump inhibitors; RRT, renal replacement therapy; TB, total bilirubin; WBC, white blood cell.

NOTE. Data are presented as mean ± standard deviation (SD) or median [interquartile range, Q1–Q3] for continuous variables, and as number (percentage) for categorical variables. Normality was assessed with the Shapiro – Wilk test and homogeneity of variances with Levene’s test. For two-group comparisons, we used Student’s t test when variances were equal, Welch’s t test when variances were unequal, and the Mann – Whitney U test for non-normal data. For three-group comparisons, we used one-way ANOVA with Tukey’s post hoc test (equal variances) or Welch’s ANOVA with Games – Howell post hoc test (unequal variances); for non-normal data, the Kruskal – Wallis test with Dunn’s post hoc test (Holm adjustment) was applied. Categorical variables were compared using the χ² test or Fisher’s exact test when expected cell counts were < 5. All tests were two-sided with α = 0.05.

CLIF-C-ACLFs : 10 * [0.33 * CLIF-organ failure score(OFs) + 0.04 * age +0.63 * ln (white blood cell [WBC]) −2].

MELDs: 3.78 * ln [TB (mg/dL)] + 11.2 * ln (INR) + 9.57 * ln[serum creatinine (mg/dL)] +6.434.

MELD-Nas: MELD +1.59 (135–serum sodium).

Intrahepatic microbiome

Nucleic acid extraction, library construction, and sequencing

Total genomic DNA and RNA were extracted from liver tissue using the NG550-TIANMicrobe Magnetic Patho-DNA/RNA Kit (TIANGEN, China). Briefly, 200–300 μl of processed liver tissue was homogenized in pre-filled grinding tubes with 150 μl tissue digestion buffer (MGHA), 150 μl lysis buffer (MGHH-A), and 30 μl Buffer ST. The mixture was vortexed for 2 min and homogenized using the TGrinder H24 tissue homogenizer (OSE-TH-01) at 6 m/s for 30 sec, followed by a 30 sec interval, repeated for 6 cycles. After centrifugation at 12,000 rpm for 1 min, the supernatant was transferred to a new tube. Protein digestion was performed by adding 20 μl Proteinase K and incubating at 65°C for 10 min. Nucleic acids were bound to magnetic beads by adding 350 μl isopropanol and 15 μl magnetic bead suspension (MGHE), followed by vortexing and incubation at room temperature for 10 min, with vortexing every 5 min. The beads were separated on a magnetic stand, and the liquid was removed. Washing steps included resuspending the beads in 750 μl buffer MGDF and 750 μl wash buffer MPWG with ethanol, each followed by vortexing and magnetic separation. The beads were air-dried for 5–10 min to remove residual ethanol. Nucleic acids were eluted in 50–100 μl RNase-Free ddH2O by incubating at 56°C for 5 min, with gentle shaking every 2 min. The eluted nucleic acids were transferred to a new tube and stored at −20°C for further use. Subsequently, the GR-RNA Nucleic Acid Detection Kit-4 (BGI, Cat. No. Z000100193) was used for metatranscriptomic library construction. The libraries were sequenced on the DNBSEQ-T10 platform with paired-end 100 bp reads.

Quality control and host removal

The average number of reads per sample was 379 million. Quality control was performed using fastp (version 0.23.2) with default parameters. This included the removal of adapter sequences, trimming of low-quality bases (quality score < 20), and filtering of reads shorter than 50 bp. Additionally, fastp generated comprehensive quality reports, including read length distribution, base quality scores, and GC content, to ensure the integrity of the sequencing data. To remove host-derived reads, the cleaned reads were aligned to the human reference genome hg38 (GenBank assembly accession: GCF_000001405.15) using BWA-MEM (version 0.7.17) with default parameters. The alignment results were processed using SAMtools (version 1.12) to extract unmapped reads. Specifically, the parameter -f 12 was used to retain only read pairs where both ends failed to align to the human genome, ensuring the removal of all host-derived sequences. On average, host-derived sequences accounted for 98% of the reads after quality control, indicating a high proportion of human genomic contamination.

Microbial alignment and taxonomic profiling

Non-human sequences were aligned and classified using Kraken2 (version 2.1.2) with default parameters and the Minikraken v2 database (a pre-built database for microbial classification). To ensure reliable microbial detection, only taxa with 10 or more aligned reads were considered positive detections. Taxonomic profiles were generated at the species, genus, and phylum levels, and relative abundances were calculated for downstream analysis.

Negative controls and water blank – based filtering

To monitor and mitigate potential reagent/environmental contamination – an essential consideration in low-biomass liver tissue profiling-three negative controls (water blanks/blank extraction controls) were processed and sequenced in parallel with the study samples in the same batch using an identical workflow (nucleic acid extraction, metatranscriptomic library preparation, and sequencing). For taxa detected in the water blanks, we applied a blank-informed normalization and filtering step. Specifically, for each taxon in each sample, a Water Normalization Score (WNS) was calculated as the relative abundance of the taxon in the sample divided by its relative abundance in the water blanks (mean across blanks). Taxa with WNS ≥ 1 were retained, whereas taxa with WNS < 1 had their read counts set to 0 in that sample to reduce background signals observed in negative controls. This procedure was used to support conservative downstream interpretation in the setting of low microbial biomass.

Functional metabolic and GO enrichment analysis

Functional metabolic profiling of microbial sequences was performed using HUMAnN3 (version 3.6) with default parameters. This included the identification of microbial metabolic pathways and gene families based on the UniRef90 and MetaCyc databases. Pathway abundances were normalized to copies per million (CPM) for comparative analysis. To functionally characterize the microbial metabolic pathways identified by HUMAnN3, we mapped UniRef90 features to their corresponding Gene Ontology (GO) terms using the clusterProfiler R package (version 4.6.0). For each GO term, we calculated aggregate statistical measures by averaging both p-values and log2 fold change values across all associated UniRef90 features.

Diversity and differentiation analysis

Alpha diversity metrics, including the Chao1 (richness), Simpson (dominance), and Shannon (diversity) indexes, were calculated based on the relative abundance of microbial taxa. Additionally, reads per million (RPM) normalization was applied to account for sequencing depth variations across samples. Beta diversity was assessed using the Bray-Curtis dissimilarity metric, and Principal Coordinate Analysis (PCoA) was performed to visualize the microbial community structure differences between groups. PCoA plots were generated using the R package vegan. To identify microbial taxa and functional pathways that were significantly different between groups, Linear Discriminant Analysis Effect Size (LEfSe) analysis was performed using the Galaxy web platform (https://huttenhower.sph.harvard.edu/galaxy/). The analysis was conducted with a linear discriminant analysis (LDA) score threshold of 2.0 and a p-value cutoff of 0.05 to highlight biologically relevant features.

Liver transcriptome

Alignment to human transcripts and quantification

The cleaned reads were aligned to human transcripts using the STAR aligner (version 2.7.10a) with default parameters. The human reference transcriptome was obtained from the GENCODE database (version 38). STAR was chosen for its high accuracy and efficiency in aligning RNA-seq data to reference transcripts. Gene expression quantification was performed using HTSeq-count (version 0.13.5) with the union mode to count reads overlapping with annotated genes. The human gene annotation file (GTF format) was obtained from the GENCODE database (version 38). This step generated raw read counts for each gene, which were used for downstream differential expression analysis.

Differential gene expression analysis and functional enrichment

Differential gene expression analysis was conducted using the DESeq2 package (version 1.36.0) in R. The raw read counts were normalized using the median of ratios method, and the negative binomial distribution was used to model the count data. Genes with an absolute log2 fold change (log2FC) >1 and an adjusted p < 0.05 were considered significantly differentially expressed. The results were visualized using volcano plots and heatmaps to highlight key genes and expression patterns. To explore the biological significance of the differentially expressed genes, Gene Ontology enrichment analyses were performed using the clusterProfiler R package (version 4.6.0). Enriched terms and pathways were identified with a significance threshold of p-value < 0.05 and q-value < 0.2. To further interpret the GO enrichment results, the pairwise_termsim function from the enrichplot R package (version 1.18.0) was used to calculate the similarity between enriched GO terms based on their semantic relationships. The resulting similarity matrix was then visualized using the emapplot function, which generates an enrichment map where nodes represent GO terms and edges represent their semantic similarity. The size of the nodes corresponds to the number of genes associated with each GO term, and the color intensity reflects the significance of the enrichment (-log10(p-value)). This approach provides a comprehensive overview of the functional relationships among the enriched GO terms.

Integration analysis

Correlations between intrahepatic microbiome and transcriptome

To explore the interaction between the intrahepatic microbiome and host gene expression, Spearman’s correlation analysis was performed in R. The relative abundance of microbial taxa (normalized to reads per million, RPM) was correlated with log2-transformed host gene expression counts (from DESeq2). Significant correlations were defined as those with a BH-FDR < 0.05 and a correlation coefficient ≥ 0.6 or ≤−0.6. The results were visualized using a heatmap (generated with the pheatmap R package, version 1.0.12).

Statistical analysis

Statistical analyses were performed using SPSS (version 26.0; IBM, Armonk, NY, USA), GraphPad Prism (version 9.3.1), and R (version 4.0.5). Comparisons between two groups were conducted using an unpaired Student’s t test or the Mann – Whitney U test, as appropriate, depending on data distribution. Categorical variables were compared using the chi-square (χ²) test. For comparisons among three or more groups, one-way analysis of variance (ANOVA) was applied, followed by the least significant difference (LSD) post hoc test for pairwise comparisons. All tests were two-sided, and p < 0.05 was considered statistically significant. For analyses involving multiple comparisons, p values were adjusted using the Benjamini – Hochberg false discovery rate (FDR) procedure, with FDR-adjusted p < 0.05 considered significant.

Results

Clinical characteristics of patients

The study enrolled 18 controls, 20 patients with HBV-DC and 24 patients with HBV-ACLF (Figure 1). The demographic and clinical profiles of the study subjects are listed in Table 1.

Figure 1.

Study design and workflow for intrahepatic metatranscriptome profiling.

Liver tissue samples were obtained from 62 participants, including Controls (Crtl, n = 18), patients with decompensated cirrhosis (DC, n = 20), and patients with acute-on-chronic liver failure (ACLF, n = 24). Total DNA and RNA were extracted and subjected to metatranscriptomic sequencing. Following quality control, host-derived sequences were removed, and microbial reads were mapped for downstream analyses. Microbial diversity and functional profiles were determined, and bacterial markers were identified. Clean reads were also mapped to the human transcriptome to perform differential gene expression and Gene Ontology (GO) enrichment analyses. Feature genes were integrated with microbial markers to conduct host – microbe association analyses.

There were no significant differences in the distribution of sex or age among the groups. As expected, the ACLF group had a more severe abnormal liver test panel, hyper-bilirubinemia and coagulopathy compared to the Ctrl and DC groups, as evidenced by laboratory results and scores of disease severity. Among the 44 HBV-ESLD patients, ascites was the most common complication at admission (28/44, 63.6%), followed by hepatic encephalopathy (24/44, 54.5%), bacterial infection (13/44, 29.5%) and gastrointestinal bleeding (7/44, 15.9%). In the HBV-ACLF group (n = 24), liver failure was most common (23/24, 95.83%), followed by coagulation (17/24, 70.83%), cerebral (15/24, 62.50%) kidney (12/24, 50.00%), respiratory (12/24, 50.00%), and circulatory failure (11/24, 45.83%). Reflecting overall severity, ACLF grade 3 predominated (16/24, 66.67%), with grade 2 in 5/24 (20.83%) and grade 1 in 3/24 (12.50%).

Antibiotics use was comparable between DC and ACLF group (DC vs ACLF: 100% v.s. 95.83%, p = 1.000). However, the ACLF group required more intensive organ support. Compared to the DC group, ACLF patients received more vasopressors (45.83% vs 0%), diuretics (66.67% vs 5.00%), artificial liver support (79.17% vs 0%), renal replacement therapy (50.00% vs 0%), and mechanical ventilation (50.00% vs 0%) (all p < 0.001).

Liver microbial composition

Sequencing of the 62 liver samples yielded a total of 23.5 Gb of raw data, with an average of 379 million reads per sample. After the quality filter, 22.1 G (357 M per sample) clean reads were retained for subsequent analysis. In all samples, 2208 bacterial species were detected, belonging to 13 phyla and 165 genera. At the phylum level, Proteobacteria and Actinobacteria were most predominant (Figure 2(A)). At the species and genus levels, E. coli and Pseudomonas were the most abundant taxa across all groups, respectively (Figure 2(B,C)). Although a slight stepwise decrease in the median Shannon index was observed from the control group to the DC and HBV-ACLF groups, no statistically significant differences were found in overall alpha diversity, including the Shannon, Chao1, and Simpson indices (Figure 2(D-F)). In contrast, PCoA of of beta diversity revealed a significant separation in microbial community structure among the three groups (Figure 2(G)). Further analysis using LEfSe identified specific bacteria that were differentially enriched among the groups (Figure 2(H)). Specifically, compared to Ctrl, Bacteroides heparinolyticus, Moraxella osloensis, and Gardnerella vaginalis exhibited relatively higher abundances at the species level in both DC and ACLF patients. Notably, the liver microbiota of ACLF patients was further enriched with Alcaligenes faecalis and Burkholderia insecticola compared to both DC patients and Ctrl, whereas Bacteroides heparinolyticus was more abundant in DC than in ACLF.

Figure 2.

Taxonomic composition and diversity of the intrahepatic microbiome across disease stages.

(A – C) Relative abundance of intrahepatic microbiota at the phylum (A), genus (B), and species (C) levels in Controls (Ctrl), decompensated cirrhosis (DC), and acute-on-chronic liver failure (ACLF).

(D – F) Comparisons of α-diversity between groups using Shannon (left), Chao1 (middle), and Simpson (right) indices. Statistical significance was evaluated using the Kruskal – Wallis test followed by Dunn’s post hoc test with Benjamini – Hochberg false discovery rate (FDR) adjustment; p values are indicated.

(G) Principal coordinates analysis (PCoA) based on Bray – Curtis dissimilarity showing clustering patterns of intrahepatic microbial communities among the three groups.

(H)Linear discriminant analysis effect size (LEfSe) identifying bacterial taxa differentially enriched in each group. Significance was determined using the default LEfSe criteria (Kruskal – Wallis and Wilcoxon rank-sum tests, p < 0.05) coupled with a logarithmic LDA score threshold of >2.0.

To evaluate the impact of blank-informed WNS filtering on key taxa, we compared taxon abundance before and after WNS correction for representative bacteria and assessed within-group differences using Wilcoxon signed-rank tests (Supplementary Table S10). Gardnerella vaginalis and Burkholderia insecticola were not detected in the water blanks, whereas all detected E. coli signals had WNS > 1 and were therefore retained. For three additional key taxa, WNS correction did not produce significant within-group abundance shifts, indicating that the blank-aware filtering step did not materially alter their group-level distribution patterns. The corrected results remained consistent with the original observations: Moraxella osloensis was enriched in DC and ACLF relative to controls, Bacteroides heparinolyticus was more enriched in DC, and Alcaligenes faecalis was more enriched in ACLF. These findings support the robustness of the major disease-associated taxonomic patterns after WNS filtering.

Contribution of intestinal, oral and respiratory tract bacteria in the livers of patients with DC and ACLF

Overall, we identified 291 bacteria species from 17 genera within the liver tissues that are known to originate from the gut, oral and respiratory tracts (Supplementary Table S1). The top 10 most frequently detected species from each of these three body sites were presented in Figure 3(A-C). Among gut-associated bacteria, E. coli was the most frequently detected (100%), followed by Salmonella enterica (98%) and Escherichia fergusonii (77%). In oral-derived microbes, Corynebacterium sphenisci showed the highest detection rate (94%), with Actinomyces sp. oral taxon 414 ranking second (85%). Among respiratory tract bacteria, Streptococcus suis had the highest prevalence (66%), followed by Streptococcus mutans (56%) and Streptococcus sobrinus (42%). Notably, the total relative abundance of these bacteria of extrahepatic origin was higher in DC patients than in ACLF patients (Figure 3(D)). Analysis of the top 10 bacterial detection rates from different body sites revealed a higher prevalence of pathogenic bacteria originating from the gut and respiratory tract. Among respiratory tract-associated bacteria, members of the Streptococcus genus were particularly dominant.

Figure 3.

Source-specific microbial taxa detected in the liver across study groups.

(A-C) Top bacterial species detected in the liver originating from the gut (A), oral cavity (B), and respiratory tract (C). Detection rates are shown for each taxon. Taxa labels are color-coded to indicate pathogenic species (red), commensal species (blue), or undefined status (black). Bar colors correspond to bacterial genera.

(D) Relative abundance of intrahepatic bacteria according to putative source (gut, oral cavity, respiratory tract, or multiple sources) in the total cohort and in Controls (Ctrl), decompensated cirrhosis (DC), and acute-on-chronic liver failure (ACLF).

Metabolic functional differences of Escherichia coli between the liver with DC or ACLF and controls

To investigate the functional potential of the intrahepatic microbiome, we performed an inferred metagenomic analysis using HUMAnN3. For this analysis, ACLF and DC patients were combined into ESLD group. Compared to Ctrl, the ESLD group showed a significant upregulation of 177 microbial gene families, while only 8 were upregulated in Ctrl (Figure 4(A)). Among these, we identified 61 HBV related and 16 E. coli related gene families with significantly higher expression in ESLD. Interestingly, while the overall abundance of E. coli did not differ significantly between the groups (Supplementary Figure S1), its associated functional gene families were markedly upregulated in the ESLD group, suggesting differential functional expression despite similar detection levels.

Figure 4.

Differential microbial gene expression and functional enrichment in end-stage liver disease (ESLD).

(A) Volcano plot showing differentially expressed microbial genes in Ctrl (n = 18) vs patients with ESLD (n = 44). Significance thresholds were set at BH-FDR – adjusted P (q-value) <0.05 and |fold change| >2.

(B) Violin plots displaying expression levels (log₂-transformed) of representative microbial genes across controls (Ctrl), decompensated cirrhosis (DC), and acute-on-chronic liver failure (ACLF), including virulence-associated genes from Enterobacter sakazakii, enterotoxigenic E. coli, and uropathogenic E. coli. Statistical differences across the three groups were assessed using the Kruskal – Wallis test with Benjamini – Hochberg FDR correction.

(C) Gene Ontology (GO) enrichment analysis of differentially expressed microbial genes, highlighting significantly enriched biological processes, molecular functions, and cellular components, with dot size representing log₂ fold change and color indicating GO category.

A closer examination of the 16 upregulated E. coli-associated gene families revealed several key functions related to bacterial adaptation and virulence (Figure 4(B)). These included the ilv operon leader peptide, which enhances bacterial survival and growth by regulating isoleucine biosynthesis [25], and a D-methionine ABC transporter ATP-binding protein, involved in nutrient acquisition and stress response. Furthermore, we also identified numerous virulence factors homologous to those from well-characterized pathogenic strains, including uropathogenic E. coli (UPEC, strain UTI89, adhesion, invasion and immune evasion factors), enterotoxigenic E. coli (ETEC, strain E24377A, toxin production and colonization genes), and Cronobacter sakazakii (Enterobacteriaceae, stress resistance and biofilm formation proteins). Notably, one protein showed homology to a Clonorchis sinensis cell wall hydrolase, potentially contributing to tissue degradation. Collectively, these findings reflected an enhanced pathogenic potential of intrahepatic E. coli in ESLD, which is characterized by [1] improved colonization via adhesion molecules and biofilm formation [2], nutrient acquisition and stress resistance adaptations [3], tissue damage through hydrolytic enzymes, and [4] potential drug resistance development via efflux mechanisms and metabolic reprogramming.

To understand the impact of these microbial symbionts on the host, we performed a GO enrichment analysis (Figure 4C). The analysis revealed significant enrichment of viral processes, particularly viral nucleocapsid formation and genome replication, suggesting active virus (e.g. HBV) activity. Key symbiont-host interaction pathways were also highlighted, including the activation of host NF-kappaB signaling, which promotes inflammation, and the induction of G2/M cell cycle arrest, which can contribute to hepatocyte dysfunction. Notably, this crosstalk was further evidenced by the disruption of host cellular compartments, with significant enrichment of terms related to nuclear and mitochondrial targeting, which implied widespread interference with energy metabolism and oxidative stress pathways. Molecular function analysis revealed symbiont exploitation of host nucleic acid binding machinery (RNA binding and DNA binding) for replication and persistence.

To evaluate whether the enhanced functional activity of Escherichia coli is associated with clinical severity, we performed Spearman’s rank correlation analyses between the abundance/expression of 16 E. coli function-related genes and a panel of clinical indices reflecting disease severity and systemic inflammation (including MELD and routine hematology/biochemistry variables). After Benjamini – Hochberg false discovery rate (FDR) correction for multiple testing, no gene – clinical association remained statistically significant (all FDR-adjusted p > 0.05; Fig. S2 and Table S11). Although several nominal associations were observed prior to correction, the corresponding effect sizes were modest (|ρ| <0.6). For instance, platelet count (PLT) showed the smallest nominal p value with the gene U_Enterobacter_sakazakii (ρ = −0.38, p = 0.002), which did not remain significant after FDR adjustment (FDR = 0.64). Collectively, these findings suggest that E. coli functional activation may exhibit weak, non-robust trends with certain clinical indices, but we did not observe strong univariate correlations under multiple-testing control in the current cohort.

To test whether the functional signals identified in the binary HC vs ESLD comparison were robust to subgroup separation, we performed pairwise DESeq2 analyses across the three groups (HC vs DC, HC vs ACLF, and DC vs ACLF) using the same thresholds as in the original HC vs ESLD analysis (Supplementary Fig. S3A – C ; Supplementary Table S12). In the comparisons against HC, 935 and 122 microbial genes were upregulated in DC and ACLF, respectively, and six E. coli-related genes overlapped with those identified in the original HC vs ESLD analysis (Fig. S3A,B), supporting the reproducibility of the previously observed ESLD-associated E. coli functional signal.GO enrichment analysis further showed that the most significant terms enriched in both HC vs DC and HC vs ACLF were highly concordant and mainly involved key steps related to symbiont/host interaction and infection-associated processes, including attachment, membrane fusion, and viral membrane structure (Fig. S3D,E). These results suggest that these pathways may represent shared core functional features across DC and ACLF relative to HC. Although the enriched GO categories were broadly similar, the ACLF vs HC comparison generally showed larger absolute log2 fold changes and stronger enrichment significance than DC vs HC, indicating progressive enhancement of these microbial functional programs with disease severity. In the DC vs ACLF comparison, 767 differentially expressed microbial genes were identified, including nine E. coli-related genes that overlapped with the previously identified ESLD-associated gene set (Fig. S3C). GO enrichment of DC – ACLF differential genes was more focused on downstream molecular processes related to viral infection, including viral genome replication, nucleic acid degradation, and regulation of gene expression (Fig. S3F), suggesting a stage-dependent shift toward more active replication/transcription-associated programs (or intensified host – virus/microbe response dysregulation) in ACLF. Collectively, these pairwise subgroup analyses support the robustness of the original HC vs ESLD findings while refining them into shared ESLD functional signatures and ACLF-enriched molecular-stage features.

Taken together, these findings highlight multifaceted host – microbe crosstalk in ESLD and nominate candidate microbial features and host pathways for future validation and therapeutic exploration.

Inter-correlation between changing liver transcriptome and altered bacteria in the livers of DC or ACLF

Transcriptional analysis of liver tissue revealed progressive gene expression alterations across disease stages. Compared to Ctrl, DC patients had modest transcriptional changes, with 14 upregulated and 47 downregulated DEGs (Figure 5(A)). In contrast, HBV-ACLF patients showed a much more dramatic dysregulation, with 1257 upregulated and 151 downregulated DEGs (Figure 5(B)). Despite this difference in scale, a core set of 18 shared DEGs (10 upregulated and 8 downregulated) was identified across both conditions, pointing to common pathological processes (Supplementary Table S2).

Figure 5.

Differential microbial gene expression and functional features in cirrhosis and ACLF compared with controls.

(A) Volcano plot of differentially expressed host genes in DC vs Ctrl using DESeq2; significance was defined as BH-FDR – adjusted P (q-value) <0.05 with |FC| >2.

(B) Volcano plot of differentially expressed microbial genes in ACLF vs Ctrl using DESeq2; significance was defined as BH-FDR – adjusted P (q-value) <0.05 with |FC| >2.

(C) Gene Ontology (GO) enrichment analysis of differentially expressed genes between DC and Ctrl groups. Point color indicates regulation direction (red: upregulated in DC; blue: downregulated in DC), shape represents gene set specificity (circles: unique to this comparison; triangles: co-regulated with another group), and size reflects Gene Ratio. X-axis shows -log₁₀(adjusted p-value).

(D) GO enrichment analysis of differentially expressed genes between ACLF and Ctrl.

(E) Heatmap of correlations between DC signature taxa and DC-associated differentially expressed host genes. Correlations were computed using Spearman’s rank correlation, and p values were adjusted across tested pairs using the Benjamini – Hochberg FDR method; only associations meeting the prespecified thresholds (|r| ≥0.6 and BH-FDR – adjusted p < 0.05) are shown.

(F) Heatmap showing correlation analysis between signature microbiota of ACLF and differentially expressed host genes in ACLF compared to Ctrl. Similar to panel E, correlations were computed using Spearman’s rank correlation with Benjamini – Hochberg FDR adjustment. Only associations meeting the thresholds of |r| ≥0.6 and FDR-adjusted p < 0.05 are shown.

Pathway enrichment analysis highlighted molecular changes common to both disease states (Figure 5(C,D)), Supplementary Table S3-S9). We detected 5 co-upregulated GO terms (mainly involving mechanical stress response, extracellular matrix remodeling, and membrane restructuring) and 8 co-downregulated terms (primarily affecting metabolic processes, particularly lipid and amino acid metabolism). Four shared KEGG pathways were activated, including signal transduction, cell-matrix interactions, immune responses, and lipid metabolism. Notably, the pathogenic E. coli infection pathway was significantly enriched in both groups, which corroborates our earlier findings on E. coli-associated functional alterations.

Comparative pathway analysis also revealed stage-specific molecular signatures in liver disease progression. DC patients exhibited unique upregulation of monoamine response pathways (e.g. cellular response to monoamine stimulus) which has been related to hepatic encephalopathy, and downregulated Fc receptor signaling associated with cirrhosis-associated immune defect. In contrast, ACLF patients exhibited more extensive pathway dysregulation, including activation of the Wnt signaling linked to the disruption of metabolic-immune-inflammatory homeostasis, as well as upregulated TGF-β and focal adhesion pathways associated with fibrogenesis, and metabolic disturbance evidenced by impaired β-oxidation, cholesterol metabolism, and glutathione detoxification systems.

Integrative analysis of LDA-selected bacterial taxa and host transcriptomes revealed distinct host-microbiome interactions in DC and ALCF patients. In DC patients, the downregulation of hepatic genes mediating protein trafficking (YIPF4), anticoagulation (PROS1), redox balance (CYB561A3), cytoskeletal integrity (FAM65A), immune differentiation (MAF), and GPCR signaling (ADRBK2) showed strong negative correlations with Rhodoferax ferrireducens abundance, suggesting microbiome-associated, multifactorial hepatic dysfunction (Figure 5(E)). Particularly noteworthy was Bacteroides heparinolyticus, whose abundance correlated positively with suppression of PCYT2 (phospholipid metabolism), implicating heparin degradation in metabolic disturbances and gut barrier compromise.

In the ACLF group, downregulation of ACAA1 (fatty acid metabolism), showed positive correlations with Alcaligenes faecalis and Gardnerella vaginalis, suggesting a potential contribution of these taxa to fatty acid metabolic disturbances in ACLF patients (Figure 5(F)). The upregulation of NFKB2, a key player in the inflammatory signaling pathway, was positively correlated with Alcaligenes faecalis. The upregulation of IDH3G, a gene linked to the tricarboxylic acid (TCA) cycle, was positively correlated with Moraxella osloensis, pointing to it potential involvement in energy metabolism disruptions. These findings suggested that these bacteria may exacerbate metabolic abnormalities, which are commonly observed in ACLF patients. Alcaligenes faecalis, Gardnerella vaginalis, and Moraxella osloensis are commensal organisms that may translocate to the liver from extrahepatic sites. Their associations with key metabolic and inflammatory genes suggest a role in amplifying metabolic, immune, and inflammatory dysregulation in ACLF.

Discussion

This study, using metatranscriptomic sequencing, provides the first comprehensive characterization of dynamic changes and functional traits of the liver tissue microbiome along the clinical spectrum from Ctrl without advanced liver disease to decompensated cirrhosis and ACLF. We demonstrate that the intrahepatic microbiome is predominantly gut-derived, with additional contributions from oral and respiratory translocation. Importantly, in end-stage liver disease, both the composition and the functional gene expression profile of the intrahepatic microbiome are markedly altered, exemplified by the functional activation of pathogenic Escherichia coli. Integrative host – microbe analyses further reveal close correlations between microbiome alterations and hepatic metabolic and inflammatory transcriptomic signatures. Together, these findings extend current understanding of host – microbe associations in end-stage liver disease and suggest potential microbiome-related candidates for future biomarker or therapeutic exploration.

Contrary to the traditional view of the liver as a sterile organ, accumulating evidence highlights the intrahepatic microbiome as an integral and dynamic component of the liver microenvironment [12,16]. Under physiological conditions, both mice and humans harbor a distinct microbiome predominantly derived from the gut, which plays a key role in shaping liver immunity, particularly through members of the Bacteroidetes phylum. Consistent with this, our study detected common gut-derived microbiota across all liver tissue samples. Notably, we also identified bacteria originating from the oral cavity, such as Corynebacterium sphenisci and Actinomyces sp. oral taxon 414, as well as respiratory tract-associated bacteria including Streptococcus suis, Streptococcus mutans, and Streptococcus sobrinus in Ctrl livers.

The presence of oral- and lung-derived bacteria in donor livers reflects a complex, multi-origin intrahepatic microbiome. As a central immune organ and blood filter, the liver is continuously exposed to microbes translocated via both the portal and systemic circulation [6], maintaining a dynamic equilibrium with low-level microbial presence [26]. Physiologically, oral secretions are continuously swallowed, and respiratory tract bacteria may disseminate through mucosal surfaces or the bloodstream, enabling transient or low-level colonization of distant organs such as the liver. However, we cannot exclude the possibility that environmental and procedural factors during organ procurement and preservation may introduce or reveal microbes from various sources.

In our study, we observed a significant enrichment of gut-derived microbiota in patients with DC, which is consistent with the pronounced bacterial translocation typically associated with portal hypertension and compromised gut barrier function [4]. Supporting this, a study utilizing 16S rRNA sequencing of samples from the portal vein, hepatic vein, central vein, and peripheral vein in patients undergoing transjugular intrahepatic portosystemic shunt (TIPS) procedures demonstrated that bacteria translocate from the gut into the portal circulation and are subsequently filtered by the liver before entering systemic circulation [27]. This bacterial translocation is closely linked to the systemic inflammation characteristic of cirrhosis. In contrast, the liver microbiota of ACLF patients was enriched with opportunistic pathogens originating from extraintestinal sites, including Alcaligenes faecalis, Gardnerella vaginalis, and Moraxella osloensis. The presence of these bacteria likely reflects the immune paralysis that follows systemic inflammation in ACLF [9,23], which impairs host immune defenses and facilitates bacterial translocation from multiple body sites [14,15,24]. Consistent with this, our previous study identified signatures of several opportunistic pathogens in the peripheral blood of patients with HBV-related ACLF [28].

Previous studies have established that the intrahepatic microbiome is linked to liver immunity rather than being a passive bystander [12,20,29]. Our findings further indicate that intrahepatic microbial alterations are associated with features of disease severity and may inform testable hypotheses regarding host – microbe interactions. Specifically, certain bacteria correlate with metabolic dysregulation – for example, Alcaligenes faecalis and Gardnerella vaginalis associate with fatty acid metabolism disorders in ACLF patients, while Moraxella osloensis links to energy metabolism disturbances. Intrahepatic microbial signatures may also be associated with immune-related injury pathways; for instance, Alcaligenes faecalis abundance positively correlates with NFKB2 upregulation, consistent with heightened inflammatory signaling [30].

Remarkably, while E. coli abundance remains similar between patients and controls, its functional gene expression is significantly upregulated in ESLD. This includes activation of survival and nutrient acquisition genes [31,32] (e.g. the ilv operon, D-methionine ABC transporter) and virulence factors from uropathogenic and enterotoxigenic strains, which facilitate adhesion, invasion, and toxin production [33,34]. Functional analyses reveal that E. coli expresses tissue-degrading enzymes and that overall microbiome activity associates with hepatocyte organelle damage and cell cycle arrest, contributing to liver dysfunction. Correspondingly, host transcriptomes show enrichment of the “pathogenic Escherichia coli infection”pathway. Together, these data suggest that in ESLD, E. coli exhibits increased transcription of virulence-associated programs, consistent with a more activated opportunistic profile; however, these findings reflect inferred functional activity and do not by themselves establish pathogenic phenotypes or direct tissue injury.

Despite the novel insights provided by our study, several limitations should be acknowledged. First, the exclusive HBV etiology, relatively small sample size, cross-sectional design may limit generalizability and preclude causal inference regarding whether intrahepatic microbial alterations contribute to the progression from DC to ACLF. Second, because the intrahepatic microbiome represents a low-biomass environment, there remains a risk of environmental or reagent-derived contamination [35], We mitigated this risk through strict experimental procedures and comprehensive bioinformatic filtering (including removal of up to 98% host-derived sequences), but residual contamination cannot be completely excluded. Third, living donor liver tissue and explanted ESLD livers differ in procurement context, ischemia time, and perioperative exposures, which may influence host RNA profiles and, in a low-biomass setting, also affect background microbial signals; therefore, case – control contrasts should be interpreted cautiously, acknowledging potential handling-related effects. Fourth, our analyses may be confounded by clinical interventions and disease severity: most ESLD patients – particularly those with ACLF – received antibiotics and intensive organ-support therapies that could alter microbial transcriptional activity and host immune/inflammatory responses. Because these exposures were highly prevalent and tightly correlated with severity in our cohort, residual treatment-related confounding cannot be fully excluded. Fifth, functional and virulence-related interpretations in this study are derived from annotation- and expression-based inference rather than direct phenotypic validation; accordingly, we describe these findings as virulence-associated signatures or inferred functional activation, and orthogonal validation will be required to confirm pathogenic phenotypes. Finally, prospective, large-scale, multicenter studies with standardized sampling/handling, granular perioperative and treatment documentation, and inclusion of additional etiologies (e.g. alcohol-related liver disease or metabolic dysfunction-associated steatotic liver disease, MASLD) will be essential to validate these microbial signatures and strengthen inference.

Although our study is observational, the heightened transcriptional activity of opportunistic taxa and their correlation with systemic inflammatory signatures highlight their potential as biomarkers for stratifying disease severity and predicting short-term risk in end-stage liver disease. Furthermore, our findings provide a rationale for future interventional trials to evaluate whether modulating microbiome-associated inflammation – such as through antimicrobial stewardship or microbiome restoration – can improve clinical outcomes [36]. These hypotheses align with the emerging paradigm that microbiome-derived signals drive systemic inflammation and organ dysfunction, underscoring the need for prospective validation of microbiome-informed biomarkers and therapies [37].

In conclusion, this study is the first to characterize the compositional and functional dysregulation of the intrahepatic microbiome in ESLD patients using metatranscriptomic sequencing. Our findings reveal that microbial alterations closely correlate with host clinicopathological features and the hepatic microenvironment, supporting the view that the intrahepatic microbiome is closely associated with disease-related host states rather than representing a purely passive bystander. These insights position the intrahepatic microbiome as a promising source of novel prognostic biomarkers and therapeutic targets in ESLD.

Funding Statement

This work was supported by the National Key R&D Program of China [No. 2022YFC2304500 to Y.Y.], the Key R&D Program of Zhejiang [No.2025C02131 to Y.Y.], the National Natural Science Foundation of China [No.82570717 to Y.S. and No. 82300690 to S.Y.], Ningbo Science and Technology 2035 Plan Project [No. 2025Z186 to H.J.] and State Key Laboratory for Diagnosis and Treatment of Infectious Diseases [No. 202503 to Y.S.].

List of abbreviations

ESLD

end-stage liver disease

DC

decompensated cirrhosis

ACLF

acute-on-chronic liver failure

HBV

Hepatitis B virus

ALT

aspartate aminotransferase

BUN

blood urea nitrogen

CLIF-C ACLF

Chronic Liver Failure-Consortium ACLF

HBsAg

Hepatitis B surface antigen

Ctrl

control

INR

international normalized ratio

MELD(−Na)

model for end-stage liver disease(−sodium)

NLR

neutrophil-to-lymphocyte ratio

PPIs

proton pump inhibitors

RRT

renal replacement therapy

TB

total bilirubin

WBC

white blood cell.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the First Affiliated Hospital of Zhejiang University School of Medicine (2022–1105).

Consent for publication

All patients or their next of kin provided written informed consent.

Data availability statement

The data reported in this paper have been deposited in the OMIX, China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/omix, accession no.OMIX011919). The original clinical data used in this article have been deposited in the Science Data Bank and are available at https://doi.org/10.57760/sciencedb.29617 [38,39].