Gut microbiome dysbiosis in hepatocellular carcinoma patients with persistent HCV viremia versus viral clearance: a cross-sectional study

This cross-sectional cohort study aimed to investigate changes in the gut microbiome composition among HCV-infected patients receiving DAA therapy and their association with HCC.

The study participants were stratified into three distinct groups according to HCV RNA status and hepatocellular carcinoma diagnosis: 1) HCC patients with confirmed HCV eradication following DAA therapy before study enrollment (THCC); 2) HCC patients with persistent HCV viremia (RHCC); and 3) well-matched healthy controls without HCC or HCV infection. The DAA-treated cohort demonstrated a median follow-up duration of 2 years post-treatment, ensuring the assessment of long-term microbial patterns rather than transient treatment effects.

The clinical investigations followed established methodological frameworks for hepatology microbiome research [24]. Our sample size determination, derived from power analyses of previous HCV microbiome studies [25], indicated that 40–50 participants per group would provide 80% power (α = 0.05) to detect moderate effect sizes (Cohen's d ≥ 0.5) in alpha diversity metrics.

Participant recruitment took place at the outpatient clinics of the Internal Medicine Department et al. Mabarrah Health Insurance Hospital in Zagazig, Egypt, between December 2023 and December 2024. The inclusion criteria were as follows: adult HCC patients (18–75 years) with documented HCV RNA status before and after DAA therapy; complete clinical records including specific DAA regimens (400 mg sofosbuvir plus 60 mg daclatasvir, with or without ribavirin following Egyptian national guidelines [26, 27]); and a minimum 2-year post-treatment interval to ensure the stabilization of treatment-related microbial changes.

Exclusion criteria were rigorously implemented to minimize confounding factors: HIV/HBV coinfection; recent antibiotic/probiotic use (within 90 days); active alcohol abuse (> 30 g/day); liver transplantation history; advanced cirrhosis (Child–Pugh C); uncontrolled diabetes; active dietary interventions; and immunomodulatory medication use. These criteria ensured that the observed microbial differences could be more confidently attributed to HCV and HCC status rather than extraneous factors.

All participants underwent comprehensive clinical assessments upon recruitment. HCC diagnosis followed international standards and Egyptian clinical guidelines, reflecting the high national burden of HCV-related HCC [29]. Clinical evaluations included liver imaging, laboratory tests, hematologic profiling, and noninvasive fibrosis assessment. Because formal fibrosis staging by biopsy or elastography was not consistently available at the time of HCC diagnosis, we utilized the FIB‑4 index and routine liver function tests (platelet count, ALT/AST, albumin, bilirubin, and the international normalized ratio (INR)) as alternative indicators of hepatic fibrosis and function. Advanced fibrosis was operationally defined as FIB‑4 > 3.25, a threshold with high specificity for advanced fibrosis in HCV cohorts. Abdominal ultrasound and transient elastography (FibroScan®, LSM > 12.5 kPa) were performed by certified operators (Echosens, France). Laboratory tests were conducted using the Roche Cobas 8000 system (Roche Diagnostics, Switzerland), and hematologic profiling was performed with Sysmex XN-series analyzers (Sysmex Corporation, Japan). Noninvasive fibrosis staging included FIB-4 scoring. Imaging-based diagnosis uses quadruple-phase CT or dynamic contrast-enhanced MRI, with lesion characterization based on the LI-RADS v2018 criteria [30, 31]. Nodules > 1 cm with arterial hyperenhancement and washout were considered diagnostic. In cirrhotic patients, AFP levels ≥ 200 ng/mL support the diagnosis [32]. Cirrhosis and portal hypertension were confirmed through consistent clinical and imaging findings.

Fresh stool samples were collected in sterile containers, transported on ice packs, and stored at − 80 °C until processing. Microbial DNA was extracted in batches using the Qiagen DNeasy PowerSoil Kit (Cat. No. 12888–100) following the manufacturer’s protocol. The integrity of the DNA was verified using 1% agarose gel electrophoresis, and the concentration was defined as the samples' absorbance values at 260 and 280 nm using a Nanodrop ND-1000 spectrophotometer (ND-1000; Thermo Scientific, Waltham, MA, USA). The hypervariable regions (V3-V4) of the bacterial 16S rRNA genes were amplified using Forward Primer 5' TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG and Reverse Primer 5 GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC (The underlined bp are Illumina adapters) [33]. Sequencing of 16S rRNA amplicon libraries was performed on an Illumina MiSeq platform according to the manufacturer’s instructions (Illumina, San Diego, CA, USA) at IGA Technology Services (Udine, Italy). Negative extraction controls and PCR blanks were included in each batch and sequenced alongside samples to ensure that no contamination occurred during DNA extraction or amplification. This quality control step confirmed that the observed microbial profiles represented true biological signals rather than artifacts introduced during sample processing.

Sequence processing utilized the DADA2 pipeline implemented in QIIME2 for denoising, chimera removal, and amplicon sequence variant (ASV) calling at 100% identity. Taxonomic assignment was conducted using the SILVA v138.1 reference database [34, 35] with a naive Bayes classifier trained at a 97% sequence identity threshold [36]. Alpha diversity metrics, Observed species (Obs), Chao1 (species richness), and the Shannon index (species richness and evenness), were calculated in QIIME2 after rarefying to 27,119 reads per sample (Additional file: Figure S1). Statistical comparisons of alpha diversity metrics and the F/B ratio across the three study groups were performed using the Kruskal–Wallis test. Pairwise comparisons between groups were conducted using the Mann–Whitney U test. All p values were corrected with Benjamini–Hochberg false discovery rate (FDR) for multiple testing to control the false discovery rate [37]. Beta diversity was assessed through Bray–Curtis distance (ASV-level composition) and both weighted and unweighted UniFrac distances (phylogenetic relationships), which were displayed through principal coordinate analysis (PCoA) plots. Statistical differences in microbial community structure between clinical groups were identified using permutational multivariate analysis of variance (PERMANOVA) with 999 permutations [38]. Differential abundance at the phylum and genus levels was analyzed using DESeq2 [39], which implements negative binomial Wald tests with Benjamini–Hochberg false discovery rate (FDR) correction. Microbial association analysis was performed using Maaslin2 (Multivariate Association with Linear Models), which applies a linear modeling framework to identify taxa significantly associated with clinical groups [40]. The genus-level OTU table was used as input data, and metadata included disease state-associated groups (Control, THCC, and RHCC), Age, and Sex. These variables were specified as fixed effects in the model to adjust for demographic confounding. Total Sum Scaling (TSS) normalization and log transformation were applied to stabilize variance and reduce compositional bias. Enterotyping was conducted using PAM clustering of Jensen-Shannon divergence distances from genus-level relative abundances, with the optimal cluster number determined by the silhouette width and Calinski-Harabasz index [41]. Microbial cooccurrences and associations with clinical parameters were assessed using Spearman correlation (r ≥ ± 0.3, p ≤ 0.05). The functional potential prediction utilized Tax4Fun with default parameters within the MicrobiomeAnalyst web platform [42‐44]. The SILVA v138 database served as the reference for taxonomic and functional annotation at the 97% sequence identity threshold for nearest neighbor identification. The predicted functional profiles were mapped to Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, resulting in relative KO (KEGG Orthology) abundance levels that were used for subsequent functional diversity analyses [45]. The icons used in Figure S4 were obtained from Pixabay (https://pixabay.com) and is free for use without attribution under the Pixabay license (https://pixabay.com/service/license/).

The identification of potential microbial classifiers associated with HCC was achieved using machine learning methods validated in microbiome research [46]. A random forest model (randomForest package; 500 trees, max depth = 5) was constructed using genus-level relative abundance data as input features. All the statistical analyses were conducted in R (v4.4.2). The key genera distinguishing the study groups were identified using the Random Forest classifier within the MicrobiomeAnalyst web platform [42, 43]. To minimize overfitting, nested cross-validation was implemented (5 outer folds, 10 inner folds) for model training and hyperparameter tuning, and performance was evaluated through AUC, precision-recall, and calibration metrics (Additional file: Table S1; Figure S2). Additional validation included an independent 30% holdout set and 1,000 permutation tests to assess model robustness.

The study cohort included three well-defined groups: HCC patients with persistent HCV viremia (RHCC, n = 46), HCC patients with HCV eradication (THCC, n = 46), and healthy controls (n = 46). Significant differences in liver function parameters were observed among the studied groups (Table 1). The control participants had normal liver function values. RHCC patients showed significant hepatic dysfunction, with significantly reduced albumin levels (3.0 ± 0.6 g/dL vs 4.3 ± 0.2 g/dL in controls; p < 0.001) and higher FIB-4 scores (6.55 ± 1.3 vs 1.13 ± 0.3; p < 0.001). THCC patients presented an intermediate phenotype, maintaining near-normal albumin (4.1 ± 0.4 g/dL) despite elevated AST (91 ± 29 IU/L).

RHCC patients exhibited the most advanced liver disease among the groups, with 100% meeting the criteria for cirrhosis, reflected in an elevated international normalized ratio (INR) of 1.58 ± 0.3 and significantly higher alpha-fetoprotein (AFP) levels of 312.64 ± 56.1 ng/mL; in contrast, THCC patients presented with residual hepatic damage, frequently presenting with advanced fibrosis, as indicated by a mean FIB-4 index of 6.04 ± 1.1. The liver function of the RHCC group was largely impaired and markedly more preserved than that of the RHCC group, as evidenced by a platelet count of 218 ± 47 × 10⁹/L, an international normalized ratio (INR) of 1.08 ± 0.1, an albumin level of 4.1 ± 0.4 g/dL, and a total bilirubin level of 1.2 ± 0.3 mg/dL.

Distinct ecological patterns were observed in the gut microbiome across the study groups. Healthy controls maintained the highest microbial diversity (Shannon index; 4.08 ± 0.39; Chao1 2909.22 ± 442.89, Kruskal–Wallis; H = 14.37, p = 0.00076). Compared with RHCC patients, THCC patients presented intermediate microbial diversity (Mann–Whitney test, U = 82.17, padj = 0.0367) but greater compositional variability (Mann–Whitney test, U = 75.5, padj = 0.0283) (Fig. 1A). RHCC patients exhibited significant microbial depletion (Chao1 2113.83 ± 580.97, Mann–Whitney U test; U = 93.6, padj = 0.00462). The Chao1 richness index effectively distinguished between HCC subgroups (THCC and RHCC, Mann–Whitney; U = 100.5, p = 0.0018; AUC = 0.81, 95% CI 0.72–0.90).

Bild vergrößern

The gut microbiome exhibited a progressive deterioration that parallels HCC clinical biomarkers, with healthy controls demonstrating optimal gut-liver axis homeostasis, as reflected by strong negative correlations between microbial diversity and HCC biomarkers (p < 0.001) (Fig. 1B). Compared with controls, RHCC patients presented more pronounced microbial dysbiosis; U = 95.5, padj = 0.00368), which correlated with marked clinical decompensation, including portal hypertension (platelets 76 ± 24 × 10³/µL, p < 0.001) and synthetic dysfunction (INR 1.58 ± 0.21, p < 0.001). THCC patients retained moderate microbial diversity (12–18% reduction versus controls), which was significantly inversely correlated with the FIB-4 score (r = −0.08, p = 0.014).

Beta diversity analysis revealed a significant separation among the groups (PERMANOVA: F = 1.5415, R² = 0.22866, p = 0.0016), which reflects the increased heterogeneity in RHCC patients compared with the tighter clustering of healthy controls and THCC patients (Fig. 1C). This compositional shift occurred independently of alpha diversity metrics and was closely linked to established clinical indicators of disease progression (FIB-4 scores; r = 0.62, p = 0.003).

A total of 2,374 bacterial ASVs were assigned to 21 phyla, 54 classes, 142 orders, 263 families, and 758 bacterial genera. Phylum-level analysis revealed significant differences among the control, THCC, and RHCC groups (Kruskal–Wallis; H = 18.76, p = 0.0034) (Fig. 2A). The relative abundance of Bacteroidetes was comparable between controls (41.5 ± 14.1%) and THCC patients (42.7 ± 15.2%) but was significantly lower in RHCC patients (32.3 ± 12.0%) (Kruskal–Wallis; H = 25.8, p = 2.49 × 10⁻⁴). Firmicutes levels were significantly greater in RHCC patients (50.2 ± 16.8%) than in both controls (40.9 ± 6.5%) and THCC patients (41.0 ± 15.9%).

Bild vergrößern

Notably, RHCC patients also presented unique microbial features, including elevated Cyanobacteria (1.1 ± 1.7% vs 0.1 ± 0.3% in controls; padj = 0.018). Furthermore, the exclusive presence of Elusimicrobia in RHCC patients (0.9 ± 1.7%; padj = 0.009 vs controls) indicates a possible microbial signature unique to HCV-persistent HCC. The most pronounced shift was observed in Spirochaetes, which were nearly undetectable in controls and minimally present in THCC (0.3 ± 1.1%) but significantly enriched in RHCC (2.9 ± 6.8%; padj = 0.007 vs RHCC). Although Proteobacteria did not differ significantly between the groups, a trend toward reduced abundance in RHCC (8.2 ± 9.0%) compared with that in the controls (14.2 ± 11.9%) was noted.

The Firmicutes/Bacteroidetes (F/B) ratio exhibited distinct disease-associated trends, with RHCC patients showing the highest F/B ratio (1.55) compared with controls (1.05, Mann–Whitney test, U = 87.32, p = 0.00079) and THCC patients (1.02) (Fig. 2B). Among THCC patients, the F/B ratio was negatively correlated with HCC-related biomarkers. On the other hand, the F/B ratio of RHCC patients was positively correlated with direct bilirubin levels (r = 0.68, p = 0.008), moderately correlated with total bilirubin (r = 0.48, p = 0.085) and exhibited distinct correlation patterns, with the F/B ratio showing a significant negative association with FIB-4 scores (r = −0.57, padj = 0.024) and AST levels (r = −0.50). These contrasting correlations imply that the pathophysiological mechanisms of gut‒liver axis dysfunction differ between THCC and RHCC disease states. The particularly strong correlation between elevated F/B ratios and increased direct bilirubin in RHCC patients may reflect specific impairments in the hepatic clearance function associated with this disease state.

Genus-level profiling revealed distinct and statistically significant microbial patterns in the gut microbiome that significantly differed between the study groups (Fig. 3A and B). Multivariate analysis using MaAsLin2, which controls for potential confounders (age and sex), identified several genera with significantly different abundances across the study groups, revealing distinct microbial landscapes associated with HCV status in hepatocellular carcinoma patients (Fig. 3C). A key finding was the profound and consistent depletion of the genus Comamonas in both HCC patient groups compared with the control group (THCC: coef = −3.83, padj = 8.41 × 10⁻⁶; RHCC: coef = −3.69, padj = 7.80 × 10⁻⁶). The RHCC exhibited a unique dysbiotic profile characterized by significant enrichment of Fournierella (coef = 4.79, padj = 7.16 × 10⁻⁴) and Asteroleplasma (coef = 7.53, padj = 0.00669) alongside depletion of Family XIII UCG-001 (coef = −2.62, padj = 0.0023), a pattern that also distinguished it from the THCC group. The RHCC-associated signature was positively associated with Erysipelotrichaceae UCG-004 (coef = 4.19, padj = 0.012), uncultured Peptococcaceae (coef = 2.26, padj = 0.030), WCHB1-41 metagenome (coef = 3.60, padj = 0.023), and uncultured Peptostreptococcaceae (coef = 2.14, padj = 0.027) and negatively associated with Finegoldia (coef = −1.52, padj = 0.010). In contrast, THCC presented a different microbial shift, marked not only by the shared depletion of Comamonas but also by a significant depletion of Collinsella (coef = −3.20, padj = 0.0035) and a distinct enrichment of Fournierella (coef = 3.76, padj = 0.0081), alongside other notable depletions, including Subdoligranulum (coef = −2.18, padj = 0.0125), Ruminococcaceae UCG-003 (coef = −2.87, padj = 0.0159), Escherichia/Shigella (coef = −3.15, padj = 0.0280), and the Eubacterium ruminantium group (coef = −4.39, padj = 0.0199). Furthermore, Random Forest machine learning identified Asteroleplasma, Moryella, Lachnoclostridium, Fournierella, Eubacterium xylanophilum, Succinivibrio, and depletion of Faecalibacterium as the top classifiers distinguishing RHCC patients from THCC patients and controls (Additional file: Figure S3).

Bild vergrößern

Furthermore, DESeq2 analysis identified Asteroleplasma as the most enriched genus in RHCC patients compared with healthy controls (log₂FC = + 2.8, padj = 0.008) and THCC patients (log₂FC = + 1.9, padj = 0.02), whereas Faecalibacterium, a key butyrate producer, was significantly depleted (log₂FC = −2.1, padj = 0.006) (Fig. 3D). These taxonomic shifts correlated strongly with clinical markers (Fig. 4), with Asteroleplasma abundance positively associated with ALT (r = 0.55) and AFP (r = 0.52), and Faecalibacterium inversely correlated with FIB-4 scores (r = −0.68, p = 0.008).

Bild vergrößern

Three distinct enterotypes were identified, each of which presented unique microbial and clinical profiles (Additional file: Table S2). The persistent HCV viremia-associated enterotype (ET-R) was predominant in RHCC patients and characterized by enrichment of Asteroleplasma and other lipopolysaccharide (LPS)-producing taxa, along with depletion of SCFA-generating bacteria. Clinically, ET-R was associated with elevated serum ALT and AST levels, increased total bilirubin, and decreased albumin. The SVR-associated enterotype (ET-T), which is more common in THCC patients, displayed transitional features with intermediate Prevotella 9 abundance, partial recovery of Ruminococcaceae, and reduced Asteroleplasma levels. This enterotype correlated with improved liver function compared with ET-R, as indicated by lower ALT/AST and bilirubin levels and higher albumin concentrations. In contrast, the control enterotype (ET-C), observed in healthy individuals, maintained a balanced microbiome dominated by Bacteroides and Faecalibacterium. ET-C is linked to normal liver enzyme levels and optimal biochemical profiles.

Analysis of microbial co-occurrence networks in the RHCC and THCC microbiomes revealed distinct yet overlapping ecological patterns, highlighting both shared and cohort-specific features of dysbiosis (Fig. 5). In the RHCC microbiomes (Fig. 5A), Bacteroides exhibited strong positive correlations with beneficial taxa, including Faecalibacterium (r = 0.6) and Lachnoclostridium (r = 0.7), while showing antagonistic relationships with potential pathobionts, such as Succinivibrio (r = −0.5) and Dialister (r = −0.3), reflecting a fragmented network structure. In contrast, the THCC microbiome demonstrated a more balanced ecological profile (Fig. 5B), with Bacteroides maintaining robust associations with Bifidobacterium (r = 0.4) and Parabacteroides (r = 0.6), along with moderate associations with Faecalibacterium (r = 0.6), suggesting partial recovery of mutualistic networks. Both cohorts shared key dysbiotic features, including antagonistic interactions between Prevotella_9 and Ruminococcaceae_UCG_002 (r = −0.6) and disrupted symbiotic relationships involving butyrate producers such as Faecalibacterium and Roseburia (r = 0.6), patterns that diverged markedly from healthy microbiomes. Pathogenic network motifs, particularly the Escherichia-Shigella-Sutterella association (r = 0.4), which may drive proinflammatory processes, although with cohort-specific variations in interaction strength, were evident in both groups. The RHCC microbiome was distinguished by pronounced ecological disruption, including strong negative correlations between commensals and pathobionts, whereas the THCC microbiome showed signs of network restoration through revived Bifidobacterium-Lactobacillus interactions (r = 0.3). Both groups exhibited hallmark features of HCC-associated dysbiosis (Fig. 5C), including depleted hepatoprotective Christensenellaceae_R_7 interactions and disrupted Ruminococcus-Lachnospira metabolic axes.

Bild vergrößern

The RHCC microbiome showed pronounced functional disruptions, including a 58% decrease in genes associated with butyrate synthesis genes (ko00620) versus controls (padj = 0.0032) and a 3.2-fold increase in genes linked to lipopolysaccharide (LPS) biosynthesis (ko00540, padj = 0.002). These functional alterations were significantly associated with clinical decline; notably, butyrate gene expression was inversely correlated with serum creatinine (r = −0.45), whereas LPS gene abundance was positively correlated with AST levels (r = 0.52). Compared with RHCC patients, THCC patients presented partially restored metabolic profiles, with a 42% recovery in butyrate biosynthesis pathways (padj = 0.013) and significant inverse correlations between Bifidobacterium abundance and ALT levels (r = −0.31, padj = 0.002).