Alterations in the Gut Microbiome in the Progression of Cirrhosis to Hepatocellular Carcinoma

Abstract

Hepatocellular carcinoma (HCC) is the second leading cause of cancer-related mortality worldwide. While cirrhosis is the main risk factor for HCC, the factors influencing progression from cirrhosis to HCC remain largely unknown. Gut microbiota plays a key role in liver diseases; however, its association with HCC remains elusive. This study aimed to elucidate microbial differences between patients with HCC-associated cirrhosis (HCC-cirrhosis) and cirrhotic patients without HCC and healthy volunteers and to explore the associations between diet, lifestyle, and the microbiome of these patients. Fecal samples and food frequency questionnaires were collected from 95 individuals (30 HCC-cirrhosis patients, 38 cirrhotic patients without HCC, and 27 age- and body mass index [BMI]-matched healthy volunteers). 16S rRNA gene sequencing was performed. Bacterial richness in cirrhosis and HCC-cirrhosis patients was significantly lower than in healthy controls. The HCC-cirrhosis group was successfully classified with an area under the curve (AUC) value of 0.9 based on the dysbiotic fecal microbial signature. The HCC-cirrhosis group had a significant overrepresentation of Clostridium and CF231 and reduced Alphaproteobacteria abundance compared to cirrhotic patients without HCC. Patients with HCC-cirrhosis who were overweight displayed significantly decreased bacterial richness and altered microbiota composition compared to their normal-weight counterparts. There was a significant correlation in the HCC-cirrhosis group between intake of artificial sweeteners and the presence of Akkermansia muciniphila A unique microbial signature was observed in patients with HCC-cirrhosis, irrespective of cirrhosis stage, diet, or treatment. BMI, dietary sugar, and artificial sweeteners were significantly associated with alterations in the microbiome of HCC-cirrhosis patients. However, the increased abundance of Clostridium and CF231 observed in HCC-cirrhosis patients was not influenced by environmental factors, implying that this change was due to development of HCC.IMPORTANCE Development of hepatocellular carcinoma in patients with cirrhosis is associated with alterations in intestinal microbiota, including an escalation of dysbiosis and reduced bacterial richness. This study demonstrates that reduced bacterial richness and dysbiosis escalate with the progression of cirrhosis from compensated to decompensated cirrhosis and to HCC-associated cirrhosis (HCC-cirrhosis). Moreover, we report for the first time the effect of environmental factors on HCC-cirrhosis. Excess weight was associated with increased dysbiosis in patients with HCC compared to their normal-weight counterparts. Moreover, fatty liver, consumption of artificial sweeteners, and high-sugar foods were associated with altered microbial composition, including altered levels of Akkermansia muciniphila in HCC-cirrhosis. We have successfully determined that levels of Alphaproteobacteria and the two genera CF231 and Clostridium are significantly altered in cirrhotic patients who develop hepatocellular carcinoma, independently of cirrhosis severity and dietary habits.

Keywords: A. muciniphila; cirrhosis; diet; gut microbiome; hepatocellular carcinoma; microbiome.

FIG 1

Gut microbiome alterations in patients with cirrhosis or HCC-cirrhosis and in controls and cirrhosis etiology distribution in study groups. (A) Box plot of α-diversity (observed features) displaying a significant decrease in bacterial richness in cirrhotic patients and patients with HCC-cirrhosis compared to healthy controls (P values = 0.014 and 0.028, correspondingly). (B) Box plot of β-diversity (unweighted UniFrac distance) displaying a significant difference in bacterial composition in cirrhotic patients and patients with HCC-cirrhosis compared to healthy controls (P values = 0.004 and 0.016, correspondingly). (C and D) Box plot of α-diversity (observed OTU indices) (C) and β-diversity (unweighted UniFrac distance matrix) (D) of cirrhosis etiologies (in cirrhosis without HCC), displaying a significant decrease in bacterial richness (P value = 0.04) and altered bacterial composition (P value = 0.03) in cirrhotic patients with NAFLD compared to HCV-cirrhosis patients. (E and F) Box plot of α-diversity (observed OTU indices) (E) and β-diversity (unweighted UniFrac distance matrix) (F) of HCC-cirrhosis etiologies (in the HCC-cirrhosis group), showing that there were no significant differences in bacterial richness (P value = 0.11) or composition (P value = 0.07) in HCC patients with NAFLD-cirrhosis compared to HCV-cirrhosis.

FIG 2

Differentially abundant taxa in patients with HCC-cirrhosis compared to the controls. (A) LDA scores computed for differentially abundant taxa in the fecal microbiomes of patients with HCC-cirrhosis (green) and healthy controls (red). Length indicates effect size associated with a taxon. P = 0.05 for the Kruskal-Wallis test; LDA score > 2. (B) A graphical representation of the classification accuracy of a machine-learning random forest model in receiver operating characteristic (ROC) curves, displayed here as ROC curves for each class (AUC of 0.9) and average ROCs and AUCs, including “microaveraging” of 0.87 (to calculate metrics globally by averaging across each sample) and “macroaveraging” of 0.94 (to give equal weight to the classification of each sample). (C) Confusion matrix displaying the classification results, with overall accuracy of 82%, baseline accuracy of 0.545, and an accuracy ratio of 1.5. (D) Important features are represented in an abundance heat map, consisting of log₁₀ frequencies of the most important taxa in each sample and group (HCC-cirrhosis and healthy controls). These are the features that maximize model accuracy, as determined using recursive feature elimination.

FIG 3

Differentially abundant taxa in patients with HCC-cirrhosis compared to cirrhotic patients without HCC. (A) LDA scores computed for differentially abundant taxa in the fecal microbiomes of patients with liver cirrhosis (red) and patients with HCC-cirrhosis (green). Length indicates effect size associated with a taxon. P = 0.05 for the Kruskal-Wallis test; LDA score > 2. (B) Taxonomic cladogram from LEfSe showing differences in fecal taxa of cirrhosis patients compared to HCC-cirrhosis patients. There were differences in the relative abundances of Alphaproteobacteria (P value = 0.039), Clostridium (P value = 0.024), CF231 (P value = 0.010), Verrucomicrobia (P value = 0.036), and Akkermansia muciniphila (P value = 0.039).

FIG 4

Gut microbiome alterations in overweight HCC-cirrhosis patients. (A) Box plot of α-diversity (Shannon’s index) displaying a significant decrease in diversity in patients with HCC-cirrhosis that were overweight (BMI > 25) compared to counterparts that were not overweight (P value = 0.024). (B) Box plot of β-diversity displaying a significant difference in bacterial composition in overweight patients with HCC-cirrhosis compared to counterparts that were not overweight (P value = 0.033). (C) LDA scores computed for differentially abundant taxa in the fecal microbiomes of overweight patients with HCC-cirrhosis. Length indicates effect size associated with a taxon. P = 0.05 for the Kruskal-Wallis test; LDA score > 2.

FIG 5

Fatty liver in HCC-cirrhosis and significant association with the relative abundance of Akkermansia. (A) Box plot of α-diversity (Faith’s phylogenetic diversity [PD]) displaying a significant decrease in diversity in patients with HCC-cirrhosis that had a fatty liver compared to counterparts without a fatty liver (P value = 0.025). (B) Box plot of β-diversity (unweighted UniFrac distance matrix) displaying a significant difference in bacterial composition in patients with HCC-cirrhosis that had a fatty liver compared to counterparts without a fatty liver (P value = 0.008). (C) LDA scores computed for differentially abundant taxa in the fecal microbiomes of patients with HCC-cirrhosis that had a fatty liver. Length indicates effect size associated with a taxon. P = 0.05 for the Kruskal-Wallis test; LDA score > 2.