Society: AASLD

Objective: To investigate the clinical relevance of fecal microbial dysbiosis in HCC development
Design: A case-control study. Fecal microbial composition and predicted genetic function inferred from high-throughput 16S ribosomal RNA sequencing were analyzed between HCC (n=53) and age-, gender- and body mass index-matched non-HCC subjects (n=53). The predicting performance of microbial dysbiosis index (MDI) of selected contrasting taxa was examined in the testing dataset and validation dataset. Further animal studies were conducted by transplantation of feces containing the selected contrasting taxa.
Results: A significant alteration of fecal microbial diversity and composition were found in HCC patients. HCC patients with early stage and liver cirrhosis had higher fecal microbial diversity but not richness. We identified the subset of 4 species (Ruminococus gnavus, Bifidobacteirum longum, Eubacteirum dolicum, and Rothia mucilaginosa)-based biomarker which achieved a great association with HCC development (area under the curve [AUC] 0.916). The AUC value increased up to 0.966 when combined with AFP values. This fecal microbial signature performed well in the validation cohort (AUC 0.920). The fecal microbiota signature in HCC was predicted to be associated with enhanced signaling pathways involving L-arginine biosynthesis, urea-cycle, heme biosynthesis, and nicotinamide adenine dinucleotide (NAD) salvage pathways; and reduced signaling pathways involving allantoin degradation and pyrimidine deoxyribonucleotide. Fecal microbial transplantation study demonstrated a significantly increased tumor growth in mice receiving feces containing the 4-species signature.
Conclusion: A distinct gut microbial profile was found in HCC patients. The specific fecal microbial signature may involve tumorigenesis and predict HCC development.

Background & Aims: Non-alcoholic fatty liver disease (NAFLD) is a rapidly growing cause of hepatocellular carcinoma (HCC) worldwide. Emerging evidence implies that dietary modification could modulate the pathogenesis of NAFLD-HCC. In this study, we aim to elucidate the roles of the fermentable fiber inulin and non-fermentable fiber cellulose in NAFLD-HCC progression and its underlying mechanisms through directly modulating gut microbiota and metabolites.
Methods: Two NAFLD-HCC models were established: C57BL/6 mice (2-3 weeks) were injected with Diethylnitrosamine (DEN) (25mg/kg) and fed with high fat high cholesterol diet (HFHCD) or choline deficient, high fat diet (CD-HFD) for 6 months. Meanwhile, 10% fermentable fiber inulin or non-fermentable fiber cellulose was added to respective diets. To track the in vivo incorporation of inulin into gut microbiota and metabolites, samples from mice labeled with ¹³C-inulin were analyzed with shotgun metagenome sequencing and non-targeted metabolomics. Human NAFLD-HCC cell lines (HKCI2 and HKCI10) and mouse NAFLD-HCC organoids were used for bio-functional study of the candidate metabolites.
Results: Inulin administration, but not cellulose, consistently suppressed NAFLD-HCC in DEN-injected mice fed HFHCD with significantly decreased tumor number and tumor load. This was confirmed in a second NAFLD-HCC mouse model induced by feeding mice with CDHFD. Metagenome sequencing combined with ¹³C DNA stable isotope labeling of fecal sample of mice identified the enrichment of beneficial microbes by inulin, including Bacteroides cellulosilyticus, Bacteroides thetaiotaomicron, Bacteroides caccae, Akkermansia muciniphila, etc. In vivo validation by oral gavage of these top enriched bacteria revealed that Bacteroides cellusilyticus as the most effective bacterium in inhibiting NAFLD-HCC formation with significantly reduced tumor number and tumor load. Meanwhile, the non-targeted metabolomics combined with ¹³C-inulin labeling identified inulin-derived bioactive metabolites pentadecanoic acid and nicotinic acid were consistently enriched in the stool and the portal vein serum of inulin-treated mice. Integrative analysis showed that these two metabolites highly correlated with Bacteroides cellusilyticus. HKCI2 and HKCI10 co-cultured with pentadecanoic acid or nicotinic acid suppressed cell viability and proliferation, and induced apoptosis (all P<0.05). Consistently, treatment with pentadecanoic acid or nicotinic acid significantly inhibited the growth of mouse NAFLD-HCC-derived organoids (all P<0.05).
Conclusion: Dietary inulin is effective in suppressing NAFLD-HCC development by enriching gut beneficial bacterium Bacteroides cellusilyticus and their associated bioactive metabolites (pentadecanoic acid and nicotinic acid).

Background: Hepatocellular carcinoma (HCC) is one of the most common causes of cancer-related death worldwide and has been increasing in incidence despite significant advancements in the treatment of viral hepatitis. These trends have largely been attributed to the coincident rise in obesity and non-alcoholic fatty liver disease (NAFLD), which are well-established risk factors for HCC. Bariatric surgery is increasingly utilized to combat lifestyle- and medication-resistant obesity, and not only results in improvements in metabolic comorbidities but has also been shown to reduce the risk of NAFLD and HCC. We aim to better characterize the effects of bariatric surgery on HCC risk in the U.S. population using a large, population-based database.

Methods: We performed a retrospective cohort analysis utilizing the National Inpatient Sample (NIS) database from October 2015 to December 2020. All adult subjects with a BMI > 40 or BMI >35 with presence of other comorbidities; hypertension (HTN), Diabetes (DM), hyperlipidemia (HLD) and/or obstructive sleep apnea (OSA) were identified and stratified into those with and without history of bariatric surgery (Roux-en-Y gastric bypass or Sleeve gastrectomy). Baseline characteristics and comorbidities among the two groups were compared using chi-squared and Wilcoxon rank-sum tests. We then performed univariable and multivariable analysis to assess the risk of HCC when adjusted for other established risk factors, including history of cirrhosis, chronic hepatitis B and C, alcohol use, NAFLD, smoking and DM, as well as various stages of obesity.

Results: Over 19 million subjects with obesity were included, 1,656,329 of which had history of bariatric surgery. The bariatric surgery group had significantly lower baseline rates of HTN, HLD, chronic kidney disease, DM, tobacco and alcohol use, cirrhosis and chronic hepatitis B and C (p<0.001), though rates of non-alcoholic fatty liver disease (NAFLD) and obstructive sleep apnea (OSA) were higher (8.3% vs 2.5% and 33% vs 25%, p<0.001, respectively). History of bariatric surgery associated with significantly lower risk of HCC (OR; 95% CI) on both univariable (0.39; 0.33-0.45) and multivariable (0.60; 0.52-0.70) analysis when adjusted for age, body mass index and other established risk factors, including cirrhosis, chronic hepatitis and NAFLD.

Conclusion: Bariatric surgery significantly associated with lower risk of HCC, independent of cirrhosis, NAFLD and chronic viral hepatitis. This study lends further evidence to the importance of treating obesity and the vital role that surgical, endoscopic and medical weight loss treatments may play in reducing the societal and economic burden of cancer. It may also help to expand considerations for bariatric surgery to those with increased risk of chronic liver disease and HCC.

Table 1. General Patient Characterestics and Comorbidities

Table 1. General Patient Characterestics and Comorbidities

Table 2. Univariable and Multivariable Analysis of Bariatric Surgery and its association with HCC

Table 2. Univariable and Multivariable Analysis of Bariatric Surgery and its association with HCC

Background: Artificial intelligence has been increasing applied in medical image analysis. We aimed to develop a deep learning model for detection and diagnosis of primary liver cancers and benign focal liver lesions (FLLs) which are commonly found in clinical practice.
Methods: 29,423 upper abdominal ultrasound still images of 4,996 patients diagnosed with FLL were retrieved from 2 tertiary hospitals and handed-labelled the diagnosis. The diagnosis of FLL including hepatocellular carcinoma (HCC), cholangiocarcinoma (CCA), simple cyst, hemangioma, focal fatty change (FFC), focal fatty sparing (FFS) and regenerative liver nodule was made by pathology, CT, MRI and clinical information. There were 12,532 images of FLLs, consisting of 1,635 HCCs, 6,454 CCAs, 2,229 cysts, 1,378 hemangiomas, 563 FFCs, 3,306 FFS, and 4,062 regenerative nodules (some images contained more than 1 FLLs). The images of FLLs were allocated independently by patient level into a training set (n=10,042), tuning set (n=670) and test set (n=1,820) in a 8 : 0.5 : 1.5 ratio. Another set of 16,891 images without FLL were allocated independently by patient level into a training set for 5% of the images FLLs (n=7,414) and 2:1 ratio in tuning set (n=6,184) and test set (n=3,293). The Yolo V5 model was trained using a supervised method. The performance of the model was evaluated on the test set with 1,457 malignant images (259 HCCs, 728 CCAs) and 363 benign FLL images (210 cysts, 197 hemangiomas, 59 FFCs, 581 FFSs and 520 regenerative nodules). Overall ROC curve (malignant vs. benign) was generated using logistic regression with multiple decision thresholds. Detection rate per image and sensitivity and specificity per lesion for diagnosis of each FLL was estimated.
Results: Detection rates of malignant and benign FLLs were 89.8% and 74.8%, respectively. With the decision threshold of 0.35 and intersection over union (IoU) threshold of 0.3, the model had a sensitivity, specificity and accuracy of 74.81%, 89.81% and 94.20% respectively, with an AUC of 0.71, for differentiating between malignant and benign FLL. In the malignant FLL group, the model had sensitivities of 78.2% and 97.1%, specificities of 97.4% and 97.9%, and NPV of 97.4% and 98.6%, respectively, for diagnosis of HCC and CCA. In the benign FLL group, the sensitivities and specificities of the model ranged from 74.3 to 100% and from 99.1 to 100% for diagnosing each type of benign FLLs (Table).
Conclusion: The deep learning model has excellent performance in diagnosis of primary liver cancers by showing a substantial NPV near 99% to exclude malignancy and common benign focal liver lesions. External validation in an independent cohort is further needed.

Sanparith Marukatat

Appearances