Search

Summary: Combination therapy (lenvatinib/programmed death-1 inhibitor) is effective for treating unresectable hepatocellular carcinoma (uHCC). We reveal that responders have better overall and progression-free survival, as well as high tumor mutation burden and special somatic variants. We analyze the proteome and metabolome of 82 plasma samples from patients with hepatocellular carcinoma (HCC; n = 51) and normal controls (n = 15), revealing that individual differences outweigh treatment differences. Responders exhibit enhanced activity in the alternative/lectin complement pathway and higher levels of lysophosphatidylcholines (LysoPCs), predicting a favorable prognosis. Non-responders are enriched for immunoglobulins, predicting worse outcomes. Compared to normal controls, HCC plasma proteins show acute inflammatory response and platelet activation, while LysoPCs decrease. Combination therapy increases LysoPCs/phosphocholines in responders. Logistic regression/random forest models using metabolomic features achieve good performance in the prediction of responders. Proteomic analysis of cancer tissues unveils molecular features that are associated with side effects in responders receiving combination therapy. In conclusion, our analysis identifies plasma features associated with uHCC responders to combination therapy.

ABSTRACT Microbiota is implicated in hepatocellular carcinoma (HCC). The spectrum of intratumoral microbiota associated with HCC progression remains elusive. Fluorescence in situ hybridization revealed that microbial DNAs were distributed in the cytosol of liver hepatocytes and erythrocytes. Viable anaerobic or aerobic bacteria were recovered in HCC tissues by fresh tissue culture. We performed a comprehensive DNA sequencing of bacterial 16S rRNA genes in 156 samples from 28 normal liver, 64 peritumoral, and 64 HCC tissues, and the DNA sequencing yielded 4.2 million high-quality reads. Both alpha and beta diversity in peritumor and HCC microbiota were increased compared to normal controls. The most predominant phyla in HCC were Patescibacteria, Proteobacteria, Bacteroidota, Firmicutes, and Actinobacteriota. phyla of Proteobacteria, Firmicutes, and Actinobacteriota, and classes of Bacilli and Actinobacteria, were consistently enriched in peritumor and HCC tissues, while Gammaproteobacteria was especially abundant in HCC tissues compared to normal controls. Streptococcaceae and Lactococcus were the marker taxa of HCC cirrhosis. The Staphylococcus branch and Caulobacter branch were selectively enriched in HBV-negative HCCs. The abundance of Proteobacteria, Gammaproteobacteria, Firmicutes, Actinobacteriota, and Saccharimonadia were associated with the clinicopathological features of HCC patients. The inferred functions of different taxa were changed between the microbiota of normal liver and peritumor/HCC. Random forest machine learning achieved great discriminative performance in HCC prediction (area under the curve [AUC] = 1.00 in the training cohort, AUC = 0.950 for top five class signature, and AUC = 0.943 for the top 50 operational taxonomy units [OTUs] in the validation cohort). Our analysis highlights the complexity and diversity of the liver and HCC microbiota and established a specific intratumoral microbial signature for the potential prediction of HCC. IMPORTANCE Gut microbiome is an important regulator of hepatic inflammation, detoxification, and immunity, and contributes to the carcinogenesis of liver cancer. Intratumoral bacteria are supposed to be closer to the tumor cells, forming a microenvironment that may be relevant to the pathological process of hepatocellular carcinoma (HCC). However, the presence of viable intratumoral bacteria remains unclear. It is worth exploring whether the metataxonomic characteristics of intratumoral bacteria can be used as a potential marker for HCC prediction. Here, we present the first evidence of the existence of viable intratumoral bacteria in HCC using the tissue culture method. We revealed that microbial DNAs were distributed in the cytosol of liver hepatocytes and erythrocytes. We analyzed the diversity, structure, and abundance of normal liver and HCC microbiota. We built a machine learning model for HCC prediction using intratumoral bacterial features. We show that specific taxa represent potential targets for both therapeutic and diagnostic interventions.