肝细胞癌的两种整合分子亚型及其预测性治疗潜力：迈向液体活检指导的精准医学

Hepatocellular carcinoma (HCC) is the third leading cause of cancer mortality worldwide, with a variety of therapeutic options in the advanced stage. Although immune checkpoint inhibitors (ICIs) and targeted therapies have transformed treatment paradigms, clinical decision-making regarding first-line therapy and subsequent lines remains challenging. Existing molecular classifications provide biological insights but often lack direct clinical translation. We aimed to identify robust, reproducible molecular subtypes of HCC that not only reflect tumor biology but also predict differential therapeutic responses, thereby advancing the potential for liquid biopsy-guided precision medicine. Transcriptomic data from the CCLE served as a discovery set and were validated with six cohorts, including TCGA. We identified two distinct molecular subtypes-METabolic and Immune-from the HCC cell line transcriptome using non-negative matrix factorization (NMF) at the REACTOME pathway level. Subtype-specific features were characterized by drug sensitivity, signaling pathway, cross-omics integration, and tumor microenvironment (TME) analyses. Therapeutic benefits from current first lines were validated across IMbrave150 and BIOSTORM. Serum proteomic and somatic mutation data from 35 patients were analyzed to identify predictive biomarkers. The METabolic subtype was characterized by elevated MET expression and enrichment of signaling pathways associated with cell survival, growth, and differentiation. In contrast, the Immune subtype exhibited prominent immune infiltration-particularly macrophages/monocytes, regulatory T cells, and CD8⁺ T cells-along with activation of PD-1/CTLA-4 pathways and poorer overall survival (OS). In the IMbrave150 cohort, the Immune subtype showed superior survival outcomes to atezolizumab-bevacizumab (p < 0.0001). Conversely, the METabolic subtype was associated with 100% recurrence after sorafenib treatment, suggesting potential benefit from MET inhibitors after sorafenib failure. A machine learning model using mutations and serum biomarkers (myoglobin, IL-6R beta, TP53 , CTNNB1 ) achieved robust performance (AUC = 0.88; accuracy = 87.1%) for subtype prediction. This study establishes clinically relevant HCC subtypes associated with distinct responses to currently recommended therapies and proposes a liquid biopsy-based approach using four predictive biomarkers. We provide rationale for the early use of MET inhibitors such as cabozantinib in the first-line setting, given its potential resistance to both sorafenib and ICI-based therapies. Additionally, our study suggests the Immune subtype is an optimal candidate for immunotherapy combined with anti-VEGF(R) agents. Further validation in larger, prospective cohorts is warranted to translate these findings into clinical practice.