Proteomic subtyping highlights tumor heterogeneity of human HCC.

Hepatocellular carcinoma (HCC) has a poor prognosis. While molecular profiling has identified subclasses with potentially druggable pathways, implementation in routine diagnostics remains challenging. Although immunohistology may aid HCC classification, multiplexed protein-based approaches have not yet been established. Proteomic heterogeneity in HCC tissue also remains poorly understood. Tissue microarrays from 58 HCC patients were analyzed using a multispectral imaging platform, enabling the detection of multiple protein biomarkers on a single tissue slide. A machine learning-based algorithm facilitated single-cell expression analysis, clustering, and spatial distribution assessment. A 4-plex immunofluorescence marker panel was designed and applied to interrogate altered signaling pathways in HCC. Unsupervised analysis revealed four factors corresponding to three HCC clusters defined by the overexpression patterns of p-S6/CRP (Cluster A), glutamine synthetase (Cluster B), and EpCam (Cluster C). Single-cell resolution uncovered substantial intratumoral heterogeneity. Only one third of HCCs showed a ≥ 0.95 purity of tumor cells in the predominant cluster. Clinically, Cluster C was associated with reduced median overall survival, while the other clinico-pathological features were not significantly different between the clusters. A protein-based subclassification of human HCC was established, characterized by three distinct subclasses (inflammation, beta-catenin/WNT signaling, progenitor-like) that align with known molecular categories. Cases with dominant progenitor features tended to have a shorter survival probability. The intratumoral heterogeneity observed in most cases may promote therapy resistance and underscores the need for precise molecular stratification to improve treatment outcomes.