CELL INTELLIGENCE · 肿瘤细胞治疗研究

BACKGROUND AND AIMS: Tertiary lymphoid structures (TLS) are associated with heterogeneous outcomes in hepatocellular carcinoma (HCC), but whether their anatomical context determines clinical impact remains unknown. We hypothesized that spatial compartmentalization of TLS influences their prognostic significance. METHODS: We developed SpatialDecoder, a deep learning pipeline that simultaneously segments TLS, classifies three maturation subtypes (Agg, Fol I, Fol II), and assigns spatial compartments (intratumoral, peritumoral, capsular), achieving accurate tissue segmentation (mean Dice similarity coefficient (DSC)=0.86), TLS segmentation (DSC=0.8609), and subtyping (macro Area Under the Receiver Operating Characteristic Curve (AUROC)=0.8810). Applied to 1188 resected HCC patients, it enabled large-scale spatial TLS mapping. RESULTS: Spatial mapping revealed a prognostic dichotomy: higher intratumoral TLS density independently predicted prolonged overall survival, whereas higher extratumoral (peritumoral and capsular) TLS density correlated with increased recurrence risk. Based on TLS distribution, we defined four spatial immune phenotypes: TLS-Enriched (high intra/low extra), TLS-Balanced (high/high), TLS-Excluded (low intra/high extra), and TLS-Deficient (low/low). These phenotypes stratified patients into a survival gradient (log-rank p<0.0001): TLS-Enriched conferred the best outcome (HR = 0.48), TLS-Excluded the worst (HR=1.80), with intermediate prognosis for the other two phenotypes. Multivariable analysis indicated independent prognostic value. Transcriptomic profiling uncovered distinct immune mechanisms across the four phenotypes. CONCLUSIONS: This study establishes TLS spatial context as a determinant of their dualistic role in HCC immunity, providing a large-scale demonstration of opposite prognostic roles based on TLS location. The SpatialDecoder pipeline and four-phenotype framework transform TLS assessment from a binary metric into a spatially informed approach for postoperative risk stratification.