利用多重免疫荧光对肿瘤微环境进行空间分析的标准化泛癌框架

A growing number of studies have shown the key role of the spatial layout and interactions between cells in the TME in determining patient prognosis and therapy response. While substantial effort has been invested in profiling individual cancers, differences in staining methods and analysis pipelines make it increasingly difficult to study spatial TME patterns in a pan-cancer setting. To address this issue, we developed standardized multiplex immunofluorescence (mIF) staining protocols and computational pipelines to profile TME architecture across tissue microarrays from endometrial, ovarian, breast, prostate, and colorectal carcinomas. Our mIF panels include up to 50 protein targets capturing major immune and stromal populations, epithelial states, and markers of stemness, allowing us to profile the TME in great detail. To achieve high-confidence cell classification, we integrated classical thresholding with pixel-based models and unsupervised clustering to produce a consensus cell-type annotation. Our analyses show that pixel-based and unsupervised clustering methods outperform thresholding by enhanced separation of ambiguous signal and improving reproducibility. Incorporating cellular shape parameters further enhanced the identification of morphologically distinct populations such as fibroblasts and macrophages. We first applied this framework to a gynecological discovery set comprising endometrial and ovarian carcinomas (n > 1600 patients, multifocal sampling). Preliminary analyses indicate that endometrial and endometrioid ovarian carcinomas, both derived from Müllerian epithelium, share similar immune and stromal profiles, whereas ovarian clear cell carcinoma shows a distinct TME composition. In addition, reduced fractions of B cells and increased fractions of dedifferentiated epithelial cells were both strongly associated with recurrence in endometrial carcinoma (two-sided t-test with FDR; both p < 0.001). Endometrioid ovarian carcinoma showed similar trends for these features, but no statistically significant associations with recurrence were found (p = 0.453 and p = 0.055, same order). Neither B-cell fraction nor dedifferentiated epithelial cells showed any association with recurrence in ovarian clear cell carcinoma (p = 0.339 and p = 0.339, respectively). These findings suggest that the cancer cell of origin may exert a stronger influence on TME composition than the anatomical site of disease. Together, our work will establish a harmonized pan-cancer mIF framework and reveal conserved TME features with clinical relevance. These standardized tools enable cross-cancer comparisons, improve the interpretability of spatial biomarkers, and may guide the development of broadly applicable therapeutic strategies.