Spatial transcriptomics reveals primary-site endothelial EMT reprogramming that drives immunosuppression in metastatic uveal melanoma and predicts poor survival

Uveal melanoma (UM) is the most common primary intraocular malignancy, and nearly 50% of patients develop liver metastases with a median survival of only 4-6 months post-metastasis. Its poor prognosis and lack of effective therapies are linked to an “immune-cold” tumor microenvironment, yet the specific cell types, spatial organization, polarization states, and signaling networks driving this environment in metastatic lesions remain poorly defined. To address this, we performed spatial transcriptomics on liver metastatic UM using 10x Visium HD. Based on PRAME expression and RNA velocity, we identified four tumor cell clusters, one of which (cluster #2) represented a downstream transitioned state with the highest cancer stem cell association. Using the cell-type marker gene database, remaining clusters were annotated as endothelial cells, hepatic stellate/cancer-associated fibroblasts (CAFs), macrophages, monocytes, and T cells. TGFB1, a key immunosuppressive cytokine, was expressed significantly only by endothelial cells and CAFs, with endothelial cells showing dominant >5-fold higher expression. Interestingly, EMT genes associated with BAP1-deficient UM showed the highest and most significant EMT module scores in endothelial cells and CAFs only. Next, we performed “immune-cold” landscape analysis which revealed that most UM clusters had a higher anti-inflammatory to pro-inflammatory marker ratio compared to the levels found in all the retinoblastoma clusters. This immune cold signature was dominated by tumor cluster #2, followed by macrophages and CAFs. These findings suggest that EMT-reprogrammed endothelial cells with elevated TGFB1 may drive both CAF formation via hepatic stellate cell fibrosis and anti-inflammatory state reprogramming of macrophages. Further supporting crucial role of EMT-reprogrammed endothelial cells, CellChat analysis revealed this cluster as the dominant driver of cell-cell communications, followed by macrophages and tumor cells. To test whether these EMT-reprogrammed endothelial cells originated from the primary ocular site, we performed UMAP and cosine similarity analyses comparing their expression profiles with ocular, HUVEC, fetal & adult liver sinusoidal, cardiac, and pulmonary endothelial cells. These analyses revealed EMT-reprogrammed endothelial cells to have highest correlation with retinal endothelial cells. To further determine whether these EMT-reprogrammed endothelial cells originate from the primary ocular site, we tested whether their high-expression gene signature could distinguish UM patients with poor prognosis using expression data from TCGA primary UM samples. Amongst all metastatic clusters, only the top marker genes from EMT-reprogrammed endothelial cells were significantly associated with poor survival (P < 0.05, log-rank test). Overall, our findings suggest that retinal endothelial cells may both drive and constitute the premetastatic niche in liver metastatic UM, with their primary-site expression linked to poor survival.