利用多重 FISH 与 AI 驱动数字病理验证膀胱癌砷暴露基因表达谱的新型空间框架

Bladder cancer exhibits marked spatial heterogeneity in gene expression and immune infiltration. In this exploratory pilot study, we integrate multiplex fluorescence in situ hybridization (mFISH) with AI-assisted digital pathology to characterize the spatial distribution of a previously validated three-gene arsenic-responsive risk model (NKIRAS2, AKTIP, HLA-DQA1). Initially identified in arsenic-exposed individuals and associated with bladder cancer risk, this gene panel achieved 94% training and 75% validation AUC in prior genomic models (PMC8760535).

We analyzed five bladder tumor specimens using whole-slide mFISH imaging and HoverNet-based nuclear segmentation to quantify gene expression at single-cell resolution. Spatial profiling revealed elevated expression scores in tumor-adjacent regions, with a strong positive correlation to tumor grade (Pearson's r = 0. 83). These gene-enriched regions exhibited spatial clustering of tumor cells.

Additionally, tumor-infiltrating lymphocyte (TIL) density was inversely correlated with tumor grade, suggesting immune exclusion in high-grade tumors.

Our findings demonstrate the feasibility of combining spatial transcriptomics with AI-driven histopathological analysis for biomarker validation. This integrative framework provides a foundation for future population-scale studies leveraging spatial omics to evaluate arsenic-associated gene signatures and assess their relevance in bladder cancer risk stratification and disease progression.