Abstract 2454: Spatial profiling of arsenic-responsive gene expression in bladder cancer models via multiplex in situ hybridization

Abstract Multiplex Fluorescent In Situ Hybridization (mFISH) enables precise identification of distinct cell phenotypes and facilitates the assessment of spatial interactions within the tumor microenvironment. Analyzing images generated from mFISH studies allows for detailed characterization and comparison of the spatial distribution of tumor cells and tumor-associated immune cells. In our prior study, we developed a gene expression-based model for predicting bladder cancer risk under arsenic exposure, identifying three key genes - NKIRAS2, AKTIP, and HLA-DQA1 that demonstrated high predictive accuracy (AUC: 94 for training and 75 for testing) (PMCID: PMC8760535). The current study employs an image-based digital Pathology approach to further validate this model, exploring the role of these potential biomarkers at single-cell resolution.Bladder cancer biopsy samples contain RNA of sufficient integrity and quality for molecular analysis (RNA Integrity Number > 7), and confirmed through positive and negative control assessments, were selected for accurate and reliable gene expression analysis in downstream. Five mFISH images were generated by ACD Bio-Techne, and hematoxylin and eosin (H&E) images were annotated by the Department of Pathology and Cell Biology at Columbia University Medical Center. Using Warpy, the mF and H&E images were overlaid to delineate tumor and non-tumor regions on whole slide images (WSI). Quantification of the WSI was conducted with QuPath, where a custom data column containing the weighted 3-gene module score for each cell was appended to the sample data. Spatial distribution of cells was analyzed in python using Sklearn, padas, numpy and matplotlib libraries.The analysis demonstrated that the previously established three-gene risk prediction model exhibited a significantly higher signal density within tumor regions across all mFISH bladder cancer samples. Tumor regions displayed a markedly closer spatial proximity between cells compared to non-tumor regions. When comparing the genomic risk prediction equation for individual cells in tumor versus non-tumor regions, the model achieved a training AUC score of 0.92 and a test accuracy of 0.85. Notably, this trend was independent of tumor grade, whereas expression intensity declined with increasing patient age.In conclusion, this study reinforces the predictive significance of the previously identified gene set comprising NKIRAS2, AKTIP, and HLA-DQA1 in assessing bladder cancer risk. The methodologies utilized highlight the potential of low-complexity, single-cell gene expression profiling through multiplex mFISH, coupled with the integration of AI-driven digital pathology tools. However, the study is limited by its small sample size and the narrow range of clinical parameters evaluated, underscoring the need for further research to validate and expand these findings. Citation Format: Sonalika Singhal, Kevin Gardner, Michael L. Miller, Anushka Dixit, Samarth Singhal, Donald Sens, Sandeep Singhal. Spatial profiling of arsenic-responsive gene expression in bladder cancer models via multiplex in situ hybridization [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2454.