O266: Transcriptomic signature from spatially inferred histologically benign ancestral clone predicts high risk prostate cancer

Abstract

Abstract Introduction Prostate cancer (PCa) is a multifocal disease driven by heterogenous spatial and temporal branching evolution of tumour clones. Spatial transcriptomic analysis has revealed, clonally distributed copy number alterations even in histologically benign regions. We aim to evaluate if the genetic signature comprised within these genetically aberrant histologically benign regions, or “ancestral clones” may predict presence of high-risk disease. Methods Fresh-frozen, radical prostatectomy specimen was profiled using the 10x Genomics Visium Spatial Gene Expression platform. Via hierarchical clustering, spatial tumour clonal relations were inferred. Differentially expressed genes (DEGs) between high-grade tumour and histologically benign ancestral clones were identified. DEGs (|log2FC|>0.75) were input into a LASSO regression algorithm with normalised mRNA expression of normal adjacent tissue from the TCGA PCa database to train a genetic risk model to predict presence of high risk PCa (Gleason>8). Derived risk score was validated on an external cohort (GSE21032). Results A total of 683 DEGs were identified. These were enriched for organic acid metabolic pathways. Regression analysis on 57 normal adjacent tissue samples from TCGA PCa dataset was performed to develop a predictive model comprising of 3 genes (SLC22A3, CRIP2, LSAMP). External validation of the genetic risk score on matched histologically benign tissue from 131 patients, had good performance for prediction of high-risk PCa (AUC 0.807; 95% CI 0.693-0.921). Conclusion Our model has good performance for predicting high-risk PCa from matched normal tissue. Developed genetic risk score has translational potential to improve detection of unsampled clinically significant prostate cancer from histologically benign biopsy cores.