Spatial recognition and semi-quantification of epigenetic events in pancreatic cancer subtypes with multiplexed molecular imaging and machine learning

Abstract

Abstract Genomic alterations are the driving force behind pancreatic cancer (PC) tumorigenesis, but they do not fully account for its diverse phenotypes. Investigating the epigenetic landscapes of PC offers a more comprehensive understanding and could identify targeted therapies that enhance patient survival. In this study, we have developed a new promising methodology of spatial epigenomics that integrates multiplexed molecular imaging with convolutional neural networks. Then, we used it to map epigenetic modification levels in the six most prevalent PC subtypes. We analyzed and semi-quantified the resulting molecular data, revealing significant variability in their epigenomes. DNA and histone modifications, specifically methylation and acetylation, were investigated. Using the same technique, we examined DNA conformational changes to further elucidate the transcriptional regulatory mechanisms involved in PC differentiation. Our results revealed that the foamy-gland and squamous-differentiated subtypes exhibited significantly increased global levels of epigenetic modifications and elevated Z-DNA ratios. Overall, our findings may suggest a potentially reduced efficacy of therapeutics targeting epigenetic regulators for these subtypes. Conversely, the conventional ductal PC subtype has emerged as a promising candidate for treatment with epigenetic modulators.