A unified prostate cancer single-cell atlas reveals molecular subtypes and lineage plasticity across disease progression
Presenter: Hanbyul Cho, BS;MS Session: Application of Bioinformatics to Cancer Biology 3 Time: 4/20/2026 2:00:00 PM → 4/20/2026 5:00:00 PM
Authors
Hanbyul Cho , Yuping Zhang , Jiayi Zhou , Rahul Mannan , Saravana M. Dhanasekaran , Xuhong Cao , Arul M. Chinnaiyan University of Michigan, Ann Arbor, MI
Abstract
Introduction: Prostate cancer displays extraordinary cellular and molecular heterogeneity in both localized and metastatic disease settings, reflecting the underlying variation in the tumor programs and cellular states that drive this disease. Methods: To examine this, we built a harmonized single-cell atlas by integrating 17 human prostate scRNA-seq studies (163 samples, 106 donors) representing various disease stages. We reprocessed the raw FASTQ data to reduce pipeline-driven batch effects. Metadata were standardized to a unified schema for donor identity, tissue site, disease state, and histologic grade. Cell type annotation was performed using a cluster-level pseudobulk correlation framework informed by Tabula Sapiens, Human Protein Atlas, and prostate-specific references. Results: Our uniform workflow yielded ~756,000 high-quality singlets representing various cell types, including ~267,000 epithelial and ~149,000 luminal cells. Data Integration using BBKNN preserved biological structure while effectively merging studies, resolving stromal (SMC, fibroblast subtypes, endothelial, Schwann), immune (T/NK, B/plasma, myeloid), and epithelial (luminal, basal, hillock, club) compartments. Within the luminal epithelium, we identified transcriptomic molecular subtypes consistent with canonical prostate cancer drivers. Expression-based scoring of ERG/TDRD1 and ETV1/ETV4 distinguished ERG+, ETV+, and ETS- tumor programs, where ETS fusion-positive tumors form distinct transcriptional neighborhoods. Inferred copy-number burden from RNA-based pseudobulks showed increased CNV burden in high-grade primary and metastatic CRPC relative to benign/normal tissue. In metastatic disease, we further resolved three reproducible phenotypic states, which are AR-driven adenocarcinoma, neuroendocrine PCa (NEPC), and double-negative (AR-/NE-) with occasional intra-tumoral coexistence, reflecting lineage plasticity relevant to treatment resistance. Transcriptional signature and module analysis revealed several additional facets of the different prostate cancer subtypes. This atlas establishes a unified framework for interrogating epithelial lineage states, molecular drivers, and tumor microenvironment programs across the clinical spectrum of prostate cancer.
Disclosure
H. Cho, None.. Y. Zhang, None.. J. Zhou, None.. R. Mannan, None.. S. M. Dhanasekaran, None.. X. Cao, None.. A. M. Chinnaiyan, None.
Cited in
Control: 1118 · Presentation Id: 3002 · Meeting 21436