Abstract
Prostate cancer (PCa) progression is driven by complex molecular reprogramming, yet distinguishing indolent from aggressive disease remains a challenge. We performed an integrative transcriptomic analysis of 1232 PCa samples spanning normal prostate and all major disease stages including primary localized tumors, metastatic hormone-sensitive PCa (mHSPC), and metastatic castration-resistant PCa (mCRPC). By integrating unsupervised consensus clustering (ATC:hclust), weighted gene co-expression network analysis (WGCNA), and explainable machine learning (ML), we identified key transcriptional programs and biomarkers associated with cancer initiation and disease progression. Our analysis revealed persistent dysregulation of mitotic control, DNA damage repair, transcriptional regulation, and cytoskeletal remodeling, underscoring their functional relevance for PCa progression. We uncovered TPX2 as a central hub gene, consistently upregulated across all disease stages and co-expressed with 21 commonly upregulated genes. ML-based gene ranking and interaction analysis identified connections among the commonly upregulated genes, highlighting CENPA-MYBL2 for primary localized PCa, EXO1-NEIL3 for mHSPC and CENPA-RRM2 for mCRPC. Stage-specific analysis further identified key drivers of distinct disease transitions including EZH2 and PLK1 as major regulators of androgen dependence in mHSPC, and TERT as a hallmark of mCRPC, highlighting its role in telomere maintenance and tumor progression. This study demonstrates that unsupervised clustering combined with WGCNA and ML enables the discovery of clinically relevant molecular signatures in PCa. Our findings establish TPX2-centered networks together with biological pathways implicated in mitotic regulation and DNA damage repair as key drivers of tumor evolution, providing a biologically informed source for biomarker development, drug testing and mechanistic studies.