Precursor-dependent steroid conversion sustains androgen receptor signaling and lineage-state divergence in prostate cancer

Castration-resistant prostate cancer (CRPC) persists despite systemic androgen deprivation, an adaptive mechanism frequently attributed to autonomous de novo intratumoral steroidogenesis. Despite the widespread clinical deployment of upstream inhibitors, the true structural and metabolic fidelity of the models underpinning this "tumor-as-gonad" dogma remains highly controversial. Here, by integrating LC-HRMS steroid profiling with transcriptomics across prostate cancer models, we show that the malignant lines examined lack transcriptionally supported autonomous steroidogenic capacity, largely because of marked CYP17A1 suppression and absence of a functional cortisol-producing branch. Instead, these models retain downstream precursor-conversion machinery, including HSD3B1, AKR1C3, and SRD5A1, consistent with efficient use of adrenal steroid precursors. Multiomic stratification of an 818-patient TCGA-PRAD clinical cohort corroborated this precursor-dependent phenotype on a global scale. We identify a critical, continuous evolutionary trajectory: an androgen receptor (AR)-High phenotype fueled by intense metabolic hyperactivity (oxidative phosphorylation and fatty acid metabolism) and an AR-low state marked by extensive lineage plasticity and an elevated iron-lipid redox stress profile. While standard linear Cox models were limited by strict multi-variable co-linearity, an explainable SHAP machine-learning framework identified AR activity and plasticity score as dominant contributors to survival-risk prediction. These findings reconcile conflicting models of intratumoral androgen synthesis by demonstrating that metabolic topology, rather than pathway completeness, dictates localized androgenic output. Our findings redefine CRPC metabolism, and support a therapeutic shift toward targeting downstream cytosolic conversion nodes and adaptive lineage-plasticity vulnerabilities.