Abstract
BACKGROUND: Prostate adenocarcinoma (PRAD) has substantial recurrence after primary treatment, and reliable prediction is clinically needed. We integrated clinical, genomic, and transcriptomic data to identify recurrence-associated genes, characterize their tumor microenvironmental context, and develop a prognostic model. METHODS: Gene expression and clinical data were obtained from TCGA (training; n = 145) and the GEO dataset GSE54460 (validation; n = 106). Single-cell RNA-seq data were analyzed to resolve gene expression across cell types. Immune infiltration was estimated using ssGSEA and the ESTIMATE algorithm, and cell–cell communication was assessed with CellChat. Recurrence-associated genes were identified by univariate Cox regression, and a LASSO Cox model was used to construct and externally validate the risk score. RESULTS: Pathological T stage, N stage, Gleason score, and TP53 mutation were linked to higher recurrence risk. We identified six risk genes (AMH, CRYBA2, CTHRC1, EFNA2, BMP6, ARHGDIG) and two protective genes (CKMT2, IP6K3) and built a risk-score model that discriminated against PFS in the training and validation cohorts. Single-cell analysis localized CTHRC1 predominantly to a fibroblast subpopulation characterized by upregulated midkine (MDK) signaling and extensive intercellular communication. Drug–gene correlation screening nominated decitabine, tegafur, calusterone, EMD-1204831, and ARQ-680; expression of several risk genes positively correlated with CTLA4, suggesting testable sensitivity to checkpoint blockade. CONCLUSIONS: Clinical variables together with transcriptomic features—particularly a CTHRC1-high fibroblast program—associate with PRAD recurrence. The externally validated risk score and the stromal context of CTHRC1/MDK provide testable avenues for prognosis and therapeutic exploration. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12957-026-04293-8.