Background: Circulating tumor cells (CTCs) is considered as an effective prognostic biomarker in malignancies. It is recently recognized that detections of programmed death-ligand 1 (PD-L1) and mesenchymal markers in CTCs may improve the predictive value. Therefore, we aimed to determine the cut-off values of these CTCs biomarkers in prostate cancer (PCa), assess the rates of PD-L1 positive and non-epithelial (NE) subgroup in CTCs among patients with high-risk (HRPC) and locally advanced PCa (LAPC) before surgery, and evaluate the clinical values of these markers. Methods: Four PCa cell lines selected from Cancer Cell Line Encyclopedia were applied in spiking experiments to establish the cut-offs of PD-L1 positive and NE CTCs. 182 HRPC and 89 LAPC patients were enrolled from June 2016 to December 2018. All patients underwent radical prostatectomy after blood sample collection. CTCs enumeration and biomarkers identification were conducted by Canpatrol CTC enrichment system and the RNA in situ hybridization technique. Lastly, using the cut-offs,we evaluated the association between CTCs markers and other variables in terms of clinical data, disease outcome, tumor tissues immunochemistry (IHC) staining, and the response of immunotherapy. Results: The cut-offs were determined as the lowest rates of NE and PD-L1⁺ CTCs among the cell lines, 45% and 25%, respectively. In HRPC patients, a higher PSA level and more advanced T stage were observed in the NE CTCs positive group (≥45%), compared with the negative group (P= 0.041 and P= 0.004). Multivariate analysis showed that the NE CTCs rate, instead of total CTCs number, was an independent risk predictor for progression-free survival (PFS) (HR = 3.85, P= 0.013). In LAPC patients, the PD-L1⁺ CTCs positive group (≥25%) presented a higher PSA level and higher percentage of lymph node metastasis than negative group (P= 0.027 and P= 0.028). The median PFSs between the two groups were 16.6 mouths vs. 21.8 mouths, respectively. (HR = 2.87, P= 0.002 ). No significant correlation was found between PD-L1 expression in CTCs and in tumor tissue IHC staining. However, we observed a significant correlation between PD-L1⁺ CTCs rate and ERG or PTEN expression in tissues (P= 0.010 and P= 0.009). In addition, patients in PD-L1⁺ CTCs positive group presented a better PSA response to Durvalumab treatment than those in negative group. Conclusions: PD-L1⁺ and NE CTCs were an effective prognosis predictors for HRPC and LAPC patients with radical prostatectomy. LAPC patients with PD-L1⁺ CTCs might benefit from adjuvant immunotherapy