High-throughput LDI MS technology decodes the distinct metabolic landscape of prostate cancer in a large-scale cohort.

BACKGROUND: Prostate cancer (PCa) remains a leading global malignancy, yet current diagnostic reliance on prostate-specific antigen (PSA) testing is limited by suboptimal sensitivity and specificity for early-stage detection. The present study aims to establish an effective high-throughput screening technique for accurate PCa diagnosis. METHODS: A large-scale cohort of 28,892 subjects was considered for inclusion in this study, and 1133 volunteers were finally selected, including 600 healthy controls, 160 patients diagnosed with other diseases of urinary system, 89 patients diagnosed with benign prostate hyperplasia (BPH), and 284 PCa patients. Discovery and internal validation phases of diagnostic models were conducted through machine learning of urine metabolic fingerprints obtained by laser desorption/ionization mass spectrometry (LDI-MS). Furthermore, the developed diagnostic model was verified in an external validation cohort. RESULTS: In retrospective cohort, the stepwise binary classification model achieved satisfactory diagnostic performance with areas under curves (AUCs) of 0.9599–0.9957 in the discovery (n = 567) and internal validation dataset (n = 284). In the external validation cohort (n = 282), AUC values from the ROC curves that differentiate Non-PD from PD, BPH from PCa, and HC from UD were 0.9815, 0.9705, and 0.9980, respectively. More than 95% significant PCa patients in the gray area (3 < tPSA < 10 ng/mL) were successfully identified from BPH subjects. Notably, four metabolite-related candidate genes were identified in this work, including AOX1, PON3, CBS and ASPA. CONCLUSIONS: This study demonstrated the clinical potential of an LDI-MS-based non-invasive urine biopsy for early prostate cancer detection, particularly in improving diagnostic accuracy for patients with tPSA levels in the gray zone (3–10 ng/mL).