Serum steroid ratio profiles in prostate cancer: A new diagnostic tool toward a personalized medicine approach

Background. Serum steroids are crucial molecules altered in prostate cancer. Mass spectrometry (MS) is currently the elected technology for the analysis of steroids in diverse biological samples. Steroids have complex biological pathways and stoichiometry is important to evaluate their quantitative ratio. MS applications to patient hormone profiling could lead to a diagnostic approach. Methods. Here, we employed the Surface Activated Chemical Ionization-Electrospray-NIST (SANIST) developed in our laboratories, to obtain quantitative serum steroid ratio relationship profiles with a machine learning Bayesian model to discriminate patients with prostate cancer. The approach is focused on steroid relationship profiles and disease association. Results. A pilot study on patients affected by prostate cancer (PCa), benign prostate hypertrophy (BPH) and control subjects (prostate specific antigen (PSA) lower than 2.5ng/mL) was done in order to investigate the classification performance of the SANIST platform. The steroid profiles of 71 serum samples (31 controls, 20 patients with prostate cancer and 20 subjects with benign prostate hyperplasia) were evaluated. The levels of 10 steroids were quantitated on the SANIST platform: Aldosterone, Corticosterone, Cortisol, 11-deoxycortisol, Androstenedione, Testosterone, dehydroepiandrosterone, dehydroepiandrosterone sulfate (DHEAS), 17-OH-Progesterone and Progesterone. We performed both traditional and a machine learning analysis. Conclusions. We show that the machine learning approach based on the steroid relationships developed here was much more accurate than the PSA, DHEAS, and direct absolute value match method in separating the PCa, BPH and control subjects, increasing the sensitivity to 90% and specificity to 84%. This technology, if applied in the future to a larger number of samples will be able to detect the individual enzymatic disequilibrium associated with the steroid ratio and correlate it with the disease. This learning machine approach could be valid in a personalized medicine setting.