Translational approach for biomarker discovery and validation workflow. (A) Candidate biomarkers are discovered using a genetic mouse model by enriching N-linked glycoproteins to sera and freshly isolated perfused prostates from wild-type and Pten cKO mice. Tryptic N-glycosites then are measured by LC-MS/MS. Identification and quantitation of proteins is performed as described. After a filtering process, candidate biomarkers are selected for the verification phase. (B) Verification phase. Highly standardized biobanking and clinical data collection are used for collecting serum and matching tissue samples from patients harboring localized PCa and control patients with BPH. N-linked glycoproteins are extracted as in A, and selected candidates from the discovery phase are measured by targeted proteomics and ELISA. At the same time, tissues are spotted as microarray and stained for the indicated antigens. Feature selection and modeling then is performed to find novel biomarkers for diagnosis, patient stratification by Gleason score, and PTEN status.

Murine prostate and serum N-linked glycoproteome. (A) Venn diagrams of the mouse prostate and serum glycoproteome identified in the wild-type and Pten cKO mice indicating proteins commonly detected or detected only in the respective genotype/organs. (B) Label-free quantification of the proteins by means of SuperHirn plotted for prostate and serum. Dots indicate the ratio for each protein between the Pten cKO and wild-type prostates or sera and indexed from the most down-modulated to the most up-regulated. (C and D) Previously unknown Pten-dependent changes in protein expression are verified by standard cell biology techniques such as Western blot (C) and immunofluorescence (D).

PTEN score status predictors modeling. (A) PTEN FISH boxplot for BPH and locPCa cases. Focal PTEN loss indicates the percent of cells with reduced PTEN FISH signals compared with the centromere signal in the analyzed epithelial cells (n = 75). (B) pSer-473-Akt immunostaining boxplot. The score indicates the average staining intensity on a scale of 0–3 multiplied by the percentage of positive epithelial cells in BPH and locPCa tissues (n = 92). (C) Boxplot showing the staining intensity of stathmin, a marker for the PTEN/PI3K signature on BPH and locPCa tissues (n = 92). Numbers indicate the percentage of positive epithelial cells. (D) Predictor variable distribution for genetic PTEN status. The predictive importance of LRP-1, THBS1, TIMP-1, CFH, Attractin (ATRN), BGN, OLFM4, Golgi membrane protein 1 (GOLPH2), ASPN, Cell adhesion molecule 1 (CADM1), Galectin-3-binding protein (LGALS3BP), Vitronectin (VTN), ECM1, Transmembrane 9 superfamily member 3 (TM9SF3), and Ceruloplasmin (CP) selected by random forests and subsequent bootstrapped exhaustive search is based on the frequency in which the candidates appear in the best 50 predicting models. (E) ROC curve for the best predicting signature for tissue PTEN status (n = 54). In boxplots (A, B, and C), the line within the box indicates the median value, the box spans the interquartile range, whiskers extend to data extremes, and asterisks are outliers >3× interquartile range.

Candidate biomarkers for diagnosis and Gleason score prediction. (A) Predictor variable distribution for Gleason score. The predictive importance for BGN, AZGP1, FN, GALNTL4, Carboxypeptidase M (CPM), ECM1, CADM1, Biotinidase (BTD), Complement factor H (CFH), Plexin B2 (PLXNB2), Lumican (LUM), Neural cell adhesion molecule L1 (L1CAM), Protein CREG1 (CREG1), ATRN, and ASPN selected by random forests and subsequent bootstrapped exhaustive search is based on the frequency in which the candidate appears in the first 50 models. (B) ROC curve for the indicated signature for prediction of Gleason score <7 or ≥7 (n = 54). (C) Predictor variable distribution for diagnosis between locPCa and BPH. (D) ROC curve showing the performance for the selected signature (green line), the signature combined with PSA (red line), and PSA alone (black line) (n = 105). (E) Independent test set for the diagnostic signature. Performance of the identified diagnostic signature (Left), PSA alone (Center), or the combination of the signature and PSA (Right) in a set of patients measured independently and not considered in the training set. Data are presented as confusion matrices with the sensitivity (sens.), specificity (spec.), and accuracy (acc.) indicated for each signature (n = 37).