Article Figures & Data
Figures
- Figure 1. Proteomic analysis of PCa samples.
(A) Overview of the study design. The dataset consists of prostatic tumour and matched benign tissue samples from 278 patients. Proteomic data were collected for 277 tumour samples and 278 benign samples in duplicate from 278 patients. A total of 1,475 MS runs were analysed in 31 batches, including tumour, benign, CTRL-A, and CTRL-B samples. The raw proteomic data were analysed by DIA-NN, quantifying 5,803 proteins. Scale bar = 100 μm. (B) tSNE projection of protein data superimposed with colour annotation of sample types. (C) Heatmap representation of the protein matrix with samples shown on the y-axis and proteins shown on the x-axis. The protein intensities were sorted first by the mass spectrometers, followed by tissue types and GGs. MS1–MS6 indicate the six mass spectrometers. (D) Volcano plot showing the up-regulated (n = 368) and down-regulated (n = 144) proteins in tumours with fold change (FC) > 1.5 and < 0.67 and the Benjamini–Hochberg (BH)-adjusted P < 0.01. Significant proteins are presented in red and blue colours, whereas other proteins are coloured in grey. (E) Analysis pipeline employed in this study and the number of proteins identified in each analysis. A total of 512 tumour-enriched proteins were identified from the comparison between tumour and benign samples, followed by stratification of the GG2 and GG3 using differential expression analysis and machine learning, and identification of a prognostic signature using survival analysis. Finally, pathway enrichment analyses were conducted for the significant sets of proteins.
- Figure S1. PPP1 study design.
Each row indicates a batch, and each column indicates PCa tissue samples. Each batch contained between 15 and 29 PCa samples, one CTRL-A, and one CTRL-B.
- Figure S2. Kaplan–Meier (KM) curves for biochemical recurrence–free survival (BCRFS) for different GGs.
Vertical lines illustrate patients who were censored at the time of their last clinical follow-up visit. The P-value shows the significance of the difference between survival estimates evaluated by the log-rank test. Coloured values represent the number of patients in each group under risk.
- Figure S3. Overview of proteomic data.
(A) Proteins and peptides quantified in tumour and benign samples (n = 1,348). A total of 53,713 peptides and 5,803 proteins are quantified. The purple colour shows the number of proteins and peptides quantified in tumour samples, and the green colour shows that in benign samples. Compared with tumour samples, a smaller number of proteins are quantified in benign samples. (B) Technical reproducibility of dataset shown by Pearson’s r among replicates. (C) Proteins quantified in different fractions of samples. ∼3,500 proteins are quantified in 50% or more samples. (D) tSNE projection of proteomic data before DIA-NN normalization. Samples analysed using different mass spectrometers (MS1-6) are shown with different colours. Clusters of tumour and benign samples are shown in different shapes. (E) tSNE projection of proteomic data after DIA-NN normalization showing no batch effects. (F) tSNE projection of tumour samples coloured by the GG. No grouping can be observed.
- Figure S4. Dysregulated pathways in tumour samples.
(A) Heatmap representation of the expression levels of differentially expressed proteins between tumour and benign samples shown in Fig 1D. Expression values are converted to z-scores. Samples are sorted according to tissue types (tumour versus benign) on the x-axis, whereas proteins are clustered on the y-axis. (B, C, D) Biological pathways including GO biological processes (B), Reactome pathways (C), and hallmark gene sets (D) enriched for the tumour versus benign differentially expressed proteins. Red bars indicate pathways enriched in up-regulated proteins, and blue bars indicate pathways enriched in down-regulated proteins. (E) PPI network components were obtained using the MCODE algorithm, showing the enriched biological processes and proteins. Up-regulated and down-regulated networks and associated proteins are coloured by FC. Up-regulated proteins are coloured in red, and down-regulated proteins, in blue. The width of the edge (between nodes) indicates the strength of the connection. A functional description is provided beside each component.
- Figure 2. Differentially expressed proteins in the GG2 versus GG3.
(A) Volcano plot showing the GG2 (n = 35) and GG3 (n = 1) enriched proteins in tumours. Significant proteins are presented in red and blue colours, whereas other proteins are coloured in grey. Only a small number of proteins were found to be significant using differential expression analysis, whereas most of them showed low FC. (A, B) Heatmap representation of the expression levels of differentially expressed proteins between GG2 and GG3 samples shown in (A). Expression data are converted to z-scores. Samples are shown on the x-axis, whereas proteins are clustered on the y-axis.
- Figure 3. Machine learning of the GG2 versus GG3.
(A) ROC curves for the best and average models for predicting GG2 and GG3 samples based on 1,000 Monte Carlo runs by XGBoost. The red dashed line represents the random guess, the blue solid curve shows the mean ROC curve over 1,000 Monte Carlo runs, the blue band represents one SD of the curves, and the orange curve shows the best ROC curve. (B) SHAP values of the top 20 most significant proteins to distinguish between GG2 and GG3 samples, sorted (from top to bottom) by their respective absolute mean SHAP values. SHAP values of proteins in different samples are shown on the horizontal axis; the top 20 proteins are sorted (by importance) from top to bottom on the y-axis. The colours from blue to red indicate protein expression levels from low to high. The vertical zero line (SHAP value = 0) is the line that has no impact on prediction, whereas the values on the left and right sides represent negative and positive impacts on prediction.
- Figure 4. Differentially expressed proteins and pathways in GG2 and GG3 PCa.
(A) GO biological processes, Reactome pathways, and hallmark gene sets enriched for the selected significant proteins. (B) PPI network components obtained using the MCODE algorithm, showing the enriched biological processes and proteins. Proteins are coloured according to the absolute mean SHAP values. The width of the edge (between nodes) indicates the strength of the connection. A functional description is provided beside each component.
- Figure 5. Survival analysis of BCR-free survival (BCRFS) of PCa.
(A) Forest plot showing the 18 proteins with their individual hazard ratios, P-values, 95% CIs, and C-index of the final multivariate Cox model. (B) Heatmap showing protein intensities sorted by a risk score and clustered for each of the two groups: risk and protective proteins. The column denotes patients, and the row indicates the 18 proteins. PCa samples with positive regression coefficients expressed risk proteins, whereas samples with negative regression coefficients expressed protective proteins. (C) Forest plot comparing the importance of the 18-protein signature with RSF-based risk score and with other clinical variables using univariate Cox models: pT stage (pT1 versus pT2), surgical margin (positive versus negative), and age at diagnosis (<64 versus ≥64). (D) Forest plot showing a simple multivariate Cox model that includes the 18-protein signature, RSF-based risk score, and other clinical variables. With recursive feature selection, the 18-protein signature remains the most important variable, with a stable C-index (from 0.96 to 0.95).
- Figure S5. Survival analysis of BCRFS.
(A) ROC curves with respective AUROCs at a 5-yr (60 mo) follow-up for the 18-protein risk score and RSF-based risk score. (B) Time-dependent AUROCs of the 18-protein risk score and RSF-based risk score.
- Figure 6. Kaplan–Meier (KM) curves for BCRFS.
KM curves with 95% CIs of the low- and high-risk groups based on the 18-protein risk score, along with respective numbers of samples corresponding to each GG. Vertical lines illustrate patients who were censored at the time of their last clinical follow-up visit. The P-value shows the significance of the difference between survival estimates evaluated by the log-rank test. Coloured values represent the number of patients in each group under risk. (A) KM curves for PCa patients in all GGs. (B) KM curves for PCa patients in the GG2 and GG3. (C) KM curves for PCa patients in the GG2 only. (D) KM curves for PCa patients in the GG3 only.
- Figure S6. Set of unique proteins.
A set of 39 unique proteins was extracted by taking the union of 18 signature proteins and 26 proteins from a univariate Cox regression model. Five proteins were found overlapping between the two sets.
- Figure 7. Significant biological pathways identified from a univariate Cox regression model.
(A) GO biological processes, Reactome pathways, and hallmark gene sets enriched for the selected significant proteins. (B) PPI network components obtained using the MCODE algorithm, showing the enriched biological processes and proteins. Proteins are coloured according to the P-values from the BCRFS analysis. The width of the edge (between nodes) indicates the strength of the connection. A functional description is provided next to each component.
- Figure S7. Fresh-frozen prostate tissue.
Biobanking of fresh-frozen prostate tissue from radical prostatectomy specimens (left; adapted with permission from Springer Nature; RH Hruban et al, Surgical Pathology Dissection, Springer Science + Business Media New York 1996) and histological evaluation of cancer and non-cancerous tissue using haematoxylin–eosin staining of fresh-frozen tissue sections (right). In detail, tissue samples (prostate cancer, benign prostatic hyperplasia) were collected as part of the ProCOC, a prospective biobank study that has now collected tumour (mainly acinar adenocarcinoma of the prostate) and non-tumour (mainly benign hyperplasia of the prostate) tissues from more than 1,000 radical prostatectomies. All patients with localized PCa scheduled for radical prostatectomy in curative intent at the University Hospital Zürich were asked to participate in the ProCOC biorepository ahead of surgery since 2008. The use of fresh-frozen tissues of the ProCOC cohort has already been reported in previous studies. The native radical prostatectomy specimens were processed in the Department of Pathology and Molecular Pathology, University Hospital Zürich, immediately after surgery. A complete cross section of each fresh, unfixed, and cooled (4°C) prostate specimen was collected from each prostatectomy specimen and divided into four quadrants; that is, the first slice after dissection of the apex was quartered and snap-frozen in four separate blocks, using a special procedure to ensure ideal sample quality. Only samples with sufficient tumour tissue (index tumour >5 mm diameter) were selected for subsequent proteomic analysis. All fresh-frozen samples were stored at −80°C. After formalin fixation overnight, the rest of the specimen was processed accordingly. The central tissue biobank of the University Hospital Zürich is specifically accredited for this purpose by the Swiss Biobanking Platform. Haematoxylin-and-eosin–stained sections of the four frozen blocks were sliced for immediate evaluation regarding tumour load and margins in synopsis with the standard formalin-fixed, paraffin-embedded histology. The size of a single fresh-frozen tissue punch was ∼1 mm3 (diameter, 1 mm; length, 1–2 mm; wet weight, ∼800 μg). Clinical and histological data: follow-up information regarding persistence of PSA or its re-emergence as biochemical relapse, and disease-specific and non–disease-specific cases of death were recorded from the electronic health records, or whether patients had their postoperative follow-up outside of our care centre at least bi-annually by contacting patients directly and their caring physicians. Patients who did not achieve a PSA nadir were excluded from the analysis. Scale bar = 100 μm.
Supplementary Materials
Table S2. Summary of 39 Signature Proteins.
Supplemental Data 1.
Supplemental Data 2.
Supplemental Data 3.
In this Issue
Subjects
Related Articles
Cited By...
- No citing articles found.