Proteomic and phosphoproteomic landscape of localized prostate cancer unveils distinct molecular subtypes and insights into precision therapeutics

Abstract

Building upon our previous investigation of genomic, epigenomic, and transcriptomic profiles of prostate cancer in China, we conducted a comprehensive analysis of proteomic and phosphoproteomic profiles of 82 tumor tissues and matched adjacent normal tissues from 41 Chinese patients with localized prostate cancer. We identified three distinct proteomic subtypes with significant difference in both molecular features and clinical prognosis. Notably, these proteomic subtypes exhibited a parallel degree of heterogeneity in the phosphoproteome, featuring unique metabolism, proliferation, and immune infiltration characteristics. We further demonstrated that a combination of proteins and phosphosites serves as the most effective biomarkers in prostate cancer to predict biochemical recurrence. Through an integrated multiomics analysis, we revealed mechanistic differences underlying different proteomic subtypes and highlighted the potential significance of Serine/arginine-rich splicing factor 1 (SRSF1) phosphorylation in promoting the malignant characteristics of prostate cancer cells. Our multiomics data provide valuable resources for understanding the molecular mechanisms of prostate cancer within the Chinese population, which have the potential to inform the development of personalized treatment strategies and enhance prognostic analyses for prostate cancer patients.

Keywords: molecular subtyping; phosphoproteomics; precision medicine; prostate cancer; proteomics.

Fig. 1.

Landscape of proteome and phosphoproteome in prostate cancer. (A) Workflow overview and processing of proteomic, phosphoproteomic, WGS, and RNA-seq data. (B and C) PCA of proteomics (B) and phosphoproteomics (C). The first two principal components are shown, with red triangles representing tumor samples, blue dots representing nontumor samples, and ellipses indicating the 0.9 CI for each type. (D and E) Volcano plot showing the difference in proteins (D) and phosphorylated sites (E) between tumor and nontumor samples. Red dots represent up-regulated proteins or phosphorylated sites [FDR < 0.01 and | log2(fold change) | ≥ log2(1.5)], while blue dots represent down-regulated proteins or phosphorylated sites [FDR < 0.01 and | log2(fold change) | ≥ log2(1.5)]. (F) GSEA to identify cancer hallmark with activity either higher (red bars) or lower (blue bars) in prostate tumor samples. (G) KSEA displays the differences in the phosphoproteome between tumor and nontumor tissues, where red bars represent kinases with higher activity in tumors, and blue bars represent kinases with lower activity in tumors.

Fig. 2.

Subtype identification based on proteomics data. (A) Consensus-clustering analysis of proteomic profiling identifies three subtypes. The heatmap depicts the abundance of signature proteins (log₂ transformed). The bars above the heatmap annotate the distribution of different clinicopathological characteristics among the three subtypes. P values are calculated by Fisher’s exact test. (B–E) Kaplan–Meier curves of BCR-free survival for each proteomic subtype in CPGEA-PP, CPGEA (S-I, n = 20; S-II, n = 12; S-III, n = 62), TCGA PRAD (S-I, n = 86; S-II, n = 80; S-III, n = 331), and cohort of Zhong et al. (S-I, n = 19; S-II, n = 32; S-III, n = 147). The proteomic subtypes of patients from the CPGEA, TCGA PRAD, and Zhong et al. are predicted using k-nearest neighbor algorithm. P values are calculated by the log-rank test. (F–H) Pathway enrichment analysis was performed for each proteomics subtype using signature proteins. P values are calculated by a hypergeometric test and adjusted by FDR. (I) Genomic profile of prostate cancer patients. The Top panel shows the top 30 mutated cancer-related genes in this prostate cancer cohort. The Middle panel shows the count of mutations with different consequences in tumor samples. The Bottom panel shows the count of different types of nucleotide substitutions in tumor samples. The Bottom bar represents three proteomic subtypes and ERG fusion status. (J) The heatmap depicts the abundance of the targets of FDA-approved drugs or candidate drugs under clinical trial. The Top bars are the same as in (A). Hazard ratios with 95% CI of these drug targets were shown in the Right panel.

Fig. 3.

Tumor microenvironment of proteomic subtypes. (A) Heatmap shows the relative abundance of 41 cell types across three proteomics subtypes. (B) Comparison of the relative abundance of 41 infiltrating cell types between tumor and nontumor samples in each subtype patient. (C–F) The Kruskal–Wallis Rank Sum test was used to assess differences between subtypes. Comparisons of T helper cells 2 (C), fibroblasts (D), regulatory T cells (E), and CD8⁺ central memory T cells (F) between tumor and nontumor samples in the S-I, S-II, and S-III subtypes. Red represents tumor samples, and blue denotes nontumor samples. P values were calculated by Wilcoxon’s rank-sum test. In the box plots, the middle bar represents the median, and the box represents the interquartile range; bars extend to 1.5× the interquartile range. P values are calculated by the two-sided Wilcoxon rank-sum test and are shown on the Top of the boxes.

Fig. 4.

Identification of protein biomarkers for prostate cancer prognostic prediction. (A) Heatmap shows the relative abundance of 13 biomarker proteins in two different risk groups of tumor samples. The Top bar represents BCR risk of patients based on the expression of biomarkers. (B) Kaplan–Meier curves of BCR-free survival for patients in CPGEA-PP. (C) The AUC of the 13-protein biomarker panel in predicting the prognosis of patients. (D–F) Kaplan–Meier curves of BCR-free survival for patients with either high or low risk in CPGEA (D), TCGA PRAD (E), and Zhong et al. (F). The BCR-free survival risk class of patients from CPGEA, TCGA, and Zhong et al. is predicted using the random forest algorithm. P values are calculated by the log-rank test.

Fig. 5.

Overview of multiomics integration analysis using the TieDIE algorithm. (A) Overview of the multiomics integration analysis workflow using the TieDIE algorithm. (B) The cancer hallmarks enriched in the integrated network. P values are calculated by a hypergeometric test.

Fig. 6.

Multiomics integration analysis reveals proteomic subtype-specific pathway networks. The proteomic subtype-specific networks are developed from the stemness pathway hallmarks for S-I (A), S-II (B), S-III (C). Edges belonging to both the subtype-specific network and the stemness pathways network are shown as thick edges, while corresponding nodes (genes) are shaded in dark gray. “Circleplot” quadrants for each gene summarize genomic, transcriptomic, and phosphoproteomic activity relevant to tumor samples in each subtype (Upper Left, TF activity; Lower Left, kinase activity; Lower Right, mutation; Upper Right, CNV; center, relative phosphorylation in tumor samples compared to normal samples). Genes and edges that are not represented in the subtype-specific networks but are in the integrated network are shown in light gray.

Fig. 7.

Impact of SRSF1 activity on PCa cells. (A) Kaplan–Meier curves of BCR-free survival in patients with either high or low SRSF1-S199 phosphorylation. (B) Western blot of SRSF1 expression in SRSF1 knockdown cells in PC-3 cell line. (C) Cell cycle analysis of scramble and SRSF1-KO PC-3 cells. (D) The proliferation assay of different PC-3 cells using EdU staining. Hoechast 33342 was used to label the position of the cell nucleus. (E) Wound healing assay of the scramble and SRSF1-KO in PC-3 cells. (F) Western blot of SRSF1 expression in the SRSF1-overexpressing cells and in SRSF1 knockdown cells after SRSF1-overexpressing in PC-3 cell line. Overexpressed SRSF1 is HA tagged. (G and H) The proliferation assay of different PC-3 cells using EdU staining (G). Hoechast 33342 was used to label the position of the cell nucleus. The statistics of the proliferation assay were shown in (H).

Fig. 8.

Inhibition of SRSF1 activity suppressed the malignant characteristics of PCa cells. (A) Chemical structure scheme of the small molecular SRPK1 inhibitor SPHINX31. (B) Proliferation curves of PC-3 cells treated with different concentrations of SPHINX31. (C) Cell cycle analysis of PC-3 cells treated with SPHINX31. (D and E) Apoptosis analysis of PC-3 cells after 24 h treatment with different concentrations of SPHINX31 (D). The statistics were shown in (E). (F) Schematic design of the PC-3 xenograft NSG mouse model. (G and H) The body weight curve (G) and the tumor growth curve (H) of PC-3 xenograft mouse model treated with SPHINX31 or vehicle. Statistics by two-way ANOVA, n = 10. (I) Images of the gross tumors from the xenograft mouse model after the treatment. (J) The plot of tumor weight from the xenograft mouse model after the treatment. (K) IHC image of the tumors from the xenograft mouse model after the treatment with SPHINX31 or vehicle, stained with Ki-67 antibody. The statistics of the Ki67 positive cells per HFP were at the Right. Statistics by the unpaired Mann–Whitney t test, n = 10. ***P < 0.001.