Identification of novel molecular subtypes and a signature to predict prognosis and therapeutic response based on cuproptosis-related genes in prostate cancer

Abstract

Background: Prostate cancer (PCa) is the most common malignant tumor of the male urinary system. Cuproptosis, as a novel regulated cell death, remains unclear in PCa. This study aimed to investigate the role of cuproptosis-related genes (CRGs) in molecular stratification, prognostic prediction, and clinical decision-making in PCa.

Methods: Cuproptosis-related molecular subtypes were identified by consensus clustering analysis. A prognostic signature was constructed with LASSO cox regression analyses with 10-fold cross-validation. It was further validated in the internal validation cohort and eight external validation cohorts. The tumor microenvironment between the two risk groups was compared using the ssGSEA and ESTIMATE algorithms. Finally, qRT-PCR was used to explore the expression and regulation of these model genes at the cellular level. Furthermore, 4D Label-Free LC-MS/MS and RNAseq were used to investigate the changes in CRGs at protein and RNA levels after the knockdown of the key model gene B4GALNT4.

Results: Two cuproptosis-related molecular subtypes with significant differences in prognoses, clinical features, and the immune microenvironment were identified. Immunosuppressive microenvironments were associated with poor prognosis. A prognostic signature comprised of five genes (B4GALNT4, FAM83D, COL1A, CHRM3, and MYBPC1) was constructed. The performance and generalizability of the signature were validated in eight completely independent datasets from multiple centers. Patients in the high-risk group had a poorer prognosis, more immune cell infiltration, more active immune-related functions, higher expression of human leukocyte antigen and immune checkpoint molecules, and higher immune scores. In addition, anti-PDL-1 immunotherapy prediction, somatic mutation, chemotherapy response prediction, and potential drug prediction were also analyzed based on the risk signature. The validation of five model genes' expression and regulation in qPCR was consistent with the results of bioinformatics analysis. Transcriptomics and proteomics analyses revealed that the key model gene B4GALNT4 might regulate CRGs through protein modification after transcription.

Conclusion: The cuproptosis-related molecular subtypes and the prognostic signature identified in this study could be used to predict the prognosis and contribute to the clinical decision-making of PCa. Furthermore, we identified a potential cuproptosis-related oncogene B4GALNT4 in PCa, which could be used as a target to treat PCa in combination with cuproptosis.

Keywords: cuproptosis; prostate cancer; signature; tumor microenvironment; unsupervised clustering.

Figure 1

The expression, prognosis, and somatic alteration of CRGs in the TCGA PCa cohort. (A) The comparison of CRGs expression between tumor and normal tissues. (B) The PFS network of CRGs and co-expression relationship between CRGs in PCa. (C) The mutation frequency of CRGs in 495 PCa samples from the TCGA cohort. (D) Histogram of the SCNA frequency of CRGs in PCa. (E) Lollipop chart of the frequency of different SCNA types. (*, p < 0.05; **, p < 0.01; ***, p < 0.001).

Figure 2

Consensus clustering of CRGs in PCa. (A) Consensus clustering matrix when k = 2. (B) The difference in PFS between the two clusters. (C) The heatmap shows the expression of CRGs between the two clusters and the correlations between the clusters and clinical features. (D) The comparison of CRGs expression between the two clusters. (E) The heatmap shows the result of GSVA between the two clusters. TNM, tumor node metastasis; p, pathology; GS, Gleason score. (*, p < 0.05; **, p < 0.01; ***, p < 0.001). ns, no significant.

Figure 3

The immune-related characteristics of cuproptosis-related molecular subtypes in the TCGA cohort. (A) The difference in immune cell infiltration between the two clusters. (B) The comparison of MHC molecules expression between the two clusters. (C) Immune checkpoint molecules expression between the two clusters. (D) The expression level of the genes that inhibit the cancer-immunity cycle between the two clusters. The comparison of the TMB score (E) and TIDE score (F) between the two clusters. (*, p < 0.05; **, p < 0.01; ***, p < 0.001). ns, no significant.

Figure 4

Development of the cuproptosis-related signature in the TCGA training cohort. (A) Sixty-three prognosis-related DEGs were identified by univariate Cox regression. The genes indicated by red arrows are the five genes involved in the construction of the prognostic model. (B) The horizontal axis represents the logarithm of the independent variable λ, and its coefficients are shown on the vertical axis. (C) The confidence interval corresponds to each lambda. (D) Coefficients of the five prognostic genes in the model. (E) Sankey diagrams displayed the correlation between cuproptosis-related subtypes, risk score, and prognosis. (F) The comparison of the expression levels of CRGs between two risk groups. (*, p < 0.05; **, p < 0.01; ***, p < 0.001).

Figure 5

Construction and internal validation of the cuproptosis-related signature. For the TCGA training cohort: Kaplan–Meier curve (A), risk score and survival status (B), the expression heat map of the 5 model genes (C), ROC curve, and AUC of the 5-gene signature (J). For the TCGA test cohort: Kaplan–Meier curve (D), risk score and survival status (E), the expression heat map of the 5 model genes (F), ROC curve, and AUC of 5-gene signature (K). For the TCGA all cohort: Kaplan–Meier curve (G), risk score and survival status (H), the expression heat map of the 5 model genes (I), ROC curve, and AUC of the 5-gene signature (L).

Figure 6

External validation of the cuproptosis-related signature. Kaplan–Meier curve as well as ROC curve and AUC of the signature in DFKZ cohort (A), MSKCC cohort (B), CPGEA cohort (C), GSE46602 cohort (D), GSE70768 cohort (E), GSE70769 cohort (F), GSE70770 cohort (G) and GSE54460 cohort (H).

Figure 7

Independent prognostic analysis as well as the development and validation of a nomogram in the TCGA cohort. The results of univariate (A) and multivariate (B) Cox regression analysis. (C) The nomogram for predicting PFS in PCa. (D) Calibration plots of the nomogram.

Figure 8

The Immune Landscape of the Signature. (A) Immune-related pathways enriched in the high-risk group. (B) Tumor-related pathways enriched in the high-risk group. (C) The difference in immune cell infiltration between the two risk groups. (D) The difference in immune-related functions or pathways between the two risk groups. (E) The comparison of MHC molecules expression between the two risk groups. (F) Immune checkpoint molecules expression between the two risk groups. (G) Stromal score, immune score, and estimate score between the two risk groups. (H) K-M analysis of the IMvigor210 cohort. (*, p < 0.05; **, p < 0.01; ***, p < 0.001). ns, no significant.

Figure 9

Somatic mutation and TMB based on the signature. Waterfall maps of the somatic mutations in the low-risk group (A) and the high-risk group (B). (C) Difference of TMB between the two risk groups. (D) Correlation between risk score and TMB. (E) Comparison in PFS between high- and low-TMB groups. (F) Comparison in PFS based on TMB and risk score. (G) Mutation frequencies of the five model genes in PCa patients from the cBioPortal database.

Figure 10

Chemotherapy response prediction and small molecule drug screening. The differences in the chemotherapy response of Cisplatin (A), Docetaxel (B), Elesclomol (C), and Bicalutamide (D) between the two risk groups. (E) Volcano plot of DEGs between the two risk groups. (F) The 3D structure of six small molecule drugs screened out from the cMap database. IC50, the half maximal inhibitory concentration.

Figure 11

The expression and regulation of these model genes and further experiments on B4GALNT4. (A–D) qRT-PCR shows the expression and regulation of model genes in prostate cell lines treated with drugs that induce cuproptosis for 24 h (n = 3). CuCl2 (2mM), Elesclomol (20 nM), both CuCl2 (2mM) and Elesclomol (20 nM). (E) Western blot showing the knockdown effect of B4GALNT4 in C4-2. (F) Experimental scheme of proteomics and transcriptomics analysis on C4-2 stable cell lines with B4GALNT4 knockdown. (G) The changes in protein levels of CRGs after B4GALNT4 knockdown. (H) The changes in mRNA levels of CRGs after B4GALNT4 knockdown. NS, P >= 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. (I) GSEA demonstrated the enrichment of tumor-related pathways after B4GALNT4 knockdown.