A Golgi apparatus-related signature for predicting prognosis and evaluating the tumor immune microenvironment of uveal melanoma

Abstract

Uveal melanoma (UVM), a highly invasive and metastatic primary eye cancer with poor prognosis, contributes significantly to melanoma-related deaths despite being less common. Despite advances in therapy, the mortality rate remains unchanged due to frequent liver metastases and limited effective prognostic biomarkers. This study employed gene expression data from The Cancer Genome Atlas and Gene Expression Omnibus databases to investigate Golgi apparatus-related gene sets (GGRGs) in UVM. Survival analysis, consensus clustering, and principal component analysis were conducted to identify GGRGs associated with patient outcomes. Additionally, tumor microenvironment was assessed using IOBR tools, and a nomogram was constructed based on Cox regression models for predicting survival probabilities. The biological function of carbohydrate sulfotransferase protein family (CHST9) was evaluated by colony formation assay, transwell invasion assay, and wound healing assay. Univariate Cox regression identified 343 GGRGs significantly correlated with UVM prognosis. Consensus clustering revealed 2 distinct subtypes (cluster1 and cluster2) differing significantly in survival, with cluster2 showing more favorable outcomes. Principal component analysis effectively separated these clusters, while Kaplan-Meier curves confirmed their survival disparity. Least Absolute Shrinkage and Selection Operator Cox regression analysis pinpointed a 5-GGRGs-based signature, termed GGRGs-derived index (GGI), composed of lunatic fringe, KDELR3, CHST9, ATP8B3, and ACAN. This GGI stratified UVM cases into High_GGI and Low_GGI groups across multiple datasets, with the Low_GGI group consistently demonstrating significantly improved survival rates compared to the High_GGI group. Notably, the Low_GGI and High_GGI groups exhibited marked differences in clinicopathological characteristics, drug sensitivities, and immune infiltration levels. Ultimately, GGI and age emerged as independent prognostic factors for UVM and were incorporated into a nomogram, which displayed outstanding performance in predicting patient prognosis. Depletion of CHST9 expression dramatically inhibited the proliferative capacity of UVM cells, concurrently suppressing their metastatic activity and invasive properties. GGRGs are promising predictors of UVM prognosis and may inform personalized treatment strategies, contributing to a deeper understanding of the molecular mechanisms driving this aggressive cancer.

Keywords: consensus clustering; drug sensitivities; golgi apparatus; immune infiltration; uveal melanoma.

Figure 1.

Consensus clustering analysis based on prognosis-related Golgi apparatus associated genes. (A) Consensus matrix heatmap (k = 2) illustrating the probability of 2 patients belonging to the same cluster across TCGA-UVM samples, as determined by prognosis-related GGRGs. (B) Consensus Cumulative Distribution Function (CDF) plot. (C) Delta area plot in consensus clustering, indicating that when the Delta area value stabilizes and the line in the plot flattens, the optimal number of clusters is deemed to be reached. (D) Kaplan–Meier (KM) survival curve analysis for clusters derived from GGRG-related classification. (E) Principal component analysis (PCA) plot based on prognosis-related GGRGs, demonstrating the separation between different clusters. (F) The distribution of status (alive/deceased), age, and stage among the identified clusters. GGI = GGRGs-derived index, GGRGs = Golgi apparatus-related gene sets, TCGA = The Cancer Genome Atlas, UVM = uveal melanoma.

Figure 2.

Construction of a GGRGs-derived index (GGI) for UVM based on prognosis-related GGRGs. (A, B) LASSO Cox regression optimization to select 5 GGRGs forming a risk signature for UVM. (C) Coefficients of the genes composing the GGI. (D) Survival analysis comparing High_GGI and Low_GGI groups in TCGA-UVM, GSE84976, and GSE22138 cohorts. (E) Principal component analysis (PCA) plot for TCGA-UVM, GSE84976, and GSE22138 cohorts based on 5 GGRGs. GGI = GGRGs-derived index, GGRGs = Golgi apparatus-related gene sets, LASSO = Least Absolute Shrinkage and Selection Operator, TCGA = The Cancer Genome Atlas, UVM = uveal melanoma.

Figure 3.

Clinical and somatic mutation characteristics of the GGI. (A) Heatmap showing the expression levels of genes composing the GGI score. (B) Comparative analysis of GGI scores across different clinical and pathological characteristic groups. (C) Top 10 most frequently mutated somatic genes in High_GGI and Low_GGI groups. (D) Comparison of Tumor Mutation Burden (TMB) between High_GGI and Low_GGI groups. (E) Scatter plot illustrating the correlation between GGI and TMB. *P < .05; **P < .01; ****P < .00001. GGI = GGRGs-derived index, ns = not significant.

Figure 4.

Biological processes and pathway features associated with the GGI. (A) GO enrichment analysis of differentially expressed genes between High_GGI and Low_GGI groups. (B) KEGG pathway enrichment analysis of differentially expressed genes between High_GGI and Low_GGI groups. GGI = GGRGs-derived index, GO = Gene Ontology, KEGG = Kyoto Encyclopedia of Genes and Genomes.

Figure 5.

Association of the GGI with drug sensitivity. (A) Differential analysis of drug sensitivity for 45 drugs between High_GGI and Low_GGI groups. (B) Correlation analysis between GGI, its 5 constituent genes, and the sensitivity to 45 drugs. *P < .05; **P < .01; ***P < .0001; ****P < .00001. GGI = GGRGs-derived index.

Figure 6.

Relationship between the GGI and the tumor microenvironment. (A) Immune infiltration difference analysis and correlation analysis of GGI with immune cell infiltration between High_GGI and Low_GGI groups. *P < .05; **P < .01; ***P < .0001. (B) Differences in Stromalscore, Immunescore, ESTIMATEscore, and Tumor purity between High_GGI and Low_GGI groups. (C) Scatter plots demonstrating the correlation between MHC_IPS, EC_IPS, SC_IPS, CP_IPS, AZ_IPS, and GGI. GGI = GGRGs-derived index.

Figure 7.

Construction of a nomogram for UVM prognosis based on the GGI. (A) Nomogram established using GGI and age as predictors. (B) ROC curve analysis demonstrating the predictive performance of the nomogram for 1-, 2-, and 3-year overall survival rates. (C) Decision curve analysis comparing the predictive utility of the nomogram against other prognostic factors for 1-year overall survival. (D) Calibration curves illustrating the agreement between the predicted and actual 1- and 2-year overall survival probabilities for the nomogram constructed with GGI and age as inputs. GGI = GGRGs-derived index, ROC = receiver operating characteristic, UVM = uveal melanoma.

Figure 8.

Evaluation of the biological function of CHST9 in UVM cells. (A) qPCR determination of the CHST9 mRNA level. (B) Western blot determination of the protein level of CHST9. (C) Comparison of the protein level of CHST9 between the si-NC and si-CHST9 groups. (D) Representative images of the transwell invasion assay. (E) Comparison of the relative invasive cells between the si-NC and si-CHST9 groups. (F) Representative images of the colony formation assay. (G) Comparison of the relative colony number between the si-NC and si-CHST9 groups. (H) Representative images of the wound healing assay. (I) comparison of the relative wound healing closure between the si-NC and si-CHST9 groups. *P < .05, **P < .01, ***P < .001, ****P < .0001. CHST = carbohydrate sulfotransferase protein family, UVM = uveal melanoma.