Establishment and validation of an autophagy-related prognostic signature for survival predicting in cutaneous melanoma

Abstract

Existing staging system for prognosis evaluating for Skin Cutaneous Melanoma (SKCM) patients had defects of subjective, inaccuracy and inconsistently, therefore, to identify specific and applicable prognostic markers and promote personalized therapeutic interventions is urgently required. This study aims to build a robust autophagy-related genes (ARGs) signature for prognosis monitoring of SKCM patients. We determined 26 ARGs as differentially expressed autophagy-related genes (DEARGs) from 103 SKCM and 23 normal skin samples in GSE15605 and GSE3189 datasets. Optimal prognostic DEARGs composed the risk model were screened and verified in 458 SKCM patients in TCGA cohort as the training cohort and 209 patients in GSE65904 as the test cohort. Finally, 4 optimal independent prognostic DEARGs (CAPNS1, DAPK2, PARP1 and PTK6) were filtered out in the training cohort to establish the risk model. A prognostic nomogram was established for quantitative survival prediction. The risk model grouped high-risk SKCM cancer patients exhibited significantly shorter survival times in both training and test cohorts. The area under the ROC curve for risk score model was 0.788 and 0.627 in the training and test cohorts indicated the risk model was relatively accurate for prognosis monitoring. Clinical correlation analysis exhibited that risk score was an independent predictor for prognosis significantly associated with T/N classification. The prognostic value of the 4 risk genes formed the risk model was also validated respectively. We identified a novel autophagy-related signature for prognosis monitoring. It has the potential to be an independent prognostic indicator and can benefit targeted therapy.

Keywords: Skin cutaneous melanoma; autophagy-related genes; prognostic risk model; survival prediction; targeted therapeutic intervention.

Figure 1

Differentially expressed autophagy-related genes (DEARGs) between 103 SKCM specimens and 23 normal skin tissues. A. Normalization of raw data in GSE15605 and GSE3189 datasets. Blue and red columns represent samples from GSE15605 and GSE3189 datasets respectively. The 23 columns on the left represent normal skin specimens and 103 columns on the right represent SKCM samples. B. Volcano plot of 232 autophagy-related genes (ARGs) in GSE15605 and GSE3189 datasets. Red spots represent up-regulated ARGs, green spots represent down-regulated ARGs. Filter criteria is |Log₂Fold Change| >1.0. C, D. Heatmap and boxplot of 26 DEARGs in SKCM and normal tissues. The depth of the color of the columns represents its expression intensity in the corresponding samples. E. PPI network of 26 DEARGs. Red or Green nodes represent up-regulated or down-regulated DEARGs. The depth of the color of the nodes was associated with log₂Fold Change. There is a negative correlation between P-value and nodes’ size and a positive correlation between the combined score of protein interaction and ligatures’ width. Square and diamond nodes stand for hub genes that interactive with >10 or >4 proteins.

Figure 2

Functional enrichment analyses of 26 DEARGs. A. Bubble plot of significant GO terms. B. Circle plot of KEGG analyses revealed significant pathways that DEARGs are involved in.

Figure 3

Identify prognostic DEARGs and build a prognostic model based on the TCGA cohort with 458 SKCM patients as the training cohort. A. Forest plots visualized the 8 prognostic DEARGs identified by univariate Cox analysis in the TCGA training cohort. B. Screening of the optimal DEARGs used for the final establishment of the prognostic risk model using multivariate Cox regression analysis. C. The prognostic nomogram for quantitative prediction of SKCM patients’ survival.

Figure 4

Validation of the prognostic risk model in TCGA and GEO independent cohort respectively. A, C. Kaplan-Meier curve of the high-risk and low-risk SKCM patients in the TCGA cohort and GEO cohort. B, D. Survival-dependent ROC curves of risk score and other clinical indicators in the TCGA cohort and GEO cohort. E, F, H, I. risk scores distribution and corresponding survival status of SKCM patients in the TCGA cohort and GEO cohort. G, J. The heatmap of 4 risk genes composed of the risk model in the TCGA cohort and GEO cohort.

Figure 5

Exploration of clinical correlations between the risk score, 4 risk genes and clinicopathological parameters. A. Pathological stage. B. T classification. C. M classification. D. N classification.

Figure 6

Evaluation of the prognostic value of the 4 risk genes (CAPNS1, DAPK2, PARP1 and PTK6) that formed the prognostic risk signature. A. Kaplan-Meier analysis of 4 risk genes. B. The mRNA expression levels of 4 risk genes in normal skin and SKCM tissues.

Figure 7

Evaluation of the prognostic value of the 4 risk genes (CAPNS1, DAPK2, PARP1 and PTK6) at the protein level. A. Immunohistochemistry staining of the 4 risk genes in SKCM and normal tissues. The antibody of CAPNS1, PARP1, PTK6 protein is HPA006872, HPA045168 and HPA036071. Immunohistochemistry staining result of DAPK2 is the absence in The Human Protein Atlas database. B. GSEA analysis of 4 risk genes in SKCM.