Development and validation of a cuproptosis-immune prognostic signature for risk stratification and personalized therapy in cutaneous melanoma Authors: Meiru Zhao, Meng Xiao, Xinmei Zhang, Junyan Zhang, Tong Liu, Huiping Wang Published: 2026 ABSTRACT Skin cutaneous melanoma (SKCM) is a highly aggressive malignancy with rising global incidence and mortality. Despite advances in immunotherapy and targeted therapies, treatment resistance remains a challenge, necessitating novel prognostic biomarkers and therapeutic strategies. Cuproptosis, a copper-dependent form of regulated cell death, and immune-related pathways have emerged as critical players in tumor progression. However, their combined prognostic potential in SKCM remains unexplored. Here, we constructed a cuproptosis-immune-related gene signature to predict SKCM prognosis and guide therapy. Using the TCGA database, we identified 474 cuproptosis-related immune genes through Pearson correlation analysis. By integrating the GTEx database, differential expression analysis revealed that 194 of these genes were significantly dysregulated in SKCM. Univariate Cox and LASSO regression analyses established a 12-gene prognostic model (C3AR1, CCL8, CCR1, CTLA4, HLA-DRB1, IFIH1, IL2RA, IRF9, KIR2DL4, TLR1, TNFRSF21, XCL2), stratifying patients into high- and low-risk groups. The model demonstrated robust predictive accuracy in training and validation cohorts. High-risk patients exhibited poorer survival, reduced immune infiltration, suppressed checkpoint expression, and lower tumor mutational burden (TMB), suggesting an immunosuppressive microenvironment. Conversely, low-risk patients showed enhanced immune infiltration, higher TMB, and increased checkpoint-related gene expression, suggesting an immune-inflamed but functionally restrained phenotype with potential relevance to immune checkpoint blockade. Drug sensitivity analysis revealed high-risk patients may benefit more from targeted therapies. A nomogram integrating risk scores and clinical factors further improved prognostic prediction, with calibration curves demonstrating strong concordance between predicted and observed survival probabilities. Single-cell RNA sequencing illustrated the cellular distribution of model genes, and functional experiments demonstrated that XCL2 suppresses melanoma cell proliferation, migration, and invasion. This study develops and validates a cuproptosis–immune integrated prognostic signature for SKCM, providing a framework to link cuproptosis-associated biology with immune microenvironmental features, risk stratification, and potential therapeutic decision-making.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-026-05396-0. Introduction Skin cutaneous melanoma (SKCM) is a malignant skin tumour that arises from the neoplastic transformation of subepidermal melanocytes [1]. SKCM is extremely aggressive and attacks melanocyte-containing tissues, particularly the skin, and accounts for approximately 4 per cent of skin cancer cases, with a mortality rate of 80 per cent [2, 3]. The incidence and mortality of SKCM have significantly grown in recent years, according to an analysis of global cancer statistics [4, 5]. Despite significant advances in therapeutic approaches, including immunotherapy and targeted therapies, these agents tend to develop resistance, necessitating the search for alternative treatment strategies [6]. New biomarkers of SKCM and effective risk models are therefore needed to improve early diagnosis, predict prognosis and guide clinical treatment. Recent studies have increasingly highlighted the pivotal role of cuproptosis-related genes and immune-related genes in cancer progression, suggesting that they may serve as important prognostic indicators. The involvement of copper ions in cancer progression has been widely studied, with evidence suggesting their participation in activating signaling pathways linked to cell growth. Cuproptosis is a newly discovered mechanism of regulated cell death triggered by excess Cu2+. The process of cuproptosis involves copper-sulfur proteins, where copper binds to lipid-acylated proteins in the tricarboxylic acid cycle, leading to the aggregation of these proteins and subsequent loss of iron-sulfur cluster proteins. This cascade induces proteotoxic stress, ultimately causing cell death [7]. Numerous studies indicate that cuproptosis promotes tumor cell death through various mechanisms, and Cu homeostasis in tumors is crucial for cancer progression [8–14]. Immunotherapeutics are the fastest-growing class of drugs with significant impact on cancer treatment and human health [15]. The level of immune cells has significant prognostic and predictive value and is closely related to the degree of response to cancer immunotherapy [16]. T lymphocytes infiltrating the tumour before and after treatment have been shown to be a favourable prognostic factor in patients with SKCM [16]. B lymphocytes also play an important role in the immunobiology of melanoma in predicting survival and response to immunotherapy [17]. Emerging evidence underscores a dynamic crosstalk between cuproptosis and tumor immunity, positioning copper metabolism as a critical regulator of the tumor microenvironment (TME) [18]. Recent studies reveal that copper homeostasis not only drives cuproptosis but also modulates immune cell activity. For instance, cuproptosis-related processes have been reported to influence antitumor immunity partly through CD8+ T cells [19]. Copper ions in tumor cells have been shown to regulate PD-L1 expression to influence immune escape [20]. These findings suggest that cuproptosis-related pathways may directly or indirectly influence immune evasion mechanisms. Although several cuproptosis-related prognostic models have been reported in different malignancies, the specific relevance of cuproptosis-associated immune features in SKCM remains insufficiently defined. This issue is particularly important because melanoma is a highly immunogenic tumor, and the clinical benefit of immune checkpoint blockade is strongly influenced by tumor-intrinsic metabolic programs, immune infiltration, T-cell functional states, and immune escape mechanisms. Therefore, an integrated analysis of cuproptosis-related and immune-related genes may provide additional biological and clinical information beyond conventional immune-only prognostic models. Materials and methods Data collection Gene expression data and corresponding clinical information for cutaneous melanoma were downloaded from The Cancer Genome Atlas (TCGA) database through the Genomic Data Commons (GDC) portal under the TCGA-SKCM project. After excluding cases with overall survival (OS) ≤ 30 days or incomplete clinical information, 406 patients were included in the training cohort. The Gene Expression Omnibus (GEO) database provided the GSE65904 dataset, and 208 patients with complete survival information and OS > 30 days were included in the validation cohort. Normal skin tissue expression data were obtained from the Genotype-Tissue Expression (GTEx) database. The inclusion criteria for all cohorts were an OS time of more than 30 days and complete clinical information, where applicable. Identification of immune-related genes co-expressed with cuproptosis-related genes From previous studies [7] we obtained 19 cuproptosis-related genes and listed them in Table S1. ImmPort (https://www.immport.org) provided the list of immune-related genes, which included 2483 genes. Pearson correlation analysis was performed between 19 cuproptosis-related genes and immune-related genes using R. In this study, cuproptosis-related immune genes (CRIGs) were defined as immune-related genes positively co-expressed with cuproptosis-related genes. Therefore, immune-related genes with a Pearson correlation coefficient > 0.3 with at least one cuproptosis-related gene were selected for subsequent analysis. Identification of differentially expressed CRIGs between cancer and adjacent normal tissues The CRIGs identified from TCGA were intersected with genes available in the GTEx dataset. TCGA-SKCM tumor samples and GTEx normal skin samples were merged according to common gene symbols. Because the TCGA and GTEx expression matrices were stored on different transformed FPKM scales, the GTEx expression data were converted to a log2(FPKM + 1)-like scale to make them comparable with the TCGA expression matrix. After merging the two datasets, between-array normalization was performed using the normalizeBetweenArrays function in the R package “limma” to reduce distributional differences across samples. The normalized expression matrix was then used for differential expression analysis between TCGA-SKCM tumor samples and GTEx normal skin samples using the “limma” package. Genes with |log2FC| > 1 and P < 0.05 were defined as differentially expressed CRIGs. Develop a prognostic model based on CRIGs Univariate Cox analysis was conducted on the differentially expressed CRIGs to identify genes with prognostic significance. The prognostic CRIGs were then subjected to LASSO-Cox regression to reduce dimensionality and construct the prognostic signature. Cross-validation was used to determine the optimal penalty parameter lambda. The LASSO coefficient profile was used to visualize the shrinkage of gene coefficients across different log(lambda) values, and the cross-validation curve was used to evaluate model performance based on partial likelihood deviance. The final gene signature was selected at lambda.min, which corresponded to the minimum cross-validated partial likelihood deviance. Genes with non-zero coefficients at lambda.min were retained for construction of the prognostic model. The risk score was calculated using the following formula: ∑ (each gene’s expression × corresponding coefficient). ROC curves were plotted using the “survivalROC” [21] package to assess the predictive accuracy of the model. Identify and analyze the differentially expressed genes between high- and low-risk groups The “limma” package was used to identify the differentially expressed genes between the high- and low-risk groups. Genes with | Log2FC | > 3 and a P-value less than 0.001 were identified as differentially expressed. These genes were then analyzed using GO (Gene Ontology) and KEGG (Kyoto Encyclopedia of Genes and Genomes) analyses with the “clusterProfiler” [22] R package. Gene Set Enrichment Analysis (GSEA) was performed using the ”c2.cp.kegg.v7.4.symbols.gmt” dataset. Conduct unsupervised clustering of CRIGs model genes The “ConsensusClusterPlus” [23] R package was used to perform unsupervised consensus clustering based on the 12 model genes in both the training and validation cohorts. The parameters were set as distance = “euclidean”, clusterAlg = “km”, and maxK = 9 for exploratory assessment of possible cluster structures. The optimal cluster number was determined based on the consensus matrix, CDF curves, delta area plot, and the reproducibility of cluster-associated survival differences across the training and validation cohorts. Develop a nomogram to predict the prognosis of SKCM patients In the TCGA-SKCM dataset, univariate Cox analysis was used to identify factors with prognostic significance, while multivariate Cox analysis was employed to identify independent prognostic factors. The ‘rms’ R package was used to create a nomogram, and a calibration curve was used to assess the prediction accuracy of nomogram. Evaluate the immune microenvironment Employ the ESTIMATE and ssGSEA algorithms to assess the infiltration levels of immune pathways and immune cells. Utilize the “ggpubr” package to assess the expression levels of immune checkpoints in TCGA-SKCM. Based on the expression profiles of immune-related genes and stromal-related genes, the Estimate algorithm calculates the EstimateScore, ImmuneScore and StromalScore for each patient [24]. The TIDE algorithm, a computational framework for modeling tumor-immune interactions, was applied to predict the likelihood of clinical benefit from immune checkpoint inhibitors (ICIs) in SKCM cohorts. Analysis of tumor mutational burden and drug sensitivity Somatic mutation data for TCGA-SKCM were downloaded from the GDC portal in Mutation Annotation Format (MAF). We did not perform de novo mutation calling from raw sequencing data; instead, publicly available GDC-processed somatic mutation MAF files were used. These files were generated through the TCGA/GDC somatic variant processing pipeline, in which primary somatic variant calling, low-confidence variant filtering, and germline masking had been performed before public release. For tumor mutational burden (TMB) calculation, non-synonymous somatic mutations were counted for each patient. The “maftools” [25] R package was used to process MAF files and visualize mutation profiles. Utilize the “pRRophetic” [26] package to calculate IC50 values for 138 drugs in TCGA-SKCM patients. Characterization of cuproptosis-related immune genes by single-cell RNA sequencing Single-cell RNA-seq data from three SKCM samples were obtained from GSE215120. Data analysis was performed using the “Seurat” R package. Quality control criteria included: (1) Each cell must contain > 300 but < 7000 detected genes; (2) UMI counts per cell > 1000 with exclusion of the top 3% cells with highest counts; (3) Mitochondrial gene expression < 10% of total genes; (4) Erythrocyte gene expression < 3% of total genes. Cell annotation was based on well-established canonical biomarkers. Cell culture and transfection Human melanoma cell lines A375 and A2058 were obtained from the American Type Culture Collection (ATCC). Cell line identities were confirmed by short tandem repeat (STR) profiling, and all cells were routinely tested and confirmed negative for mycoplasma contamination before use. Cells were cultured in DMEM supplemented with 10% fetal bovine serum and maintained in a humidified incubator at 37 °C with 5% CO₂. Transfection was performed using Lipofectamine 2000 reagent with small interfering RNA (siRNA) targeting XCL2. The siRNA sequences were as follows: forward, 5′-GGAGUGAAGUCUCACAUAGTT-3′; Reverse, 5′-CUAUGUGAGACUUCACUCCTT-3′. RNA extraction, reverse transcription, and quantitative real-time PCR Total RNA was extracted using the SPARK-easy Cell RNA Kit. cDNA synthesis was performed using the GoScript™ Reverse Transcriptase System (Promega, USA). Quantitative PCR amplification was carried out using the SLAN-96P Real-Time PCR System (Shanghai Hongshi Biotechnology, China). The primer sequences for XCL2 were as follows: Forward: 5′-GTAGGGAGTGAAGTCTCACATAG-3′; Reverse: 5′-CCACGTGGCTTGTGGATCA-3′. Cell counting kit-8 assay A375 and A2058 cells were seeded into 96-well plates at a density of 5,000 cells per well. After 24 h of incubation at 37 °C with 5% CO₂, 100 µL of 10% CCK-8 diluted solution (Beyotime Biotechnology, Shanghai, China) was added to each well. The plates were then protected from light and incubated for 1 h at 37 °C with 5% CO₂. Absorbance was measured at a wavelength of 450 nm using a microplate reader. Each sample was analyzed in triplicate. Migration and invasion assays A375 and A2058 cells suspended in serum-free medium were seeded into each Transwell chamber (Corning, Corning, NY, USA) at a density of 50,000 cells per chamber. The lower chambers were filled with complete medium supplemented with 20% fetal bovine serum. After 48 h of incubation, non-migrated cells on the upper surface of the membrane were removed by swabbing. The migrated cells on the underside of the membrane were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet. Three random microscopic fields per chamber were imaged and quantified. For invasion assays, the chambers were pre-coated with 20% Matrigel matrix (Corning), with subsequent procedures identical to those described for the migration assay. Colony formation assays A375 and A2058 cells were seeded into 6-well plates at a density of 500 cells per well. After 2 weeks of incubation at 37 °C with 5% CO₂, the cells were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet. Colonies containing more than 50 cells were counted. Statistical analysis Differences in immune cell infiltration, immune pathway scores, and checkpoint-related gene expression between the high- and low-risk groups were assessed using the two-sided Wilcoxon rank-sum test, followed by Benjamini–Hochberg FDR correction for multiple comparisons. All in vitro experiments, including RT-qPCR, CCK-8, Transwell migration/invasion, and colony formation assays, were performed in three independent biological replicates. Quantitative data are presented as mean ± SD. For comparisons between si-NC and si-XCL2 within each cell line, statistical significance was determined using a two-tailed unpaired Student’s t-test. P < 0.05 was considered statistically significant. Results Identify differentially expressed immune genes associated with cuproptosis Pearson correlation analysis was performed on 19 cuproptosis-related genes and 1739 immune-related genes in TCGA data, identifying 474 immune genes with a correlation coefficient greater than 0.3, referred to as CRIGs (Fig. 1A). The list of CRIGs is presented in Table S2. After merging the TCGA and GTEx datasets and normalizing the data, we intersected the merged dataset with CRIGs, obtaining a total of 455 CRIGs. Differential expression analysis between normal skin tissue expression data from the GTEx database and the melanoma expression data from TCGA identified a total of 194 differentially expressed genes (Fig. 1B). Fig. 1Identification of cuproptosis-related immune genes. A Pearson correlation analysis identified immune genes associated with cuproptosis-related genes. B The volcano plot shows the differentially expressed CRIGs between cancer and normal skin tissues Identification of cuproptosis-related immune genes. A Pearson correlation analysis identified immune genes associated with cuproptosis-related genes. B The volcano plot shows the differentially expressed CRIGs between cancer and normal skin tissues Develop and validate a prognostic model based on CRIGs First, univariate Cox analysis was performed on the 194 differentially expressed genes, resulting in 156 prognostic CRIGs. Next, these 156 prognostic CRIGs were entered into LASSO-Cox regression for dimensionality reduction. The LASSO coefficient profiles showed the shrinkage of candidate gene coefficients across different log(lambda) values (Fig. 2A). Cross-validation based on partial likelihood deviance was used to select the optimal lambda value (Fig. 2B). At lambda.min, 12 genes retained non-zero coefficients and were used to construct the prognostic signature. Using the previously described algorithm, risk scores were determined for every melanoma patient based on these coefficients. Melanoma patients were classified into high-risk and low-risk groups based on the median risk score. The same method was applied to classify melanoma patients in the validation set GSE65904. Heatmaps displayed the expression levels of the 12 model genes in both the training and validation sets (Fig. 2C, D). The survival status distribution plot shows that the proportion of deceased patients in the high-risk group is higher in both the training and validation sets (Fig. 2E-H). According to survival analysis, high-risk patients in both the training and validation cohorts had worse outcomes (Fig. 2I, K). The ROC curves demonstrated AUC values of 0.726, 0.709, and 0.699 for 1-, 3-, and 5-year survival rates in the training set (Fig. 2J), while the validation set showed corresponding values of 0.684, 0.648, and 0.595 (Fig. 2L). Fig. 2Establishment of a prognostic model based on CRIGs. A, B The number of genes in the model was determined through LASSO-Cox regression. C, D The heatmap displays the expression of model genes in high-risk and low-risk groups (C: TCGA; D: GSE65904). E-H Scatter plots of survival status distribution with increasing risk scores (E, G: TCGA; F, H: GSE65904). I, K Kaplan-Meier curves demonstrating inferior survival in high-risk group (I: TCGA; K: GSE65904). J, L ROC curves demonstrating AUC values for 1-, 3-, and 5-year survival (J: TCGA, 0.726, 0.709, 0.699; L: GSE65904, 0.684, 0.648, 0.595) Establishment of a prognostic model based on CRIGs. A, B The number of genes in the model was determined through LASSO-Cox regression. C, D The heatmap displays the expression of model genes in high-risk and low-risk groups (C: TCGA; D: GSE65904). E-H Scatter plots of survival status distribution with increasing risk scores (E, G: TCGA; F, H: GSE65904). I, K Kaplan-Meier curves demonstrating inferior survival in high-risk group (I: TCGA; K: GSE65904). J, L ROC curves demonstrating AUC values for 1-, 3-, and 5-year survival (J: TCGA, 0.726, 0.709, 0.699; L: GSE65904, 0.684, 0.648, 0.595) Identify and analyze the differentially expressed genes between the two risk groups Differential expression analysis between the two risk groups identified 447 differentially expressed genes, with 442 genes downregulated in the high-risk group and only 5 genes upregulated in the high-risk group. GO and KEGG enrichment analyses based on the differentially expressed genes between the two risk groups showed predominant enrichment of immune-related terms. In the enrichment dot plots, the x-axis represents GeneRatio, bubble size indicates the number of enriched genes, and color indicates enrichment significance. GO analysis highlighted immune activation and immune receptor-related terms, including B cell receptor signaling pathway, antigen receptor-mediated signaling pathway, immune response-activating signal transduction, lymphocyte-mediated immunity, immunoglobulin production, T cell receptor complex, chemokine receptor binding, cytokine activity, and MHC protein binding (Supplementary Fig. 1A). KEGG analysis further identified immune-related pathways, including PD-L1 expression and PD-1 checkpoint pathway in cancer, Th1 and Th2 cell differentiation, natural killer cell-mediated cytotoxicity, antigen processing and presentation, chemokine signaling pathway, T cell receptor signaling pathway, and cytokine–cytokine receptor interaction (Supplementary Fig. 1B). GSEA, which was performed using the ranked gene list, showed that immune-related pathways were more active in the low-risk group, whereas metabolic pathways were relatively enriched in the high-risk group. These results suggest that risk stratification is associated with both immune microenvironmental differences and metabolic remodeling (Supplementary Fig. 1C). Unsupervised clustering of model-related genes Consensus clustering was carried out using 12 model-related genes. The most noticeable variations between subgroups occurred when K = 2, indicating that the 12 model genes effectively divide melanoma patients into two clusters (Fig. 3A-D). According to survival analysis, cluster 2 had a better prognosis (Fig. 3E, F). The Sankey diagram indicated that the low-risk group comprised the great majority of individuals in cluster 2 (Fig. 3G, H). Fig. 3Unsupervised clustering of CRIGs model genes. A, C SKCM patients were grouped into two molecular clusters when k = 2, based on the model gene profile (A: TCGA; C: GSE65904). B, D Empirical cumulative distribution function plot displaying consensus distributions for each k value (from 2 to 9) (B: TCGA; D: GSE65904). E, F Kaplan-Meier analysis of the prognosis of SKCM patients belonging to two different molecular clusters (E: TCGA; F: GSE65904). G, H Alluvial diagram shows the interrelationship between molecular clusters, survival status, and risk groups in SKCM patients (G: TCGA; H: GSE65904) Unsupervised clustering of CRIGs model genes. A, C SKCM patients were grouped into two molecular clusters when k = 2, based on the model gene profile (A: TCGA; C: GSE65904). B, D Empirical cumulative distribution function plot displaying consensus distributions for each k value (from 2 to 9) (B: TCGA; D: GSE65904). E, F Kaplan-Meier analysis of the prognosis of SKCM patients belonging to two different molecular clusters (E: TCGA; F: GSE65904). G, H Alluvial diagram shows the interrelationship between molecular clusters, survival status, and risk groups in SKCM patients (G: TCGA; H: GSE65904) Develop and validate a nomogram Initial screening through univariate Cox regression assessed clinical variables and risk scores, followed by multivariate Cox regression to identify independent prognostic factors, including age, T stage, N stage, and risk score (Fig. 4A, B). Using the “rms” R package, construct a nomogram based on independent prognostic factors to predict the 1-, 3-, and 5-year survival rates of melanoma patients (Fig. 4C). Calibration curves show high agreement between nomogram predictions and actual survival results (Fig. 4D-F). Fig. 4Independent prognostic analysis of the CRIGs prognostic model and the construction of a nomogram and calibration. A The hazard ratio (HR) and 95% confidence interval of risk score and all clinical features were calculated using the univariate Cox regression analysis. B The HR and 95% confidence interval of risk score and clinical features were calculated using the multivariate Cox regression analysis. C The nomogram that includes the risk score and independent clinical factors predicted the probability of the 1-, 3-, and 5-year OS. D The calibration curves for the predicted 1-year OS rates. E The calibration curves for the predicted 3-year OS rates. F The calibration curves for the predicted 5-year OS rates Independent prognostic analysis of the CRIGs prognostic model and the construction of a nomogram and calibration. A The hazard ratio (HR) and 95% confidence interval of risk score and all clinical features were calculated using the univariate Cox regression analysis. B The HR and 95% confidence interval of risk score and clinical features were calculated using the multivariate Cox regression analysis. C The nomogram that includes the risk score and independent clinical factors predicted the probability of the 1-, 3-, and 5-year OS. D The calibration curves for the predicted 1-year OS rates. E The calibration curves for the predicted 3-year OS rates. F The calibration curves for the predicted 5-year OS rates Estimation of tumor microenvironments To investigate the roles of these 12 CRIGs in the immune microenvironment, Box plots show the infiltration levels of 23 immune cells, with results indicating low-risk group had a higher level of immune cell infiltration (Fig. 5A). Box plots show the infiltration levels of immune-related pathways, with results showing high infiltration of all immune-related pathways in the low-risk group (Fig. 5B). The expression of immune checkpoint-related genes was further compared between the two risk groups. The low-risk group showed higher expression of multiple checkpoint-related genes (Fig. 5C). Considering that this group also exhibited increased immune cell infiltration and enhanced immune-related pathway activity, the elevated checkpoint expression may reflect an immune-inflamed microenvironment accompanied by adaptive inhibitory regulation, rather than a purely immunosuppressive state. We then used the TIDE algorithm to further evaluate immune-related functional states. Patients in the low-risk group had higher TIDE scores and Dysfunction scores, but lower Exclusion scores (Fig. 5D-F), suggesting that immune cells may effectively infiltrate these tumors, whereas T-cell function may remain restrained by checkpoint-related inhibitory mechanisms. Together, these findings indicate that the low-risk group may represent an immune-inflamed but checkpoint-restrained phenotype [27]. Additionally, the Estimate algorithm showed that the low-risk group had higher EstimateScore, ImmuneScore, and StromalScore (Fig. 5G-I). Fig. 5TME, and checkpoint analysis in SKCM. A The box plots of immune cells between the high risk and low risk groups. B The box plots of immune related pathways between the high risk and low risk groups. C The box plots of checkpoint related genes between the high risk and low risk groups. D-F The violin plot shows the differences in TIDE scores (D), Dysfunction scores (E), and Exclusion scores (F) between high-risk and low-risk groups. G-I The boxplot shows the differences in ESTIMATE scores (G), Immune scores (H), and Stromal scores (I) between high-risk and low-risk groups. For immune cell, immune pathway, and checkpoint-related gene comparisons, statistical significance was determined using Benjamini–Hochberg FDR-adjusted P values. For single-score comparisons, nominal P values are shown. *P or FDR < 0.05, **P or FDR < 0.01, ***P or FDR < 0.001 TME, and checkpoint analysis in SKCM. A The box plots of immune cells between the high risk and low risk groups. B The box plots of immune related pathways between the high risk and low risk groups. C The box plots of checkpoint related genes between the high risk and low risk groups. D-F The violin plot shows the differences in TIDE scores (D), Dysfunction scores (E), and Exclusion scores (F) between high-risk and low-risk groups. G-I The boxplot shows the differences in ESTIMATE scores (G), Immune scores (H), and Stromal scores (I) between high-risk and low-risk groups. For immune cell, immune pathway, and checkpoint-related gene comparisons, statistical significance was determined using Benjamini–Hochberg FDR-adjusted P values. For single-score comparisons, nominal P values are shown. *P or FDR < 0.05, **P or FDR < 0.01, ***P or FDR < 0.001 Tumor mutation burden analysis Previous research suggests that individuals with higher TMB are more likely to show better outcomes when treated with immune checkpoint inhibitors. The TMB analysis reveals that the group with lower risk scores exhibits a higher TMB (Fig. 6A), suggesting greater tumor immunogenicity and potential relevance to immune checkpoint blockade. However, this finding should be interpreted together with TIDE, immune infiltration, and checkpoint-related gene expression, rather than as direct evidence of clinical immunotherapy response. Survival analysis indicates that individuals with high TMB tend to have a more favorable prognosis (Fig. 6B). When TMB and risk scores are considered together, patients with high TMB and low-risk scores show the most favorable outcomes (Fig. 6C). The waterfall plot shows the mutation patterns of the 20 most frequently mutated genes in the TCGA-SKCM dataset (Fig. 6D, E). In general, the low-risk group shows a higher frequency of mutations in these genes. This implies that individuals with lower risk scores generate a greater number of antigens, improving the immune system ability to recognize and target tumor cells, which contributes to a better prognosis. Fig. 6Determination of the relationship between the prognostic model and immunotherapy. A The TMB difference in the high risk and low risk groups tested by the Wilcoxon rank-sum. B Kaplan–Meier plots of overall survival for patients in the high TMB group and low TMB group. C Kaplan–Meier plots of overall survival for patients in the risk groups and TMB groups. D, E The waterfall plot shows the top 20 genes mutated and their difference in the TCGA cohort high risk and low risk groups Determination of the relationship between the prognostic model and immunotherapy. A The TMB difference in the high risk and low risk groups tested by the Wilcoxon rank-sum. B Kaplan–Meier plots of overall survival for patients in the high TMB group and low TMB group. C Kaplan–Meier plots of overall survival for patients in the risk groups and TMB groups. D, E The waterfall plot shows the top 20 genes mutated and their difference in the TCGA cohort high risk and low risk groups Drug sensitivity analysis The IC50 values for 138 drugs were computed with the pRRophetic package, and a heatmap was generated for 33 drugs (P < 0.05) and 12 model genes (Fig. 7A). The six drugs with the most significant differences were then displayed in box plots and analyzed for correlation (Fig. 7B-M). Drug sensitivity analysis based on pRRophetic predicted differential IC50 values for several compounds between the high- and low-risk groups. The high-risk group showed lower predicted IC50 values for multiple targeted agents, suggesting potential sensitivity to these compounds at the computational prediction level. These results should be interpreted as hypothesis-generating and require further experimental and clinical validation. Fig. 7Efficacy of the prognostic model in predicting drug sensitivity. A The relationship between drugs, risk score, and model genes. B-M Boxplots of the comparison of IC50 of drugs between high risk and low risk groups, and correlation between the IC50 and risk score in TCGA cohort. *p < 0.05, **p < 0.01, ***p < 0.001 Efficacy of the prognostic model in predicting drug sensitivity. A The relationship between drugs, risk score, and model genes. B-M Boxplots of the comparison of IC50 of drugs between high risk and low risk groups, and correlation between the IC50 and risk score in TCGA cohort. *p < 0.05, **p < 0.01, ***p < 0.001 Overview of the scRNA-seq data generated from SKCM After performing quality control on the data, we obtained a total of 20,744 cells. Based on the expression patterns of marker genes across different cells, we identified five distinct cell clusters: cancer cells, T cells, B cells, fibroblasts, and endothelial cells (Fig. 8A, B). The bubble plot demonstrates the marker gene expression profiles of the five cell populations (Fig. 8C, D). Subsequently, we analyzed the expression of 12 model genes across different cell types. Among these, CTLA4, IL2RA, KIR2DL4, and XCL2 exhibited significantly higher expression in T cells; HLA-DRB1 showed prominent expression in B cells; TLR1 was markedly upregulated in cancer cells; TNFRSF21 demonstrated elevated expression in fibroblasts; while no significant differences were observed for the remaining genes (Fig. 8E-P). Fig. 8Verification of model genes through sc-RNA seq. A, B tSNE plots of cells generated from Cutaneous tissue. The plots are colored by cell cluster, and the cells are clustered into 5 sub-clusters. Each dot represents a SKCM cell. C, D The bubble plot displays the expression of marker genes in different cell clusters. E-P The expression of signature genes in SKCM visualized in tSNE Verification of model genes through sc-RNA seq. A, B tSNE plots of cells generated from Cutaneous tissue. The plots are colored by cell cluster, and the cells are clustered into 5 sub-clusters. Each dot represents a SKCM cell. C, D The bubble plot displays the expression of marker genes in different cell clusters. E-P The expression of signature genes in SKCM visualized in tSNE XCL2 is associated with favorable prognosis and regulates melanoma cell malignant phenotypes in vitro XCL2, a member of the XC chemokine family, is encoded by a gene located on human chromosome 1 and was initially identified by Yoshida et al. [28, 29]. Previous studies have demonstrated that XCL2 is expressed in immune cells, including natural killer cells and activated T cells [29, 30], which is consistent with our single-cell RNA-seq results showing predominant enrichment of XCL2 in T cells. We therefore further evaluated XCL2 at the tissue level and in melanoma cell lines. In our differential expression analysis, XCL2 was significantly upregulated in melanoma tissues compared with normal tissues, with a log2FC of 3.6142 and a P value of 7.38 × 10^-98 (Supplementary Fig. 2A). This result was further validated in the independent GSE46517 dataset, which included 8 normal skin samples and 31 primary melanoma samples, where XCL2 was also significantly increased in melanoma tissues (P = 0.0164) (Supplementary Fig. 2B). Pan-cancer analysis using GEPIA also showed that XCL2 expression was higher in most tumor types, including SKCM, than in corresponding normal tissues (Supplementary Fig. 2C). Given its prognostic significance in our model and its association with favorable survival, we further investigated the potential role of XCL2 in melanoma using tissue-level expression analyses and in vitro functional assays. Kaplan–Meier survival analysis based on the TCGA database revealed that high XCL2 expression was correlated with better prognosis (Fig. 9A), supporting its favorable prognostic relevance in melanoma. We further compared XCL2 expression between the high- and low-risk groups and found that XCL2 expression was significantly higher in the low-risk group (Supplementary Fig. 2D), supporting the consistency between XCL2 expression and the favorable-risk phenotype. To further evaluate the relationship between XCL2 and the risk score, patients were stratified into four subgroups according to risk group and XCL2 expression. The low-risk/XCL2-high subgroup showed the most favorable survival. In contrast, within the high-risk group, both XCL2-high and XCL2-low subgroups showed poor outcomes, with no obvious survival difference between them (Supplementary Fig. 2E). These results indicate that combined risk/XCL2 stratification further separated patients with favorable prognosis, particularly within the low-risk group. To support the biological rationale for XCL2 knockdown in melanoma cell lines, we assessed basal/control XCL2 expression using the same RT-qPCR experiment performed for knockdown validation. The si-NC group was used to represent the basal/control expression level of XCL2. XCL2 mRNA was readily detectable in both A375 and A2058 cells, with average Ct values of 25.12 and 24.98, respectively, whereas β-actin showed average Ct values of 13.49 and 13.47. These data indicate clear and reproducible basal/control expression of XCL2 in melanoma cells. RT-qPCR further confirmed efficient XCL2 knockdown in both cell lines after si-XCL2 transfection (Fig. 9B). Functional assays demonstrated that XCL2 knockdown significantly enhanced the proliferation rate of melanoma cells, as shown by CCK-8 assay (Fig. 9C, D). Transwell assays indicated that XCL2 depletion markedly promoted the migration and invasion capabilities of melanoma cells (Fig. 9E-G). Furthermore, colony formation assays revealed that XCL2 knockdown substantially increased the proliferative capacity of melanoma cells (Fig. 9H, I). In summary, our findings indicate that XCL2 is associated with favorable prognosis in melanoma, and that XCL2 depletion promotes proliferation, colony formation, migration, and invasion of melanoma cells under in vitro conditions. Fig. 9XCL2 is associated with favorable prognosis and regulates melanoma cell malignant phenotypes in vitro. A Kaplan–Meier survival analysis comparing melanoma patients with high versus low XCL2 expression. B RT-qPCR analysis of XCL2 mRNA levels in A375 and A2058 cells transfected with si-NC or si-XCL2. C, D Proliferation rates of A375 and A2058 cells transfected with si-NC or si-XCL2. E-G Representative images and quantification of Transwell migration and invasion assays using A375 and A2058 cells transfected with si-NC or si-XCL2. H, I Representative images and quantification of colony formation assays in A375 and A2058 cells transfected with si-NC or si-XCL2. Quantitative data are presented as mean ± SD from three independent biological replicates. Statistical significance was determined using a two-tailed unpaired Student’s t-test for comparisons between si-NC and si-XCL2 within each cell line. *P < 0.05, **P < 0.01, ***P < 0.001 XCL2 is associated with favorable prognosis and regulates melanoma cell malignant phenotypes in vitro. A Kaplan–Meier survival analysis comparing melanoma patients with high versus low XCL2 expression. B RT-qPCR analysis of XCL2 mRNA levels in A375 and A2058 cells transfected with si-NC or si-XCL2. C, D Proliferation rates of A375 and A2058 cells transfected with si-NC or si-XCL2. E-G Representative images and quantification of Transwell migration and invasion assays using A375 and A2058 cells transfected with si-NC or si-XCL2. H, I Representative images and quantification of colony formation assays in A375 and A2058 cells transfected with si-NC or si-XCL2. Quantitative data are presented as mean ± SD from three independent biological replicates. Statistical significance was determined using a two-tailed unpaired Student’s t-test for comparisons between si-NC and si-XCL2 within each cell line. *P < 0.05, **P < 0.01, ***P < 0.001 Discussion A growing body of research suggests that cuproptosis significantly contributes to the initiation and progression of cancer [31]. The regulatory effects of copper oxidase on cancer cells can be achieved through multiple pathways, including influencing cell metabolism, promoting tumor cell proliferation or inducing apoptosis. In addition, the immune characteristics within the tumor microenvironment significantly influence tumor development [32]. The extent of infiltration and the functional status of the immune cells directly impact tumor growth and metastasis. Emerging studies have identified a distinct interaction between cuproptosis and the immune response. Cuproptosis can influence the function of immune cells in the tumour microenvironment by altering their patterns of infiltration and activation state, thereby modifying the immune response against tumors. Conversely, the immune system can also impact the appearance of cuproptosis through the secretion of cytokines and the modulation of the redox state in the tumor microenvironment. Therefore, exploring the connection between cuproptosis and the immune response is crucial for unraveling the mechanisms underlying tumor development and for identifying novel therapeutic targets. In this study, we have effectively constructed a prognostic model that combines cuproptosis-related and immune-related genes, offering fresh perspectives for the diagnosis and management of patients with SKCM. An important distinction between our signature and previously reported immune-only prognostic models is that this model was designed to integrate immune-related features with cuproptosis-associated biology. Immune-only signatures mainly reflect the abundance or activation status of immune components in the tumor microenvironment, whereas our CRIG-based model may additionally capture tumor-intrinsic metabolic and regulated cell death programs related to copper homeostasis and mitochondrial function. Consistent with this concept, the risk score in our study was associated with immune infiltration, immune checkpoint expression, TIDE scores, TMB, and predicted drug sensitivity. These results suggest that the cuproptosis–immune integrated model may provide complementary information for prognosis prediction and therapeutic stratification in SKCM. Utilizing expression profiles of cuproptosis-related genes, we employed Pearson correlation analysis to pinpoint immune-related genes exhibiting a correlation coefficient exceeding 0.3, which were subsequently classified as CRIGs. Following this, a series of analytical techniques were applied to construct a prognostic model integrating cuproptosis and immune response. This model includes 12 genes: C3AR1, CCL8, CCR1, CTLA4, HLA.DRB1, IFIH1, IL2RA, IRF9, KIR2DL4, TLR1, TNFRSF21, and XCL2. Evaluations through survival analysis, ROC curve assessment, and calibration plots demonstrate the model’s superior predictive accuracy for SKCM patient prognosis. Furthermore, external validation using the GSE65904 dataset confirms the model’s generalizability. The use of ImmPort-derived immune genes enabled us to focus on immune regulatory programs, including cytokines, chemokines, immune receptors, antigen presentation, and adhesion-related molecules. Together with the GO/KEGG enrichment results, which highlighted cytokine signaling, chemokine activity, T cell receptor signaling, B cell receptor signaling, leukocyte activation, and antigen presentation, these findings support the biological connection between cuproptosis-related genes and immune regulation in SKCM. Inspired by pathway-guided prognostic signature studies such as the TGF-β signaling-related lncRNA model reported by Duan et al. [33], future studies integrating ImmPort with MSigDB HALLMARK and other curated pathway gene sets may further improve the precision and interpretability of CRIG selection. Numerous genes have been identified as playing important roles in the progression of various tumors. Among these, C3AR1, a member of the G protein-coupled receptor family, has been established as a prognostic indicator in tumors such as ovarian and renal carcinoma [34, 35]. Similarly, CCL8 has been recognized for its role in modifying the tumor microenvironment, serving as a prognostic marker in diffuse large B-cell lymphoma [36]. The CCR1 gene contributes to tumor development by altering the tumor microenvironment in hepatocellular carcinoma via the CCL16-CCR1 signaling pathway [37]. CTLA4, known for its role as an immune checkpoint, is crucial in modulating T cell activation and immune responses, making it a target for cancer therapy across multiple cancer types [38]. IL2RA has been validated as an important immune-related biomarker in ovarian cancer [39]. Research has also highlighted IRF9 as a factor influencing survival outcomes in pancreatic ductal adenocarcinoma [40]. Additionally, TLR1 has been linked to prognostic outcomes in low-grade gliomas [41], while XCL2 is associated with adverse prognosis and enhanced immune cell infiltration in clear cell renal cell carcinoma [42]. The relationship between DEG-based GO/KEGG enrichment and GSEA should be interpreted according to their different analytical principles. GO and KEGG analyses were performed using threshold-defined DEGs and therefore mainly highlighted the most strongly altered immune-related genes between the two risk groups. In contrast, GSEA considers the ranked expression changes of all genes and can detect coordinated pathway-level shifts that may not be captured by the DEG cutoff. Therefore, the immune-dominant GO/KEGG results and the high-risk metabolic enrichment observed by GSEA are not contradictory. Rather, they indicate that the high-risk phenotype is characterized by attenuated immune-related transcriptional activity together with broader metabolic remodeling. Analysis of the tumor immune microenvironment revealed that the low-risk group exhibited elevated levels of immune cell infiltration, immune-related pathway activity, and immune checkpoint-related gene expression compared with the high-risk group. These findings suggest that low-risk patients may have a more immune-inflamed tumor microenvironment with pre-existing antitumor immune activity, which may partly contribute to their improved prognosis. However, the higher expression of immune checkpoint-related genes in this group requires cautious interpretation. Although inhibitory checkpoint molecules are commonly associated with T-cell dysfunction or exhaustion, their elevated expression in bulk transcriptomic data may also reflect abundant immune infiltration and ongoing immune activation. Activated T cells physiologically upregulate inhibitory receptors, such as PD-1, CTLA4, LAG3, and BTLA, as part of a normal immune negative-feedback mechanism to prevent excessive immune responses. In the tumor microenvironment, persistent antigen stimulation may lead to sustained checkpoint expression and functional restraint of T cells. Therefore, the low-risk group may represent an immune-inflamed but checkpoint-restrained phenotype, rather than a purely immunosuppressive state. Additionally, low-risk patients demonstrated a higher TMB, suggesting an increased likelihood of generating tumor-associated neoantigens. These neoantigens may enhance tumor immunogenicity and contribute to the higher immune cell infiltration and immune pathway activity observed in this group. This interpretation is also consistent with the clinical observation that, in some cancer types, high PD-L1 expression can predict benefit from PD-1/PD-L1 blockade, as it indicates that this inhibitory pathway is present and may represent an important mechanism of tumor immune escape [43]. Thus, low-risk patients may have a more favorable immune context for immune checkpoint blockade than immune-cold tumors; however, the increased checkpoint-related inhibitory signals also indicate that T-cell functional restraint should be considered. Furthermore, drug sensitivity analysis identified 33 compounds with differential IC50 values between the risk groups, primarily targeting signaling pathways, kinases, or proteasomes. The pRRophetic-based drug sensitivity analysis should be interpreted cautiously. Because pRRophetic predicts IC50 values by applying models trained on pharmacogenomic cancer cell-line datasets, such as GDSC, to tumor transcriptomic profiles, the predicted IC50 values represent computational estimates rather than direct clinical drug responses [26]. Cancer cell lines may not fully capture the complexity of SKCM tumors in patients, including tumor heterogeneity, stromal components, immune-cell interactions, drug metabolism, and pharmacokinetic factors. Therefore, the finding that high-risk patients showed lower predicted IC50 values for several targeted agents should be considered hypothesis-generating. Future hybrid models integrating transcriptomic features, genomic alterations, pathway activity, immune biomarkers, drug chemical properties, patient-derived organoids, PDX models, and clinical treatment-response data may further refine targeted therapy prediction for SKCM [44]. It is noteworthy that XCL2 emerged as a favorable prognostic gene in our model and was further supported by tissue-level expression analyses and in vitro functional validation. The favorable prognostic association of XCL2 in SKCM is consistent with its role as a favorable component of the CRIG-based model. XCL2 expression was significantly higher in the low-risk group, and combined stratification showed that patients with low-risk/XCL2-high tumors had the best prognosis. However, within the high-risk group, both XCL2-high and XCL2-low patients showed poor outcomes, suggesting that the additional discriminative value of XCL2 may be limited within the high-risk context. Because XCL2 is included in the CRIG-based model, XCL2 expression and the overall risk score may provide partially overlapping prognostic information. Recent studies have emphasized that memory-like immune states and durable immune persistence are important for sustained antitumor responses, supporting the possibility that XCL2-high tumors may represent a more favorable immune contexture [45]. Future studies using formal Cox interaction analyses will help determine whether XCL2 modifies the prognostic effect of the CRIG-based risk score. Although the single-cell RNA-seq analysis showed predominant enrichment of XCL2 in T cells, this pattern should be interpreted as a relative distribution across cell populations and does not exclude its detectable expression in melanoma cells. In line with this, our RT-qPCR analysis showed clear and reproducible basal/control expression of XCL2 in A375 and A2058 melanoma cells, supporting the rationale for subsequent knockdown experiments. Functionally, XCL2 depletion promoted melanoma cell proliferation, colony formation, migration, and invasion under in vitro conditions, suggesting that XCL2 can restrain malignant phenotypes of melanoma cells in a cell-culture setting. Previous studies have also reported tumor-tissue expression and cancer-cell-related functional relevance of XCL2 in other solid tumors. In lung cancer, XCL2 was shown to be significantly upregulated in tumor tissues compared with adjacent non-cancerous tissues, and this finding was further supported by immunohistochemical evidence showing XCL2 protein expression in lung cancer specimens [30]. This tissue-level protein evidence is particularly relevant, as it supports that XCL2 can be detected in solid tumor tissues at the protein level and should not be regarded solely as a transcriptomic immune-cell-associated signal. In clear cell renal cell carcinoma, XCL2 was also reported to be upregulated in tumor tissues, and functional manipulation of XCL2 in 786-O and Caki-1 cells significantly affected malignant cellular phenotypes [42]. These findings support the biological plausibility of investigating XCL2 function in cancer cell lines, although its functional direction may be context-dependent across different tumor types. M stage was not incorporated into the final nomogram because the proportion of patients with M1 disease was very low in the analyzed cohort, accounting for only 5.42%. Including such an imbalanced variable may lead to unstable hazard ratio estimation, wide confidence intervals, and potential overfitting. Therefore, M stage was excluded to maintain the stability and reliability of the nomogram. In addition, consensus clustering was performed using the 12 model genes with maxK set to 9 for exploratory assessment. Although k = 2 showed the most interpretable separation and was reproducible in the validation cohort, the choice of cluster number was mainly based on the consensus matrix, CDF curves, delta area plot, and survival relevance. Recent spatial clustering studies, including those by Zhang et al. [46] and Li et al. [47], have emphasized the value of complementary quantitative metrics, such as silhouette width and gap statistic, to improve the objectivity and robustness of cluster-number determination. Therefore, although our k = 2 classification was supported by consensus clustering patterns and validation-cohort reproducibility, additional quantitative metrics may further strengthen the objective validation of the optimal cluster number in future studies. Another limitation is that the immunotherapy relevance of the risk model was inferred from computational indicators, including TIDE, TMB, immune infiltration, and checkpoint-related gene expression. Although the low-risk group showed higher immune infiltration, higher TMB, lower Exclusion scores, and higher Dysfunction scores, these findings should be interpreted as indicating an immune-inflamed but functionally restrained phenotype rather than direct evidence of immunotherapy benefit. Because an independent immunotherapy-treated melanoma cohort with responder/non-responder annotation was not included in this study, the predictive value of the model for PD-1/PD-L1 blockade response remains to be validated in future studies. Future integrative models incorporating TMB/MSI status, TIDE-related immune evasion scores, digital pathology, and clinical immunotherapy response data may further improve therapeutic prediction [48, 49]. Another limitation relates to the interpretation of XCL2. XCL2 is a chemokine with important immune regulatory functions, and its enrichment in T cells suggests that its prognostic value in bulk melanoma transcriptomic data may partly reflect immune infiltration and immune activation within the tumor microenvironment. Therefore, the risk-model weight of XCL2 should be interpreted as a tissue-level signal integrating both immune microenvironmental components and melanoma-cell-associated expression, rather than as a purely tumor-cell-autonomous variable. Although our cell-line experiments support a melanoma-cell-related role of XCL2 under in vitro conditions, they cannot fully recapitulate the non-cell-autonomous and paracrine functions of XCL2 in vivo. Future studies using XCL2 gain-of-function approaches, T cell–tumor cell co-culture systems, and immune-competent animal models will be required to further clarify whether XCL2 mainly acts through direct tumor-cell effects, immune-mediated mechanisms, or both. In addition, CRIGs in this study were defined based on positive correlations with cuproptosis-related genes. This strategy allowed us to focus on immune genes co-expressed with cuproptosis-related programs, but it did not include negatively correlated immune genes. Therefore, the current model cannot fully characterize potential antagonistic relationships between cuproptosis-related programs and immune features. Recent studies have used bidirectional signed-correlation strategies with more stringent thresholds and FDR correction to identify cell-death-related signatures [50, 51]. Future studies applying bidirectional correlation analyses may further clarify whether immune–cuproptosis antagonism contributes to SKCM progression and treatment response. In summary, this study not only provides a reliable prognostic prediction tool but, more importantly, through experimental validation, reveals the critical role of XCL2 in melanoma, laying a theoretical foundation for the development of new treatment strategies. These findings hold promise for translation into more personalized therapeutic approaches for melanoma patients in the future. Conclusions In this study, we developed and validated a CRIGs-based prognostic model for risk stratification in SKCM. This model may provide useful information for prognosis prediction, immune microenvironment characterization, and potential therapeutic stratification. Nevertheless, several limitations should be acknowledged, including the retrospective nature of public-database analyses, potential residual batch effects between datasets, the absence of an independent immunotherapy-treated cohort with responder/non-responder annotation, and the need for further experimental and clinical validation. Future studies using larger prospective cohorts, multi-omics integration, immunotherapy response datasets, and mechanistic experiments will be important to refine this model and clarify the biological roles of key genes such as XCL2. Overall, our findings provide a framework for linking cuproptosis-associated biology with immune features and may contribute to more personalized management of melanoma patients. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Supplementary Material 1.