Lipid metabolism-related molecular typing and prognostic characteristics of cutaneous melanoma

Introduction

Cutaneous melanoma exhibits a high recurrence rate and strong metastatic potential, ranking among the most lethal malignancies (1,2). Emerging therapies targeting MEK, BRAF, and immune checkpoints such as cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) and programmed cell death protein 1 (PD-1) have significantly improved clinical outcomes in selected patients (3,4). Nevertheless, both primary and acquired resistance remain major clinical obstacles, as many patients experience disease progression during initial treatment or fail to respond at all (5). Consequently, overcoming therapeutic resistance requires the identification of novel molecular drivers and the development of strategies that can potentiate the efficacy of current targeted and immunotherapeutic strategies.

The current Tumor-Node-Metastasis (TNM) staging system for skin cutaneous melanoma (SKCM), as defined by the American Joint Committee on Cancer (AJCC)/and the Union for International Cancer Control (UICC), classifies disease progression according to primary tumor thickness, lymph node involvement, and distant metastasis (6,7). However, this system fails to incorporate molecular mechanisms underlying tumor heterogeneity and distinct biological behaviors, thereby limiting its prognostic accuracy and its ability to predict therapeutic response to immunotherapy (8).

Beyond clinicopathologic staging, melanoma progression is governed by complex molecular regulatory networks and extensive metabolic reprogramming (9). Metabolic reprogramming contributes substantially to melanoma pathogenesis and intratumoral heterogeneity (10). Among these alterations, dysregulated lipid metabolism and abnormal lipid accumulation have emerged as defining metabolic hallmarks (11). Lipid metabolism underpins essential biological processes—including energy production, membrane biosynthesis, and intracellular signaling—through major lipid classes such as fatty acids, phospholipids, and sterols (12). Mounting evidence indicates that dysregulated lipid metabolism is intricately linked to tumor progression and immune suppression, with aberrant alterations in lipid pathways emerging as a hallmark of malignancy (13). Specifically, enhanced fatty acid synthesis and uptake reprogram the metabolic landscape of melanoma cells. Elevated fatty acid synthase (FASN) expression, a key lipogenic enzyme, correlates with increased invasiveness and unfavorable prognosis (14). Moreover, fatty acid β-oxidation, exogenous lipid uptake, and intracellular lipid storage have been implicated in promoting melanoma cell migration and survival (15). Multiple lipid metabolic pathways have also been demonstrated to promote resistance to targeted and immune-based therapies (16,17). Hence, elucidating the specific regulators and mechanisms governing lipid metabolic reprogramming is vital for developing interventions that limit tumor plasticity and restore metabolic homeostasis.

Despite accumulating evidence, the overall impact of lipid metabolism on SKCM remains incompletely understood, given the pronounced heterogeneity of melanoma and the complexity of lipid metabolic pathways. Therefore, a comprehensive characterization of the fundamental associations between lipid metabolism and SKCM prognosis, the TME, and immunotherapy response are warranted. In this study, we established novel molecular subtypes of SKCM based on lipid metabolism-associated genes. We then delineated the defining characteristics of these subtypes from multiple perspectives. Collectively, our findings provide a new molecular framework for precision classification and therapeutic stratification in SKCM, offering potential to enhance treatment efficacy and patient survival. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2548/rc).

Methods

Data acquisition

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.Bulk Ribonucleic Acid sequencing (RNA-seq) and clinical data from the The Cancer Genome Atlas-SKCM (TCGA-SKCM) cohort were obtained from the UCSC Xena database (https://xenabrowser.net/). Normal tissue RNA-seq data were retrieved from the Genotype-Tissue Expression (GTEx) database for comparative analysis. To ensure robust external validation, we systematically searched the GEO (https://www.ncbi.nlm.nih.gov/geo/) database using the keywords “melanoma” and “RNA-seq” or “gene expression profiling”. Datasets were included if they met the following criteria: (I) contained human cutaneous melanoma samples with available gene expression profiles; (II) included sufficient sample size (>20 samples) and complete clinical or survival information; and (III) were generated using standardized platforms (Affymetrix, Illumina, or RNA-seq). Studies based on cell lines, single-cell sequencing, or lacking clinical metadata were excluded. Based on these criteria, five independent SKCM datasets—GSE3189, GSE15605, GSE65904, GSE19234, and GSE78220—were selected for validation analyses (18). Among these, GSE65904 and GSE19234 provided comprehensive survival information for prognostic validation, while GSE78220 contained data from patients treated with immune checkpoint blockade (ICB), enabling assessment of immunotherapy response. For immunotherapy response analysis, we additionally incorporated the Liu SKCM cohort (18), which consists of melanoma patients who received immune checkpoint blockade. Lipid metabolism-related genes were obtained from the PathCards module of the GeneCards database using the search term “Metabolism of lipids”, resulting in a curated gene list of 755 genes. This curated lipid gene set was employed for subsequent molecular classification and pathway enrichment analyses. Chemosensitivity-associated gene sets were assembled from the published literature to evaluate potential therapeutic vulnerabilities (19). All the data were obtained by December 2024.

Differential expression analysis

Differential expression analysis was conducted using the R package limma (v3.56.2), which applies a linear modeling framework with empirical Bayes moderation to account for variance heterogeneity. Statistically significant differentially expressed genes (DEGs) were defined by a dual threshold of absolute log2-transformed fold change (|log2FC|) >1.0 and Benjamini-Hochberg adjusted P value <0.05, thereby controlling the false discovery rate (FDR). Subsequently, WGCNA was implemented through systematic preprocessing and parameter optimization (20): expression matrices were subjected to stringent quality control, including the removal of low-expressed genes (detected in <90% of samples) and the exclusion of outlier samples identified through Pearson correlation-based hierarchical clustering (cutoff height =0.25). The soft-thresholding power (β=8) was chosen to satisfy the scale-free topology criterion (R²>0.85), after which a signed topological overlap matrix (TOM) was constructed to quantify gene-gene interaction strength. Modules were identified using dynamic hybrid tree cutting (deepSplit =2, minClusterSize =30), followed by eigengene-based merging of highly correlated modules (Pearson correlation >0.75) to obtain the final co-expression modules.

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis

GO and KEGG pathway enrichment analyses were systematically performed using the clusterProfiler package (v4.4.4). The enrichGO function was employed to identify significant over-representation in three GO domains: biological processes (BP), molecular functions (MF), and cellular components (CC), while pathway-level enrichment was assessed using the enrichKEGG function (21).

Molecular subtyping and clinicopathological correlation analysis

Molecular subtyping was performed using the ConsensusClusterPlus package with the following parameters: 80% of samples and 100% of genes were randomly subsampled over 100 iterations to ensure robustness. Sample similarity was computed using the Partitioning Around Medoids algorithm with Euclidean distance as the metric. The maximum number of clusters was set to maxK=9, and the optimal number of subtypes was determined by evaluating consensus matrices and cumulative distribution function (CDF) curves to ensure cluster stability.

To assess the association between molecular subtypes and clinicopathological characteristics, a chi-square test was employed. The distribution of clinical features across subtypes was examined, and statistical significance was determined at P<0.05.

Survival analysis

Survival outcomes across molecular subtypes were assessed using Kaplan-Meier analysis implemented in the survival R package (v3.5-7). Statistical comparisons between subtype survival curves were conducted via the log-rank test, with p-values reported to four decimal places for enhanced precision.

Comprehensive immune microenvironment analysis

The immune landscape was systematically characterized through an integrated computational approach. Immune cell infiltration levels of 22 distinct subtypes were quantified using CIBERSORT (v1.06) with LM22 signature matrix deconvolution of TCGA-SKCM transcriptomic data (TPM format), employing 1,000 permutations for robust estimation. Comparative analysis between groups was performed using Wilcoxon rank-sum tests (two-sided, FDR-adjusted q<0.05) (22).

The tumor microenvironment (TME) was further evaluated through a multi-algorithm framework: ESTIMATE (v1.0.13) calculated TumorPurity, StromalScore, and ImmuneScore, with feature selection via Filter-Common genes (50% cohort detection threshold); TIDE (v2.0) modeled immune evasion potential; while TCIA-derived Immunophenoscores (v3.0) predicted CTLA-4/PD-1 blockade response. Immune checkpoint gene expression patterns (including co-inhibitory/co-stimulatory molecules, antigen presentation components, and cytokine mediators) were analyzed using Wilcoxon tests with Hochberg FDR correction (q<0.05). All analyses were conducted in R (v4.2.1) using standardized pipelines to ensure reproducibility.

Machine learning methodology

Three distinct machine learning algorithms were employed for prognostic gene selection: least absolute shrinkage and selection operator (LASSO) regression, random forest (RF), and extreme gradient boosting (XGBoost).

The LASSO-Cox proportional hazards model was implemented using the glmnet package (v4.1-6) to identify prognosis-associated genes and construct predictive models (21,22). A 12-fold cross-validation procedure was performed to determine the optimal penalty parameter (λ), with the minimum λ (lambda.min) selected based on evaluation metrics including the concordance index (C-index) and deviance. The coefficient shrinkage process was visualized through λ trajectory plots, and genes retaining non-zero coefficients at the optimal λ were selected as predictive features.

The RF algorithm was executed with 600 decision trees (ntree =600) to evaluate variable importance using the MeanDecreaseGini index. Model optimization was achieved by identifying the point of minimal classification error (err.rate). Feature selection was performed by retaining genes demonstrating importance scores exceeding 2.0.

The XGBoost algorithm was configured with a Cox proportional hazards objective function (objective = “survival:cox”) for survival prediction modeling. Hyperparameters included a learning rate (eta) of 0.1 and 100 boosting iterations (nrounds =100), with model performance assessed using the negative partial log-likelihood (COX-Nloglik) metric. Feature importance was quantified through the Gain value metric via the xgb. Importance function, with the top 15 highest-ranking genes selected for further analysis.

Prognostic model construction and validation

A stepwise multivariate Cox regression analysis was performed to establish an optimized prognostic model, yielding a four-gene lipid metabolism-related molecular (LMM) scoring system: LMM score = −0.178 × LCP2 expression + 0.060 × GPR143 expression + 0.009 × TSPAN13 expression − 0.101 × ACOX2 expression. Model validation was conducted using independent GEO datasets (GSE65904 and GSE19234), where patients were stratified into high- and low-risk groups based on optimal LMM score cutoffs, followed by survival analysis to verify prognostic discrimination.

The prognostic independence of the LMM score was assessed through univariate and multivariate Cox analyses in the TCGA-SKCM cohort, while receiver operating characteristic (ROC) curves compared its predictive performance against conventional clinical parameters. A clinically applicable nomogram integrating the LMM score with key clinicopathological variables (age, T/N stage) was developed using the “rms” package (v6.7-0) to predict 1-, 3-, and 5-year overall survival (OS) probabilities, with model calibration evaluated via Hosmer-Lemeshow goodness-of-fit tests and corresponding calibration curves.

Construction of protein-protein interaction (PPI) networks

The PPI network was constructed using the STRING database (https://cn.string-db.org/) with a confidence score threshold of >0.7 (high confidence) (22). Molecular interactions were subsequently visualized and analyzed using Cytoscape software (version 3.9.1), with network topology parameters calculated to identify hub nodes.

Statistical analysis

Statistical analyses were performed in R (v4.2.1). Differential expression was determined using the limma package with |log₂FC| >1.0 and adjusted P<0.05. Survival was analyzed by Kaplan–Meier method and log-rank test. Associations involving categorical variables were examined using Chi‑squared test, and group comparisons using Wilcoxon rank-sum test. Prognostic modeling was conducted via LASSO‑Cox, random forest, XGBoost and stepwise multivariate Cox regression. Enrichment, subtyping, immune infiltration and PPI network analyses were performed using corresponding packages and databases. A two‑sided P<0.05 was deemed statistically significant.

Results

Molecular subtypes were constructed based on lipid metabolism-related differentially expressed genes (LMDEGs)

Figure 1 illustrates the overall analytical framework of this study. We obtained mRNA expression profiles from 472 TCGA-SKCM tumor samples and 813 normal skin tissues derived from GTEx via the UCSC Xena database, and further integrated tumor/normal transcriptomic data from the GEO datasets (GSE3189 and GSE15605). Differential expression analysis was performed using thresholds of |log2FC| >1 and FDR <0.05, identifying 16,108 (TCGA + GTEx), 2,050 (GSE3189), and 2,784 (GSE15605) DEGs (Figure S1A-S1E). To identify functionally relevant modules and genes associated with SKCM, we performed WGCNA on mRNA expression profiles from the two independent SKCM datasets (GSE3189 and GSE15605). A soft-threshold power of 0.9 was applied to ensure the construction of a scale-free co-expression network. By assessing the correlation coefficients and P values of gene modules between normal and SKCM samples, we merged related modules, resulting in the GSE3189_WGCNA and GSE15605_WGCNA gene sets (Figure S1B-S1E). The intersection of these WGCNA-derived module genes with the three DEG datasets produced 490 high-confidence DEGs between tumor tissue and normal tissue (DEGs_NT) (Figure S1F). Functional enrichment analysis (GO and KEGG) revealed that these DEGs were mainly enriched in biological processes related to skin development, immune regulation, lipid metabolism, and tumor-associated signaling pathways (Figure S1G,S1H). These findings suggest that the identified DEGs play critical roles in the interplay between lipid metabolism and melanoma progression.

Figure 1 Outline of the analyses performed in this study. DEGs, differentially expressed genes; EMT, epithelial-mesenchymal transition; GO, Gene Ontology; GTEx, Genotype-Tissue Expression; IPS, Immunophenotype Score; K-M, Kaplan-Meier; KEGG, Kyoto Encyclopedia of Genes and Genomes; LMDEGs, lipid metabolism-related differentially expressed genes; OS, overall survival; ROC, receiver operating characteristic; SKCM, skin cutaneous melanoma; ssGSEA, single sample gene set enrichment analysis; TCGA, The Cancer Genome Atlas; TIDE, Tumor Immune Dysfunction and Exclusion; WGCNA, weighted gene co-expression network analysis.

To identify clinically relevant lipid metabolism-associated genes in melanoma, we performed univariate Cox regression analysis on all protein-coding genes from patient RNA-seq data, identifying 5,326 genes significantly associated with OS. By intersecting these survival-associated genes with 755 lipid metabolism-related genes (LMGs) curated from the GeneCards database and our previously identified DEGs_NT, we obtained eight high-confidence LMDEGs (Figure 2A). Unsupervised consensus clustering of the 472 SKCM samples based on these eight LMDEGs revealed the optimal cluster number (k=2), showing the highest intra-group consistency. Accordingly, the samples were classified into Subtype A (n=238) and Subtype B (n=234) (Figure 2B,2C). Kaplan-Meier analysis demonstrated significantly worse OS for Subtype B compared to Subtype A (log-rank P<0.001) (Figure 2D). The heatmap (Figure 2E) revealed distinct molecular patterns between subtypes, with Subtype B showing more aggressive clinical features including greater tumor thickness and invasion depth. Functional enrichment analysis further revealed that subtype-specific genes were predominantly involved in adaptive immune response, immune receptor signaling, immune cell adhesion/activation, and antigen processing/presentation pathways (Figure 2F,2G).

Figure 2 Molecular subtyping based on LM-DEGs. (A) Venn diagram illustrating the intersection among LMGs, DEGs, and OS-associated genes. (B) CDF curves for consensus clustering analysis across k values (ranging from 2 to 9). (C) Consensus clustering matrix demonstrating sample classification into two distinct molecular subtypes (k=2). (D) K-M curve of the survival difference between cluster A and cluster B. (E) Unsupervised clustering heatmaps show correlations between two subtypes and clinical phenotypes. (F,G) Functional enrichment analysis of the LM-DEGs: (F) GO terms and (G) KEGG pathways. Dot size reflects the number of enriched genes, and color indicates significance. *, P<0.05; **, P<0.01; ***, P<0.001. CDF, cumulative distribution function; GO, Gene Ontology; K-M, Kaplan-Meier; KEGG, Kyoto Encyclopedia of Genes and Genomes; LM-DEGs, lipid metabolism-related differentially expressed genes; LMGs, lipid metabolism genes; OS, overall survival.

Immunological landscape of molecular subtypes

Given the close association between lipid metabolism and the immune microenvironment, as well as the significant enrichment of LMDEGs in immune-related signaling pathways, we systematically evaluated the immune cell infiltration patterns between the two subtypes. Using CIBERSORT, we analyzed the infiltration levels of 22 immune cell subsets and observed that Subtype A exhibited significantly higher proportions of plasma cells, CD8⁺ T cells, activated CD4⁺ T cells, and M1 macrophages compared to Subtype B (P<0.01). Conversely, resting CD4⁺ T cells and M2 macrophages were more abundant in Subtype B. Notably, CD8⁺ T cells and macrophages constituted the dominant immune populations in Subtype A (Figure 3A). Cross-platform validation confirmed significant differences in immune infiltration between the two clusters (Figure S2). Furthermore, Subtype A demonstrated markedly higher stromal, immune, and composite scores compared with Subtype B (Figure 3B), indicating an immunologically active TME accompanied by enhanced stromal remodeling. To assess the clinical implications of these differences, we employed the TIDE algorithm to predict immune checkpoint inhibitor (ICI) efficacy. Subtype A exhibited a higher TIDE score, suggesting a greater likelihood of immune evasion and poorer response to ICI therapy (Figure 3C). Additionally, Immunophenotype Score (IPS) analysis was performed to predict patient responsiveness to anti-CTLA-4 and anti-PD-1 blockade. Subtype A showed significantly higher IPS values than Subtype B, indicating a potentially enhanced response rate to immune checkpoint inhibitors (Figure 3D). Finally, expression profiling of canonical immune checkpoint genes further showed consistent upregulation in Subtype A (Figure 3E). Collectively, these findings reveal distinct immune phenotypes between subtypes: Subtype A represents an immunologically “hot” but potentially immune-evasive phenotype, whereas Subtype B exhibits a more immunosuppressive TME, which may have important implications for precision immunotherapy.

Figure 3 Comparative analysis of TME characteristics between molecular subtypes. (A) Comparative analysis of TME characteristics between molecular subtypes. The x-axis indicates immune cell subtypes, while the y-axis represents normalized infiltration scores. (B) Violin plots comparing TME scores between clusters, including Stromal Score, Immune Score, and ESTIMATE Score. Scores were calculated using the ESTIMATE algorithm, with significance levels annotated. (C) TIDE score distribution between subtypes. Higher scores indicate greater immune evasion potential. (D) IPS analysis demonstrating differential immunogenicity between clusters. The IPS was derived from MHC molecules, immunomodulators, effector cells, and suppressor cell features. (E) Box plots comparing expression levels of key immune checkpoint genes between Cluster A and Cluster B. **, P<0.01; ***, P<0.001; ****, P<0.0001. IPS, Immunophenoscore; MHC, major histocompatibility complex; TIDE, Tumor Immune Dysfunction and Exclusion; TME, tumor microenvironment.

Construction of a prognostic prediction model

To elucidate the molecular mechanisms underlying lipid metabolism to clinical outcomes in SKCM patients, we performed differential gene expression analysis between Subtype A and Subtype B (Figure 4A). By intersecting the 910 subtype-specific DEGs (DEGs_cluster), 5,326 OS-related genes (OS) identified through univariate Cox regression, and 490 DEGs between tumor and normal tissues (DEGs_NT), we obtained 21 OS-associated DEGs (Figure 4B). We employed three machine learning approaches to identify robust prognostic biomarkers. The LASSO regression analysis demonstrated optimal model fit at lambda.min =9 (Figure 4C,4D), identifying key genes including LCP2, DLL3, and GPR143. Subsequent RF algorithm analysis pinpointed 14 signature genes, notably TUBB4A, SLC45A2, PMEL, and GPR143 (Figure 4E). XGBoost algorithm further extracted the top 15 feature genes ranked by importance score (Figure 4F). Through integrative analysis, four candidate genes (LCP2, GPR143, TSPAN13, and ACOX2) were selected for prognostic modeling (Figure 4G). A multivariable Cox regression model incorporating these genes was then developed to establish the LMM score (Figure 5A).

Figure 4 Comprehensive identification of molecular subtype-specific signature genes. (A) Volcano plot of the DEGs between subtypes A and B. (B) Three-way Venn diagram identifying overlapping genes among: (I) tumor vs. normal DEGs (DEGs_NT), (II) inter-subtype DEGs (DEGs_cluster), and (III) genes related to OS. (C,D) Coefficient distribution plots of log (lambda) sequences (C) and selection of optimal parameters (lambda) in the LASSO model (D). (E) Key genes screened and characterized using the random forest model. (F) Feature importance derived from the XGBoost model. (G) Venn diagram showing crossover genes after the analyses of XGBoost, RF and LASSO. DEGs, differentially expressed genes; DEGs_NT, differentially expressed genes between tumor and normal tissue; FC, fold change; LASSO, least absolute shrinkage and selection operator; OS, overall survival; RF, random forest; XGBoost, extreme gradient boosting.

Figure 5 Construction and validation of the LMM prognostic score. (A) Multivariate Cox regression analysis for model construction. (B,E,G) Distribution of risk scores, clinical outcomes, and the expression of model genes expression in low- vs. high-risk groups from TCGA (B), GSE65904 (E), and GSE19234 (G) datasets. (C,F,H) The Kaplan-Meier survival curve for the low-risk and high-risk groups of patients in the TCGA (C), GSE65904 (F) and GSE19234 (H) dataset. (D) Risk score distribution in subtypes A and B. (I,J) Univariate (I) and multivariate (J) Cox regression analyses of clinical characteristics and LMM score for OS in the TCGA-SKCM cohort. (K) ROC curves evaluating the predictive performance of clinical parameters, LMM score, and nomogram. (L) Nomogram integrating T stage, N stage, and risk score to predict 1-, 3-, and 5-year OS probabilities. (M) The calibration curves for nomogram. The ideal nomogram is shown by a dashed diagonal line. *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. LMM, lipid metabolism molecular; OS, overall survival; ROC, receiver operating characteristic; SKCM, skin cutaneous melanoma; TCGA, The Cancer Genome Atlas.

Validation of LMM score and establishment of predictive nomogram

To evaluate predictive performance, TCGA-SKCM patients were stratified into high- and low-risk groups based on their LMM scores. Higher LMM scores correlated with shorter OS (Figure 5B). Kaplan-Meier survival analysis demonstrated markedly worse OS outcomes in the high-risk group compared to the low-risk group (Figure 5C). Consistent with molecular subtyping results, Subtype B displayed significantly higher LMM scores than Subtype A (Figure 5D). To further validate the robustness of the LMM score, we obtained two independent datasets (GSE65904 and GSE19234) from the GEO database. Consistent with our primary findings, patients classified into the high-risk group based on the LMM score exhibited inferior OS compared to those in the low-risk group in both validation cohorts (Figure 5E-5H). The LMM score demonstrates robust prognostic capacity, effectively stratifying SKCM patients into distinct risk groups. Its consistent performance in both primary cohort analysis and independent external validation underscores its translational potential.

Univariate Cox regression analysis demonstrated that the LMM score was significantly associated with OS, with a hazard ratio (HR) of 2.859 [95% confidence interval (CI): 1.819-4.494; P<0.001; Figure 5I]. Multivariate Cox regression analysis confirmed the independent prognostic value of the LMM score (HR =2.173, 95% CI: 1.344-3.512; P<0.005; Figure 5J). Additionally, three other parameters were identified as independent prognostic indicators: age (HR =1.015, 95% CI: 1.002-1.029; P=0.021), T stage (HR =1.663, 95% CI: 1.321-2.095; P<0.001), and N stage (HR =1.829, 95% CI: 1.349-2.481; P<0.001; Figure 5J).

We further evaluated the predictive performance of age relative to the LMM score. Time-dependent ROC analysis revealed that age (AUC =0.648) exhibited comparable predictive accuracy for OS to the LMM score (AUC =0.653), and outperformed other clinical factors including T stage (AUC =0.684), and N stage (AUC =0.597) (Figure 5K). Based on these four independent prognostic factors (LMM score, age, T stage, and N stage), we developed a nomogram to predict the 1-, 3-, and 5-year OS probabilities for SKCM patients in the TCGA cohort (Figure 5L). Calibration plots for 1-, 3-, and 5-year survival were subsequently constructed to evaluate the predictive accuracy of the nomogram (Figure 5M). The close alignment between the predicted and observed survival probabilities, as indicated by the overlap of the model prediction line with the ideal reference line, demonstrated good predictive performance of the nomogram across the entire cohort. Furthermore, the 3-year ROC curve analysis revealed that the nomogram provided superior predictive capability compared to using clinical features or the risk score alone (Figure 5I). Together, these findings demonstrate that the LMM-based nomogram is a powerful tool for individualized prognosis prediction in SKCM.

Differences in TME characteristics and immunotherapy responses between high-risk and low-risk groups

To elucidate the distinct immune landscape of the TME between risk subgroups, we systematically compared the TME characteristics of high-risk and low-risk SKCM patients. Using the CIBERSORT algorithm, we quantified the relative abundance of 22 immune cell subsets (Figure 6A). The analysis revealed that the high-risk subgroup exhibited significantly lower proportions of memory B cells, plasma cells, CD8⁺ T cells, CD4⁺ memory T cells, and M1 macrophages compared to the low-risk subset (Figure 6A). In contrast, the high-risk group demonstrated markedly elevated infiltration levels of M2 macrophages and eosinophils. Further TME characterization using the IOBR package revealed that high-risk tumors exhibited an immunosuppressive, exclusionary, and functionally exhausted phenotype (Figure 6B-6D). Comparative evaluation of TME scores revealed that the low-risk group had significantly higher stromal, immune, and ESTIMATE scores than the high-risk group (Figure 6E). These findings were corroborated by validated using Tumor Immune Estimation Resource (TIMER)-based immune infiltration analysis, which consistently demonstrated reduced immune infiltration in the high-risk group relative to the low-risk group (Figure 6F). The correlation between immune cell infiltration and the four model genes was evaluated (Figure 6G). Follicular helper T cells (Tfh cells) and M2 macrophages were found to be positively correlated with the LMM score, whereas CD8⁺ T cells and M1 macrophages showed negative correlations with it (Figure S3A). Comparative analyses of immune cell infiltration across different platforms further revealed significant discrepancies between the high-risk and low-risk groups (Figure S4A-S4C). The low-risk group exhibited higher infiltration of anti-tumor immune cells (e.g., CD8⁺ T cells and M1 macrophages), yet the TME was characterized as “immunosuppressive and exhausted”. Further comparative analyses of immune checkpoint expression revealed significant upregulation of immune checkpoint molecules in the low-risk group (Figure 7A). Notably, correlations were identified between the expression of four model genes and immune checkpoint molecules (Figure 7B). The low-risk group, despite exhibiting higher quantities of CD8⁺ T cells and other anti-tumor immune cells, demonstrated functional suppression due to elevated expression of immunosuppressive molecules, resulting in an overall “immunosuppressive/exhausted” TME. Scatter plots depicting the relationship between risk scores and immune checkpoint expression revealed significant negative correlations between the risk score and nearly all examined immune checkpoints (Figure S3B).

Figure 6 TME characteristics between high- and low-risk groups. (A) Comparison of the infiltration of immune cells between two subgroups. (B-D) Comparison of immune suppression, exclusion and exhaustion features between two subgroups. (E) Box plots of stromal score, immune score, and ESTIMATE score of the two risk groups, respectively. (F) Box plot of the difference in the infiltration ratio of six immune cell types between the two risk groups estimated by the TIMER algorithm. (G) Heatmap of the correlation between model genes and 22 immune cell types (CIBERSORT-derived infiltration proportions); estimated by the TIMER algorithm. *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001; ns, not significant. TIMER, Tumor Immune Estimation Resource; TME, tumor microenvironment.

Figure 7 The performance of the model in predicting immunotherapeutic response in ICB-treated cohorts. (A) Comparison of the expression of checkpoints between two subgroups. (B) Correlations between the prognostic model genes and the immune checkpoint genes. (C) Survival analysis between risk groups. (D) liu2018SKCM different immunotherapy response groups. (E) GSE78220 different immunotherapy response groups. *, P<0.05; **, P<0.01; ***, P<0.001. CR, complete remission; ICB, immune checkpoint blockade; PD, disease progression; PR, partial response; SD, stable disease.

To further evaluate the predictive potential of the LMM score for immunotherapy outcomes, two independent immunotherapy cohorts were analyzed. In the Liu SKCM cohort, patients with higher LMM scores exhibited significantly shorter OS (Figure 7C), and the complete response (CR) group exhibited markedly lower risk scores compared to partial response (PR), stable disease (SD), and progressive disease (PD) groups (Figure 7D). Analysis of the GSE78220 cohort similarly demonstrated significantly lower risk scores in the CR group versus PR and PD groups. These findings suggest that the LMM score may serve as a potential predictive biomarker for immunotherapy efficacy, potentially aiding in the identification of treatment-sensitive populations (Figure 7E). Collectively, these findings indicate that the LMM score may serve as a promising biomarker for predicting immunotherapy responsiveness and identifying treatment-sensitive patient subpopulations.

In clinical management, immune checkpoint blockade and BRAF/MEK-targeted therapy have become the standard of care and main determinants of prognosis in advanced melanoma, while conventional chemotherapy plays a very limited role. Using in vitro drug sensitivity data from the GDSC2 database, we systematically analyzed differences in drug response between high- and low-risk groups. The low-risk group displayed greater sensitivity to conventional chemotherapeutics, including temozolomide, platinum compounds, and vinblastine, compared to the high-risk group (Figure 8A). Notably, ten agents exhibited significantly lower half-maximal inhibitory concentrations (IC50) in high-risk patients, comprising five extracellular signal-regulated kinase (ERK)-targeted drugs, two mitogen-activated protein kinase kinase (MEK)-targeted agents (Figure 8B; Figure S5A,S5B), and three compounds respectively targeting transforming growth factor-β receptor (TGFβR), mammalian target of rapamycin (mTOR), and insulin-like growth factor 1 receptor (IGF1R) (Figure S6A,S6B).

Figure 8 Analysis of drug sensitivity and genetic correlation. (A-C) Box plots showing drug sensitivity differences of chemotherapeutic agents (A), ERK targeted drugs (B) and MEK targeted drugs (C) between high-risk and low-risk groups. (D,E) Correlation networks of characteristic genes with chemotherapy sensitivity-related genes (D) and targeted drug sensitivity-related genes (E), with correlation coefficients and significance marked. (F) Violin plot of pathway activity scores distribution in high-risk vs. low-risk groups. ***, P<0.001. ERK, extracellular signal-regulated kinase; MEK, mitogen-activated protein kinase kinase.

Correlation analyses revealed that LCP2, ACOX2, and TSPAN13 expression were positively associated with chemosensitivity-related genes (P<0.05), whereas GPR143 expression was negatively correlated (P<0.05) (Figure 8D). AKR1C1, EGFR, HOXA9, and TBX5 expression demonstrated negative correlations with the LMM score (P<0.05) (Figure S5B). Target analysis of therapeutic agents indicated MAP2K1 and TGFBR2 expression positively correlated with LCP2, ACOX2, and TSPAN13 but negatively with GPR143, while MAP2K2 showed inverse patterns (Figure 8E, Figure S6C). Notably, MAP2K1 and TGFBR2 were negatively associated with the LMM score (P<0.05), whereas MAP2K2 showed a positive association (P<0.05) (Figure S6D). Additionally, the low-risk group displayed significantly lower MAPK pathway activation scores compared to the high-risk group (P<0.05) (Figure 8F).

These findings collectively demonstrate distinct chemotherapeutic sensitivity patterns between risk groups: the low-risk group showed enhanced sensitivity to conventional chemotherapeutic agents, while the high-risk group may exhibit greater therapeutic potential for specific targeted drugs.

Single-gene analysis of prognostic signature genes

To clarify the molecular mechanisms underlying the prognostic and immunological implications of the LMM score, we conducted systematic single-gene analyses of the four model genes. Kaplan-Meier survival analysis showed that low expression levels of LCP2, TSPAN13, or ACOX2 were significantly linked to poorer OS in SKCM patients (Figure 9A-9F). Expression landscape analysis indicated consistently higher levels of LCP2, TSPAN13, and ACOX2 in the low-risk group compared to the high-risk group, with LCP2 exhibiting the most notable difference (Figure 9A-9F). We also examined the relationships between these three model genes (LCP2, TSPAN13, and ACOX2) and immune cell infiltration, revealing that LCP2 and TSPAN13 had stronger correlations with both immune cell infiltration and immune checkpoint expression (P<0.01) (Figure 9G, Figure S7A-S7D). Single-cell analysis showed widespread distribution of LCP2 across various immune cell populations within the TME (Figure S8). To explore potential functional mechanisms, we used the STRING database for protein-protein interaction analysis, identifying direct interactions between LCP2 and both chemosensitivity genes and targeted drug substrates (Figure 9H,9I). Notably, PLCG1, a key partner of LCP2, is involved in the phosphatidylinositol signaling pathway (Figure 9J), which plays a critical role in lipid metabolism regulation. Notably, LCP2 is not included in existing lipid metabolism gene sets, suggesting that it may represent a novel regulatory node bridging immune signaling and lipid metabolic reprogramming in melanoma.

Figure 9 Prognostic and immunological characteristics of model genes in melanoma. (A,C,E) Kaplan-Meier survival analysis for LCP2 (A), TSPAN13 (C), and ACOX2 (E). (B,D,F) Density plots illustrating expression patterns of LCP2 (B), TSPAN13 (D), and ACOX2 (F) in high- and low-risk groups. (G) Heatmap displaying Spearman correlation coefficients between model genes and immune cell infiltration (red: positive; blue: negative; gray: non-significant). (H) Protein-protein interaction network of LCP2 and genes related to the MAPK signaling pathway. (I) Protein-protein interaction network of LCP2 and genes associated with chemotherapy sensitivity. (J) Protein-protein interaction network of LCP2 generated from STRING database, highlighting key immune-related interactors. MAPK, mitogen-activated protein kinase.

Discussion

Lipid metabolism and lipid oxidation are increasingly recognized as core regulators of melanoma progression and clinical outcome (13,23). A large body of experimental evidence from preclinical melanoma models has established a strong causal link between dysregulated lipid metabolic programs, enhanced lipid oxidation, reactive oxygen species (ROS) accumulation, and malignant progression. Metabolic reprogramming toward elevated lipid catabolism not only fuels melanoma cell proliferation and survival (24,25) but also triggers oxidative stress by promoting ROS generation, which in turn drives genomic instability, invasion, and metastatic dissemination (26,27). Notably, lipid metabolism and ROS signaling are also key determinants of metastatic tropism in melanoma, directly governing the preferential colonization of tumor cells in visceral organs versus lymph node sites (13,28). Distinct patterns of lipid oxidation and ROS balance have been functionally associated with organ-specific metastatic cascades, with certain lipid metabolic profiles favoring dissemination to visceral organs (such as lung, liver, and brain) while others are more closely linked to lymph node metastasis (29). These experimental observations high33light that lipid metabolism and oxidative stress are not merely byproducts of tumorigenesis, but act as central regulators that shape melanoma aggressiveness, metastatic destination, and ultimately patient prognosis (13,28).

Oxidative stress is implicated in tumorigenesis, metastasis, and immune regulation (30). Accumulating evidence suggests that lipid metabolic reprogramming profoundly affects tumor progression, drug resistance, TME remodeling, and immune evasion by ROS generation and ferroptosis (30-32). While dysregulated lipid metabolism has been well-documented in melanoma (13), the metabolic heterogeneity across melanoma subtypes remains insufficiently characterized. This study systematically identified LMDEGs in SKCM and investigated their prognostic significance, interactions within the immune microenvironment, and therapeutic implications. By integrating multi-omics data from TCGA and GEO databases, we established a robust lipid metabolism-related molecular signature (LMM score). This signature effectively stratified SKCM patients into distinct risk groups exhibiting divergent clinical outcomes, immune infiltration patterns, and drug sensitivities. Collectively, our findings highlight the crucial interplay between lipid metabolism and tumor immunity, providing new mechanistic insights into SKCM pathogenesis and potential directions for precision immunometabolic therapy.

The current driver gene-based (BRAF/NRAS/KIT) classification system for melanoma exhibits several inherent limitations. Firstly, inadequate molecular coverage creates therapeutic blind spots. BRAF-mutant classification encompasses only approximately 40–50% of cutaneous melanomas (non-chronic sun-damaged type), while the mutation rate falls below 20% in acral and mucosal subtypes. Secondly, tumor heterogeneity poses a significant challenge to genotypic stratification: Intratumoral subclonal coexistence is prevalent. For instance, approximately 28% of BRAF-mutant patients harbor concurrent NRAS-activated subclones, frequently leading to acquired resistance to targeted therapy within 3–6 months (33). Moreover, more than 50% of mucosal melanomas lack identifiable driver mutations, placing them beyond the scope of current genotypic frameworks and without well-defined therapeutic targets (34,35). Critically, existing immunophenotypic classification systems also have critical shortcomings: they fail to capture the dynamic and context-dependent nature of the TME and exhibit limited predictive accuracy for immunotherapy response. This limitation is underscored by the modest 40–50% objective response rate to PD-1 inhibitors and the inability of current classifications to elucidate either intrinsic or adaptive resistance mechanisms (23).

Emerging evidence increasingly implicates lipid metabolism as a promising source of novel therapeutic targets for overcoming treatment resistance. Aberrant lipid accumulation within tumor cells (e.g., 27-hydroxycholesterol) has been shown to suppress T-cell activity, while upregulated CD36 expression correlates positively with resistance to anti-PD-1 therapy (23). Furthermore, long-chain fatty acids induce CD8⁺ T cell exhaustion (characterized by a ~60% reduction in granzyme B secretion) via the PPARγ signaling pathway, contributing to the failure of PD-1 inhibitors (23). Lipid metabolites, such as ganglioside GD3, promote an immunosuppressive microenvironment by inhibiting HLA-DR expression on dendritic cells and inducing regulatory T cell (Treg) expansion (36). Conversely, inhibition of the cholesterol esterification enzyme acyl-CoA: cholesterol acyltransferase 1 (ACAT1) increases membrane cholesterol content in CD8⁺ T cells, thereby enhancing T-cell receptor (TCR) clustering and signaling, facilitating immune synapse formation, and ultimately promoting cytokine secretion, cytolytic granule release, proliferation, and cytotoxic function (37). Critically, these mechanistic insights into lipid metabolism-driven immune evasion and resistance remain unincorporated into any major melanoma classification system. Through rigorous intersectional analysis of survival-associated genes and lipid metabolism genes (LMGs), we identified eight key LMDEGs (LMDEGs: HSD11B1, ACOX2, HSD11B2, CIDEA, PIK3C2G, AACS, ST3GAL5). Stratification of SKCM patients into two distinct subtypes (A and B) based on these LMDEGs revealed significantly divergent survival outcomes. Strikingly, GO and KEGG pathway analyses demonstrated that these LMDEGs primarily regulate immune-related pathways, extending beyond their canonical metabolic functions and suggesting their dual regulatory roles in SKCM progression.

Immune cell subpopulations constitute a highly intricate network within the TME, engaging in diverse and dynamic crosstalk mechanisms (38). Comprehensive immune landscape analysis revealed pronounced differences in immune signatures between the molecular subtypes defined by LMDEGs. Subtype A was characterized by enrichment of CD8⁺ T cells and M1 macrophages, accompanied by elevated immune and stromal scores, indicative of an immunologically “hot” TME. In contrast, Subtype B displayed polarization towards M2 macrophages and exhaustion of CD4⁺ T cells, characteristic of an immunosuppressive TME. These findings align with recent research on lipid metabolism regulating immune cell function: Fatty acid (FA) accumulation within the TME can activate PPARα signaling in CD8⁺ T cells, enhancing their lipid metabolism and sustaining effector functions (39). Promoting FA oxidation metabolism in CD8⁺ T cells has been shown to augment their anti-tumor activity. Furthermore, PPAR-γ, a key regulator of lipid metabolism, drives the differentiation of naïve T cells into effector T cells (Teffs) by promoting FA oxidation and inhibiting Teff apoptosis, concurrently increasing the activity and abundance of effector/memory CD8⁺ T cells in draining lymph nodes and tumors (40). Moreover, the expression levels of immune checkpoint molecules (e.g., PD-1, CTLA-4) were significantly higher in the lipid metabolism-active Subtype A compared to Subtype B, further supporting the distinct immunological profiles.

The observed inverse correlation between the LMM score and the expression of immune checkpoint molecules (PD-1, CTLA-4) further supports the pivotal role of lipid metabolism in modulating immunotherapy responsiveness. Lipid scarcity in CD8⁺ T cells impairs their proliferation and cytotoxic function, manifesting as reduced IFN-γ secretion and upregulated PD-1 expression (41,42). Mechanistically, PD-1 signaling impedes the TCR/CD28-mediated PI3K/AKT/mTOR pathway, consequently promoting lipolysis and fatty acid oxidation (FAO) (43). Studies indicate that PD-1 antibody therapy reduces the FAO capacity of T cells, and its excessive activation may trigger cellular apoptosis (43). Our findings position LCP2 as a pivotal immunometabolic regulator through its interaction with phospholipase Cγ1 (PLCG1). The PLCγ family catalyzes the hydrolysis of phosphatidylinositol 4,5-bisphosphate (PIP2) into two secondary messengers: inositol 1,4,5-trisphosphate (IP3) and diacylglycerol (DAG) (44,45). DAG mediates the activation of protein kinase C (PKC), triggering downstream cascades including NF-κB, ERK, and MAPK signaling, whereas IP3 induces intracellular Ca²⁺ release to activate nuclear factor of activated T cells (NFAT) (46,47). This dual signaling mechanism enables the LCP2–PLCG1 axis to coordinately regulate both immunologic activation (via NF-κB/NFAT) and metabolic reprogramming (via MAPK/ERK-mediated lipid metabolism), establishing LCP2 as a molecular bridge linking immune activation with metabolic regulation in melanoma progression.

Consistent with previous clinical and experimental studies, our analyses identified age as an independent prognostic factor for SKCM. Time-dependent ROC analysis revealed that age exhibited comparable predictive accuracy for OS to our LMM score, suggesting that age and LMM characteristics may jointly influence melanoma outcome. Emerging evidence indicates that aging is associated with profound alterations in lipid metabolic profiles, including increased lipid oxidation and ROS accumulation, which can modulate melanoma cell proliferation, invasion, and metastatic behavior. Notably, preclinical models have demonstrated that age-related metabolic dysregulation may also shape metastatic tropism, influencing the preferential colonization of melanoma cells in visceral organs versus lymph nodes. Our finding that age is an independent prognostic indicator, alongside the LMM score, underscores the importance of considering both chronological age and LMM features when evaluating melanoma prognosis and therapeutic strategies.

Nonetheless, several limitations of this study should be acknowledged. Firstly, most analyses and conclusions were derived from publicly available genomic and transcriptomic databases; although validated across multiple independent cohorts, inherent selection and sampling biases remain unavoidable. Secondly, while the prognostic signature genes were identified through robust and well-validated bioinformatic pipelines, the precise molecular mechanisms-particularly how these genes modulate lipid metabolism across tumor, immune, and stromal compartments-require further elucidation through comprehensive in vitro and in vivo experiments. Thirdly, the potential mechanistic interplay between dysregulated lipid metabolism and chemotherapeutic resistance in SKCM was not comprehensively investigated in this study. Finally, while we evaluated the prognostic significance of age in our cohort, the precise molecular mechanisms underlying the interplay between age, lipid metabolism, and melanoma metastatic behavior were not fully elucidated. Future studies incorporating age-stratified analyses and functional experiments are needed to clarify these relationships.

Conclusions

In conclusion, we molecularly typed patients with SKCM based on their LMDEGs and revealed the immunologic and genetic characteristics of the different molecular subtypes. In addition, we constructed an LMM score that can be used to predict the prognosis and efficacy of immunotherapy for patients with SKCM. This score has been validated in multiple datasets and shows the potential for guiding individualized and precise diagnosis and treatment of SKCM.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2548/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2548/prf

Funding: This research was funded by the National Natural Science Foundation of China (No. 82372272) and the Tengfei Project of Third Affiliated Hospital of Naval Medical University (No. TF2024XSYJ05).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2548/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Welch HG, Mazer BL, Adamson AS. The Rapid Rise in Cutaneous Melanoma Diagnoses. N Engl J Med 2021;384:72-9. [Crossref] [PubMed]
2. Tasdogan A, Sullivan RJ, Katalinic A, et al. Cutaneous melanoma. Nat Rev Dis Primers 2025;11:23. [Crossref] [PubMed]
3. Grover P, Lo SN, Li I, et al. Efficacy of adjuvant therapy in patients with stage IIIA cutaneous melanoma. Ann Oncol 2025;36:807-18. [Crossref] [PubMed]
4. Joshi UM, Kashani-Sabet M, Kirkwood JM. Cutaneous Melanoma: A Review. JAMA 2025;334:2113-25. [Crossref] [PubMed]
5. Revach OY, Cicerchia AM, Shorer O, et al. Overcoming resistance to immunotherapy by targeting CD38 in human tumor explants. Cell Rep Med 2025;6:102210. [Crossref] [PubMed]
6. Crocetti E, Stanganelli I, Mancini S, et al. Evaluation of the agreement between TNM 7th and 8th in a population-based series of cutaneous melanoma. J Eur Acad Dermatol Venereol 2019;33:521-4. [Crossref] [PubMed]
7. Keung EZ, Gershenwald JE. Clinicopathological Features, Staging, and Current Approaches to Treatment in High-Risk Resectable Melanoma. J Natl Cancer Inst 2020;112:875-85. [Crossref] [PubMed]
8. Tang B, Chi Z, Chen Y, et al. Safety, Efficacy, and Biomarker Analysis of Toripalimab in Previously Treated Advanced Melanoma: Results of the POLARIS-01 Multicenter Phase II Trial. Clin Cancer Res 2020;26:4250-9. [Crossref] [PubMed]
9. Kuranaga Y, Hatem Y, Grossniklaus HE, et al. The emerging role of metabolic interventions in uveal melanoma. Semin Cancer Biol 2025;117:23-37. [Crossref] [PubMed]
10. Li Y, Ming R, Zhang T, et al. TCTN1 Induces Fatty Acid Oxidation to Promote Melanoma Metastasis. Cancer Res 2025;85:84-100. [Crossref] [PubMed]
11. Wang JX, Choi SYC, Niu X, et al. Lactic Acid and an Acidic Tumor Microenvironment suppress Anticancer Immunity. Int J Mol Sci 2020;21:8363. [Crossref] [PubMed]
12. Ping Y, Fan Q, Zhang Y. Modulating lipid metabolism improves tumor immunotherapy. J Immunother Cancer 2025;13:e010824. [Crossref] [PubMed]
13. Gurung S, Budden T, Mallela K, et al. Stromal lipid species dictate melanoma metastasis and tropism. Cancer Cell 2025;43:1108-1124.e11. [Crossref] [PubMed]
14. Yang K, Wang X, Song C, et al. The role of lipid metabolic reprogramming in tumor microenvironment. Theranostics 2023;13:1774-808. [Crossref] [PubMed]
15. Webb BA, Chimenti M, Jacobson MP, et al. Dysregulated pH: a perfect storm for cancer progression. Nat Rev Cancer 2011;11:671-7. [Crossref] [PubMed]
16. Gil-Ordóñez A, Martín-Fontecha M, Ortega-Gutiérrez S, et al. Monoacylglycerol lipase (MAGL) as a promising therapeutic target. Biochem Pharmacol 2018;157:18-32. [Crossref] [PubMed]
17. Fang Y, Xu X, Lu R, et al. TEAD3 + high-risk melanoma cells crosstalk with GAS6 + macrophages via the GAS6-TYRO3 ligand-receptor axis to modulate propionate metabolism and drive melanoma progression. J Exp Clin Cancer Res 2025;44:279. [Crossref] [PubMed]
18. Liu D, Schilling B, Liu D, et al. Integrative molecular and clinical modeling of clinical outcomes to PD1 blockade in patients with metastatic melanoma. Nat Med 2019;25:1916-27. [Crossref] [PubMed]
19. Chen H, Yang W, Li Y, et al. Leveraging a disulfidptosis-based signature to improve the survival and drug sensitivity of bladder cancer patients. Front Immunol 2023;14:1198878. [Crossref] [PubMed]
20. Ma X, Du W, Bai M, et al. Integrating WGCNA and machine learning algorithm to identify ACSM5 as a prognostic biomarker and therapeutic target for predicting immunotherapy efficacy in non-small cell lung cancer. Transl Cancer Res 2026;15:52. [Crossref] [PubMed]
21. Li H, Tan J, Chu H, et al. IL20RB as a prognostic and immune-related biomarker in lung cancer: association with immune infiltration, tumor progression, and potential therapeutic targeting. Transl Cancer Res 2026;15:8. [Crossref] [PubMed]
22. Ling H, Wu G, Tang L, et al. CREM regulates the tumor immune microenvironment and predicts prognosis in thyroid carcinoma. Transl Cancer Res 2026;15:15. [Crossref] [PubMed]
23. Xiong L, Cheng J. Rewiring lipid metabolism to enhance immunotherapy efficacy in melanoma: a frontier in cancer treatment. Front Oncol 2025;15:1519592. [Crossref] [PubMed]
24. Hsu CY, Ahmed YK, Mohammed S, et al. Metabolism at the core of melanoma: From bioenergetics to immune escape and beyond. Semin Oncol 2025;52:152413. [Crossref] [PubMed]
25. Pellerin L, Carrié L, Dufau C, et al. Lipid metabolic Reprogramming: Role in Melanoma Progression and Therapeutic Perspectives. Cancers (Basel) 2020;12:3147. [Crossref] [PubMed]
26. Wilcock DJ, Badrock AP, Wong CW, et al. Oxidative stress from DGAT1 oncoprotein inhibition in melanoma suppresses tumor growth when ROS defenses are also breached. Cell Rep 2022;39:110995. [Crossref] [PubMed]
27. Han B, Si H, Yang S, et al. SMYD2 promotes oxidative stress-responsive lipid metabolism in melanoma by regulating H3K4 tri-methylation. J Dermatol Sci 2026;121:11-21. [Crossref] [PubMed]
28. By Staff Writer. A surprising look at the mechanics of metastasis | Harvard T.H. Chan School of Public Health [Online]. 2024 [zuletzt aufgerufen am 19.02.2026]. Available online: https://hsph.harvard.edu/news/olive-oil-cancer-metastasis/
29. Boitor A. Highlights in Cancer Research: December 2025 [Online]. The Cancer Researcher. 2025 [zuletzt aufgerufen am 19.02.2026]. Available online: https://magazine.eacr.org/highlights-in-cancer-research-december-2025/
30. Hayes JD, Dinkova-Kostova AT, Tew KD. Oxidative Stress in Cancer. Cancer Cell 2020;38:167-97. [Crossref] [PubMed]
31. Sies H, Jones DP. Reactive oxygen species (ROS) as pleiotropic physiological signalling agents. Nat Rev Mol Cell Biol 2020;21:363-83. [Crossref] [PubMed]
32. Zheng M, Zhang W, Chen X, et al. The impact of lipids on the cancer-immunity cycle and strategies for modulating lipid metabolism to improve cancer immunotherapy. Acta Pharm Sin B 2023;13:1488-97. [Crossref] [PubMed]
33. Biersack B. Editorial for the Special Issue-"Recent Advances of Novel Pharmaceutical Designs for Anti-Cancer Therapies". Int J Mol Sci 2023;24:8238. [Crossref] [PubMed]
34. Babu S, Chen J, Baron CS, et al. Specific oncogene activation of the cell of origin in mucosal melanoma. Nat Commun 2025;16:6750. [Crossref] [PubMed]
35. Li Y, Cui Z, Song X, et al. Single-Cell Transcriptomic Landscape Deciphers Intratumoral Heterogeneity and Subtypes of Acral and Mucosal Melanomas. Clin Cancer Res 2025;31:2495-514. [Crossref] [PubMed]
36. Mukhatayev Z, Dellacecca ER, Cosgrove C, et al. Antigen Specificity Enhances Disease Control by Tregs in Vitiligo. Front Immunol 2020;11:581433. [Crossref] [PubMed]
37. Sugi T, Katoh Y, Ikeda T, et al. SCD1 inhibition enhances the effector functions of CD8(+) T cells via ACAT1-dependent reduction of esterified cholesterol. Cancer Sci 2024;115:48-58. [Crossref] [PubMed]
38. Picard E, Verschoor CP, Ma GW, et al. Relationships Between Immune Landscapes, Genetic Subtypes and Responses to Immunotherapy in Colorectal Cancer. Front Immunol 2020;11:369. [Crossref] [PubMed]
39. Wang H, Franco F, Tsui YC, et al. CD36-mediated metabolic adaptation supports regulatory T cell survival and function in tumors. Nat Immunol 2020;21:298-308. [Crossref] [PubMed]
40. Chowdhury PS, Chamoto K, Kumar A, et al. PPAR-Induced Fatty Acid Oxidation in T Cells Increases the Number of Tumor-Reactive CD8(+) T Cells and Facilitates Anti-PD-1 Therapy. Cancer Immunol Res 2018;6:1375-87. [Crossref] [PubMed]
41. Angelin A, Gil-de-Gómez L, Dahiya S, et al. Foxp3 Reprograms T Cell Metabolism to Function in Low-Glucose, High-Lactate Environments. Cell Metab 2017;25:1282-1293.e7. [Crossref] [PubMed]
42. Hu B, Lin JZ, Yang XB, et al. Aberrant lipid metabolism in hepatocellular carcinoma cells as well as immune microenvironment: A review. Cell Prolif 2020;53:e12772. [Crossref] [PubMed]
43. Haku Y, Kitaoka K, Ichimaru K, et al. Active aldehydes accelerate CD8(+) T cell exhaustion by metabolic alteration in the tumor microenvironment. Nat Immunol 2026;27:281-94. [Crossref] [PubMed]
44. Yang Z, Kim S, Mahajan S, et al. Phospholipase Cγ1 (PLCγ1) Controls Osteoclast Numbers via Colony-stimulating Factor 1 (CSF-1)-dependent Diacylglycerol/β-Catenin/CyclinD1 Pathway. J Biol Chem 2017;292:1178-86. [Crossref] [PubMed]
45. Chylek LA, Holowka DA, Baird BA, et al. An Interaction Library for the FcεRI Signaling Network. Front Immunol 2014;5:172. [Crossref] [PubMed]
46. Fu G, Chen Y, Yu M, et al. Phospholipase C{gamma}1 is essential for T cell development, activation, and tolerance. J Exp Med 2010;207:309-18. [Crossref] [PubMed]
47. Yang YR, Choi JH, Chang JS, et al. Diverse cellular and physiological roles of phospholipase C-γ1. Adv Biol Regul 2012;52:138-51. [Crossref] [PubMed]