Background
Melanoma is the primary cause of skin tumor-related death with increasing annual cases of morbidity and mortality [
1,
2]. While patients with melanoma undergo a combination of surgery, chemotherapeutic drugs, and molecular-targeted therapeutic drugs, the effectiveness of current therapies varies among individuals [
3,
4]. The emergence of immunotherapy has offered hope to patients with advanced melanoma, but its benefits are limited [
5‐
7]. Due to the inconsistent treatment outcomes and varying response rates, considerable effort has been devoted to identifying prognostic and therapeutic biomarkers for melanoma, particularly those capable of predicting the efficacy of immunotherapy [
8‐
10].
G protein-coupled receptors (GPCRs) are membrane protein receptors that bind chemicals in the cellular environment and activate a series of intracellular signaling pathways that ultimately lead to changes in cellular state [
11‐
13]. GPCRs have also been implicated in carcinogenesis and metastasis [
14‐
17] and have emerged as important targets for drug therapy due to their wide distribution in the body [
18]. In addition, they play a pivotal role in shaping the tumor microenvironment (TME). Santagata et al. found that two GPCRs, CXCR4 and CXCR7, orchestrate the recruitment of immune and stromal cells [
19]. In the context of melanoma, Ridky et al. demonstrated that a combination of anti-PD-1 and G-1, a G protein-coupled estrogen receptor-selective agonist, could effectively inhibit tumor growth [
20]. Nevertheless, there is a noticeable absence of comprehensive and sophisticated studies delving into the role of GPCRs in melanoma. Furthermore, the precise mechanism through which GPCRs influence the TME and the response to immunotherapy in melanomas remain unknown.
The emergence of high-throughput sequencing techniques, particularly single-cell RNA sequencing (scRNA-seq), has propelled tumor research into the era of precision. The integration of scRNA-seq and bulk sequencing (bulk-seq) allows researchers to dissect the contribution of individual genes in tumorigenesis and progression, both at the tissue and single-cell levels [
21‐
23]. In the current study, we have harnessed the power of GPCRs in conjunction with the intricate cellular landscape of the TME to construct a GPCR–TME classifier for better clinical classification and therapeutic strategies. This innovative approach enhances clinical classification and informs the development of more effective therapeutic strategies. To a significant extent, our findings address the limitations of current clinical staging methodologies and offer valuable insights for the precise management of melanoma.
Materials and methods
Sources of data
Gene expression and survival data of melanoma cohorts were obtained from two publicly available data sets (The Cancer Genome Atlas (TCGA)–SKCM and GSE65904). For scRNA-seq melanoma data, we utilized data from GSE189889 to visualize the GPCR score of different cell types within the TME. Pan-cancer data were downloaded from Xena [
24]. Furthermore, we tested the utility of the GPCR–TME classifier as an immunotherapy predictor using data from three melanoma cohorts (GSE35640, GSE91061, and GSE145996) with available immunotherapy response data.
Data preprocessing
For RNA-seq data, the normalization was performed using the R package ‘DESeq2’ based on the downloaded count data. For microarray data, ‘affy’ package was used for background correction and normalization. For scRNA-seq data, the ‘NormalizeData’ function of ‘Seurat’ package was used for normalization.
Quantification of GPCRs and TME cells
The list of GPCRs was downloaded from the Molecular Signatures Database (GOMF_G_PROTEIN_COUPLED_RECEPTOR_ACTIVITY). Gene expression matrices of the melanoma cohorts were then extracted based on the 870 GPCRs. CIBERSORT enables the calculation of 22 immune cell types through a deconvolution algorithm using the bulk-seq data. Prior to applying CIBERSORT, we followed standard preprocessing steps of normalization of RNA-seq and microarray data, to ensure the comparability and reliability of our gene expression data. The enrichment scores calculated by CIBERSORT were utilized for developing the TME score [
25].
Establishment and verification of the GPCR score at bulk and single-cell level
Based on the survival data of the TCGA–SKCM cohort, we employed univariate Cox regression analysis with a bootstrap algorithm (resampling = 1000) to screen for GPCRs related to overall survival (OS). A significance threshold of
P < 0.001 was used as the cutoff. Subsequently, we performed the least absolute shrinkage and selection operator (LASSO) regression analysis using the R package “glmnet,” to further refine the selection of prognostic GPCRs. Finally, we utilized multivariate Cox regression analysis with a bootstrap algorithm (resampling = 1000) to identify the GPCRs most correlated with OS. For comparative analysis of the included GPCRs across pan-cancers, we utilized Xiantao (
www.xiantao.love/). Survival analysis of individual GPCRs in the TCGA–SKCM cohort was conducted using GEPIA [
26]. To experimentally validate the expression levels of the included GPCRs, real-time quantitative polymerase chain reaction (qPCR) was employed to assess their expression in A375 and HaCaT cell lines. For result stability, we defined the bootstrap coefficient of each included GPCR as: bootstrap coefficient =
\(\frac{\text{coefficient}}{\text{bootstrap standard deviation}}\). The GPCR score was calculated using the formula: GPCR score =
\({\sum }_{i=1}^{n}\text{bootstrap coefficient} \, \left(\text{included }{\text{GPCR}}_{i}\right) \times \text{expression level (included }{\text{GPCR}}_{i}\text{)}\). To categorize patients into low- or high-GPCR score groups, we utilized the median as the cutoff point. Differences in OS between these two GPCR score groups within the TCGA–SKCM cohort were investigated using the “survival” package. In addition, CIBERSORT was employed to analyze the differences in immune cell composition between the two groups. We extended the evaluation of the GPCR score’s prognostic impact to pan-cancer scenarios.
For the scRNA-seq data, we retained only those cells that exhibited more than 200 detected genes, less than 20% of mitochondrial genes, and fewer than 3% of red blood cell genes. Subsequently, we employed the R package “Seurat” to identify highly variable genes, perform principal component analysis, conduct graph-based clustering, and execute t-distributed stochastic neighbor embedding (t-SNE) analysis. The annotation of individual cells was based on classical marker genes. To validate the annotation of melanoma cells, we utilized the “inferCNV” package. To compute the GPCR score of each cell, we employed the “AddModuleScore” function, and the results were visualized by ‘FeaturePlot’ and ‘VlnPlot’ functions.
Establishment of the TME score
For the TME score, we first calculated the abundance of immune cells in melanoma using CIBERSORT and obtained quantitative data from 22 immune cell types. Patients were divided into high- and low-infiltration groups based on the infiltration of each immune cell, and survival analysis was performed. Prognostic immune cells were defined as those exhibiting a different OS between the two groups. Furthermore, we utilized multivariate Cox regression analysis with a bootstrap algorithm (resampling = 1000) to calculate the bootstrap coefficient of the prognostic immune cells. The TME score was defined as: TME score = \({\sum }_{i=1}^{n}\text{bootstrap coefficient} \, \left(\text{prognostic }{\text{immune cell}}_{i}\right) \times \text{infiltration level (prognostic }{\text{immune cell}}_{i}\text{)}\). Patients were classified into low- or high-TME score groups based on the median, and a survival analysis was conducted to investigate the difference in OS between the two TME groups. Subsequently, we combined the GPCR score with the TME score to develop the GPCR–TME classifier. Melanoma patients were divided into four subgroups: GPCRlow/TMElow, GPCRhigh/TMElow, GPCRlow/TMEhigh, and GPCRhigh/TMEhigh based on the median of GPCR and TME score. A survival analysis was performed to investigate the difference in OS between the four subgroups. Furthermore, we assessed the precision of the GPCR–TME classifier using the area under the curve (AUC) of 1-, 3-, and 5-year receiver operating characteristic curves (ROC) with the R packages “timeROC” and “survivalROC.”
Robustness and independence of the GPCR–TME classifier
Survival analysis was utilized to investigate the differences in OS among the subgroups in the TCGA–SKCM cohort. In addition, Cox regression analyses were conducted to assess whether the GPCR–TME classifier could function as an independent prognostic factor for melanoma in the TCGA–SKCM cohort. These findings were further validated using the GSE65904 cohort.
Enrichment analysis of the GPCR–TME classifier
Gene set enrichment analysis (GSEA) was conducted to elucidate potential pathways associated with the high-/low-GPCR and high-/low-TME groups. Weighted gene co-expression network analysis (WGCNA) was employed to cluster genes with similar expression profiles using an unsupervised analysis method [
27,
28]. Subsequently, Metascape was utilized to explore the enrichment results of the genes within the key modules identified through WGCNA [
29]. Subsequently, we employed the R package “fgsea,” to perform GSEA among the subgroups. Ultimately, we applied the tracking tumor immunophenotype (TIP) to explore the anticancer immune status of the subgroups based on the tumor immune cycle in seven stages [
30].
Decoding the GPCR–TME classifier at the genome level
Tumor mutational burden (TMB) has the potential to drive effective anti-tumor immune responses, ultimately leading to sustained clinical responses to immunotherapy [
31]. We calculated the TMB of each patient in the TCGA–SKCM cohort using previously established methods [
32] and compared TMB levels among the subgroups.
Genomic mutation data (mutect2) for the TCGA–SKCM was retrieved using the R package “TCGAbiolinks.” We utilized the “maftools” package to investigate and visualize the top 20 genes with the highest gene frequencies in both the GPCRhigh/TMElow and GPCRlow/TMEhigh groups.
Prediction of the immunotherapy response rate among GPCR–TME subgroups
Furthermore, we conducted a comparative analysis of the expression levels of antigen presentation genes and immune checkpoints across the subgroups. Finally, we constructed the GPCR–TME classifier for the three melanoma immunotherapy cohorts and investigated the response rate to immunotherapy in each subgroup.
Statistical analysis
All statistical analyses were carried out using R 4.1.1, including the Student’s t test, Wilcoxon rank-sum test, Fisher’s’ exact test, log-rank test, and Cox regression analyses. For multiple groups comparison, the Bonferroni method was employed for multiple testing correction. The cutoff was set at P < 0.05 unless otherwise stated.
Discussion
In recent years, the increased number of studies dedicated to GPCRs and their interaction with the TME has significantly enhanced our understanding of their critical roles in the prognosis and therapeutic approaches for cancer patients [
33‐
36]. For instance, Zhang et al. observed a substantial down-regulation of GPRASP1 in head and neck cancers, which was notably associated with the infiltration of CD8 T cells [
37]. In a separate study, Yu et al. identified GNAI2 as a risk factor for gastric cancer as it appeared to promote the accumulation of Tregs [
38]. In addition, Yu et al. found that P2RY12 was downregulated in lung adenocarcinoma and exhibited a significant correlation with M2 macrophage and dendritic cell infiltration [
39]. Nevertheless, these investigations primarily focused on individual GPCR for other cancer types. Studies utilizing multi-omics data, coupled with GPCRs and the TME, to predict immunotherapy response rate and OS remain relatively scarce. In our comprehensive study, we systematically integrated extensive melanoma data sets, enabling us to thoroughly explore the crosstalk between the GPCRs and the TME. The outcomes of this effort, the GPCR–TME classifier, has proven to be an exceptionally effective predictor for both the OS and immunotherapy response rate of melanoma.
Using a variety of machine learning algorithms, we identified 12 prognostic GPCRs and constructed the GPCR score. Notably, a high GPCR score signifies a poor OS for melanoma; however, it may serve as a protective factor in other cancers. This discrepancy could be attributed to the distinct roles played by these 12 GPCRs in the tumorigenesis of various cancers. Furthermore, the widespread distribution of GPCRs in vivo may account for our inability to detect differential expression levels between HaCaT and A375 cells. Considering prior research highlighting the varying impact of GPCRs on TME [
40‐
42], we delved into the immune microenvironment of the high- and low-GPCR score groups using CIBERSORT. Our findings revealed that the low-GPCR group exhibits heightened infiltration levels of immunologic effector cells, including CD8 T cells and M1 macrophages, thus providing some insight into their improved OS.
Given that bulk-seq analysis treats all cells in the TME as homogeneous and may result in the loss of crucial information, we took a closer look at the GPCR score at the single-cell level. In the TME of melanoma, we identified seven distinct cell types, including T cells, melanoma, fibroblasts, macrophages, endothelial cells, B cells, and plasma cells. Notably, immune cells have exhibited significantly higher GPCR scores when compared to tumor and stroma cells, further substantiating the intricate relationship between GPCR and TME. Consequently, we proceeded to construct the TME score to provide a quantitative assessment of the TME in patients with melanoma. This score integrated the impact of five immune cell types that were found to be prognostically relevant in melanoma. As expected, a higher TME score demonstrated a favorable OS, and we combined the GPCR and TME scores to develop a GPCR–TME classifier. The GPCRlow/TMEhigh subgroup had the best OS, whereas the GPCRhigh/TMElow subgroup had the worst OS. To further underly reasons for diverse prognoses among these subgroups, we conducted various enrichment analyses. The results shed light on the underlying mechanism of superior OS in the GPCRlow/TMEhigh subgroup, which included the activation and recruitment of immune effector cells and the positive regulation of the immune response.
Subsequently, we sought to unravel the GPCR–TME classifier at the genomic level. Since TMB has gained widespread acceptance as a fundamental biomarker influencing responses to immunotherapy [
43‐
45], we conducted a comparative analysis of TMB across the various GPCR–TME subgroups. Intriguingly, the GPCR
low/TME
high subgroup, which had the best OS, had the highest TMB, whereas the GPCR
high/TME
low subgroup displayed a significantly lower TMB. This result indicated that different GPCR–TME subgroups may have different immunotherapy response rates. In addition, we found that the mutation rate of
BRAF was higher in the GPCR
low/TME
high subgroup, implying the potential efficacy of
BRAF inhibitors, such as vemurafenib and dabrafenib, for these patients [
46,
47].
Having established that GPCRs exhibit crosstalk with the TME and that the GPCR
low/TME
high subgroup exhibited higher TMB, we speculated that the GPCR–TME classifier might serve as a predictor of immunotherapy response for melanoma. Initially, we compared the expression levels of antigen presentation genes among the GPCR–TME subgroups, which all displayed upregulation in the GPCR
low/TME
high subgroup. This suggests that dendritic cells may more effectively recognize tumor cells and initiate tumor eradication via CD8 T cell activation [
48]. Subsequently, we investigated the expression levels of classical immune checkpoints and observed elevated levels of CTLA-4, CD274, PDCD1, TIGIT, CD86, CD209, IDO1, and LAG3 in the GPCR
low/TME
high subgroup. Pul et al. found that the local delivery of anti-CTLA-4 could reduce the systemic Treg populations and activate effector T cells in melanoma [
49]. CD274, also known as PD-L1, has been implicated in inducing immune evasion by tumor cells. Research by Ribas et al. demonstrated that a combination of anti-PD-L1 and dabrafenib can enhance immune infiltration and elicit a durable response in advanced melanoma [
50]. Anti-PDCD1 (PD-1) therapy is a well-established immunotherapy approach for melanoma. Tjulandin et al. found that the novel PD-1 inhibitor prolgolimab could mediate significant anti-tumor effects and an endurable safety profile in advanced melanoma [
51]. TIGIT is an inhibitory receptor expressed by Tregs [
52]; research by Shusuke et al. suggests that the TIGIT/CD155 axis mediates resistance to ICIs in melanoma [
53]. CD86 is expressed on the cell membrane of melanoma and activates the T cells, which enhances the anti-tumor effect [
54]. CD209 can regulate dendritic cell trafficking and transient T-cell binding [
55]; however, its role in melanoma remains unexplored. IDO1, was initially observed in plasmacytoid-shaped cells within melanoma. Kevin et al. found that IDO1 was correlated with intra-tumoral CD8 T cells and Th1-related genes, suggesting that IDO1 could act as a biomarker of immunologic tumor control [
56]. LAG3 is expressed on the surface of activated CD4 and CD8 T cells [
57], and Nicolas et al. found that the combination of anti-LAG-3 and anti-PD-1 enhances the cytotoxic capacity of CD8 T cells and leads to the anti-tumor effect [
58]. Combining this information above, we constructed a GPCR–TME classifier for three melanoma immunotherapy cohorts. The outcomes demonstrated that the GPCR
low/TME
high subgroup exhibited significantly higher immunotherapy response rate compared to others in the GSE91061 cohort. In the GSE145996 and GSE35640 cohorts, these differences did not reach statistical significance, potentially due to the bias introduced by the limited sample sizes under stringent filtering criteria. Nevertheless, it was noteworthy that we still observed a higher immune therapy response rate in GPCR
low/TME
high subgroup compared to others in these two cohorts.
In a clinical application, the GPCR–TME classifier holds the potential to enhance the refinement of molecular subtyping and treatment strategies for melanoma. Specifically, following the surgical removal of a patient’s melanoma specimen, bulk-seq can be employed. Based on the gene expression data, calculations can be made to determine both the GPCR score and TME score. These scores aid in categorizing the patient into a specific GPCR–TME subgroup, enabling the prediction of the patient’s OS and their potential response rate to immunotherapy.
We acknowledge several limitations in our study. First, due to constraints related to tumor specimens, our research primarily relies on bioinformatics. We anticipate that future experiments, such as flow cytometry and immunohistochemistry, will help validate our findings. Second, to enhance the robustness of our conclusions, we recommend utilizing an internal cohort that includes gene expression data, survival data, and immunotherapy response data to further assess the performance of the GPCR–TME classifier.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.