Bioinformatics analysis of the tumor microenvironment in melanoma – Constructing a prognostic model based on CD8+ T cell-related genes: An observational study

Abstract

This research endeavor seeks to explore the microenvironment of melanoma tumors and construct a prognostic model by focusing on genes specific to CD8+ T cells. The single-cell sequencing data of melanoma underwent processing with the Seurat package, subsequent to which cell communication network analysis was conducted using the iTALK package and transcription factor analysis was performed using the SCENIC package. Univariate COX and LASSO regression analyses were utilized to pinpoint genes linked to the prognosis of melanoma patients, culminating in the creation of a prognostic model through multivariate COX analysis. The model was validated using the GSE65904 and GSE35640 datasets. Multi-omics analysis was conducted utilizing the maftools, limma, edgeR, ChAMP, and clusterProfiler packages. The examination of single-cell sequencing data revealed the presence of 8 cell types, with the transcription factors RFXAP, CLOCK, MGA, RBBP, and ZNF836 exhibiting notably high expression levels in CD8+ T cells as determined by the SCENIC package. Utilizing these transcription factors and their associated target genes, a prognostic model was developed through COX and LASSO analyses, incorporating the genes GPR171, FAM174A, and BPI. This study validated the model with independent datasets and conducted additional analysis involving multi-omics and immune infiltration to identify a more favorable prognosis for patients in the low-risk group. The findings provide valuable insights into the tumor microenvironment of melanoma and establish a reliable prognostic model. The integration of multi-omics and immune infiltration analyses enhances our understanding of the pathogenesis of melanoma. The identification of specific genes holds promise as potential biomarkers for individuals with melanoma, serving as important indicators for predicting patient outcomes and determining their response to immunotherapy.

1. Introduction

Melanoma, a malignancy arising from melanocytes, presents a significant challenge in oncology due to its aggressive nature and propensity for metastasis.^[1] Despite advancements in surgical and chemotherapeutic interventions, the prognosis for melanoma patients remains unfavorable.^[2] However, recent developments in immunotherapy have revolutionized the treatment of melanoma by harnessing the immune system to target the disease.^[3]

Central to melanoma immunotherapy is the intricate interplay between tumors and T cells.^[4] T cells play a crucial role in adaptive immunity, particularly CD8+ cytotoxic T lymphocytes which possess the ability to effectively combat tumor growth by targeting and eliminating cancerous cells.^[5,6] Nonetheless, the tumor microenvironment in melanoma is characterized by a complex interplay of regulatory T cells, M2 polarized macrophages, and other factors that hinder T cell activity, thus dampening self-immunity while also facilitating immune evasion, proliferation, and metastasis of tumor cells.^[7] A comprehensive understanding of the intricate dynamics between melanoma cells and T cells is imperative for the development of efficacious therapeutic interventions.

This study utilized single-cell sequencing to investigate the tumor microenvironment of melanoma, specifically focusing on the T cell status. Furthermore, a novel risk model was developed to assess its efficacy as a prognostic biomarker for melanoma patients, offering valuable insights for the field of melanoma immunotherapy.

2. Materials and methods

2.1. Data source and download

The single-cell transcriptome sequencing dataset of untreated melanoma samples from melanoma patients was retrieved from the gene expression omnibus (GEO) database (GSE115978), totaling 32 samples.^[8] We acquired The Cancer Genome Atlas (TCGA) melanoma patient transcriptome sequencing data, methylation sequencing data, and gene mutation data from the Xena database (https://xena.ucsc.edu/). Then we integrated the sequencing data with clinical information, eliminated duplicate and incomplete samples, and curated a cohort of 455 melanoma samples.

2.2. Single-cell sequencing data processing

The “Seurat” package was employed to process single-cell datasets, with the parameters min.cell = 5 and min.genes = 2000 specified in the CreateSeuratObject function for filtering purposes. The dataset was standardized through principal component analysis (PCA), with 20 dimensions selected based on the ElbowPlot results.^[9] Subsequently, dimensionality reduction analysis was conducted using t-distributed stochastic neighbor embedding (t-SNE). The FindAllMarkers function was utilized to identify differentially expressed genes (DEGs), followed by Kyoto Encyclopedia of Genes and Genomes (KEGG) and GO enrichment analysis on the DEGs using the “clusterProfiler” package.^[10]

2.3. Cell communication and transcription factor analysis

The “iTALK” package is used to analyze the signaling network between cells, and its built-in database can be used to screen identified ligand–receptor pairs for matching.^[11] The “SCENIC” package is a method based on co-expression and motif analysis to reconstruct gene regulatory networks and identify cell states.^[12] In this study, we first used GRNBoost to derive the co-expression regulatory network, then used RcisTarget for motif enrichment and target gene prediction, followed by quantitative analysis of the activity of transcription factors and target genes in gene regulatory network modules using AUCell, and finally selected specific high expression transcription factors and their corresponding target genes in each cell type.

2.4. Survival analysis

Utilizing transcriptome sequencing data from melanoma patients in TCGA and their survival data, we performed univariate COX analysis on transcription factors highly expressed in CD8+ T cells and their target genes (P < .05) to ascertain potential prognostic genes in melanoma patients. Subsequently, key genes were identified through LASSO analysis and multivariate COX analysis.

2.5. Establishment of prognostic model for melanoma

A prognostic model for melanoma was developed utilizing multifactor COX analysis, with individual risk scores calculated accordingly. Patients were stratified into high-risk and low-risk categories based on the median score. Time-dependent receiver operating characteristic (ROC) analysis was employed to assess the model’s predictive accuracy for patient survival. The prognostic efficacy of the model was further confirmed through validation in the GSE65904.^[13] Furthermore, an investigation was conducted to assess the predictive capacity of this model on the effectiveness of immunotherapy for melanoma patients utilizing the dataset GSE35640.^[14]

2.6. Multi-omics analysis

The gene mutation, mRNA, and methylation data of TCGA melanoma patients were obtained from Xena. The “maftools” package was employed for the analysis and visualization of gene mutation data.^[15] Subsequently, the mRNA data underwent differential expression analysis using the “limma” and “edgeR” packages, with DEGs being identified based on |logFC| > 0.7 and P < .05.^[16,17] Methylation data are analyzed using the “ChAMP” package. Enrichment analysis of related genes is conducted using the “clusterProfiler” package.^[18]

2.7. Immune infiltration analysis

The Estimation of STromal and Immune cells in MAlignant Tumor tissues using Expression data (ESTIMATE) tool is utilized for assessing tumor immune infiltration status by determining the proportions of tumor cells, immune cells, and stromal cells in tumor tissues through the analysis of gene expression data.^[19] Cell-type identification by estimating relative subsets of RNA transcripts (CIBERSORT), on the other hand, is a bioinformatics technique employed for analyzing transcriptome sequencing data to predict the relative abundance of various immune cell types based on the transcriptome sequencing data of samples.^[20] The tumor microenvironment was stratified into 4 distinct subtypes by molecular functional portrait (MFP), taking into account tumor malignancy and microenvironment characteristics. Patients belonging to these subtypes exhibited notable variations in clinical outcomes and treatment responses.^[21] This study employed ESTIMATE, CIBERSORT, and MFP methodologies to investigate the influence of risk scores on immune infiltration, therapeutic interventions, and prognostic outcomes in high-risk and low-risk patient cohorts.

3. Results

3.1. Composition of melanoma tumor microenvironment

Using elbow plot to analyze the single-cell transcriptome sequencing data of melanoma, 20 principal components were determined (Fig. 1A). Dimensionality reduction analysis of the sequencing data was performed using t-SNE, resulting in 18 cell clusters (Fig. 1B). Cell annotation of these 18 cell clusters was conducted based on marker genes: CD8+ T cells (CD3E, CD8A), CD4+ T cells (CD3E, CD4), macrophages (CD14), melanoma cells (PMEL, MITF, DCT), endothelial cells (VWF), fibroblasts (COL1A1, COL3A1), B cells (CD79A), and NK cells (CCL5, GZMB) (Fig. 1C and D). Furthermore, differential expression analysis of each cell type identified genes that were specifically highly expressed in each cell type (Fig. 1E).

Figure 1.

Single-cell sequencing analysis for classification of different cells in melanoma patients. (A) Principal component analysis of single-cell sequencing results; (B) dimension reduction analysis using t-SNE, dividing 3630 cells in dataset GSE115978 into 18 cell clusters; (C) expression of various cell markers in different cell clusters; (D) annotation of 18 cell clusters into 9 cell types; (E) heatmap of genes with high specificity expression in each cell type. t-SNE = t-distributed stochastic neighbor embedding.

Differential expression genes of CD4+ T cells and CD8+ T cells were extracted for GO enrichment analysis. The results showed that the biological processes of specifically highly expressed genes in CD4+ T cells were mainly enriched in pathways such as “cytoplasmic translation”; the cellular components were mainly concentrated in “cytosolic ribosome,” “cytosolic large ribosomal subunit,” etc; and the molecular functions were mainly concentrated in “structural constituent of ribosome,” etc (Fig. 2A and C). The biological processes of specifically highly expressed genes in CD8+ T cells were mainly enriched in pathways such as “cytoplasmic translation,” “T cell receptor signaling pathway,” “leukocyte cell–cell adhesion,” “lymphocyte differentiation,” etc; the cellular components were mainly concentrated in “cytosolic ribosome,” “focal adhesion,” etc; and the molecular functions were mainly concentrated in “structural constituent of ribosome,” “GTPase regulator activity,” etc (Fig. 2D–F). CD4+ T cells and CD8+ T cells shared a large number of enriched biological pathways for specifically highly expressed genes, but CD8+ T cells had higher pathway scores.

Figure 2.

GO enrichment analysis of highly expressed genes specific to CD4+ T cells and CD8+ T cells. (A) Enriched pathways related to biological processes in CD4+ T cells with high expression of specific genes; (B) enriched pathways related to cellular components in CD4+ T cells with high expression of specific genes; (C) enriched pathways related to molecular functions in CD4+ T cells with high expression of specific genes; (D) enriched pathways related to biological processes in CD8+ T cells with high expression of specific genes; (E) enriched pathways related to cellular components in CD8+ T cells with high expression of specific genes; (F) enriched pathways related to molecular functions in CD8+ T cells with high expression of specific genes.

3.2. Analysis of cell communication and transcription factors in the tumor microenvironment of melanoma

Using the iTALK package to screen for receptor–ligand pairs based on the average expression levels of various cell types, it was found that the highly expressed ligands related to cytokines in CD4+ T cells are CCL5, with receptors CXCR3, CXCR4, and CD4, while in CD8+ T cells the highly expressed ligands related to cytokines are CCL4 and CCL5, with receptors CXCR3, CXCR4, CXCR6, and CCR5 (Fig. 3A); the highly expressed ligands related to checkpoints in CD4+ T cells are TNFSF14, LGALS9, BTLA, CD40LG, with receptors SIGLEC10, CD247, HAVCR2, CTLA4, CD28, and ITGB2, while in CD8+ T cells the highly expressed ligands related to checkpoints are TNFSF14, LGALS9, BTLA, with receptors CD28, SIGLEC10, CD247, CTLA4, ITGB2, and HAVCR2 (Fig. 3B).

Figure 3.

Analysis of cell communication and transcription factors in various cell types. (A) Analyze the communication of cytokines between various types of cells using the “iTALK” package; (B) analyze the checkpoint communication between various types of cells using the “iTALK” package; (C) analyze the specific high expression transcription factors in various types of cells using the “SCENIC” package.

To further explore the differential expression of transcription factors in various cell types, the cells were analyzed using the “SCENIC” package. The results showed that the transcription factors DLX4, NR5A2, PRDM14, MAFK, and USF1 were specifically highly expressed in CD4+ T cells; while the transcription factors RFXAP, CLOCK, MGA, RBBP, and ZNF836 were specifically highly expressed in CD8+ T cells (Fig. 3C).

3.3. Establishment of prognostic model for melanoma patients

A large number of studies suggest a closer relationship between CD8+ T cells and tumors. Therefore, this study aimed to explore the specific transcription factors with high expression in CD8+ T cells and their target genes. A total of 725 target genes were selected by setting importance > 0.8, and single-factor COX analysis revealed that 113 genes were associated with the prognosis of melanoma patients (Supplementary Table 1, Supplemental Digital Content, http://links.lww.com/MD/N201). Through LASSO regression analysis, the key genes were further reduced to 8 (Fig. 4A), and finally, GPR171, FAM174A, and BPI were determined to be independent factors for the prognosis of melanoma patients using multifactor COX analysis (Fig. 4B). Our results showed that compared to melanoma patients with poor prognosis, those with better prognosis had higher expression of GPR171, FAM174A, and BPI (Fig. 4C and E).

Figure 4.

Screening key genes associated with prognosis in melanoma patients. (A) LASSO regression analysis screens genes related to prognosis in single-factor COX analysis; (B) random forest plot shows GPR171, FAM174A, and BPI as independent prognostic factors for melanoma patients; (C) survival analysis of melanoma patients based on the expression of GPR171; (D) survival analysis of melanoma patients based on the expression of FAM174A; (E) survival analysis of melanoma patients based on the expression of BPI.

An ideal tumor prognosis model can predict patients’ future risk of death and response to treatment. By combining multifactor COX analysis, we established a melanoma prognosis model using GPR171, FAM174A, and BPI: Melanoma risk score = (−0.41503 * GPR171) + (−0.33039 * FAM174A) + (−0.88386 * BPI). Based on this model, calculate the risk score for each patient, rank them in ascending order, and found that as the risk score increases, the likelihood of patient death also increases (Fig. 5A and C). Furthermore, we analyzed the effectiveness of the model through time-dependent ROC analysis. The area under the curve at 12, 24, 36, 48, and 60 months was 0.65, 0.66, 0.64, 0.66, and 0.69, respectively (Fig. 6A), supporting the high accuracy of the prognosis model in predicting melanoma patient death over several years.

Figure 5.

Construction of a prognostic model for melanoma patients based on GPR171, FAM174A, and BPI. (A) According to the prognostic model for melanoma patients, each melanoma patient is scored and ranked in ascending order. The horizontal axis represents each patient, and the vertical axis represents the risk score; (B) patients are ranked in ascending order on the horizontal axis based on their risk scores, the vertical axis represents the duration of patient follow-up, red indicates patients who are alive, blue indicates patients who are deceased; (C) heatmap showing the expression of GPR171, FAM174A, and BPI in melanoma patients, patients are ranked from left to right based on their risk scores. Red indicates high expression, green indicates low expression.

Figure 6.

Validation of the model’s ability to predict prognosis and immune response in melanoma patients. (A) The ROC analysis shows the effectiveness of the model in predicting the prognosis of melanoma patients at 12, 24, 36, 48, and 60 mo; (B) the effectiveness of the model was validated in the GSE65904 dataset, where red represents the high-risk group and blue represents the low-risk group; (C) according to the model, patients in the GSE35640 dataset were divided into high-risk and low-risk groups, and the chi-square test shows that patients in the low-risk group have a higher response rate to immune checkpoint inhibition therapy. ROC = receiver operating characteristic.

In addition, we validated the above models using the GSE65904 dataset, and as expected, patients in the low-risk group had better prognosis (Fig. 6B). To explore the effectiveness of the model in predicting immunotherapy, we calculated risk scores for patients in the GSE35640 dataset, divided them into high-risk and low-risk groups based on the scores, and found that patients in the low-risk group had significantly better response to immunotherapy, as determined by chi-square test (Fig. 6C).

3.4. Multi-omics analysis between high and low-risk groups of melanoma patients

Using the melanoma prognosis model, melanoma patients in TCGA were divided into high-risk and low-risk groups, and the differences in mRNA, methylation, and gene mutations between the 2 groups were analyzed. Compared to patients in the low-risk group, patients in the high-risk group showed that highly expressed mRNAs were associated with pathways such as “Tyrosine metabolism” and “Ras signaling pathway,” while lowly expressed mRNAs were associated with pathways such as “Th17 cell differentiation,” “Th1 and Th2 cell differentiation,” and “Cytokine–cytokine receptor interaction” (Fig. 7A); compared to patients in the low-risk group, patients in the high-risk group with high methylation genes were associated with pathways such as “Th17 cell differentiation,” “Th1 and Th2 cell differentiation,” and “Cytokine–cytokine receptor interaction,” while genes with low methylation were associated with pathways such as “Tyrosine metabolism” and “Ras signaling pathway” (Fig. 7B), which is consistent with the mRNA analysis results.

Figure 7.

Multi-omics analysis based on risk score. (A) KEGG enrichment analysis of differentially expressed genes between high and low-risk groups; (B) KEGG enrichment analysis of differentially methylated genes between high and low-risk groups; (C) variant allele frequency in the high-risk group; (D) variant allele frequency in the low-risk group. KEGG = Kyoto Encyclopedia of Genes and Genomes.

The higher the variant allele frequency (VAF), the higher the heterogeneity and purity of the tumor. When analyzing 2 groups of gene mutations, it was found that the VAF of mutated genes in the high-risk group was significantly higher than that in the low-risk group (Fig. 7C and D), indicating that the heterogeneity and purity of melanoma in the high-risk group are higher.

3.5. Analysis of the correlation between melanoma risk score and tumor immunity

Explore the correlation between risk score and tumor immunity through ESTIMATE, CIBERSORT, and MFP analysis. In the ESTIMATE analysis, it was found that the high-risk group had lower stromal score, lower immune score, and lower ESTIMATE score compared to the low-risk group, but higher tumor purity score (Fig. 8A–D), which is consistent with the results obtained from VAF analysis. The ESTIMATE analysis results showed that melanoma patients in the low-risk group had more immune cells and stromal cells, and relatively lower tumor malignancy. Compared to patients in the high-risk group, patients in the low-risk group often have better prognosis.

Figure 8.

Correlation analysis between risk score and immune microenvironment. (A–D) Use ESTIMATE to analyze the differences in StromalScore (A), ImmuneScore (B), ESTIMATEScore (C), and TumorPurity (D) between high and low-risk groups; (E) use MFP analysis to classify TCGA patients into 4 subtypes and compare the risk scores between the 4 subtypes; (F) use CIBERSORT to analyze the high and low-risk groups, obtain the proportions of various cell types, and make pairwise comparisons (** indicates P < .01; **** indicates P < .0001). CIBERSORT = cell-type identification by estimating relative subsets of RNA transcripts, ESTIMATE = Estimation of STromal and Immune cells in MAlignant Tumor tissues using Expression data. MFP = molecular functional portrait, TCGA = The Cancer Genome Atlas.

Using MFP analysis, melanoma can be classified into 4 subtypes: immune-enriched, nonfibrotic (IE); immune-enriched, fibrotic (IE/F); fibrotic (F); immune-depleted (D). It is well known that the tumor microenvironment can influence patients’ clinical outcomes and response to treatment. Typically, patients with IE and IE/F subtypes have stronger antitumor capabilities and can benefit more from immune checkpoint inhibition therapy. In the TCGA melanoma dataset, D group melanoma had the highest risk scores, followed by the F group, while the IE and IE/F groups had the lowest scores with no significant difference between the 2 groups (Fig. 8E).

The CIBERSORT analysis results show that in the low-risk group, the proportions of initial B cells, plasma cells, CD8+ T cells, activated CD4+ memory T cells, activated NK cells, monocytes, and M1 macrophages are significantly higher than those in the high-risk group; while in the high-risk group, the proportions of resting CD4+ memory T cells, M0 macrophages, M2 macrophages, and resting mast cells are significantly higher than those in the low-risk group (Fig. 8F). The above analysis is consistent with previous results, indicating a close relationship between risk score and immune cell infiltration. Lower scores are associated with increased immune cell infiltration in the tumor microenvironment, which may indicate a better prognosis.

4. Discussions

Melanoma is a prevalent malignant neoplasm worldwide, with treatment typically involving surgical intervention, chemotherapy, and other modalities. Prognosis for patients with metastatic melanoma remains poor, with high mortality rates. Consequently, the search for efficacious treatment strategies is imperative, with immunotherapy demonstrating notable therapeutic benefits in melanoma by effectively inhibiting cancer cell proliferation.^[22] T cells are integral components of the immune system, primarily responsible for mediating cellular immune responses, with CD8+ T cells serving a pivotal function in eradicating tumor cells.^[23] Consequently, it is imperative to conduct additional research on the correlation between T cells and melanoma, as well as develop novel risk assessment models, in order to gain fresh perspectives on the management of melanoma.

The tumor microenvironment (TME) is composed of tumor cells, immune cells, stromal cells, etc, and they interact with each other.^[24] By analyzing single-cell transcriptome sequencing data of melanoma tissues, 8 cell types were identified: CD8+ T cells, CD4+ T cells, macrophages, melanoma cells, endothelial cells, fibroblasts, B cells, and NK cells. CD8+ T cells are closely related to tumors and can exert antitumor activity by interacting with tumor antigens to indirectly or directly lyse cells.^[23,25] This study found that the transcription factors RFXAP, CLOCK, MGA, RBBP, and ZNF836 are specifically highly expressed in CD8+ T cells. Further analysis using single-factor COX analysis of these 5 transcription factors and their target genes revealed 113 genes associated with the prognosis of melanoma patients. Through LASSO regression and multifactor COX analysis, it was determined that high expression of GPR171, FAM174A, and BPI is associated with a good prognosis in melanoma patients. Previous studies have shown that GPR171 can promote appetite generation and its receptor has been proven to be a drug target for treating eating disorders such as anorexia nervosa or bulimia.^[26] Activation of GPR171 can also effectively alleviate abnormal pain and hyperalgesia.^[27] FAM174A has been found to be associated with cholesterol and Parkinson disease.^[28,29] Overexpression of BPI in T cells can promote self-inflammatory responses.^[30] However, the functions of GPR171, FAM174A, and BPI in melanoma are not yet clear and they may serve as potential prognostic genes for melanoma.

A prognostic model based on GPR171, FAM174A, and BPI can classify melanoma patients into high-risk (RS) or low-risk (RS) groups, and the effectiveness of this model has been validated across multiple datasets. Differential analysis of mRNA, methylation, and gene mutations between the high-risk and low-risk groups was conducted through multi-omics analysis. It was found that compared to patients in the low-risk group, the low-expressed mRNAs in the high-risk group were associated with pathways such as “Th17 cell differentiation” and “Th1 and Th2 cell differentiation.” Consistent with the mRNA analysis results, hypermethylated genes in the high-risk group were also associated with the aforementioned pathways. Studies have shown that transplanting Th17 cells can promote the reduction of tumor cells in mice.^[31] The mechanism by which Th17 cells reduce tumor cells is not fully understood, but it may be closely related to CD8+ T cells.^[32] In addition, melanoma patients who respond effectively to treatment show a significant increase in IL17 expression in their bodies compared to those who do not respond to PD-1 therapy.^[33] Th1 cells can produce IFN-γ, TNFα, etc, participating in antitumor immune responses, while Th2 cells dominate at tumor sites, promoting tumor growth and metastasis.^[34,35] The imbalance between Th1 and Th2 cells is a common mechanism of immune escape by tumor cells, closely related to the occurrence, development, and prognosis of cancer.^[36]

The above research demonstrates a close relationship between prognosis models and immunity, so this study further explores the correlation between risk scores and immunity. Through ESTIMATE analysis, it was found that compared to the high-risk group, the low-risk group has more immune cells and stromal cells, and the malignancy of the tumor is relatively lower, proving that patients in the low-risk group have a better prognosis. CIBERSORT analysis results are consistent with ESTIMATE analysis, showing that the low-risk group has more antitumor CD8+ T cells, active CD4+ memory T cells, and active NK cells infiltrating, while the high-risk group has more pro-tumor M2 macrophages infiltrating.

Although this study provides a certain theoretical basis and research reference, there are still limitations. First, the study is based on public datasets, and the predictive ability needs further validation through prospective clinical studies. Second, this study does not include any in vitro or in vivo experiments, and further exploration is needed on the potential molecular mechanisms of prognosis models and immunotherapy.

5. Conclusion

In summary, this study identified 3 genes, GPR171, FAM174A, and BPI, related to CD8+ T cells through single-cell transcriptome sequencing, t-SNE dimensionality reduction analysis, univariate COX analysis, LASSO regression, and multivariate COX analysis. A melanoma prognosis model was established to provide accurate survival prediction for melanoma patients, making these genes potential prognostic markers and therapeutic targets for melanoma. However, further validation of all results from this study is needed in experimental and clinical practice.

Author contributions

Validation: Zhenghao He.

Visualization: Zhenghao He.

Writing – original draft: Zhenghao He, Manli Chen.

Writing – review & editing: Zhijun Luo.

Supplementary Material

medi-103-e38924-s001.docx ^{(26.8KB, docx)}