Crosstalk of four kinds of cell deaths defines subtypes of cutaneous melanoma for precise immunotherapy and chemotherapy

Abstract

Background: Cell death patterns can give therapeutic and biological clues that facilitate the development of individualized treatments for this lethal form of skin cancer.

Methods: We employed unsupervised clustering to establish robust classifications based on the four kinds of cell death-associated gene expression of 462 melanoma patients in the Cancer Genome Atlas (TCGA) and tested their reproducibility in two independent melanoma cohorts of 558 patients. We then used dimensionality reduction of graph learning to display the different characteristics of cell death patterns and immune microenvironments.

Results: We examined 570 cell death-associated gene expression data of melanoma patients for exploration, independent verification, and comprehensive classification of five reproducible melanoma subtypes (CS1 to CS5) with different genomic and clinical features. Patients in death-inactive subtypes (CS1, CS2, and CS5) had the least immune and stromal cell infiltration, and their prognosis was the poorest. A death-active subtype (CS4), on the other hand, had the highest infiltrated immune and stromal cells and elevated immune-checkpoints. As a result, these patients had the highest response to immunotherapy and the best prognosis. An additional subtype (CS3) had more diversified cell death and immune characteristics with moderate prognoses. Based on graph learning, we successfully divided the CS3 subtype into two subgroups (group A and group B) with distinct survival outcomes and immune features. Finally, we identified eight potential chemical drugs that were specifically targeted for the therapy of melanoma subtypes.

Conclusions: This research defines the intrinsic subtypes of melanoma based on the crosstalk of four kinds of cell deaths, which affords a blueprint for clinical strategies and guiding precise immunotherapy and chemotherapy for melanoma patients.

Keywords: cell deaths; chemotherapy; cutaneous melanoma; immunotherapy; subtype.

Figure 1

Identification of melanoma subtypes based on cell death-associated gene expression. (A) Calculating clustering prediction index (blue line) and Gaps-statistics (red line) in the cutaneous melanoma of The Cancer Genome Atlas data (TCGA_SKCM) cohort to determine the best cluster number. (B) Consensus heat map according to the findings of 10 integrative clustering methods with a cluster number of five. (C) Comprehensive heat map of distinctive molecular patterns across apoptosis, ferroptosis, necroptosis, and pyroptosis with annotation of genes. (D) Top 20 mutant genes in the TCGA_SKCM cohort. (E) Mutational OncoPrint of five identified melanoma subtypes in the TCGA_SKCM cohort.

Figure 2

Independent verification of melanoma subtypes. (A) Bar plot of fraction genome altered among five identified melanoma subtypes in the cutaneous melanoma of The Cancer Genome Atlas data (TCGA_SKCM cohort). (B) Comparison of tumor mutation burden and single-nucleotide variants among five melanoma subtypes in the TCGA_SKCM cohort. (C) Kaplan–Meier survival curve of five melanoma subtypes in the TCGA_SKCM cohort. (D) Heat map of subtype-specific upregulated biomarkers for five melanoma subtypes in the TCGA_SKCM cohort. (E) Heat map of nearest template prediction in the meta-Gene Expression Omnibus (GEO) cohort using subtype-specific upregulated biomarkers identified from the TCGA_SKCM cohort. (F) Kaplan–Meier survival curve of five melanoma subtypes in the meta-GEO cohort. **** means p < 0.0001.

Figure 3

Consistency evaluation of different prediction approaches. (A) Consistency heat map using Kappa statistics among different prediction approaches. (B) Agreement of the predicted five subtypes of melanoma with Clark level and tumor stage classification in the cutaneous melanoma of The Cancer Genome Atlas data cohort. (C) Heat map of subtype-specific functional pathways based on upregulated genes for five identified melanoma subtypes.

Figure 4

Association between cell deaths and tumor microenvironment. (A) Distribution of four cell death indices (apoptosis, ferroptosis, necroptosis, and pyroptosis) among five identified melanoma subtypes in the cutaneous melanoma of The Cancer Genome Atlas data (TCGA_SKCM) cohort. (B) Distribution of tumor microenvironment-related predictors (immune score, stromal score, tumor purity, ESTIMATE score, and mRNAsi score) among five identified melanoma subtypes in the TCGA_SKCM cohort. (C) Correlation coefficients between cell death indices and tumor microenvironment-related predictors in the TCGA_SKCM cohort. (D) Distribution of four cell death indices among five identified melanoma subtypes in the meta-Gene Expression Omnibus (GEO) cohort. (E) Distribution of tumor microenvironment-related predictors among five identified melanoma subtypes in the meta-GEO cohort. (F) Correlation coefficients between cell death indices and tumor microenvironment-related predictors in the meta-GEO cohort.

Figure 5

Landscape of cell deaths for five identified melanoma subtypes. (A) Comprehensive heat map of tumor-associated pathways, immune-microenvironment signatures, and expression of immune-checkpoint molecules among five identified melanoma subtypes. (B) Box plot of tumor-associated pathways, immune-microenvironment signatures, and expression of immune-checkpoint molecules among subtypes. (C) t-SNE plot for melanoma subtype distribution. (D) Graph learning-based dimensionality reduction plot of melanoma subtypes; each color represents a subtype corresponding to the previously defined subtype. (E) The intra-cluster heterogeneity within CS3 subtype, which was further divided into two subgroups according to their location in graph learning. (F) Kaplan–Meier survival curve of subgroups A and B in CS3 subtype. **** means p < 0.0001.

Figure 6

Validation of heterogeneity within CS3 subtype and immunotherapy response. (A) Kaplan–Meier survival curve of subgroups A and B in CS3 subtype at the meta-Gene Expression Omnibus (GEO) cohort. (B) Kaplan–Meier survival curve of subgroups A and B in CS3 subtype at the meta-immune response cohort. (C) Box plot of CD8 T cell effector, DNA damage response, antigen processing and presenting machinery, immune-checkpoint, tumor microenvironment score, INFG signature, MHC classes I and II, ICB resistance, and T cell exhaustion between subgroups (A) and (B, D) Immunotherapy response rate of five identified melanoma subtypes in the cutaneous melanoma of The Cancer Genome Atlas data (TCGA_SKCM) cohort. (E) Immunotherapy response rate of five identified melanoma subtypes in the meta-immune response cohort. (F) Rate of immunotherapy response for subgroups A and B in CS3 subtype at the TCGA_SKCM cohort. (G) Rate of immunotherapy response for subgroups A and B in CS3 subtype at the meta-immune response cohort.

Figure 7

Drug identification in cell death-associated gene classified clusters of melanoma. Immunotherapeutic response and potential compounds. (A) Heat map of the differential expression of cell death-associated gene classified clusters in melanoma cells at the Genomics of Drug Sensitivity in Cancer database. (B) Box plot of the area under the curve of ACY-1215, CHIR-99021, EHT-1864, ELESCLOMOL, FTI-277, NILOTINIB, TUBASTATIN A, and TWS-119 among five clusters. (C) Cell viability curves and estimated IC50 values of ACY-1215, tubastatin A, and EHT-1864. (D) Wound healing assay in A375 cells was performed after treatment with ACY-1215 (10.33 μm), tubastatin A (17.77 μm), and EHT-1864 (32.83 μm) at a 48-h recovery period. (E) Flow cytometry analysis of A357 cells which were stained with Annexin V-FITC and propidium iodide after 48 h of ACY-1215 (10.33 μm), tubastatin A (17.77 μm), and EHT-1864 (32.83 μm) treatment.