3.1 TME subtypes and their prognostic significance

To comprehensively understand the functional and cellular components of melanoma and its TME, we analyzed 51 TME-related gene signatures, which were categorized into five groups: anti-tumor microenvironment, pro-tumor microenvironment, angiogenesis/fibrosis, malignant cell characteristics, and unknown (Fig. 1A). In the TCGA-SKCM dataset, Pearson’s correlation analysis of these signatures identified two major distinct clusters. One cluster was enriched in both anti-tumor and pro-tumor immune components, while the other was associated with stromal components, blood vessels development, and tumor-promoting cytokines (Fig. 1B). Neither cluster correlated with intrinsic tumor cell features, such as proliferation and metastasis.

Fig. 1

Three melanoma microenvironment subtypes via 51 TME signatures correlate with distinct prognostic outcomes. (A) Functional grouping of the 51 TME signatures included. (B) Pearson’s correlation between gene signature scores in TCGA cutaneous melanoma (TCGA-SKCM) tumor samples. Heatmap of unsupervised clustering results for TCGA-SKCM (C) and GSE65904 (D) samples, classified into three TME subtypes based on the 51 TME signatures. (E). Overall survival (OS) of melanoma patients from TCGA-SKCM stratified by TME subtype. (F, G). Distant metastasis-free survival (DMFS) (F) and disease-specific survival (DSS) (G) of melanoma patients from GSE65904 stratified by TME subtype

We classified patients from two independent melanoma datasets (TCGA-SKCM, n = 463; GSE65904, n = 214) using unsupervised clustering based on the ssGSEA scores of the 51 TME signatures. Consistent results across both datasets revealed three distinct microenvironment subtypes: TME-A, TME-B, and TME-C (Fig. 1C, D, Supplementary Fig. S1A). TME-A melanoma was characterized by elevated anti-tumor immune signature level, indicating a robust intra-tumoral immune response. In contrast, TME-C melanoma exhibited high expression of malignant cell-associated signatures, marking it as the most aggressive subtype. TME-B displayed intermediate features between these two subtypes. Survival analysis revealed that patients with the TME-A subtype had significantly longer OS, disease-specific survival, and distant metastasis-free survival than those with TME-B and TME-C subtypes (Fig. 1E-G). Patients with the TME-C subtype had the poorest prognosis. To validate our TME classification system, we applied it to a combined dataset including TCGA-SKCM, GSE65904, GSE19234, and GSE54467. This classification maintained consistent expression patterns and prognostic significance, confirming its reliability (Supplementary Fig. S1B-E).

3.2 Immune cell infiltration of TME subtypes

We used the ImmuCellAI algorithm to estimate immune cell abundance in melanoma tissues (TCGA-SKCM dataset), aiming to fully illustrate the association between the three TME subtypes and survival outcomes. Notably, patients with the TME-A subtype had significantly higher infiltration scores for major immune cell populations. These included neutrophils, dendritic cells, B cells, natural killer cells, CD4+ T cells, Tr1 cells, and cytotoxic T cells (Fig. 2A). In sharp contrast, the TME-C subtype showed reduced immune cell infiltration, with macrophages, CD8+ naïve T cells, and central memory T cells as the dominant populations.

Fig. 2

Multi-dimensional characterization of immune cell infiltration differences among TME-A, TME-B, and TME-C subtypes. (A). Immune cell infiltration across the three TME subtypes. (B). Immune subtype distribution among the three TME subtypes. (C–H). Comparative analysis of immune-related features across the three TME subtypes: IFN-γ response gene expression (C), leukocyte fraction (D), TIL regional fraction (E), BCR richness (F), TCR richness (G), and tumor purity (H)

We applied the immune subtype classification method proposed by Thorsson et al. [25], which revealed that the TME-A subtype was predominantly reclassified as C2 (IFN-γ dominant), indicative of immune cell-rich and characterized by a robust type I immune response (Fig. 2B). In particular, the TME-A subtype demonstrated significantly increased expression of IFN-γ response-associated genes (Fig. 2C). It also showed a higher leukocyte fraction, greater regional TIL presence, a richer BCR repertoire, and a more diverse TCR repertoire (Fig. 2D–G). In contrast, the TME-C subtype was marked by an inactive immune microenvironment and the highest tumor purity (Fig. 2C-H). Using GSVA score, this analysis further indicated that immune-related KEGG pathways and GO functions were notaly enhanced in the TME-A subtype compared with TME-B and TME-C subtypes (Supplementary Fig. S2A–D).

3.3 MSGRS construction based on TME Subtypes

To explore the underlying mechanisms of the TME subtypes, we performed WGCNA using the TCGA-SKCM dataset to identify hub genes correlated with both the TME subtypes and corresponding survival outcomes. We determined an optimal soft threshold (β = 16) was for constructing a scale-free signed network, which generated 12 distinct gene modules (Fig. 3A, Supplementary Fig. S3A). Using screening criteria of MM > 0.6 and GS > 0.5, we extracted a total of 564 genes from the three modules most strongly associated with the TME subtypes: the brown module (R = 0.89, P = 3 × 10− 161), green-yellow module (R = 0.62, P = 8 × 10− 51), and magenta module (R = 0.61, P = 7 × 10− 49) (Supplementary Fig. S3B–D). KEGG pathway analysis revealed that these genes were enriched in immune-related pathways, including hematopoietic cell lineage, cytokine–cytokine receptor interaction, and viral protein interaction with cytokine and cytokine receptor (Supplementary Fig. S3E). Further intersection with 1,620 immune-related genes from the ImmPort database narrowed the candidate gene set to 173 genes (Supplementary Fig. S3F). After univariate Cox regression analysis, we selected 12 prognostic genes for subsequent LASSO regression analysis (Fig. 3B). Ultimately, three hub genes (CXCR6, PD-1, and CD45) were identified to construct the MSGRS. The MSGRS for each patient was calculated using the following formula: MSGRS = (-0.1539284317 × CXCR6) + (-0.0611122903 × CD45) + (-0.0007949123 × PD-1).

Fig. 3

The MSGRS predicts prognosis in melanoma patients and their outcome to ICIs. (A). Identification of 12 gene modules correlated with TME subtype via WGCNA. (B). Selection of 3 hub TME subtype-related genes using LASSO regression analysis. (C). Distribution of risk scores, MSGRS, survival outcomes, and expression profiles of the three hub genes (CD45, CXCR6, PD-1) in the TCGA-SKCM dataset. (D). OS of melanoma patients from TCGA-SKCM stratified by MSGRS. (E). DFS of melanoma patients from GSE65904 stratified by MSGRS. (F, G). OS of melanoma patients from GSE54467 and GSE19234 stratified by MSGRS. (H, I). Differences in immune checkpoint expressions (H) and exhausted T (Tex) (I) cell infiltration between high- and low-MSGRS groups. (J, K). OS of ICIs- treated melanoma patients from DFCI2015 (J) and DFCI2019 (K) stratified by MSGRS

3.4 MSGRS and immunotherapy outcomes

As shown in Fig. 3C, the risk of death increased concomitantly with higher MSGRS values. To validate the prognostic performance of the MSGRS, we stratified untreated melanoma patients from TCGA-SKCM, GSE65904, GSE19234, and GSE54467 cohorts into high-MSGRS (top 25%) and low-MSGRS (bottom 25%) groups. Patients in the low-MSGRS group had significantly longer OS and disease-free survival (DFS) than those in the high-MSGRS group (Fig. 3D–G). Multivariate Cox regression analysis further confirmed the MSGRS as an independent prognostic factor for melanoma patients (Supplementary Table S3).

The low-MSGRS group exhibited elevated expression of immune checkpoints, including CD272, PD-L1, CD152, TIM-3, LAG-3, PD-1, and PD-L2, consistent with the abundant infiltration of exhausted T (Tex) cells in this cohort (Fig. 3H, I). To further assess the clinical relevance of the MSGRS in the context of immunotherapy, we divided melanoma patients treated with ICIs from DFCI2015 and DFCI2019 cohorts into low- and high-MSGRS groups. Kaplan–Meier survival curve analysis consistently showed significantly prolonged OS for patients with low-MSGRS group (Fig. 3J, K), indicating that the MSGRS retains robust prognostic value in melanoma patients receiving ICI therapy.

3.5 Effect of CXCR6 expression on prognosis and immune cell infiltration

To further clarify the clinical significance of the MSGRS model, we used clinical samples from the SYSUCC cohort (Table 1, Supplementary Table S4) and performed mIHC experiments to evaluate the association among the risk model, patient prognosis and immune cell infiltration. Based on the median protein expression level of each hub gene, we stratified melanoma patients into high- and low- expression groups. Patients with high CXCR6 expression had significantly longer OS and PFS than those with low CXCR6 expression (Fig. 4A, B). In contrast, no significant prognostic differences were observed according to PD-1 and CD45 expression levels (Supplementary Fig. S4A–D).

Fig. 4

CXCR6 expression correlates with prognosis and the immune cell infiltration in melanoma patients. (A-B). OS (A) and PFS (B) of untreated melanoma patients from the SYSUCC cohort, stratified by CXCR6 expression level. C-L. Differences in immune cell infiltration between CXCR6 high- and low-expression groups: immune cell density (C), CD45+ leukocytes (D), CD4+ T cell density (E), CD8+ T cell density (F), CD69+ immune cell density (G), CD4+CXCR6+ T cell density (H), CD8+CXCR6+ T cell density (I), CD8+CD69+ T cell density (J), CD8+PD-1+ T cell density (K), and CD8+LAG-3+ T cell density (L)

Given CXCR6 prominence in the MSGRS, we investigated its association with the density of various intratumoral immune cell populations (Fig. 4C). The high-CXCR6 expression group displayed significantly higher infiltration of CD45+ leukocytes, CD4+ T cells, CD8+ T cells, and CD69+ immune cells in melanoma tissues (Fig. 4D–G). Notably, CD69 serves not only as an early marker of T cell activation [25] but also as a marker of resident memory T (Trm) cells [26], indicating enhanced local immune protection. We further stratified the CD4 + T cell population into CD4+CXCR6+ and CD4+CXCR6− subpopulations (Fig. 4H; Supplementary Fig. S4E). Similarly, we subdivided the CD8+ T cell compartment into eight subpopulations: CD8+CXCR6+, CD8+CD69+, CD8+PD-1+, CD8+LAG-3+, CD8+CXCR6−, CD8+CD69−, CD8+PD-1−, and CD8+LAG-3− (Fig. 4I-L; Supplementary Fig. S4F–I). Although the enrichment of CD8+LAG-3+ T cells in the high-CXCR6 expression group did not reach statistical significance, a trend toward increased presence was observed (Fig. 4L). This finding suggests that melanoma patients with high CXCR6 expression may derive greater benefit from ICI therapy [27].

3.6 Effect of CXCR6 expression on immune cell distribution and spatial organization

We subsequently investigated how CXCR6 expression regulates the spatial distribution of various immune cells at the tumor invasive margin. In the high-CXCR6 expression group, immune cells were abundant and displayed distinct distribution patterns in both intratumoral and peritumoral regions. Notably, although the difference did not reach statistical significance, the densities of CD8+ T cells and CD69+ immune cells tended to be higher within the tumor than in the peritumoral region—particularly within 100 μm of the tumor boundary (Fig. 5A, B). In contrast, the low-CXCR6 expression group had relatively few immune cells, with significantly lower densities of CD45+ leukocytes and CD4+ T cells in the intratumoral region compared with the peritumoral region (Fig. 5C, D).

Fig. 5

CXCR6 expression modulates the spatial distribution of immune cells in melanoma. (A-D). Differences in the distribution of CD45+ leukocytes (A), CD4+ T cells (B), CD8+ T cells (C), and CD69+ immune cells (D) between intratumoral and peritumoral regions. (E-F). Comparative analysis of spatial associations: (E) Spatial distribution of CXCR6+ cells and CD8+ T cells between CXCR6-high and CXCR6-low expression groups: (F) Spatial distribution of CXCR6+ cells and CD69+ T cells between CXCR6-high and CXCR6- low expression groups

To further characterize the spatial relationship between CXCR6+ cells and intratumoral immune populations, we performed a proximity analysis. This analysis defined cells within 20 μm as being in close spatial association [28]. Compared with the low-CXCR6 expression group, the high-CXCR6 expression group showed a trend toward higher proximity frequencies between CXCR6+ cells and both CD8+ T cells and CD69+ immune cells; however, these differences did not reach statistical significance (Fig. 5E, F). Collectively, these findings indicate that CXCR6 expression is primarily associated with enhanced immune cells infiltration and improved spatial organization within the TME.