Checkpoint Gene Signatures Reveal Three Melanoma Immune Types With Sharply Different Survival Odds

Immune checkpoint inhibitors have transformed the treatment of advanced melanoma, yet some patients live for years after therapy while others derive little benefit. A new re-analysis published in Frontiers in Genetics argues that part of that heterogeneity may be visible in the coordinated expression of checkpoint genes themselves. By clustering 472 cutaneous melanoma tumors from The Cancer Genome Atlas according to a 47-gene immune checkpoint panel, the researchers identified three tumor-immune phenotypes — ICG-high, ICG-mid and ICG-low — and found that patients whose tumors belonged to the high-expression group lived substantially longer overall. Because the work is a re-analysis of public cohorts rather than a prospective trial, the findings describe associations, not predictive tests, but the scale and internal consistency of the results make them difficult to ignore.

The team, led by dermatologists and a plastic surgeon at hospitals in Sichuan, Shandong and Jiangsu, standardized each of the 47 checkpoint-related genes across tumors and applied unsupervised consensus clustering over cluster counts from two to six. Based on the cumulative distribution function of consensus matrices and the gains in stability at each step, the largest jump occurred at three clusters, so three phenotypes were retained. The final groups were ordered by mean checkpoint z-score: 136 tumors, or 28.8 percent, fell into the ICG-high group, 203 tumors, or 43.0 percent, into ICG-mid, and 133 tumors, or 28.2 percent, into ICG-low. Mean consensus within phenotypes was 0.958, compared with only 0.020 between them, indicating that the three groups were internally coherent and clearly separated.

The first hint that these phenotypes capture something biologically meaningful came from the immune landscape. The checkpoint score correlated strongly with a CD8 T-cell signature measured by single-sample gene set enrichment analysis, with a Spearman correlation of 0.828 and a P value of 5.06 × 10⁻¹²⁰. Mean CD8 T-cell scores declined steadily across the phenotypes, from 0.211 in ICG-high through −0.011 in ICG-mid to −0.134 in ICG-low, and T-cell signatures showed the strongest separation of any measure tested, with a Kruskal–Wallis P value of 3.95 × 10⁻⁷⁰. In other words, tumors with high checkpoint expression were not simply more immunosuppressed; they were also the tumors with the strongest signs of immune-cell infiltration, a pattern consistent with the established concept of adaptive immune resistance, in which an active antitumor response induces its own regulatory brakes.

Survival analysis sharpened the picture. After selecting one tumor per patient, 454 of the 469 unique patients had usable follow-up, and median overall survival differed dramatically across phenotypes: 4,930 days for ICG-high, 2,192 days for ICG-mid and 1,487 days for ICG-low, with a log-rank P value of 1.50 × 10⁻⁵. Because checkpoint expression also varied with AJCC stage, the investigators built multivariable Cox models adjusting for stage, age and sex. Among 404 complete cases with 190 deaths, ICG-high versus ICG-low tumors carried a hazard ratio of 0.34 (95 percent confidence interval 0.22–0.51; P = 2.68 × 10⁻⁷), and the mid group also fared better than the low group (hazard ratio 0.57; 95 percent CI 0.41–0.80). A sensitivity model stratifying the baseline hazard by stage gave a nearly identical estimate, 0.35. The authors are careful to frame this as a prognostic association in a retrospective cohort, not evidence that checkpoint expression predicts who will benefit from immunotherapy.

An exploratory look at somatic mutations added texture but also caution. Mutation annotations were available for only 66 tumors, including just four in the ICG-high group, so the comparisons carry wide uncertainty. Within that subset, BRAF mutation rates were 50.0 percent in ICG-high, 55.6 percent in ICG-mid and 42.3 percent in ICG-low, while median total mutation counts trended higher in the high phenotype — 581.5 versus 283.0 and 235.0 — without reaching statistical significance. UV-associated mutation fractions were similarly distributed. The authors present these results as descriptive rather than conclusive.

To introduce a temporal dimension, the team turned to GSE91061, a published nivolumab cohort containing 109 RNA-sequencing samples from 65 patients, including 43 patients with paired pretreatment and on-treatment biopsies. Within-patient paired testing showed that the checkpoint score itself rose during treatment, with a median increase of 0.115 (Wilcoxon signed-rank P = 0.012). CD8 T-cell, interferon-gamma and exhaustion signatures rose in parallel: median CD8 T-cell scores increased from 0.142 to 0.209, interferon-gamma scores from 0.304 to 0.358, and exhaustion scores from 0.047 to 0.115, all statistically significant. The strongest paired change was a cytotoxicity signature. Concurrent increases in effector and exhaustion signals suggest that nivolumab treatment triggers immune activation accompanied by regulatory feedback, rather than a simple unleashing of antitumor activity.

The third cohort, GSE120575, offered cellular resolution. The researchers re-analyzed all 16,291 CD45-positive immune cells from 48 biopsies representing 32 patients treated with anti-PD-1, anti-CTLA4 or combination therapy, using the deposited log-transformed expression values and marker-based annotation into broad populations: 7,242 general T cells, 4,403 CD8 T cells, 2,438 B cells, 1,182 myeloid cells, 912 NK cells, 104 dendritic cells and 10 mast cells. Crucially, the analysis accounted for repeated biopsies from the same patient, adjusted for time point and treatment regimen, and applied patient-clustered robust inference. Among 20 responder biopsies from 14 patients and 28 nonresponder biopsies from 18 patients, immune-cell interferon-alpha and interferon-gamma scores were significantly higher in nonresponders after adjustment — interferon-alpha scores averaged 1.343 in responders versus 2.084 in nonresponders (adjusted difference −1.147; P = 1.29 × 10⁻⁴) and interferon-gamma scores 1.104 versus 1.657 (adjusted difference −0.556; P = 0.004). Official Hallmark gene sets reproduced the direction with even tighter confidence intervals.

That nonresponders carried higher interferon scores may seem paradoxical, since interferon-gamma is classically a pro-immune signal. The baseline-only interferon-gamma estimate was directionally lower in responders but imprecise (P = 0.055), whereas the on-treatment estimate was more precise (P = 0.012), a pattern the authors interpret cautiously as compatible with an on-treatment interferon-high or feedback-regulated state in nonresponders — persistent immune stimulation or resistance-associated signaling rather than effective tumor control. They note that acute interferon-gamma signaling induces antigen-presentation and checkpoint molecules in melanoma cells, while sustained tumor-cell interferon signaling has been linked experimentally to adaptive resistance. To test the acute arm directly, they treated two human melanoma cell lines, A375 and SK-MEL-28, with recombinant human interferon-gamma at 10 or 100 ng/mL for 24 or 48 hours and measured transcripts by RT-qPCR. Relative expression of CD274, the gene encoding PD-L1, rose from 1.00 to 4.16 in A375 and to 3.41 in SK-MEL-28; the chemokine CXCL10 showed the largest induction, reaching 9.23-fold and 7.24-fold respectively; and HLA-A increased to 2.42-fold and 1.73-fold. All six comparisons remained significant after Holm correction, with adjusted P values below 0.001.

The authors emphasize that these four lines of evidence — bulk prognostic clustering, longitudinal treatment transcriptomics, single-cell response comparison and cell-line stimulation — address different questions in different patients, compartments and time points, so their effect directions are not expected to align along a single axis. The TCGA analysis estimates prognosis in a heterogeneous cohort; the nivolumab cohort measures within-patient change after therapy begins; the single-cell cohort compares immune-cell expression between response groups at the biopsy level; and the qPCR experiment validates pathway inducibility in isolated tumor cells. None of them establishes causality, and the cohorts were not linked at the patient level.

The limitations are laid out plainly: the single-cell cohort profiled CD45-positive immune cells rather than tumor cells, so cell composition may partly explain biopsy-level interferon differences despite within-cell-type analyses; the imprecise baseline-only estimate shows that pretreatment separation of responders is uncertain; and TCGA survival associations cannot establish predictive value for checkpoint blockade benefit. What the study does deliver is a carefully documented framework — full gene-set definitions, one-tumor-per-patient survival modeling, sensitivity analyses excluding enrichment-labelled cells, and analysis of every one of the 16,291 immune cells — for reading checkpoint and interferon programs as a coordinated language of the tumor-immune interface. For a disease in which immunotherapy outcomes remain frustratingly variable, that framework offers researchers a richer descriptive vocabulary, even as the hunt for a true clinical biomarker continues.

Subject of Research: Immune checkpoint gene-based tumor-immune phenotyping in cutaneous melanoma

Article Title: Immune checkpoint gene-based tumor-immune phenotypes stratify prognosis and immunotherapy-associated remodeling in cutaneous melanoma: a multi-cohort transcriptomic and single-cell re-analysis

Article References: Mao, Q., Yang, S., & Gong, M. (2026). Immune checkpoint gene-based tumor-immune phenotypes stratify prognosis and immunotherapy-associated remodeling in cutaneous melanoma: a multi-cohort transcriptomic and single-cell re-analysis. Frontiers in Genetics, 17, Article 1978135. https://doi.org/10.3389/fgene.2026.1978135

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1978135

Keywords: cutaneous melanoma, immune checkpoint genes, tumor-immune phenotypes, overall survival, immunotherapy, interferon signaling, single-cell RNA sequencing, TCGA, nivolumab, biomarkers, tumor microenvironment, Immune