GBP2 condensates promote ferroptosis to sensitize anti-PD-L1 immunotherapy in melanoma Authors: Xin Liu, Shilin Fang, Xianghai Ren, Yuntian Hong, Jianchun Guo, Xiaoping Liu, Liping Gao, Haihang Nie, Yumei Ning, Meng Zhang, Fei Xu, Lan Liu, Jun Fang, Yu Pang, Qiu Zhao, Fan Wang, Haizhou Wang Published: 2025 ABSTRACT AbstractThe mechanisms underlying the limited response to immune checkpoint blockade (ICB) remain unclear. One explanation is that tumors escape cytotoxic T cell killing by disrupting interferon-gamma (IFN-γ) signaling. Here we show that guanylate-binding protein 2 (GBP2), an IFN-γ–inducible molecule, functions as a general amplifier of T cell–mediated cytotoxicity by promoting ferroptosis in melanoma. GBP2 enhances STAT1 activation and suppresses SLC7A11, thereby sensitizing tumor cells to ferroptotic death. Upon IFN-γ stimulation, GBP2 undergoes phase separation through an intrinsically disordered region, forming condensates that sequester SHP1 and sustain STAT1 activation. Disrupting GBP2 phase separation impairs ferroptosis, accelerates tumor growth, and weakens T cell–driven tumor control. GBP2 also increases HMGB1 release from ferroptotic cells, promoting cytotoxic T cell infiltration. These findings identify GBP2 as a key mediator linking IFN-γ signaling to ferroptosis and demonstrate that enhancing this pathway can improve tumor responsiveness to ICB immunotherapy.Subject terms: Melanoma, Cell death and immune response, Tumour biomarkers, Gene regulation Introduction Immune checkpoint blockade (ICB), exemplified by inhibitors targeting programmed cell death 1 (PD-1) or programmed death ligand 1 (PD-L1), has emerged as a transformative therapeutic approach1. ICB therapy restores or enhances the function of effector CD8+T cells (cytotoxic T lymphocytes, CTL) and drives interferon-γ (IFN-γ) production to eliminate tumor cells. Despite substantial clinical success, a significant subset of melanoma patients exhibit suboptimal responses. The pre-existing immunosuppressive tumor microenvironment (TME) remains a major barrier to efficacy, driven by factors such as low expression of immune checkpoints and neoantigens, genomic instability, tumor mutational burden, limited infiltration of tumor-infiltrating lymphocytes (TILs), and attenuated IFN-γ gene signature2,3. These challenges underscore the urgent need for combinatorial strategies capable of overcoming TME‑associated resistance and broadening the benefit of ICB. One promising approach to elicit this transformation is to induce ferroptosis in tumor cells. Ferroptosis is an iron‑dependent form of cell death driven by lethal accumulation of lipid peroxides4. Multiple strategies have been employed to trigger ferroptosis, including inhibition or degradation of ferroptosis suppressor glutathione peroxidase 4 (GPX4), development of novel sensitizers for ferroptosis suppressor protein 1 (FSP1), blockade of system Xc−, and therapeutic cancer nanovaccines—all of which have shown synergistic effects with ICB therapy in preclinical cancer models5. Recently, Wang et al. demonstrated that type II IFN signaling serves as an effective entry point to amplify ferroptosis6. However, the precise molecular mechanisms by which IFN-γ enhances ferroptosis remain poorly understood. Emerging evidence suggests that many molecular assemblies form membraneless biomolecular condensates via phase separation, offering an insight into previously unexplained biological phenomena7–9. Dysregulated phase separation has been increasingly recognized as a key contributor to cancer progression, including modulation of immune responses8,10,11. Moreover, recent studies indicate that phase separation can facilitate ferroptosis, raising the possibility that this process may serve as a regulatory hub12,13. Nonetheless, the mechanisms by which phase separation selectively compartmentalizes signaling components to control ferroptotic activity remain largely elusive. In this study, to identify candidate targets for combinatorial immunotherapy, we applied weighted correlation network analysis (WGCNA) to screen hub genes associated with IFN-γ response and CTL cytolytic activity. Guanylate binding protein 2 (GBP2) was emerged as a key gene of interest. We observed that GBP2 expression positively correlated with CTL cytotoxicity and infiltration. GBP2 overexpression enhanced, whereas GBP2 deletion attenuated, ferroptosis induced by ICB‑activated CTLs. Mechanistically, we reveal an IFN‑γ–dependent pathway in which tumor cell GBP2 undergoes phase separation, sequestering SHP1 to promote STAT1 phosphorylation. This signaling cascade drives immunogenic ferroptosis, thereby potentiating the efficacy of anti–PD‑L1 immunotherapy. Our findings establish GBP2 phase separation as a previously unrecognized molecular circuit that mediates IFN-γ-dependent ferroptosis and highlight GBP2 as a promising target for combinatorial melanoma immunotherapy. Results WGCNA identified GBP2 as a CTL cytolytic activity associated gene in melanoma ICB therapy primarily eliminates tumor cells by enhancing the effector function of CTL. Activated CTLs engage multiple cytolytic pathways, including IFN-γ, perforin-granzyme, and Fas-Fas ligand signaling to eradicate tumor cells14. However, tumors frequently evade this immune pressure by impairing CTL cytolytic activity—either through inducing T cell dysfunction in CTL-high tumors or by preventing T cell infiltration in CTL-low tumors6. To identify molecular targets associated with CTL cytolytic activity, we employed WGCNA, a robust strategy for screening candidate genes linked to phenotypic traits15. Fifteen co‑expression modules were identified (Figs. 1A and S1). Among them, the yellow module exhibited the strongest correlation with 6-gene-IFN-γ signature (R = 0.89, P = 4e-29), cytolytic activity signature (R = 0.81, P = 7e-21) and two independent cytotoxic cell signatures (R = 0.86, P = 9e-26 and R = 0.81, P = 3e-20), as determined by module eigengene analysis (Fig. 1B and Supplementary Dataset 1). These signatures collectively predict CTL cytolytic activity16–19. Gene Ontology biological process (GO-BP) analysis further revealed that yellow module genes were significantly enriched in immune regulation, cell killing and IFN-γ response (Fig. 1C). These findings indicate that the yellow module represents a central network associated with CTL cytotoxicity, warranting further interrogation of its constituent genes for hub gene identification. Identification of GBP2 as a CTL cytolytic activity associated gene in melanoma through WGCNA. A Gene dendrogram from the GSE8401 dataset clustered according to dissimilarity (1–TOM). B Heatmap showing correlations between module eigengenes and CTL cytolytic activity–related signatures in GSE8401 dataset. C The Gene Ontology (GO) analysis based on genes in the yellow module. D Gene set enrichment analysis (GSEA) of “positive regulation of GTPase activity” and “guanyl ribonucleotide binding” signatures between responder and no-responder patients according to the PRJEB23709 cohort. E Venn diagram of intersection of GTP binding proteins and genes in the yellow module. F Correlations between the intersected genes and overall survival of patients in SKCM cohort assessed by Univariate Cox analysis (n = 470). GBP2 level (HR = 0.778, 95%CI: 0.722–0.838). G Spatial transcriptomics was employed to analyze one tissue section, identifying clusters and accurately aligning them with morphological features as indicated by hematoxylin and eosin staining and cluster mapping. This slice can be divided into three areas: malignant, boundary, and non-malignant. H The left chart displays the cellular composition of spatial spots in the whole sample; the right one compares the cellular compositions across all regions, presented as bar charts. I The spatial feature plot that visualizes the proportion of the indicated cell types and spatial expression patterns of GBP2 in spatial spots (n = 1). TOM: topological overlap matrix. SKCM: skin cutaneous melanoma. All p value < 0.05 as statistic difference. p value by a two-sided Wald test (F) or two-sided permutation test (D) or one-sided hypergeometric test (C). Source data are provided as a Source Data file. GO molecular function (MF) analysis showed additional enrichment of GTP binding and GTPase activity in the yellow module (Fig. 1C). GTP-binding proteins regulate a wide range of cellular processes, and elevated GTP levels have recently been linked to be associated with tumor progression, chemotherapy resistance, and ICB resistance20–22. Gene set enrichment analysis (GSEA) of the PRJEB23709 cohort demonstrated that GTPase activity and GTP binding were significantly positive enriched in ICB responders (Fig. 1D). Based on this, all yellow module genes were intersected with known GTP-binding proteins, yielding 18 candidates (Fig. 1E). Univariate Cox analysis identified guanylate binding protein 2 (GBP2) as the gene most significantly associated with overall survival (OS) in melanoma patients (Fig. 1F). This finding was validated in GSE8401, GSE59455, and GSE54467 cohorts, with consistent results. (Fig. S2A–C). In the GSE65904 cohort, patients with low GBP2 expression exhibited shorter distant metastasis free survival (DMFS) and disease-specific survival (DSS) compared to those with high GBP2 expression (Fig. S2D). We next examined the relationship between GBP2 expression and clinical features. In the GSE3189 dataset, GBP2 expression was significantly higher in normal tissues than in melanoma tissues (Fig. S2E). Spatial transcriptomic analysis of a melanoma specimen, segmented into malignant, boundary and non-malignant regions and annotated with 10 cell types, revealed that GBP2 was predominantly expressed in normal epithelial cells, with markedly lower levels in malignant regions (Fig. 1G–I). Analysis of major pathological prognostic factors demonstrated that higher GBP2 expression was associated with favorable clinical features, including reduced Breslow thickness, lower Clark level and ulceration—all indicators of better prognosis (Fig. S2F)23,24. To explore the potential biological functions of GBP2 in melanoma, we conducted GSEA and GO analyses. Across six independent cohorts, patients with high GBP2 expression showed significant enrichment in IFN-γ response, regulation of immunity and cell killing (Figs. S3 and S4). Together, these bioinformatic findings in clinical samples identify GBP2 as a potential prognostic biomarker and reveal its close association with CTL cytolytic activity in melanoma. GBP2 deletion inhibited the CTL-induced ferroptosis to blunt the efficacy of ICB We next investigated the relationship between GBP2 and ICB responsiveness in melanoma. GBP2 knockout (GBP2−/−) lines were generated using CRISPR/Cas9 in B16-ovalbumin (OVA). GBP2 overexpressing (GBP2oe) lines were established via lentiviral transduction in both B16-ovalbumin (OVA) and A375 cells (Fig. S5A, B and Supplementary Table 1). To investigate the role of GBP2 in antitumor immunity, we first assessed its function in immunodeficient NOD. Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice. GBP2 knockout exhibited growth kinetics comparable to control tumors, indicating that GBP2 does not influence tumor progression in the absence of adaptive immunity (Fig. S5C). This finding suggests that GBP2 may exert its antitumor function through modulating adaptive immune responses, particularly CD8⁺ T cell–mediated cytotoxicity. To assess the contribution of GBP2 to T cell–mediated tumor control, OVA‑specific CD8+ (OT‑1) T cells were adoptively transferred into mice bearing either control or GBP2−/− B16‑OVA tumors. OT‑1 transfer effectively suppressed the growth of control tumor growth; however, this effect was markedly diminished in GBP2−/− tumors (Fig. 2A, B). These findings suggest that GBP2 is essential for optimal CD8+ T‑cell–mediated tumor rejection, prompting evaluation of its role in ICB efficacy. Knockout of GBP2 blunted the efficacy of ICB and inhibited the CTL-induced ferroptosis. A Tumor growth in mice subcutaneously injected with sgCtrl or GBP2−/− B16-OVA cells and treated with OT‑1 T cells or PBS every 7 days. B Tumor inhibition rate of GBP2−/− relative to sgCtrl tumors under OT-1 treatment (n = 6). C Tumor growth in mice bearing sgCtrl or GBP2−/−B16-OVA tumors treated with anti–PD‑L1 (200 μg/mouse) or IgG every 3 days. D Tumor inhibition rate of GBP2−/− versus sgCtrl tumors following anti–PD-L1 treatment (n = 6). E Control or GBP2-overexpressing tumor cells were co-cultured with OT-1 cells (E:T = 1:1, 24 h) with or without emricasan (20 μM), ferrostatin-1 (Fer-1, 10 μM), necrostatin-1 (Nec-1, 20 μM), or disulfiram (20 μM) (n = 3). F GSEA of ferroptosis between high- and low-GBP2 patients according to the SKCM cohort. Lipid ROS in B16-OVA tumors treated with OT-1 (G) or anti–PD-L1 (H), quantified in CD45− cells by flow cytometry. Representative lipid ROS images (I) and viability (J) of Vec or GBP2oe B16-OVA cells co-cultured with OT-1 (E:T = 5:1, 24 h), followed by Fer-1 (10 μM) for 24 h. K IFN-γ levels in tumor lysates from anti–PD-L1 or IgG-treated mice measured by ELISA. L Viability of Vec or GBP2oe cells after co-culture (24 h, E:T = 1:1) with OT-1 T cells transduced with sgCtrl, sgIFNγ, sgTNFα, or sgPRF1 (n = 3). M Vec or GBP2oe B16-OVA cells were subcutaneously implanted into C57BL/6 mice. When tumors reached ~125 mm³, both Vec and GBP2oe groups were respectively treated with either anti-IFN-γ or control IgG antibodies every 3 days. Tumor-bearing mice from both groups received anti-PD-L1 therapy every 3 days. All p value < 0.05 as statistic difference. n = 3 independent experiments (E, J, L); n = 6 independent experiments (A–D, G–I, K, M). The data are represented as mean ± SD. p value by unpaired two-tailed t-test (B, D, G, H) or one-way ANOVA (I–K) or two-way ANOVA (A, C, E, L, M) or two-sided permutation test (F). Source data are provided as a Source Data file. To this end, mice bearing GBP2−/− or sgCtrl B16-OVA tumors were treated with anti-PD-L1 blockade. GBP2 deficiency significantly reduced the antitumor efficacy of anti–PD‑L1 therapy compared to sgCtrl tumors (Fig. 2C, D). Similar results were observed in GBP2‑knockout MC38 tumors, in which GBP2 loss attenuated the response to combined OT‑1 transfer and anti–PD‑L1 treatment (Fig. S5D–H). These data indicate that GBP2 sensitizes tumor cells to CD8+T cell-mediated cytotoxicity, thereby enhancing ICB efficacy in vivo. Considering that CD8+ cytotoxic T lymphocytes (CTLs) eliminate tumor cells through multiple cell death pathways, we next sought to identify which CTL‑dependent cell death pathways are influenced by GBP2. OT-1 cells were co-cultured with GBP2oe B16-OVA cells in the presence of pathway‑specific inhibitors. Notably, treatment with the ferroptosis inhibitor ferrostatin‑1 (Fer‑1) significantly attenuated OT‑1-induced cytotoxicity in GBP2oe tumor cells, whereas inhibitors of pan‑caspases (emricasan), necroptosis (necrostatin‑1), or pyroptosis (disulfiram) had no effect (Fig. 2E). Consistently, GSEA revealed a strong positive correlation between GBP2 expression and ferroptosis signatures (Figs. 2F and S6A–E). These findings indicate that GBP2 facilitates CTL-mediated ferroptosis in tumor cells. To further confirm this association, we measured lipid ROS—a hallmark of ferroptosis—in B16‑OVA tumors. GBP2 knockout tumors from both OT‑1 transfer and anti-PD-L1 treatment groups exhibited reduced lipid ROS levels relative to sgCtrl tumors (Figs. 2G, H and S6F, G). In vitro, GBP2 overexpression enhanced OT-1–induced cytotoxicity and intracellular lipid ROS, effects that were reversed by Fer-1 (Fig. 2I, J). Together, these results demonstrate that GBP2 promotes CTL‑induced ferroptosis and thereby enhances the antitumor efficacy of ICB therapy. We next investigated the upstream signals through which GBP2 promotes ferroptosis. Given the known role of CD8⁺ T cell secreted cytokines in ICB-induced tumor cell death25, we measured IFN-γ and TNF-α levels in tumors with or without anti–PD-L1 treatment. Under basal conditions, GBP2 loss had no effect on cytokine levels. At day 7 of anti–PD-L1 therapy, both IFN-γ and TNF-α showed moderate increases in treated tumors, regardless of GBP2 status (Figs. 2K and S6H). Strikingly, in the late phase of treatment (Day 14), cytokine levels were markedly elevated in control tumors, whereas this increase was blunted in GBP2-deficient tumors, suggesting that GBP2 supports the sustained release of T cell–derived cytokines during late-phase ICB response (Figs. 2K and S6H). To dissect which cytokine mediates the GBP2-dependent cell death, we co-cultured GBP2-overexpressing tumor cells with OT-1 cells lacking IFN-γ, TNF-α, or perforin. Loss of IFN-γ—but not TNF-α or perforin, abrogated OT‑1-mediated killing of GBP2oe tumor cells (Fig. 2L). These results support previous findings that IFN-γ can trigger ferroptosis in tumor cells and suggest that GBP2 may act within this pathway6,26. Thus, we performed in vivo neutralization experiments. GBP2 overexpression enhanced both anti-PD‑L1 efficacy and lipid ROS accumulation, effects that were completely abolished by anti-IFN‑γ monoclonal antibody treatment, confirming the dependence on IFN-γ signaling (Figs. 2M and S6I). Together, these data suggest that GBP2 acts as a general amplifier of CD8+ T cell–mediated cytotoxicity rather than an ICB-specific effector. GBP2 deficiency attenuated OT-1–driven tumor control even in the absence of PD-L1 blockade, highlighting its fundamental role in IFN-γ–dependent ferroptotic killing of tumor cells. Thus, GBP2 represents a core mediator of T cell cytotoxicity that potentiates the therapeutic efficacy of immune checkpoint blockade. GBP2 promotes tumor ferroptosis through STAT1/SLC7A11 We next sought to elucidate the molecular mechanism by which GBP2 regulates IFN-γ-induced ferroptosis. Long-chain-fatty-acid-CoA ligase 4 (ACSL4) is a well-established determinant of ferroptosis27, however, GBP2 depletion did not significantly alter ACSL4 expression (Fig. S7A). The cystine/glutamate antiporter system Xc−, consisting of the light-chain subunit (SLC7A11) and the heavy-chain subunit (SLC3A2), mediates extracellular cystine uptake for antioxidant glutathione (GSH) synthesis, thereby limiting lipid peroxidation and suppressing ferroptosis28,29. To explore whether system Xc⁻ participates in GBP2-mediated ferroptosis regulation in melanoma cells, we examined the expression of its subunits. Compared with sgCtrl cells, sgGBP2 cells displayed a marked attenuation of IFN-γ-induced downregulation of SLC7A11 (Fig. 3A), whereas SLC3A2 expression remained unaffected by GBP2 deletion (Fig. S7B). In situ multi-color immunofluorescence staining of clinical melanoma samples revealed that tumors with high GBP2 expression exhibited lower SLC7A11 levels (Fig. 3B). To determine whether GBP2 regulates ferroptosis via SLC7A11, we silenced SLC7A11 expression in GBP2−/− cells using siRNA (Fig. S7C and Supplementary Table 2). Upon treatment with the ferroptosis inducer RSL3, SLC7A11 knockdown restored the lipid ROS levels that were reduced by GBP2 deficiency in IFN-γ–primed B16-OVA cells (Fig. 3C). Similarly, GBP2 knockout markedly increased intracellular GSH content and cell viability—effects that were reversed by SLC7A11 silencing (Fig. 3D). In vivo, SLC7A11 overexpression significantly blunted the tumor-suppressive effect of GBP2 overexpression, reduced intratumoral levels of lipid ROS, and abolished its ability to potentiate ICB efficacy (Figs. 3E, F and S7D). GBP2 promotes ferroptosis through STAT1/SLC7A11. A Western blot analysis of sgCtrl and GBP2−/− of B16-OVA and A375 cells were treated with 10 ng/mL IFN-γ for 12 h, respectively. B Representative images of GBP2 and SLC7A11 in situ immunofluorescent staining in human melanoma samples (n = 3). Scale bars, 50 μm. C, D The sgCtrl and GBP2−/− B16-OVA cells primed with IFN-γ (10 ng/ml) for 12 h and knockdown SLC7A11 expression using small interfering RNA (siRNA), then treated with 0.1 μM RSL3 for 12 h. Relative lipid ROS were quantified by flow cytometric analysis (C); intracellular GSH and cell viability were assessed by ELISA and ATP assays (D). E Control, oeGBP2, or oeGBP2 + oeSLC7A11 B16-OVA cells (1 × 106) were subcutaneously injected into C57BL/6 mice, followed by anti–PD-L1 or IgG treatment every 3 days after tumors reached 125 mm3. F Lipid ROS in CD45− tumor cells from mice in (E) were quantified by flow cytometry. G Western blot analysis of sgCtrl and GBP2−/− B16-OVA and A375 cells were treated with 10 ng/mL IFN-γ for 12 h, respectively. H Confocal imaging of p-STAT1 in sgCtrl and GBP2−/− B16-OVA cells treated with IFN-γ (10 ng/mL, 12 h). Nuclear fluorescence intensity was quantified using ImageJ. Scale bar, 25 μm. I GBP2oe or GBP2oe + STAT1−/− (1 × 106) cells subcutaneously injected into C57BL/6 mice. Mice were adoptively transferred with OT-1 cells (5 × 106 per mouse) once per 7 days. J Tumor inhibition rate for GBP2oe cells relative to GBP2oe + STAT1−/− cells under OT-1 treatment, respectively (n = 6). K Relative lipid ROS of GBP2oe or GBP2oe + STAT1−/− tumors in mice treated with OT-1 were quantified in CD45− cells by flow cytometric analysis. All p value < 0.05 as statistic difference. n = 3 independent experiments (A–D, G); n = 6 independent experiments (E, F, H–K). The data are represented as mean ± SD. p value by unpaired two-tailed t-test (H, J, K) or one-way ANOVA (C, D, F) or two-way ANOVA (E, I). Source data are provided as a Source Data file. We then investigated the upstream mechanism by which GBP2 represses SLC7A11. Previous studies have implicated the STAT1 signaling pathway in SLC7A11 transcriptional regulation30. We therefore assessed whether GBP2 modulates SLC7A11 expression through STAT1 activation in melanoma. As anticipated, GBP2 depletion markedly reduced phosphorylated STAT1 (p-STAT1) levels in B16-OVA and A375 cells (Fig. 3G). Immunofluorescence imaging and nuclear–cytoplasmic fractionation following IFN-γ stimulation further demonstrated impaired nuclear translocation of p-STAT1 in GBP2−/− cells (Fig. 3H). STAT1 is a central regulator of antitumor immunity, orchestrating key ICB response pathways, including MHC-I antigen presentation, checkpoint ligand expression, and cytotoxic lymphocyte (CD8+ T/NK cell) recognition25. To assess whether STAT1-dependent functions were altered by GBP2 deletion, we examined MHC-I levels in B16-OVA and MC38-OVA cells using the PresentER system, which constitutively co-overexpresses OVA and MHC-I components via peptide/MHC complexes31. As a result, these cells constitutively overexpress both OVA antigen and MHC-I molecules, eliminating the possibility that reduced immunotherapy efficacy in the GBP2-knockout context is due to altered MHC-I presentation. Flow cytometry confirmed comparable MHC-I expression between sgCtrl and sgGBP2 cells (Fig. S7E, F). We next assessed PD-L1 expression. Western blot analysis of sgGBP2 cells revealed no significant difference between sgCtrl and sgGBP2 cells under basal or IFN-γ–stimulated conditions (Fig. S7G). To evaluate potential changes in immune recognition, we employed FucoID technology, which quantifies cell-cell interactions via interaction-dependent fucosyl-biotinylation32. No significant difference in CD8+ T/NK cell–tumor cell interaction frequency was detected between sgCtrl and sgGBP2 groups (Fig. S7H). These findings support that while GBP2 modulates STAT1 activation, it does not affect MHC-I expression, PD-L1 levels, or direct lymphocyte–tumor cell interactions in these models. Based on these findings, we hypothesize that the STAT1/SLC7A11 axis mediates GBP2-driven ferroptosis. STAT1 has been reported to bind to the SLC7A11 promoter, repressing its transcription6. Chromatin immunoprecipitation (ChIP) qPCR demonstrated that GBP2 loss abrogated IFN-γ–induced STAT1 binding to the transcription start site of SLC7A11 in B16-OVA and A375 cells (Fig. S7I). More strikingly, STAT1 knockout in GBP2oe B16-OVA cells abolished the increase in lipid ROS, leading to accelerated tumor growth (Figs. 3I–K and S7J). To exclude off-target effects, GBP2 re-expression in GBP2−/− B16-OVA cells significantly rescued SLC7A11 expression and STAT1 phosphorylation (Fig. S7K). Furthermore, enforced GBP2 expression of GBP2 in GBP2-deficient tumor cells restored lipid ROS accumulation and reinstated sensitivity to anti-PD-L1 in vivo (Fig. S7L, M). Together, these results demonstrate that GBP2 promotes ferroptosis by regulating the STAT1/SLC7A11 axis. GBP2 undergoes phase separation to form condensates Next, we investigated how GBP2 regulates the STAT1–SLC7A11 axis to promote ferroptosis. Recent studies have implicated phase separation as a critical mechanism triggering ferroptosis, with IFN-γ shown to induce phase separation of key proteins at high concentrations12,13,33,34. Given that GBP2 is an interferon-stimulated gene and that our results demonstrated GBP2 sensitizes tumor cells to IFN-γ-induced ferroptosis, we hypothesized that GBP2 might form biomolecular condensates via phase separation. To test this, we first analyzed the amino acid sequence of the GBP2 protein using the Predictor of Natural Disordered Regions (PONDR)35, which identified residues 436–579 (GBP2436–579) as an intrinsically disordered region (IDR) (Fig. 4A). To visualize of GBP2 condensates, we generated B16-OVA and A375 cells with a mCherry tag knocked in at the C terminus of endogenous GBP2. Upon IFN-γ stimulation, endogenous GBP2 formed prominent droplet-like condensates that exhibited fusion and aggregation over time (Figs. 4B, C and S8A, B). We also assessed the dynamic properties of these condensates using fluorescence recovery after photobleaching (FRAP), observing >60% fluorescence recovery within approximately 2 min, indicative of liquid-like mobility (Figs. 4D and S8C). To validate our FRAP methodology, glycogen condensates served as a positive control and mCherry alone as a negative control36,37. Glycogen displayed rapid fluorescence recovery, consistent with known liquid-like dynamics (Fig. S8D), whereas mCherry showed negligible recovery, confirming its non-condensing behavior (Fig. S8E). These controls substantiate the dynamic and fluid nature of GBP2 condensates. GBP2 forms biological condensates. A Amino acids 436–579 of GBP2 was predicted as an intrinsically disordered region (IDR) using a web-based software PLAAC. B Representative images from live GBP2-mCherry B16-OVA cells stimulated with or without 10 ng/ml IFN-γ for 6 h. Scale bar: 20 μm. C Representative images from live cell images of B16-OVA cells expressing GBP2-mCherry stimulated with 10 ng/ml IFN-γ for 6 h. Scale bar: 10 μm. The arrows point to the representative condensates formed by GBP2 that fused over time. D Representative images showed that FRAP assay of the droplets formed in B16-OVA expressing GBP2-mCherry stimulated with 10 ng/ml IFN-γ for 6 h. The data are plotted as the mean ± SD (n = 6). E Representative images showed that GBP2-mCherry B16-OVA cells were treated with or without 5% 1,6-hexanediol (1,6-HD) for 5 min stimulated with 10 ng/ml IFN-γ for 6 h. F Representative images from B16-OVA cells expressing GBP2-mCherry and GBP2-mCherry IDR deletion (△IDR) stimulated with 10 ng/ml IFN-γ for 6 h. n = 3 independent experiments (B, C, E, F); n = 6 independent experiments (D). The data are represented as mean ± SD. Source data are provided as a Source Data file. Moreover, treatment with 1,6-hexanediol (1,6-HD), which disrupts liquid–liquid phase separation by interfering with hydrophobic interactions, effectively abolished GBP2 puncta formation (Figs. 4E and S9A). To define the role of the IDR, we generated a deletion mutant lacking residues 436–579 (ΔIDR) in the mCherry-GBP2 construct and expressed it in B16-OVA and A375 cells. Considering the potential destabilization effect of removing this segment, recombinant full-length (FL) and ΔIDR GBP2 proteins were purified from E. coli, with Coomassie staining confirming proper folding of the ΔIDR mutant (Fig. S9B). In GBP2−/− B16-OVA cells transfected with FL or ΔIDR constructs, treatments with proteasomal (MG132), autophagic (BafA1), or lysosomal (chloroquine) inhibitors did not significantly affect GBP2 expression levels (Fig. S9C). Importantly, deletion of GBP2436–579 completely abolished condensate formation (Figs. 4F and S9D). Collectively, these data demonstrate that GBP2 forms the dynamic and reversible phase-separated condensates through its intrinsically disordered region, a property likely integral to its function in regulating ferroptosis. GBP2 condensates retains SHP1 to protect STAT1 activation to promote ferroptosis Previous studies have reported that the biomolecular condensates formed via phase separation can spatially restrict protein functions, including nuclear regulation or protein–proteins interactions38. To further elucidate whether GBP2 phase separation facilitates ferroptosis, we performed mass spectrometry to identify GBP2-interacting proteins. As a result, SHP1, a known phosphatase involved in dephosphorylating STAT1 in the IFN-γ signaling pathway39, was detected in GBP2-immunoprecipitates (Fig. 5A). This interaction was validated at both endogenous and exogenous levels by co-immunoprecipitation assays (Figs. 5B and S9E). Functionally, SHP1 knockdown in GBP2−/− cells substantially restored STAT1 phosphorylation and further suppressed SLC7A11 expression (Figs. 5C, S9F and Supplementary Table 2). We hypothesized that GBP2 modulates SHP1 activity through phase separation. Confocal microscopy revealed that GBP2 and SHP1 co-localize within droplet-like condensates in IFN-γ–treated B16-OVA cells, supporting a spatial sequestration model (Fig. 5D). GBP2 sequesters SHP1 via IDR-mediated phase separation to augment ferroptosis. A B16-OVA cells treated with IFN-γ (10 ng/mL, 12 h) were immunoprecipitated with anti-GBP2 for mass spectrometry (MS); SHP1 was selected. B IP assay shows endogenous GBP2 interact with SHP1 in B16-OVA and A375 cells. C sgCtrl and GBP2−/− of B16-OVA and A375 cells were treated with IFN-γ (10 ng/mL, 12 h), followed by SHP1 knockdown using siRNA for 24 h; cell lysates were analyzed by western blot. D Representative live-cell imaging of B16-OVA cells expressing GBP2-mCherry stimulated with 10 ng/ml IFN-γ for 6 h, and transfected with SHP1-mEGFP plasmid for 36 h. The line plots show GBP2-mCherry and SHP1-mEGFP colocalization of areas. E Western blot analysis of B16-OVA GBP2−/− and A375 GBP2−/− cells treated with 10 ng/ml IFN-γ for 12 h, followed by transfection with vector, GBP2-FL, or GBP2 ΔIDR plasmids for 24 h. F, G Representative live-cell imaging and high-resolution confocal demonstrated colocalization of GBP2-mCherry with SHP1-mEGFP, but not GBP2ΔIDR, upon IFN-γ stimulation. H In vitro kinase assays revealed enhanced STAT1 phosphorylation in the presence of GBP2 condensates. I, J In vitro phase separation assays compared condensate formation between GBP2-WT, GBP2ΔIDR, and GBP2-MBP, confirming IDR-dependent condensate activity and STAT1 phosphorylation enhancement. K GBP2−/− B16-OVA cells expressing GBP2-FL or GBP2ΔIDR (1 × 106) were subcutaneously injected into C57BL/6 mice and treated with anti–PD-L1 antibody (200 μg/mouse) or IgG every 3 days. L Lipid ROS in CD45⁻ tumor cells were quantified by flow cytometry. M Tumors from GBP2-FL and GBP2ΔIDR groups were immunoprecipitated with anti-SHP1 and analyzed by western blot. N Representative immunofluorescence images of GBP2 in GBP2-FL and GBP2ΔIDR tumors treated with anti–PD-L1 (n = 3). All p value < 0.05 as statistic difference. n = 3 independent experiments (B–J, M, N); n = 6 independent experiments (K, L). The data are represented as mean ± SD. p value by unpaired two-tailed t-test (L) or two-way ANOVA (K). Source data are provided as a Source Data file. To investigate the functional relevance of GBP2-SHP1 condensates, we compared full-length GBP2 (FL) with the IDR deletion mutant in GBP2−/− cells. The ΔIDR mutant exhibited diminished IFN-γ–induced phase-separated compartments, accompanied by reduced STAT1 tyrosine phosphorylation and impaired nuclear translocation (Figs. 5E, F and S9G). Correspondingly, IDR deletion led to decreased lipid ROS levels and elevated intracellular GSH upon RSL3 treatment (Fig. S9H, I). Co-immunoprecipitation experiments further demonstrated direct SHP1–STAT1 binding, which was significantly enhanced in GBP2 knockout cells (Fig. S9J). High-resolution imaging confirmed endogenous co-localization of GBP2 and SHP1 condensates, indicating that GBP2 physically sequesters SHP1 under physiological conditions (Fig. 5G). These findings support our central hypothesis that GBP2 condensates function as spatial regulators by enriching and isolating SHP1. To dissect this mechanism in vitro, we purified recombinant GBP2, SHP1, JAK2, and STAT1 proteins from E. coli, 293 T, and Sf9 systems. Pre-formed GBP2 condensates were incubated with SHP1 before assessing STAT1 phosphorylation. Without GBP2 condensates, p-STAT1 levels rapidly decreased due to SHP1-mediated dephosphorylation. In contrast, the presence of GBP2 condensates preserved STAT1 phosphorylation, indicating effective SHP1 sequestration and impaired phosphatase activity (Fig. 5H). Similar results were observed with the ΔIDR mutant, which failed to form condensates and hence could not sequester SHP1 (Fig. 5I). Furthermore, fusion of GBP2 with a hydrophilic maltose-binding protein (MBP) tag to reduce self-association disrupted condensate formation and SHP1 sequestration in vitro, resulting in decreased STAT1 phosphorylation compared to wild-type GBP2 (Fig. 5J). Collectively, these results demonstrate that GBP2 phase separation is essential for sequestering SHP1 and sustaining STAT1 signaling, robustly supporting our claims regarding the functional role of phase separation. In vivo, introducing GBP2436-579 significantly impaired GBP2-FL-mediated lipid ROS accumulation, accelerated tumor growth, and attenuated anti–PD-L1 therapy efficacy (Fig. 5K, L). Comparable results were observed in OT-1 adoptive cell therapy models (Fig. S10K–M). Co-immunoprecipitation from PD-L1-treated tumors revealed enhanced STAT1-SHP1 binding and decreased STAT1 phosphorylation in GBP2-ΔIDR tumors (Fig. 5M). Furthermore, in vivo phase separation could observe aggregation foci under immunofluorescence36. We performed immunofluorescence staining on these tumor specimens and visualized distinct GBP2 aggregate foci in GBP2-FL tumors following anti-PD-L1 treatment, whereas such structures were absent in GBP2-ΔIDR tumors (Fig. 5N). These findings convincingly demonstrate that CTL-secreted IFN-γ promotes GBP2 phase separation, facilitating SHP1 sequestration, which sustains STAT1 activation, represses SLC7A11 transcription, and induces ferroptosis. GBP2 promotes the intratumoral CD8+T cell infiltration through enhancing the immunogenicity of ferroptotic cells Next, we aimed to investigate the impact of GBP2 on the tumor immune microenvironment (TME) under ICB therapy. We first analyzed the correlation between GBP2 and TME in melanoma patients using the ssGSEA algorithm to estimate the abundance of 28 immune cell types across six cohorts. GBP2 expression was strongly associated with an inflamed immune microenvironment, showing a significant positive correlation primarily with CD8⁺T cells (Fig. S10A). To validate these findings experimentally, we performed flow cytometry on tumor-infiltrating leukocytes from anti‑PD‑L1–treated sgGBP2 and sgCtrl tumors. Applying the established gating strategies40, we quantified the major immune subsets, including CD8⁺ T cells, CD4⁺ T cells, NK cells, B cells, neutrophils, monocytes/macrophages, and dendritic cells (DCs) (Fig. S10B). Anti-PD-L1 treatment significantly increased the proportions of CD8⁺ T cells, CD4⁺ T cells, NK cells and B cells. Critically, GBP2 deletion substantially reduced infiltration of CD8⁺ T cells regardless of PD-L1 treatment, while CD4⁺ T cells, DCs, B cells, NK cells, macrophages, and neutrophils were unaffected. (Fig. 6A). CD4⁺ T cells, and dendritic cells Similarly, flow cytometry revealed that CD8+OT-1 cell was markedly lower in GBP2-/- tumors compared to controls (Fig. 6B, C). Given the pivotal role of CD8+T cells in antitumor immunity, we depleted CD8+T cells in vivo and observed that tumor growth in the oeGBP2 group was increased (Fig. 6D), underscoring the dependence of GBP2-mediated tumor suppression on CD8⁺ T cells. Furthermore, flow cytometric analysis of CD8⁺ T cell activation markers showed that GBP2 ablation markedly attenuated IFN-γ⁺ and TNF-α⁺ CD8⁺ T cell frequencies (Fig. 6E). Analysis of a single-cell RNA sequencing data (GSE115978) from 32 melanoma patients using t-distributed stochastic neighbor embedding (t-SNE) identified 25 cell clusters into 8 major immune cell types based on canonical marker expression (Fig. S10C). Notably, the five patients with highest GBP2 expression exhibited significantly increased CD8⁺ T cell infiltration (Figs. 6F and S10D). Immunofluorescence staining of further corroborated GBP2-FL tumors following anti-PD-L1 harbored greater CD8⁺ T cell infiltration (Fig. 6G). Collectively, these data demonstrate that GBP2 shapes an immunostimulatory TME, potentiating CD8+T cell-mediated antitumor immunity. GBP2 promotes the intratumoral CD8+T cell infiltration through enhancing the immunogenicity of ferroptotic cells. A The relative percentage of tumor infiltrating immune subsets between sgCtrl and GBP2−/− tumors with or without anti-PD-L1 in B16-OVA-bearing mice. n = 3 mice. B Representative dot blot of flow cytometric analysis of CD45+CD8+ OT-1 cells. C The relative quantitative percentage of CD45+CD8+OT-1 cells between Vec and GBP2 knockout tumors were shown. D 1 × 106 vector or GBP2oe B16-OVA cells were subcutaneously injected into C57BL/6 mice. Mice with a tumor size of 125 mm3 were treated with anti-CD8α antibody (200 μg per mouse) or IgG once per 3 days. E Representative images and statistical quantitation of the flow cytometric analysis of the percentage of IFN-γ+CD8+ and TNF- α+CD8+ TILs from sgCtrl and sgGBP2 xenografts. F t-Distributed stochastic neighbor embedding (t-SNE) plot from five lowest GBP2 expression group (left) and highest GBP2 expression melanoma patients (right). G Representative immunofluorescent staining images of GBP2 and CD8A in GBP2-FL or GBP2 ΔIDR tumors under anti-PD-L1 treatment (n = 3). Scale bars, 50 μm. All p value < 0.05 as statistic difference. n = 3 independent experiments (A, E, G); n = 6 independent experiments (C, D). The data are represented as mean ± SD. p value by unpaired two-tailed t-test (C, E) or two-way ANOVA (A–D). Source data are provided as a Source Data file. Recent studies have revealed that ferroptotic tumor cells release damage associated molecular patterns (DAMPs), conferring an immunogenic cell death (ICD)-like phenotype41. Importantly, ICD promotes the recruitment and activation of DCs, which facilitate tumor antigen presentation to CTL, enhancing CD8+T cell infiltration and activation42. Since GBP2 significantly potentiates CTL-induced ferroptosis, we hypothesized that GBP2 promotes CTL-mediated killing by triggering the ICD feature in ferroptotic tumor cells. To test this, we measured levels of the DAMP high-mobility group box 1 (HMGB1) in tumor tissues from mouse models. ELISA revealed that GBP2 deficiency significantly reduced HMGB1 release under both anti-PD-L1 and OT-1 therapies (Fig. S11A). Concordantly, GBP2−/− tumors exhibited decreased expression of DC maturation markers, including the co-stimulatory molecules CD80 and CD86 (Fig. S11B). Analysis of single-cell data also showed elevated CD86 expression in patients with high GBP2 expression (Fig. S11C). Deletion of STAT1 in GBP2-overexpressing tumors significantly reduced HMGB1 levels, while SLC7A11 overexpression rescued the GBP2-induced HMGB1 upregulation (Fig. S11D, E). Furthermore, HMGB1 knockout in GBP2-overexpressing B16-OVA cells accelerated tumor growth in vivo, accompanied by diminished CD8⁺ OT-1 T cell infiltration and reduced CD80/CD86 expression (Fig. S11F–I). Together, these results suggest that GBP2 enhances the immunogenicity of ferroptotic tumor cells through HMGB1 release, thereby promoting CD8⁺ T cell infiltration in melanoma. GBP2 facilitates ICB response in tumor patients Next, we sought to validate the correlation between GBP2 expression and ICB response in human melanoma. We compared three established predictive signatures for ICB response: T cell inflamed gene expression profile (GEP), innate anti-PD-1 resistance (IPRES) and immuno-predictive score (IMPRES) signatures43–45. Higher enrichment scores of all three signatures were observed in the high GBP2 expression group across multiple melanoma cohorts (Figs. 7A and S12A–D). Using SubMap analysis to compare GBP2 expression profiles with a published melanoma immunotherapy dataset46, the high GBP2 group demonstrated significant similarity to the PD-1 responder group across five cohorts (Figs. 7B and S12E–H). Additionally, we analyzed a cohort comprising pre-treatment (PRE) biopsies taken prior to immunotherapy (n = 73), and early during treatment (EDT) samples (7–14 days post-treatment initiation, n = 18) (Supplementary Dataset 2). PRE samples with high GBP2 expression exhibited markedly improved clinical responses to ICB compared to low GBP2 samples. Conversely, patients with high SLC7A11 expression showed increased ICB resistance (Figs. 7C and S13A). More importantly, stratifying PRE samples by GBP2 and SLC7A11 expression yielded three groups: group 1 (high GBP2/low SLC7A11), group 2 (low GBP2/low SLC7A11 or high GBP2/high SLC7A11), and group 3 (low GBP2/high SLC7A11). Group 1 displayed the highest percentage of complete or partial responses (CR/PR), while group 3 had the lowest (Fig. 7D). Due to limited EDT samples, direct comparison revealed elevated GBP2 expression in ICB responders versus non-responders (Fig. S13B). Survival analyses indicated that among ICB-treated patients, the high GBP2 group had superior prognosis compared to the low GBP2 group, whereas high SLC7A11 predicted poorer outcomes (Figs. 7E and S13C). Moreover, group 1 exhibited the most favorable survival outcome, followed by group 2, with group 3 demonstrating the shortest OS (Fig. 7F). Similar patterns of elevated GBP2 expression in ICB responders were observed in metastatic urothelial carcinoma and non-small cell lung cancer cohorts47,48, where high GBP2 expression also predicted better prognosis (Fig. S13D, E). GBP2 acts as an immunotherapeutic target in patients with cancer. A Box plots showing expression of ICB response-related signatures between high and low GBP2 group in the SKCM cohort (n = 470). IMPRES: Minimum (high vs. low: 0.064 vs. 0.001); Median (high vs. low: 0.668 vs. 0.332); Maximum (high vs. low: 1.0 vs. 0.751). IPRES: Minimum (high vs. low: 0.0 vs. 0.0); Median (high vs. low: 0.57 vs. 0.419); Maximum (high vs. low: 1.0 vs. 1.0). T cell-inflam GEP: Minimum (high vs. low: 0.047 vs. 0.0); Median (high vs. low: 0.707 vs. 0.316); Maximum (high vs. low: 0.941 vs. 0.732). B Submap analysis demonstrates high GBP2 group in SKCM samples closely align with the PD1-response (PD1-R) group defined in the melanoma cohort. C Treatment response rates to anti-PD-1 or combination immunotherapy in pre-therapy (PRE) patients stratified in high or low GBP2 expression groups. SD stable disease, PD progressive disease, CR complete response, PR partial response. D PRE patients were stratified into three groups based on SLC7A11 and GBP2 expression, with corresponding immunotherapy response rates shown. Kaplan–Meier survival analysis for GBP2-high vs. -low groups (E), and for the three SLC7A11/GBP2-based groups (F). G Spatial transcriptomics of two hepatocellular carcinoma sections (one responder, one non-responder) identified morphology-aligned clusters, annotated as malignant, boundary, or non-malignant regions. H The spatial feature plot that visualizes the proportion of the CD8+T cells and spatial expression patterns of GBP2 and SLC7A11 in spatial spots (n = 2). I The tissue sections from 12 patients with melanoma were immunohistochemically stained with anti-GBP2 or anti-SLC7A11 antibody. The correlation between the expression of GBP2 and SLC7A11 is shown (right). Scale bars, 200 μm. All p value < 0.05 as statistic difference. p value by unpaired two-tailed Wilcoxon test (A) or two-sided permutation test (B) or Pearson’s chi-square test (D) or Log-rank test (E, F) or the correlation coefficient and two-tailed p value were calculated using Pearson’s correlation analysis (I). Source data are provided as a Source Data file. To explore spatial heterogeneity, we analyzed spatial transcriptomics data from a hepatocellular carcinoma (HCC) dataset, dividing tissue slices into malignant, boundary, and non-malignant regions (Figs. 7G and S13F). Responders exhibited higher GBP2 and lower SLC7A11 expression in malignant areas, accompanied by increased CD8⁺ T cell infiltration, compared to non-responders (Figs. 7H and S13G). Finally, immunohistochemistry (IHC) of melanoma samples validated that high GBP2 expression correlated with reduced SLC7A11 levels (Fig. 7I). Together, these findings indicate that elevated GBP2 predicts improved OS and better ICB response in tumor patients. Discussion Despite the remarkable clinical efficacy of current cancer immunotherapies, the majority of patients either fail to respond or eventually develop resistance to ICB. It is therefore critical to elucidate the mechanisms underlying immune resistance within the tumor microenvironment to identify novel therapeutic targets and devise more effective immunotherapies49. In this work, we identified GBP2 as a pivotal mediator that promotes immunogenic ferroptosis driven by ICB-activated CTLs in melanoma. Notably, we discovered that IFN-γ-induced phase separation of GBP2 sequesters SHP1, thereby enhancing STAT1 phosphorylation, which suppresses SLC7A11 expression, triggers ferroptosis, and improves the efficacy of anti–PD-L1 treatment (Fig. S114). This mechanism reveals an unrecognized CTL-mediated link between GBP2 and phase separation, unveils a regulatory axis of immunogenic ferroptosis and suggests GBP2 as a combinatorial biomarker for melanoma immunotherapy. To identify genes associated with CTL cytolytic activity, we performed WGCNA. Intriguingly, we found that the module genes of interest were strongly associated with GTP binding and GTPase activity. Enhanced synthesis and utilization of nucleotide triphosphates (NTPs) represent a critical and universal metabolic dependency in cancer cells across diverse cancer types and genetic backgrounds. Previous studies have demonstrated that elevated nucleotide metabolism promotes uncontrolled proliferation, chemotherapy resistance, immune evasion, and metastasis of tumor cells50. For instance, Huang et al. reported significantly increased purine synthesis in chemotherapy resistant small-cell lung cancer cells, with GTP showing the most pronounced elevation21. Our GSEA further revealed a positive correlation between ICB-responsive melanoma patients and gene signatures involved in GTP binding and GTPase activity. Based on these findings and intersection with GTP-binding proteins, GBP2 was selected as a gene of interest. Emerging evidence supports a broader role of the GBP family involvement in modulating immunotherapeutic responses across cancer. Immunohistochemical confirmed elevated GBP6 expression in ICB-responsive small cell lung cancer patients51. Pancreatic cancer organoids with high GBP4 expression exhibited increased sensitivity to PD-1 blockade, correlating with enhanced infiltration of IFN-γ⁺CD8⁺ T cell and tumor cell death52. Li et al. further demonstrated that the IFN-γ-inducible lncRNA LIMIT cis-activates GBP clusters, thereby amplifying tumor immunogenicity and potentiating checkpoint therapy efficacy53. Collectively, these findings underscore the GBP family as critical immunotherapeutic regulators. Recent studies revealed that GBP2’s involvement in tumor progression, albeit with context-dependent roles54–56. Our data confirmed that elevated GBP2 expression predicts favorable prognosis across multiple melanoma cohorts and enhances the efficacy of OT-1 transfusion and anti-PD-L1 therapy. Further bioinformatics analysis revealed that compared to patients who did not respond to ICB, GBP2 expression was elevated in patients who responded to ICB. These data suggest that GBP2 as a promising immunotherapeutic target in melanoma, though its underlying mechanisms largely unexplored. It is well established that CTLs eliminate target cells primarily through perforin–granzyme–caspase-mediated apoptosis. However, emerging evidence indicates that ferroptosis—a distinct form of regulated cell death characterized by iron-dependent lipid peroxidation—also plays a critical role in suppressing tumor progression and modulating the efficacy of ICB. Notably, IFN-γ signaling, enhanced by ICB-activated CTLs, has been shown to induce and amplify tumor ferroptosis, thereby contributing to effective tumor regression6,26,49. Therefore, triggering immunogenic ferroptosis may represent a promising therapeutic strategy, particularly for tumors resistant to apoptosis or necroptosis. As GBP2 is an interferon-stimulated gene, we hypothesized and subsequently confirmed that GBP2 enhances the sensitivity of melanoma cells to ICB by promoting ferroptosis. Furthermore, our findings provide compelling evidence that GBP2 acts as a crucial regulator of IFN-γ–induced ferroptosis in melanoma via the STAT1/SLC7A11 axis. Mechanistically, we demonstrate that GBP2 facilitates STAT1 activation and nuclear translocation upon IFN-γ stimulation. This enhanced p-STAT1 activity is essential for suppressing SLC7A11 transcription and driving ferroptotic cell death. Notably, in GBP2-deficient cells, the transcriptional repression of SLC7A11 by STAT1 is impaired, leading to ferroptosis resistance, increased GSH levels, and decreased lipid ROS. Importantly, our in vivo data reveal that the tumor-suppressive and immunotherapeutic benefits of GBP2 are largely dependent on this ferroptotic pathway. Restoration of GBP2 expression re-sensitized tumors to anti–PD-L1 therapy, whereas SLC7A11 overexpression or STAT1 knockout nullified these effects. These observations position GBP2 as an immune-regulated ferroptosis facilitator and underscore the therapeutic potential of targeting the GBP2–STAT1–SLC7A11 axis to overcome resistance to ICB. In immunology, phase separation plays a critical role in forming immune signaling condensations and regulating signal transduction downstream of T cell receptor and B cell receptor10. However, phase separation is concentration- and condition- dependent, and only specific molecules can undergo this process, often driven by intrinsically disordered regions (IDRs)9. Both type I and type II interferons, when present at high concentrations, have been reported to induce phase separation events that modulate immune responses, albeit often leading to distinct outcomes. While previous studies have predominantly explored phase separation associated with type I IFN signaling components57, our findings extend this paradigm by revealing a phase separation mechanism linked to type II IFN signaling. Specifically, we identified an IDR within GBP2—a known ISG—and demonstrated that IFN-γ stimulation induces GBP2 condensate formation. Given that GBP2 expression can be upregulated in response to IFN-γ, we hypothesized and subsequently confirmed that GBP2 regulates the STAT1-SLC7A11 signaling axis through phase separation-mediated sequestration of SHP1. The GBP2-SHP1 condensates formed in an IDR-dependent manner (GBP2436–579), and disruption of this region markedly impaired ferroptosis induction and weakened the efficacy of ICB. These findings highlight a previously unappreciated layer of regulation wherein IFN-γ–induced GBP2 phase separation modulates downstream signaling pathways critical for ferroptosis and antitumor immunity. Moreover, our data contribute to the emerging concept that phase separation can exert both positive and negative regulatory effects on ICB responsiveness, depending on the molecular context and specific signaling cascades involved. This work underscores the complexity of IFN-mediated immune regulation and suggests that targeting phase separation dynamics of ISGs like GBP2 may offer therapeutic avenues to enhance immunotherapy outcomes. Collectively, these data expand our understanding of the interplay between IFN-γ signaling, ferroptosis, and immunotherapy, offering a promising strategy for enhancing anti-tumor immunity in melanoma. In addition to generating abundant lipid ROS, ferroptotic tumor cells can also release tumor-associated antigens and DAMPs to represent an immunogenic cell death (ICD)-like feature, which promote DC recruitment and activation, ultimately enhancing CD8⁺ T cell infiltration and cytotoxic activity42. HMGB1, a prototypical DAMP, is known to translocate and be released in response to elevated ROS levels. During ferroptosis, HMGB1 is frequently secreted and acts as an immunogenic signal, contributing to the activation and recruitment of CD8⁺ T cells58. This forms a positive feedback loop that amplifies antitumor immunity. Our findings reveal that GBP2 not only triggers ferroptosis but also promotes HMGB1 expression and release. Importantly, the STAT1/SLC7A11 axis not only regulates ferroptotic sensitivity but also modulates HMGB1 dynamics, positioning this pathway at the convergence of regulated cell death and immune activation. The loss of therapeutic efficacy in CD8⁺ T cell–depleted or HMGB1-deficient models further underscores the dependence of GBP2-mediated antitumor immunity on immunogenic ferroptosis. Taken together, our results highlight a previously unrecognized role for GBP2 in orchestrating an immunostimulatory tumor microenvironment through ferroptosis-induced immunogenicity. Given that IFN-γ is predominantly secreted by activated effector T cells, enhanced T cell infiltration and elevated IFN-γ signaling could further reinforce ferroptotic responses in tumors. Thus, leveraging GBP2 to trigger immunogenic ferroptosis represents a compelling therapeutic strategy, particularly in tumors resistant to apoptosis or necroptosis. The current study has limitations, including not fully elucidating the intrinsic mechanisms by which GBP2 promotes adaptive immunity. Notably, we did not investigate the regulatory mechanisms governing GBP2 expression. Emerging evidence implicates epigenetic dysregulation as a conserved mechanism controlling GBP family expression in cancer52,59. Nonetheless, our current focus was to dissect GBP2’s regulatory mechanism in shaping ICB response. Future studies will aim to rigorously explore GBP2’s epigenetic regulation. Additionally, we identified GBP2’s dual role in enhancing T cell cytotoxicity and improving ICB efficacy; further research into GBP2’s modulation of T cell function may extend its therapeutic application to CAR-T cell therapies. Moreover, additional preclinical and clinical trials are warranted to validate GBP2 as a feasible immunotherapeutic target. Given that most patients do not respond to ICB, our current work raises the compelling possibility of augmenting ICB efficacy via GBP2-mediated ferroptosis enhancement. Collectively, our findings indicate that GBP2 functions as a general enhancer of CD8⁺ T cell–mediated cytotoxicity rather than an ICB-specific regulator. GBP2 facilitates T cell–derived IFN-γ–dependent ferroptosis in tumor cells, a process that can occur independently of immune checkpoint blockade. The observation that GBP2 deficiency diminishes OT-1–mediated tumor suppression in the absence of PD-L1 inhibition further supports its broader immunomodulatory role. Nonetheless, GBP2-driven ferroptotic cell death substantially amplifies the therapeutic efficacy of ICB, suggesting that while GBP2 acts through a fundamental T cell cytotoxic mechanism, it also synergistically enhances ICB responsiveness by strengthening IFN-γ–induced ferroptotic signaling within the tumor microenvironment. Methods Human samples and animals All melanoma cases were pathologically confirmed by two independent pathologists, and only treatment-naïve samples from patients without other malignancies were included. Only clinical samples collected in excess of those needed for diagnostic and therapeutic procedures were used for research. All participants provided written informed consent for the collection and subsequent analysis of their samples and clinical data. Twelve resected human melanoma tissues were obtained from patients at Zhongnan Hospital of Wuhan University. The information of patients is shown in Supplementary Table 3. Six-week-old female and male C57BL/6 mice and NOD. Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice were purchased from the Hubei BIONT Biological Technology Co., LTD. OT-1 mice were purchased from the Jackson Laboratory. Sex specific differences were not investigated as the study focused on general treatment efficacy rather than sex-related outcomes. Animal model and treatment protocols C57BL/6 mice were subcutaneously injected with sgCtrl and GBP2−/− B16-OVA or MC38-OVA cells (1 × 106/ml). When the tumors had grown to 125 mm3, the C57BL/6 mice were randomized into different groups. The anti-PD-L1 antibody (200 μg per mouse) and IgG isotype were administered intraperitoneally every 3 days until the end of the observation period. For CTL killing in vivo, OT-1 cells are first isolated and enriched. In detail, after grinding the spleen to obtain a single-cell suspension, the suspension was placed in 2 ml of red blood cell lysis buffer (Sigma Aldrich) for 1 min. Then, CD8+ OT-1 cells were purified using EasySep mouse CD8+ T Cell Isolation Kit (Stemcell). The CD8+ OT-1 cells were cultured at 2 × 106/ml in RPMI culture medium containing 2 nM OVA257-264 (SIINFEKL), 10 ng/mL human recombinant IL-2, and 50 μM 2-mercaptoethanol. The cells were incubated at 37 °C for 5 days. Finally, C57BL/6 mice were subcutaneously injected with cells as indicated. Enriched OT-1 cells (5 × 106/ml per mouse, every 7 days) were resuspended into 200 μl PBS and intravenously injected into tumor-bearing mice. All the tumor volume (mm3) was measured and calculated as follows: formula: V = a × b × b/2. Cell culture and reagents A human melanoma cell line (A375), mouse melanoma cell line (B16-OVA), mouse colon carcinoma cell line (MC38-OVA) and human HEK 293 T cells were obtained from the Cell Resource Centre of Peking Union Medical College (Beijing, China). All cells were cultured in RPMI 1640 medium (Hyclone, USA) with 10% fetal bovine serum (Gibco, Australia) and 1% penicillin-streptomycin at 37 °C with 5% CO2. All cell lines were authenticated by short tandem repeat (STR) fingerprinting and tested to confirm no mycoplasma contamination and used in the experiments were within 20 passages. Human and murine recombinant IFN-γ were purchased from PeproTech. The in vivo murine anti-PD-L1 antibody (BE0101) and IgG isotype (BE0089) were purchased from BioXCell. The ferroptosis agonizts RSL3 (S8155) and inhibitor ferrostatin-1 (Fer-1, S7243) were purchased from Selleck. CRISPR–Cas9-mediated genome editing The CRISPR/Cas9 based on pX459 plasmid was constructed. The guide RNA sequences were listed in Supplementary Table 1. For CRISPR/Cas9 plasmid transfection, B16-OVA and A375 cells were transfected according to instructions for the lipofectamine 2000 and sorted by puromycin to acquire a monoclonal cell line. Finally, the GBP2−/−, HMGB1−/− and STAT1−/− cells were identified via western blotting and cultured for further analysis. Stable overexpressing cell line construction To generate stable GBP2 overexpressing cells, Pspax2, pVSV-G and constructed Plvx-IRES-Puro or empty vector were transfected into HEK 293 T cells using lipofectamine 2000. The supernatants containing viruses were added in B16-OVA with multiplicity of infection (MOI) at 5:1 and with the concentration of Polybrene at 8 μg/ml and the infection lasted for 12 h. After 48 h infection, puromycin was added to cells and the cells were treated with concentration at 1.5 μg/ml. The treatment lasted for 3 days, and the cells were kept in puromycin in half of the concentration for stable selection. Lipid ROS detection For the detection of lipid ROS in vivo, tumor tissues were chopped, ground, and filtered to form a single-cell suspension. Then, the density gradient centrifugation (Ficoll, Sigma-Aldrich) was used to pre-enrichment. Finally, the cell pellet was stained with anti-CD45 (30-F11) and anti-OVA257–264-H2Kb (25-D1.16) antibodies. Afterwards, 5 μM BODIPY-C11 (D3861, Thermo Fisher Scientific) staining were performed and incubated at 37 °C for 30 min. The signal of oxidized C11 was immediately analyzed using flow cytometry. For the detection of lipid ROS in vitro, B16-OVA cells (2 × 104/ml) co-cultured with activated OT-1 (B16-OVA: OT-1 ratio of 5:1) for 24 h followed by treatment with 10 μM Fer-1 for an additional 24 h. OT-1 cells were removed via gently washing several times with PBS. Then, the 5 μM C11-BODIPY 581/591 working solution was added and incubated with cells for 30 min in the incubator. Ready-to-use DAPI staining solution (C1005, Beyotime, China) was used to stain the nuclei for 5 min. After washing twice, the cells were covered with HBSS, and the images of oxidized C11 (wavelength 505–550 nm) were captured using a fluorescence microscope. Cell viability detection Cell viability was measured using a CellTiter-Glo Luminescent Cell Viability Assay kit (Promega), which is based on the luciferase reaction to measure the amount of ATP from viable cells. The amount of ATP in the cells correlates with cell viability. Western blot First, total protein fractions of melanoma cells were obtained using NP40 lysis buffer (Beyotime, China). The BCA kits (Beyotime, China) were used to determine the concentration of the protein fractions. Then, 30 μg of protein from each group was run on an 10% SDS-PAGE gel and transferred to polyvinylidene fluoride (PVDF) membranes (Millipore, USA). The PVDF membranes were blocked with 5% fat-free milk and incubated overnight at 4 °C with the appropriate primary antibodies: GBP2 (11854-1-AP, Proteintech, USA), p-STAT1 (#9167, CST, USA), STAT1 (#14994, CST, USA), SLC7A11 (#98051, CST, USA), SLC3A2 (15193-1-AP, Proteintech, USA), ACSL4 (22401-1-AP, Proteintech, USA), PD-L1 (17952-1-AP, Proteintech, USA), HMGB1 (10829-1-AP, Proteintech, USA), SHP1 (24546-1-AP, Proteintech, USA), and GAPDH (10494-1-AP, Proteintech, USA). Finally, enhanced chemiluminescence reagents (Thermo, USA) were applied to detect the bands after incubation with secondary antibody (GB23303, Servicebio, China). Flow cytometric analysis of tumor-infiltrating immune cells B16-OVA cells were subcutaneously injected into mice, which were then treated with anti–PD-L1 antibody as described in the animal section. On day 21 after treatment, tumors were harvested, weighed, mechanically minced, and digested with collagenase IV (2 mg/mL; Sigma-Aldrich) and DNase I (50 µg/mL; Sigma-Aldrich) for 1 h. The resulting cell suspension was triturated and passed through a 70 µm cell strainer to obtain a single-cell suspension. Cells were blocked with anti-mouse CD16/32 antibody (BioLegend) and stained with the indicated antibodies on ice for 30 min. Live/Dead Aqua dye (1:1,000; Thermo Fisher Scientific) was added together with surface antibodies to exclude dead cells. After washing, cells were fixed and permeabilized using the Foxp3/transcription factor staining buffer set (eBioscience) according to the manufacturer’s instructions, blocked with mouse and rat serum, and then stained with intracellular antibodies. All stained cells were analyzed on an Attune NxT systems and the data was analyzed with FlowJo version 10 software. ELISA assay Cytokine production in the homogenate supernatant from tumor tissue was quantified by ELISA kits (R&D Systems) according to the manufacturer’s protocol. Protein expression, purification, and in vitro kinase assay To obtain recombinant wild-type (WT) proteins, their corresponding mutants, MBP-tagged, or fluorescent GBP2 proteins, E. coli BL21 (DE3) cells harboring the respective constructs (pGEX vector) were cultured in LB medium supplemented with 50 μg/mL kanamycin. Protein expression was induced overnight at 18 °C with 0.1 mM IPTG once the OD600 reached 0.8. Cells were lysed in a buffer containing 20 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM DTT, 1 mM PMSF, and 1 mg/mL lysozyme. Fusion proteins were purified by affinity chromatography using glutathione-Sepharose resin (GE Healthcare Life Sciences, USA). The purity of the purified GBP2 was evaluated by electrophoresis on a 10% polyacrylamide gel, followed by Coomassie Blue staining and immunoblotting. Human SHP1 was expressed in E. coli BL21 (DE3) cells using an N-terminal strep-tag construct. Protein expression was induced with IPTG at 18 °C overnight, and the protein was purified by Ni2+-affinity chromatography followed by size-exclusion chromatography. JAK2 was expressed in HEK293T cells by transient transfection, and the protein was isolated from cell lysates using anti-FLAG affinity resin and further purified by gel filtration. STAT1 was produced in Sf9 insect cells using the baculovirus expression system, and purified through Ni2+affinity chromatography and size-exclusion chromatography. For in vitro kinase assays, purified STAT1 was incubated with JAK2 in kinase buffer (50 mM HEPES pH 7.5, 10 mM MgCl₂, 1 mM DTT, 100 μM ATP) at 30 °C. When indicated, GBP2 or SHP1 was added after phosphorylation to examine dephosphorylation. Reactions were terminated by adding SDS sample buffer and analyzed by SDS‑PAGE and immunoblotting with phospho-specific antibodies. In vitro T cell killing assay Differentially labeled, B16-OVA cells were seeded into 12-well plates in DMEM supplemented with 10% FBS. After 24 h, the medium was replaced with RPMI containing 10% FBS and preactivated OT‑I T cells at the indicated effector-to-target ratios. After 24 h of co-culture, tumor cells were harvested and analyzed by flow cytometry to assess cell viability. In vitro and cellular phase separation assay Phase separation of recombinant wild-type or MBP-tagged mCherry‑GBP2 proteins with recombinant GFP‑tagged SHP1 was examined in LLPS buffer (40 mM Tris-HCl pH 7.4, 150 mM NaCl, and 1.25% PEG8000). Protein mixtures were gently mixed and incubated at room temperature, and 10 µL of the reaction was placed on a glass slide, covered with a coverslip, and immediately imaged using a confocal microscope. Generation of endogenously tagged GBP2-mCherry cell line CRISPR–Cas9 with a single guide RNA (sgRNA) targeting the endogenous GBP2 locus was used to insert a mCherry sequence immediately before the stop codon in B16-OVA. An oligonucleotide encoding this sgRNA (TGAAGCAAAACAAATCGTCT) was cloned into pX330, which also encodes Cas9 (Addgene no. 42230). Six hundred thousand B16-OVA cells were seeded and transfected with 0.5 μg pX330/sgRNA vector and 1 μg repair template, consisting of mCherry flanked by 800-base pair (bp) homology arms using lipo2000. One week later, mCherry-positive cells were collected by sorting and single-cell clones were generated. Successful targeting was verified by sequencing. Live cell imaging Cells were grown on poly-l-lysine-treated no. 1.5 coverglass chambers (Lab-Tek) for 24 h. Before imaging, culture medium was replaced with RPM-1640 (ThermoFisher) supplemented with 10% FBS and 1× GlutaMAX (ThermoFisher). Cells were imaged with a Nikon A1R+ confocal microscope with a 60× oil objective and a Nikon A1 camera equipped with a heated stage (37 °C) and 5% CO2 environment. Fluorescence recovery after photobleaching FRAP was performed using a Nikon A1R+ microscope with a 488-nm or 561-nm laser. For FRAP experiments on GBP2 condensates in living cells, the entire punctum, typically less than 1 μm in diameter, was photobleached using 70% laser power, before time-lapse imaging. Fluorescence intensity in the photobleached region was measured using ImageJ and values were normalized to pre-bleach time points. Coimmunoprecipitation (CoIP) Total protein fractions were collected for CoIP with indicated antibody according to the manufacturer’s protocol using an IP/CoIP kit (Absin, China). In brief, 293T, A375, and B16-OVA cells were lysed with IP buffer (P0013J, Beyotime, China) containing protease and phosphatase inhibitors for 45 min at 4 °C. The cell lysates were precleared with protein A/G-agarose beads for 1 h and then incubated with IgG, indicated antibody, overnight at 4 °C with continuous inversion. The cell lysates were precipitated with protein A/G-agarose beads for 4 h and then the immunoprecipitated beads were washed five times the following day. Bound proteins were eluted and subjected to immunoblot analysis with indicated antibodies. siRNA transfection The human and mouse SHP1 and SLC7A11 siRNAs were purchased from Ribobio (Guangzhou, China). B16-OVA and A375 cells was transfected with siRNA using lipofectamine 2000 (Invitrogen, USA). Scramble siRNA was used as a control. Sequences of siRNAs were listed in the Supplementary Table 2. Chromatin immunoprecipitation assay (ChIP) A Sonication ChIP kit (Abclonal, China) was used to perform the ChIP assay according to the manufacturer’s instructions. Briefly, 1 × 107 cells were cross-lined with 1% formaldehyde for 8 min at room temperature. Then, fixed cells were harvested, lysed, and sonicated on ice with several pulses and incubated for 30 s on ice between pulses. The sheared chromatin fragments were immunoprecipitated with antibodies against STAT1 (5 μl per 1 mg total protein) and rabbit IgG (2 μl per 1 mg total protein). After elution and decross-linking, the enriched DNA fragments were purified by universal DNA purification kit (DP214, TIANGEN). The primer sequences used for ChIP-qPCR are listed in the Supplementary Table 4. Multiplex immunofluorescence analysis and IHC analysis Multiplex immunofluorescence staining was obtained using PANO 3-plex IHC kit (Panovue, Beijing, China). The primary antibodies of GBP2 (27299-1-AP, Proteintech, USA), SLC7A11 (ab307601; Abcam, USA), and CD8A (A11856, ABclonal, USA) were sequentially applied, followed by horseradish peroxidase-conjugated secondary antibody incubation and tyramide signal amplification (TSA). The slides were microwave-treated after each TSA operation. Nuclei were stained with DAPI after all the human antigens had been labeled. In addition, standard immunohistochemistry protocol was also performed using the above antibodies60. Immunofluorescence Collected cells were spun down onto a coverslip, fixed in 4% paraformaldehyde and permeabilized with 0.2% Triton X-100. Fixed cells were blocked in 5% BSA and probed with anti-p-STAT1 (#9167, CST, USA) antibody in 5% BSA, followed by Alexa Fluor® 488 Goat anti-rabbit (ab150077, Abcam, USA) and the slides were further counterstained with DAPI and mounted for confocal analysis and the fluorescent image was observed under a fluorescent microscope. The glutathione (GSH) assay The reduced (GSH) assay kit (Nanjing Jiancheng) was used to quantitatively analyze the levels of intracellular GSH according to the supplier’s instructions. Datasets collection In this work, 7 independent melanoma cohorts were selected, including a total of 1076 melanoma patients: skin cutaneous melanoma (SKCM) from The Cancer Genome Atlas (TCGA); GSE6590461; GSE840162; GSE5945563; GSE318964; GSE11597865 and PRJEB2370966. The upper quartile fragments per kilobase of transcript per million mapped reads (FPKM-UQ) for SKCM were downloaded from the UCSC Xena browser (https://xenabrowser.net/datapages/). The PRJEB23709 cohort was obtained from the Tumor Immunotherapy Gene Expression Resource database (http://tiger.canceromics.org/#/download)67. The remaining five cohorts were all downloaded from the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/). Detailed information of above datasets is shown in Supplementary Dataset 3. Bioinformatic analysis The relative abundance of 28 immune cell infiltration in each melanoma sample was quantified via ssGSEA (Supplementary Dataset 4)68,69. We used the “limma” R package to screen the differentially infiltrating cells in the tumors separated by median GBP2 expression (high versus low). An unsupervised subclass mapping method (SubMap) (GenePattern module ‘SubMap’) (https://cloud.genepattern.org) was used to identify common subgroups between independent cohorts despite their technical differences70. A melanoma dataset including anti-PD-1 and anti-CTLA4 therapy was applied to measure the similarity with high- and low-GBP2 groups in other melanoma datasets included in this study46. The spatial tumor microenvironment (SpatialTME) database was used to analyze the spatial transcriptome of one melanoma samples (https://www.spatialtme.yelab.site/#!/search/Visium/Melanoma_10x/Melanoma_IFStainedFFPE_CytAssist_10x)71. A spatial transcriptome dataset of hepatocellular carcinoma (HCC) containing ICB therapy was also used to validate our results72. For survival analysis, Kaplan–Meier survival curves and log-rank tests were used to evaluate the outcomes of melanoma patients in the with high- and low-GBP2 groups according to “survival” R package. Gene set enrichment analysis (GSEA) and gene ontology (GO) functional annotation GSEA was used to enrich the significant gene sets between the high- and low-GBP2 groups. The gene sets of “H.all.v7.4” were downloaded from the Molecular Signatures Database and used as the reference set to represent well-defined biological states or processes. Additionally, Gene Ontology (GO) functional enrichment analysis was performed using the DAVID database (https://david.ncifcrf.gov/)73. The GTP binding protein were collected from the GO term of GTP binding (GO: 0005525) according to the quickGO website (https://www.ebi.ac.uk/QuickGO/) (Supplementary Dataset 5). Weighted gene coexpression network analysis (WGCNA) First, raw data of GSE8401 were preprocessed with RMA background correction, log2 transformation, and quantile normalization, and then the median-polish probeset summarization was applied through the “Affy” R package74. The probes were annotated through the “hgu133a.db” R package. GSE8401 was assessed by sample clustering according to the distance between different samples in Pearson’s correlation matrices. Using a cutoff height of 60, no samples were removed (Fig. S1A). Genes were filtered by median absolute deviation (MAD) since low-expressed or non-varying genes usually represent noise. The “WGCNA” R package was used to construct a coexpression network according to the top 5000 genes sorted by MAD74,75. Then, we calculated the Pearson’s correlation matrices for all paired genes. A weighted adjacency matrix was further constructed using a power function amn = |cmn | β (cmn = Pearson’s correlation between gene m and gene n; amn = adjacency between gene m and gene n). β was a soft thresholding parameter that could emphasize strong correlations between genes and penalize weak correlations. The power of β = 6 was selected as a soft threshold to ensure a scale-free network (Fig. S1B–E). Moreover, the adjacency was transformed into a topological overlap matrix (TOM) that could measure the network connectivity of a gene and is defined as the sum of its adjacency with all other genes for network generation76. The average linkage hierarchical clustering was used to classify genes with similar expression profiles into different modules with a minimum gene group size of 3077. Finally, we used the module eigengenes (MEs) to identify gene modules of interest related to phenotype. MEs were deemed as the first principal component in the principal component analysis for each gene module, which summarizes the expression patterns of all genes into a single characteristic expression profile within a given module. Statistical analysis and reproducibility All experiments were performed at least three independent experiments. All results are shown as means ± SD. Comparisons between two unpaired groups were evaluated by two-tailed Student’s t test. Two-tailed unpaired Student’s t-test and ANOVA with multiple comparisons were used to calculate P values. Correlations between groups were determined by Pearson’s correlation test. The numbers of samples and independent biological experimental repeats are indicated in the figures or figure legends. The P < 0.05 was considered statistically significant. The analysis was conducted using the GraphPad 8.0 software and R 4.1.3 software. Ethical statement The study was conducted in accordance with the principles outlined in the Declaration of Helsinki and was approved by the Ethics Committee of Zhongnan Hospital of Wuhan University (approval No. ZN2024028K). C57BL/6 mice, NOD. Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice and OT-1 mice, C57BL/6-Tg (TcraTcrb) 1100Mjb/J were housed under specific pathogen free conditions. These animals were specific pathogen-free and maintained in 6 animals per group on a 12/12-h light/dark cycle at 20–26 °C with around 40–70% humidity, and given ad libitum access to standard food and water. All procedures involving mice and research protocols were approved by the Experimental Animal Welfare Ethics Committee, Zhongnan Hospital of Wuhan University (approval No. ZN2024072) and the Ethics Committee of the Experimental Animal Care of Beijing Chest Hospital, Capital Medical University (approval No.XK-2025-059). The maximal tumor size permitted by the ethics committee is 1500 mm3. Throughout this study, this tumor size limit was strictly monitored and was not exceeded in any of the experimental animals. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Dataset 1 Supplementary Dataset 2 Supplementary Dataset 3 Supplementary Dataset 4 Supplementary Dataset 5 Reporting Summary Transparent Peer Review File Acknowledgements The work was supported from the National Natural Science Foundation of China (82303181 to H.Z.W., 82403279 to F.W.). Program of Excellent Doctoral (Postdoctoral) of Zhongnan Hospital of Wuhan University (grant no. ZNYB2022003 to H.Z.W.). Advantageous discipline construction funds (XKJS202020 to Q.Z.). Fundamental Research Funds for the Central Universities (grant no. 2042023kf0076 to H.Z.W.). Science and Technology Innovation Cultivation Fund of Zhongnan Hospital of Wuhan University (grant no. CXPY2023018 to H.Z.W.). Clinical Medicine + Youth Lift Program of the Wuhan University (grant no. 413000555 to H.Z.W.). Beijing Municipal Science & Technology Commission (Z221100007422056&Z221100007422064 to Y.P.). Key project of Science and Technology Commission of Tongzhou District, Beijing (KJ2024CX028 to Y.P.). Author contributions H.Z.W., X.L., and F.W. contributed to conception and designed the experiments; X.L., S.L.F., X.H.R., and Y.T.H. performed most of the experiments. L.P.G., H.H.N., Y.M.N., and M.Z. helped with the in vitro experiments. X.P.L. and J.C.G. helped with the sample collection, immunohistochemistry and immunofluorescence experiments. L.L. helped with project design and data analysis. H.Z.W. and X.L. analyzed the data and wrote manuscript; Y.P., Q.Z., and F.W. revised the manuscript; all authors read and approved the manuscript. Peer review Peer review information Nature Communications thanks anonymous reviewer(s) for their contribution to the peer review of this work. [A peer review file is available]. Data availability All data associated with this study are present in the paper or the Supplementary Information. TCGA-SKCM dataset was obtained from UCSC Xena [https://xenabrowser.net/datapages/?dataset=TCGA-SKCM.star_fpkm-uq.tsv&host=https%3A%2F%2Fgdc.xenahubs.net&removeHub=https%3A%2F%2Fxena.treehouse.gi.ucsc.edu%3A443]. Previously published sequencing data of melanoma that were obtained from Gene Expression Omnibus (accession numbers: GSE3189, GSE59455, GSE65904, GSE115978, GSE8401). Melanoma cohort with immunotherapy of PRJEB23709 was obtained from TIGER database [http://tiger.canceromics.org/#/download]. The key raw data of protein mass spectrometry have been deposited to the iProX [https://www.iprox.cn//page/project.html?id=IPX0011042000]. Source data are provided with this paper. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Xin Liu, Shilin Fang, Xianghai Ren, Yuntian Hong, Jianchun Guo. Contributor Information Jun Fang, Email: xhfangjun@163.com. Yu Pang, Email: pangyupound@163.com. Qiu Zhao, Email: qiuzhao@whu.edu.cn. Fan Wang, Email: fanndywang@foxmail.com. Haizhou Wang, Email: whzcj@outlook.com. Supplementary information The online version contains supplementary material available at 10.1038/s41467-025-67690-9. References Associated Data Supplementary Materials Supplementary Information Description of Additional Supplementary Files Supplementary Dataset 1 Supplementary Dataset 2 Supplementary Dataset 3 Supplementary Dataset 4 Supplementary Dataset 5 Reporting Summary Transparent Peer Review File Data Availability Statement All data associated with this study are present in the paper or the Supplementary Information. TCGA-SKCM dataset was obtained from UCSC Xena [https://xenabrowser.net/datapages/?dataset=TCGA-SKCM.star_fpkm-uq.tsv&host=https%3A%2F%2Fgdc.xenahubs.net&removeHub=https%3A%2F%2Fxena.treehouse.gi.ucsc.edu%3A443]. Previously published sequencing data of melanoma that were obtained from Gene Expression Omnibus (accession numbers: GSE3189, GSE59455, GSE65904, GSE115978, GSE8401). Melanoma cohort with immunotherapy of PRJEB23709 was obtained from TIGER database [http://tiger.canceromics.org/#/download]. The key raw data of protein mass spectrometry have been deposited to the iProX [https://www.iprox.cn//page/project.html?id=IPX0011042000]. Source data are provided with this paper.