Fibroblast-specific ablation of MHC-I antigen presentation via the IGF2 axis cripples CD8⁺ T cell surveillance in lung cancer

Abstract

How cancer-associated fibroblasts (CAFs) dictate CD8⁺ T cell dysfunction during lung cancer remains unclear. Through single-cell analysis of 84 human NSCLC samples, we identified insulin-like growth factor 2 (IGF2) as a key immunosuppressive factor in a specific CAF subset. IGF2⁺ CAF infiltration correlated with CD8⁺ T cell dysfunction. Fibroblast-specific IGF2 knockout enhanced antigen presentation cell-autonomously via major histocompatibility complex class I (MHC-I), boosting CD8⁺ T cell effector function, tumor suppression, survival, and PD-1 blockade synergy. Mechanistically, IGF2 sustained MYC signaling to up-regulate DNA methyltransferase 1 (DNMT1), which methylated the STAT1 promoter, epigenetically silencing STAT1. This impaired MHC-I presentation and CD8⁺ T cell activation. Clinically, high IGF2 levels associated with reduced CD8⁺ T cell cytotoxicity and poor outcomes. An IGF2⁺ fibroblast signature predicted worse response to immunotherapy in multiple cohorts. Our findings establish IGF2 as a central regulator of CAF-mediated immunosuppression and a stromal target for enhancing immunotherapy in lung cancer.

IGF2 signaling suppresses fibroblast MHC-I antigen presentation and CD8⁺ T cell surveillance in lung cancer.

INTRODUCTION

The lung tumor microenvironment (LTME) is a complex ecosystem where dynamic interactions between cancer cells and stromal components critically shape antitumor immunity and therapeutic outcomes (1). Cancer-associated fibroblasts (CAFs) are a dominant stromal population known to exhibit profound functional heterogeneity (2). Specific subsets of CAFs actively suppress immune responses, primarily through the inhibition of cytotoxic CD8⁺ T cell function. These T cells serve as key effector cells mediating direct tumor cell elimination via cytokine secretion, such as interferon-γ (IFN-γ), and the release of cytolytic molecules, including granzyme B and perforin (3, 4). Despite the established importance of CD8⁺ T cell dysfunction in tumor immune evasion, the specific CAF subpopulations and molecular mediators driving this immunosuppression within lung cancer remain incompletely defined, limiting the development of effective stromal-targeting strategies.

The functional heterogeneity of CAFs stems from distinct cellular subtypes, each characterized by unique molecular profiles and spatial organization within the TME (5). Myofibroblastic CAFs (myCAFs), predominantly localized at the tumor-invasive front, exhibit elevated expression of α–smooth muscle actin (α-SMA) and fibroblast activation protein (6). These cells drive extracellular matrix remodeling through excessive collagen deposition, generating physical barriers that impede CD8⁺ T cell infiltration while fostering immune evasion (7, 8). Inflammatory CAFs cluster within hypoxic tumor niches and secrete protumorigenic cytokines [e.g., interleukin-6 (IL-6) and C-C motif chemokine ligand 2 (CCL2)] via signal transducer and activator of transcription 3 (STAT3)/Akt–mediated pathways. This paracrine signaling recruits immunosuppressive populations such as myeloid-derived suppressor cells (MDSCs) and regulatory T cells (T_reg cells), establishing a chronic inflammatory milieu that dampens cytotoxic T lymphocyte (CTL) activity (9, 10). A unique subset, antigen-presenting CAFs (apCAFs), constitutively expresses major histocompatibility complex class II (MHC-II), and CD74 yet lacks critical costimulatory molecules (e.g., CD80/CD86), rendering them incapable of fully activating CD4⁺ T cells and potentially inducing T cell anergy (9, 11). Moreover, apCAFs modulate CD8⁺ T cell function indirectly through nicotinamide N-methyl transferase–regulated secretion of complement components, which facilitate MDSC recruitment, thereby amplifying immunosuppression within the TME (12). However, the key molecular determinants governing the antigen-presenting capacity of CAFs remain poorly defined.

MHC-I is indispensable for antitumor immunity by presenting tumor-derived peptides to CD8⁺ T lymphocytes (13). Malignant cells frequently evade this immune surveillance through MHC-I down-regulation, thereby impairing T cell recognition and promoting tumor progression (14, 15). In contrast, MHC-I expression on nontransformed stromal cells can initiate potent CD8⁺ T cell–mediated antitumor responses. Substantial evidence confirms that stromal populations such as macrophages and endothelial cells activate CD8⁺ T cells via MHC-I–dependent antigen presentation. This process enhances cytotoxic granzyme B and perforin release, resulting in direct killing of adjacent tumor cells (16, 17). Nevertheless, the immunological function of MHC-I in CAFs remains virtually unexplored, despite their abundance and established role in creating immunosuppressive TME. This notable knowledge gap impedes a mechanistic understanding of stromal-immune interactions in cancer.

Insulin-like growth factor 2 (IGF2) belongs to the insulin-like growth factor family and functions as a secreted protein regulating cellular proliferation, survival, and differentiation (18, 19). While mounting evidence links elevated IGF2 levels to tumor aggressiveness, current understanding remains largely confined to its direct tumor cell–autonomous effects (20, 21). Crucially, the mechanisms by which endogenous IGF2 signaling mediates CAF-dependent immunosuppression and facilitates immune evasion in lung cancer are yet to be elucidated. To bridge this knowledge gap, we integrated multiomic profiling, used fibroblast-specific IGF2 knockout murine models, and implemented selective IGF2 pharmacological blockade. Our comprehensive approach revealed that IGF2-mediated cellular signaling pathways in CAFs drive CAF-dependent immune suppression within the LTME.

RESULTS

IGF2-expressing CAFs correlate with CD8⁺ T cell dysfunction in NSCLC

To investigate key CAF genes regulating CD8⁺ T cell antitumor function within the LTME, we integrated four single-cell RNA sequencing (scRNA-seq) datasets, encompassing 84 non–small cell lung cancer (NSCLC) samples from our in-house NSCLC cohort and the Gene Expression Omnibus (GEO) database (Fig. 1A). Following batch correction, dimensionality reduction, and clustering analysis, CAF and T cell populations were extracted from these tumor tissues. On the basis of distinct differential gene expression patterns, CAFs were categorized into five functionally divergent subpopulations (CAF_C1, CAF_C2, CAF_C3, CAF_C4, and CAF_C5). Similarly, we identified four major T cell subsets: CD8⁺ T cells, CD4 effector T cells (CD4Teff), conventional T (Tconv) cells, proliferating T (Tprolif) cells, and T_reg cells (fig. S1, A to C).

Fig. 1. IGF2-expressing fibroblast is correlated to CD8⁺ T dysfunction in lung cancer.

(A) Workflow of scRNA-seq analysis of GEO lung cancer cohorts. (B) CTL scores of CD8⁺ T cells in the TME. Patients were stratified into high- and low-CTL groups by the median score. (C) UMAP showing CAF and T cell clusters. (D) Cell-cell communication network between CAF subsets and CD8⁺ T cells. (E) Heatmap of interaction frequencies in tumors enriched for CD8⁺ T^dys or CD8⁺ T^act cells. (F) Differential interaction network between CD8⁺ T^dys and CD8⁺ T^act tumors; red and blue indicate increased and decreased interactions, respectively. (G) Differentially expressed genes in CAF_C1 between CD8⁺ T^dys and CD8⁺ T^act tumors. (H) Overlap between CAF_C1 marker genes and differentially expressed genes (DEGs) and Cox regression analysis of prognostic significance. CAF_C1 markers were defined by one-vs-rest analysis versus CAF_C2–CAF_C5. (I) Gene correlation network in TCGA-LUAD. (J and K) IGF2 expression in LLC tumors assessed by dot plot and flow cytometry. (L) Representative immunofluorescence images showing colocalization of IGF2 and α-SMA in human LUAD. Scale bars, 50 μm. (M) Flow cytometry quantification of IGF2 in human LUAD specimens. (N) Correlation between IGF2 and TGFB1 in TCGA-LUAD. (O and P) IGF2 expression in LLC- or LUAD-derived CAFs treated with tumor-conditioned medium from siNC- or siTGFB1-transfected tumor cells, with or without recombinant TGF-β. Data are mean ± SEM; statistical significance was determined by one-way analysis of variance (ANOVA) unless otherwise indicated. FC, fold change; HR, hazard ratio; CI, confidence interval; rTGF-β, recombinant TGF-β.

To determine which CAF subsets predominantly influence CD8⁺ T cell function within the LTME, we stratified the lung cancer samples based on the median score of CTL-associated gene signatures in CD8⁺ T cells. Samples scoring above the median were classified as CD8⁺ T cell–activated (CD8⁺ T^act), while those scoring below were defined as CD8⁺ T cell–dysfunctional (CD8⁺ T^dys) tumors (Fig. 1B). Compared to CD8⁺ T^act tumors, CD8⁺ T^dys tumors exhibited a substantially increased proportion of CAF subsets, alongside markedly reduced proportions of Tprolif and CD8⁺ T cell subsets (Fig. 1C and fig. S1D). Critically, expression levels of CD8⁺ T cell functional markers [interferon gamma (IFNG), granzyme B (GZMB), and perforin 1 (PRF1)] were notably lower in CD8⁺ T^dys tumors, indicating impaired antitumor activity (fig. S1E).

Cell-cell communication analysis revealed notable interactions between CAF subsets and CD8⁺ T cells. Notably, both the interaction frequency between CAF_C1 and CD8⁺ T cells were markedly increased in CD8⁺ T^dys tumors compared to CD8⁺ T^act tumors, suggesting that CAF_C1 may act as a key regulator of CD8⁺ T cell function (Fig. 1, D to F). To identify key regulatory factors within CAF_C1, we performed differential gene expression analysis and identified 748 genes substantially up-regulated in CAF_C1 derived from CD8⁺ T^dys tumors (Fig. 1G). Integrative intersection analysis with the 1259 genes specifically enriched in the CAF_C1 cluster, compared with other clusters in CD8⁺ T^dys tumors, narrowed the candidates to eight molecules: SLC26A3, SFTPB, CCN2, CCN1, IGF2, SFTPA1, SYN3, and SFTPC (Fig. 1H). Furthermore, we performed deconvolution analysis to evaluate the prognostic impact of gene-positive CAFs in patients with lung cancer from The Cancer Genome Atlas (TCGA) cohort. The results showed that a higher abundance of IGF2⁺ CAFs was notably associated with poorer patient survival (Fig. 1H). Furthermore, IGF2 expression demonstrated a positive correlation with fibroblast marker genes in TCGA lung cancer transcriptomic datasets (Fig. 1I). Validation in murine Lewis lung carcinoma (LLC) tumors using scRNA-seq and flow cytometry confirmed predominant IGF2 expression in fibroblasts relative to other cell types (Fig. 1, J and K). Similarly, in our in-house lung adenocarcinoma (LUAD) cohort, IGF2 expression was also found to be enriched in fibroblasts relative to other cell clusters (fig. S1F). Immunofluorescence staining of human lung cancer tissues revealed IGF2 colocalization with the fibroblast marker α-SMA, and flow cytometry consistently demonstrated elevated IGF2 expression in CAFs (Fig. 1, L and M).

To assess the potential role of IGF2 in modulating CD8⁺ T cell–mediated antitumor immunity, we further performed a transcriptomic analysis of TCGA-LUAD and TCGA-LUSC datasets. Notably, this revealed a substantial inverse association between elevated IGF2 expression and critical CD8⁺ T cell effector molecules, including cytotoxic granule components granzymes A/B and perforin, the cytokine IFN-γ, and lymphocyte activation markers CD8A and NKG7 (fig. S1, G and H). Consistent with this, scRNA-seq analysis of human LUAD specimens revealed elevated CD8⁺ T cell infiltration and up-regulated cytotoxic molecules (GZMB, GZMK, GZMA, and PRF1) in IGF2-low tumors relative to IGF2-high counterparts (fig. S1, I and J).

To investigate the potential mechanisms leading to high IGF2 expression in CAFs in NSCLC, we analyzed transcriptomic data from TCGA-LUAD cohorts. Bioinformatic analysis confirmed a statistically notable positive correlation between IGF2 and TGFB1 expression, which encodes the immunosuppressive cytokine transforming growth factor–β (TGF-β; Fig. 1N). In vitro coculture experiments demonstrated that genetic silencing of TGFB1 in LLC cells or human A549 lung adenocarcinoma cells markedly suppressed IGF2 expression in CAFs. Conversely, supplementation with recombinant TGF-β enhanced IGF2 levels in these stromal populations (Fig. 1, O and P). Collectively, these findings suggest that tumor-derived TGF-β may enhance IGF2 expression in CAFs and that IGF2-expressing CAFs are positively associated with impaired CD8⁺ T cell function and an immunosuppressive LTME.

IGF2 loss in CAFs potentiates CD8⁺ T cell–dependent antitumor immunity

To evaluate whether CAF-derived IGF2 contributes to immune suppression in LTME, we further generated S100a4-creERT–mediated IGF2 deletion mice (Igf2^flox/floxS100a4^creERT, referred to as Igf2^cKO) by crossing Igf2^flox/flox mice with S100a4^creERT mice (Fig. 2A). This strategy was supported by our in-house LUAD cohort and prior studies (22, 23), which identified S100A4/FSP1 as a fibroblast-associated marker and showed its predominant enrichment in fibroblast-lineage cells (fig. S2A). Given that the IGF2⁺ CAF signature identified in LLC tumors was also highly enriched in CAFs from human lung tumors (fig. S2B), these findings support the notion that the LLC model recapitulates the key functional features of IGF2⁺ CAFs in lung cancer. We therefore used the LLC model for subsequent mechanistic studies. Upon subcutaneous LLC tumor implantation, immunofluorescence analysis confirmed efficient deletion of IGF2 in CAFs from Igf2^cKO tumors compared with control tumors (Fig. 2, B and C). In contrast, scRNA-seq analysis showed no appreciable changes in IGF2 expression in other cell populations, including cancer cells, epithelial cells, and macrophages (fig. S2C). Given that systemic genetic alterations might affect T cell homeostasis, we assessed T cell populations in lymph nodes, spleens, and peripheral blood. Notably, Igf2^cKO mice exhibited no notable changes in T cell homeostasis compared to wild-type (WT) controls (fig. S2, D to I). Flow cytometry analysis of the LTME demonstrated that IGF2 deletion in CAFs notably enhanced the expression of CD8⁺ T cell effector molecules [IFN-γ and tumor necrosis factor–α (TNF-α)] (Fig. 2, D and E). These findings were functionally corroborated by the observation that Igf2^cKO mice exhibited substantially reduced tumor growth and tumor burden, suggesting enhanced CD8⁺ T cell–mediated antitumor immunity upon IGF2 ablation (Fig. 2, F to H). In the murine tissue culture-1 (TC-1) lung cancer model, we similarly found that, compared with the control group, loss of IGF2 expression in CAFs markedly enhanced the expression of functional molecules in tumoral CD8⁺ T cells and reduced tumor burden (Fig. 2, I to L).

Fig. 2. IGF2 deficiency in CAFs enhances CD8⁺ T cell function to promote tumor regression.

(A) Genotyping of WT, Igf2^flox/flox, and Igf2^cKO mice. (B) Schematic of the subcutaneous LLC model. (C) Representative immunofluorescence images of IGF2 expression in CAFs from LLC tumors. Scale bars, 50 μm. (D and E) Flow cytometry quantification of IFN-γ⁺ and TNF-α⁺ CD8⁺ T cells in LLC tumors. (F to H) Tumor growth and weight of subcutaneous LLC tumors in WT and Igf2^cKO mice. (I) Schematic of the TC-1 tumor model. (J to L) IFN-γ⁺ CD8⁺ T cells, tumor growth, and tumor weight in TC-1 tumors. (M) Schematic of the orthotopic LLC model. (N and O) Representative H&E images and quantification of tumor area in orthotopic LLC tumors. (P) Frequencies of IFN-γ⁺ and TNF-α⁺ CD8⁺ T cells in orthotopic tumors. (Q) Frequencies of circulating CD8⁺ and CD4⁺ T cells in peripheral blood of LLC-bearing mice. (R and S) Tumor growth and weight in WT and Igf2^cKO mice treated with or without anti-CD8 antibodies. (T) Kaplan-Meier survival analysis of LLC-bearing mice. Data are mean ± SEM; n values are indicated in the panels. Statistical significance was determined by unpaired two-tailed t test, one-way ANOVA, two-way ANOVA, or log-rank test, as appropriate. sc, subcutaneous.

Furthermore, we established an orthotopic LLC lung cancer model by intravenously injecting LLC cells via the tail vein to further investigate the role of IGF2 in tumor immunity (Fig. 2M). Flow cytometry analysis revealed no notable difference in CAF infiltration among the subcutaneous LLC, TC-1, and orthotopic LLC tumor models (fig. S2, J and K). In the orthotopic LLC model, deletion of IGF2 in CAFs markedly increased the expression of cytotoxic molecules in CD8⁺ T cells within lung tumors and reduced the tumor burden in the lungs compared with the control group (Fig. 2, N to P).

To confirm that this phenotype depended on CD8⁺ T cells, we treated WT and Igf2^cKO mice with a CD8-neutralizing antibody. Flow cytometry analysis of peripheral blood confirmed that this treatment substantially depleted circulating CD8⁺ T cells (Fig. 2Q). As expected, CD8⁺ T cell depletion completely reversed the antitumor effects of IGF2 deletion, including suppression of tumor growth, reduction in tumor burden, and extension of survival (Fig. 2, R to T, and fig. S2L). These rescue experiments confirmed that the antitumor effects of IGF2-deficient CAFs are CD8⁺ T cell dependent. In summary, our findings demonstrate that IGF2 depletion in CAFs potently enhances CD8⁺ T cell functionality, thereby amplifying antitumor immunity and restricting tumor progression.

IGF2 inhibition in CAFs promotes CD8⁺ T cell precursor-like plasticity and enhances antitumor cytotoxicity

To further elucidate the regulatory role of IGF2 in LTME, we initially assessed the relationship between IGF2 expression and CD8⁺ T cell infiltration using tumor immune estimation resource (TIMER) database analysis, which revealed an inverse correlation between IGF2 levels and CD8⁺ T cell abundance (fig. S3A). Moreover, using cell-type identification by estimating relative subsets of RNA transcripts (CIBERSORT) algorithm analysis on RNA-seq data from the TCGA-LUAD cohort, we consistently demonstrated that elevated IGF2 expression exhibited positive association with immunosuppressive M0 macrophage infiltration while displaying negative correlations with antitumor immune effectors including CD8⁺ T cells and natural killer (NK) cells (fig. S3B).

Next, we performed scRNA-seq analysis on LLC tumors derived from both WT and Igf2^cKO mice (Fig. 3A). On the basis of canonical marker genes, we identified 12 distinct cell populations, including M2-like macrophages (M2_Mac), M1-like macrophages (M1_Mac), adipose-like macrophages (Adip_Mac), conventional monocytes (Mono_conv), Tconv cells, neutrophils (Neutro), proliferative monocytes (Prolif_mono), type 1 dendritic cells (cDC1), CD8⁺ T cells (CD8T), type 2 dendritic cells (cDC2), NK cells, and B cells (Fig. 3B and fig. S3, C and D). When comparing Igf2^cKO with WT tumors, we observed a notable increase in the proportion of antitumor immune cells, particularly M1_Mac, CD8⁺ T, NK, and B cells, along with a concomitant reduction in protumor immune cells, including neutrophils and M2_Mac (Fig. 3C). To further investigate T cell populations, we conducted subcluster analysis and identified five major subsets: effector CD8⁺ T (CD8Teff) cells, T_reg cells, stem-like CD8⁺ T (CD8Tstem) cells, CD4Teff, and innate-like T cells (fig. S3E). As expected, IGF2 deletion in CAFs markedly increased the frequencies of CD8Teff and CD8Tstem while decreasing T_reg cell proportion (Fig. 3D and fig. S3F). CytoTRACE analysis revealed that CD8Teff exhibited a relatively low differentiation state, followed by CD8Tstem (Fig. 3, E and F). Notably, IGF2 ablation in CAFs markedly elevated the CytoTRACE scores across T cell populations (Fig. 3G), suggesting its role in maintaining their undifferentiated state, which may contribute to sustained antitumor effector function.

Fig. 3. IGF2 loss in CAFs promotes CD8⁺ T cell precursor-like plasticity and enhances antitumor cytotoxicity.

(A) Schematic of scRNA-seq analysis of LLC tumors from WT and Igf2^cKO mice. (B) UMAP of CD45⁺ immune cells in LLC tumors. (C and D) Quantification of immune cell clusters and T cell subsets in WT and Igf2^cKO tumors. (E) UMAP showing CytoTRACE-predicted developmental potential of T cell subsets. (F and G) CytoTRACE-based pseudotemporal analysis of T cell differentiation across all subsets and by genotype. (H) CTL, cytolytic activity, and cytokine effector scores in tumor-infiltrating CD8⁺ T cells. (I) GSEA of differentially enriched pathways in CD8⁺ T cells from WT and Igf2^cKO tumors. (J) t-SNE of immune cell distributions in LLC tumors. (K) Flow cytometry analysis of Ki67 expression in CD8⁺ T cells. (L to P) Frequencies of CD8⁺ T cells, CD4⁺ effector T cells, exhausted CD8⁺ T cells, myeloid cells, and T_reg cells within the CD45⁺ compartment. (Q) Schematic of the in vitro CAF-CD8⁺ T cell coculture system. (R and S) Proliferation and cytotoxic molecule expression of CD8⁺ T cells cocultured with WT or Igf2^−/− CAFs. Data are mean ± SEM; statistical significance was determined by unpaired two-tailed t test or one-way ANOVA, as appropriate. CFSE, carboxyfluorescein diacetate succinimidyl ester; NES, normalized enrichment score.

Consistent with above observation, CD8Teff cells from Igf2^cKO tumors displayed enhanced antitumor activity scores and up-regulated T cell receptor (TCR) signaling pathway activation compared to their WT counterparts (Fig. 3, H and I). Flow cytometry analysis reinforced these findings, demonstrating that IGF2-deficient CAFs promoted CD8⁺ T cell proliferation within LLC tumors while increasing the proportions of immunostimulatory populations such as CD4Teff and CD8⁺ T cells (Fig. 3, J to M, and fig. S4A). Conversely, immunosuppressive cell types, including exhausted CD8⁺ T cells, myeloid cells, and T_reg cells, were notably reduced, aligning well with the scRNA-seq data (Fig. 3, C and N to P).

Given the observed change in T_reg cell abundance, we next asked whether deletion of IGF2 in CAFs influences the differentiation trajectory of CD4⁺ T cells toward T_reg cells. To address this, we reclustered the CD4⁺ CD8A⁻ T cell compartment after dimensionality reduction (fig. S4B). T_reg cell populations were annotated by scoring a T_reg cell gene signature using the Seurat’s AddModuleScore function, leading to the identification of T_reg cell–like and CD4_nonTreg subsets (fig. S4, C and D). Pseudotime analysis revealed no notable change in the expression of T_reg cell differentiation–associated genes (fig. S4E). Consistent with this, density distribution analysis along the CD4–to–T_reg cell pseudotime trajectory showed a highly overlapping pattern between the WT and Igf2^cKO groups, with no substantial overall shift (fig. S4F). These findings suggest that IGF2 deletion in CAFs does not markedly reshape the differentiation trajectory of CD4⁺ T cells toward a T_reg cell state, although it may still influence the overall abundance of T_reg cells within the tumor microenvironment.

In vitro coculture experiments provided mechanistic support, showing that WT CAFs potently suppressed CD8⁺ T cell proliferation, reduced the expression of cytotoxic molecules, and promoted CD8⁺ T cell exhaustion, whereas IGF2 deletion in CAFs markedly reversed these inhibitory effects (Fig. 3, Q to S, and fig. S4G). Together, these results demonstrate that targeted IGF2 ablation in CAFs enhances CD8⁺ T anergy, preserves their less differentiated state, and ultimately potentiates their antitumor cytotoxicity in LLC tumors.

IGF2 inhibition potentiates CD8⁺ T cell cytotoxicity through enhanced antigen presentation in CAFs

To elucidate how IGF2 regulates CAF-mediated suppression of CD8⁺ T cell function, we isolated CAFs from LLC tumors of WT and Igf2^cKO mice for transcriptomic profiling. Comparative analysis demonstrated marked up-regulation of MHC-I–mediated antigen processing in IGF2-deficient CAFs (Fig. 4, A and B). Consistently, IGF2 ablation markedly elevated the expression of antigen presentation machinery components in CAFs (fig. S5A). The scRNA-seq of LLC tumor further demonstrated a strong positive correlation between low IGF2 expression and enrichment of MHC-related pathways in CAFs (fig. S5B). In human CAFs, IGF2 knockdown markedly increased the mRNA expression of antigen presentation–related molecules, including human leukocyte antigen A (HLA-A), HLA-B, TAP1, TAP2, and B2M (fig. S5C). scRNA-seq analysis of LLC tumors further revealed that, compared with controls, deletion of IGF2 in CAFs notably increased B2M expression. By contrast, IGF2 deletion did not up-regulate the expression of MHC-II–associated molecules (Cd74 and H2-Ab1) or the costimulatory molecules Cd80 and Cd86 in CAFs (fig. S5, D to F and H). Moreover, type I IFN signaling was not markedly activated following IGF2 deletion (fig. S5G). Together, these data suggest that IGF2 primarily regulates the antigen-presenting capacity of CAFs by modulating MHC-I–related machinery rather than MHC-II or costimulatory pathways. In vitro experiments showed that IGF2 inhibition enhanced surface expression of B2M and MHC-I, an effect further potentiated by IFN-γ stimulation (Fig. 4, C and D). Similarly, pharmacological inhibition of IGF2 with Chromceptin in WT CAFs elevated levels of antigen-presenting molecules (B2M and MHC-I) (Fig. 4, E and F, and fig. S5I). Mirroring these findings, flow cytometry analysis of both subcutaneous LLC tumors and orthotopic LLC lung tumors from Igf2^cKO mice demonstrated increased B2M and MHC-I expression in CAFs (Fig. 4, G and H).

Fig. 4. IGF2 deficiency in CAFs potentiates CD8⁺ T cell antitumor responses by enhancing intrinsic MHC-I expression.

(A) GSEA of differentially expressed genes in WT and Igf2^−/− CAFs showing the top enriched pathways. (B) Gene set variation analysis (GSVA) scores for the MHC-I antigen presentation pathway in WT and Igf2^−/− CAFs. (C and D) B2M and MHC-I expression in CAFs treated with IFN-γ. (E and F) Dose-dependent effects of Chromceptin on B2M and MHC-I expression in CAFs. (G and H) B2M and MHC-I expression in CAFs from subcutaneous and orthotopic LLC tumors of WT and Igf2^cKO mice. (I) Schematic of the LLC model used to assess antigen acquisition by CAFs. (J) Frequency of ZsGreen⁺ CAFs in LLC tumors from WT and Igf2^cKO mice. (K) Schematic of the LLC model used to assess CAF-mediated tumor antigen presentation. (L to N) LLC tumor growth, H2-Kb expression in CAFs, and frequencies of IFN-γ⁺ CD8⁺ T cells in the indicated groups. (O and P) B2M and MHC-I expression in WT or Igf2^−/− CAFs treated with or without rIGF2. Data are mean ± SEM; statistical significance was determined by unpaired two-tailed t test, one-way ANOVA, or two-way ANOVA, as appropriate. shNC, shRNA negative control; Chr., Chromeceptin.

To investigate whether IGF2-mediated modulation of antigen presentation influences CD8⁺ T cell effector function, we leveraged scRNA-seq data from LLC tumors and conducted CellChat-based computational analysis of intercellular communication between CAF subpopulations and CD8⁺ T cells. Our analysis identified four major CAF subtypes in the TME: apCAFs, myCAFs, epithelial-like CAFs, and vasculature-associated CAFs (fig. S5J). Notably, information flux analysis revealed markedly enhanced signaling from apCAFs to CD8⁺ T cells under IGF2-deficient conditions compared to controls. This increased signaling activity involved multiple pathways including MHC-I–mediated antigen presentation and immunomodulatory factors [adhesion G protein-coupled receptor E (ADGRE) and amyloid precursor protein (APP)] (fig. S5K). Notably, cell-cell interaction network analysis demonstrated that IGF2 deletion substantially increased the interaction frequency between apCAFs and CD8Teff cells of MHC-I–mediated signaling events (fig. S5L). Ligand-receptor pairing analysis revealed strengthened antigen-presenting interactions (e.g., H2-t22-CD8a and H2-k1-CD8a) in Igf2^cKO tumors (fig. S5M). In the in vivo ZsGreen-LLC tumor model, deletion of IGF2 in CAFs notably increased the proportion of ZsGreen⁺ CAFs relative to controls (Fig. 4, I and J), indicating enhanced uptake of tumor-derived material by CAFs. To further test whether CAFs can present antigen and specifically activate CD8⁺ T cells in vivo, we isolated WT and Igf2^−/− CAFs from LLC tumors, transfected them with ovalbumin (OVA) and enhanced green fluorescent protein (EGFP) plasmids, and coinoculated them with LLC cells into CD45.2⁺ C57BL/6 mice. Activated OT-I cells isolated from CD45.1⁺ OT-I mice were then adoptively transferred into tumor-bearing recipients (Fig. 4K). Subsequent analyses showed that Igf2^−/− CAFs notably suppressed tumor growth, increased H2-Kb-SIINFEKL expression on EGFP⁺ CAFs, and enhanced cytotoxic effector molecule expression in transferred CD45.1⁺ OT-I cells (Fig. 4, L to N). Collectively, these data support that IGF2 deletion in CAFs enhances the acquisition of tumor-derived material and promotes antigen presentation to activate CD8⁺ T cells in vivo.

To directly assess the immunomodulatory effects of WT and IGF2-deficient CAFs, we established an in vitro coculture system with CD8⁺ T cells. Compared to WT CAFs, Igf2^−/− CAFs exhibited markedly elevated expression levels of both MHC-I and B2M (Fig. 4, O and P). This up-regulation was effectively normalized when either CAF type was treated with recombinant IGF2 (rIGF2) protein, demonstrating the reversibility of this regulatory mechanism (Fig. 4, O and P). Parallel assessment of CD8⁺ T cell function revealed that coculture with Igf2^−/− CAFs substantially enhanced the expression of cytolytic effectors (fig. S5, N and O). Consistent with the observed effects on antigen presentation machinery, the augmented cytotoxic potential of CD8⁺ T cells was completely abrogated upon IGF2 reconstitution (fig. S5, N and O). To assess the physiological relevance of these findings, we performed functional antigen presentation assays using the OVA_257–264 peptide. These experiments confirmed that IGF2-deficient CAFs exhibited markedly improved antigen-presenting capacity compared to their WT counterparts (fig. S5P). Supporting this observation, coculture experiments with OT-I cells revealed that Igf2^−/− CAFs triggered markedly higher cytokine production in OT-I cells, demonstrating their enhanced ability to activate cytotoxic T cell responses (fig. S5Q). In the breast tumor microenvironment, IGF2 has been reported to suppress CD8⁺ T cell function by regulating programmed death-ligand 1 (PD-L1) expression (24). In contrast, in the LTME, inhibition of IGF2 in CAFs did not substantially alter PD-L1 expression (fig. S5R). Together, genetic or pharmacological blockade of IGF2 in CAFs enhances their antigen-presenting capacity, leading to CD8⁺ T cell activation and amplified cytotoxic effector function.

IGF2 deficiency in CAFs promotes intrinsic MHC-I antigen presentation via MYC signaling suppression

To elucidate how IGF2 modulates antigen-presenting machinery in CAFs, we performed gene set enrichment analysis (GSEA) on transcriptomic data from WT and Igf2^−/− CAFs. IGF2 deficiency showed a notable positive correlation with activation of the MYC repression pathway, as consistently demonstrated by scRNA-seq analysis (Fig. 5A and fig. S6A). To identify potential transcriptional regulators of B2M, we systematically analyzed multiple transcription factor databases [knockdown/out of transcription factor (KnockTF), encyclopedia of DNA elements (ENCODE), human transcription factor target (hTFtarget), ChIP_Atlas, and gene transcription regulation database (GTRD)]. Intersection analysis via UpSet plots revealed 11 candidate regulators (GATA3, MYC, NFYB, RFX5, STAT1, YY1, GATA1, FOXA1, STAT3, TFAP2C, and GTF2B) (Fig. 5B). Given the association between high IGF2 expression and MYC activation, we hypothesized that MYC might serve as the key transcriptional mediator linking IGF2 to B2M regulation. In TCGA-LUSC and TCGA-LUAD datasets, tumor tissues exhibited elevated MYC expression compared to normal counterparts, with high MYC levels predicting poor patient prognosis (fig. S6, B to D). Immune deconvolution using CIBERSORT demonstrated an inverse correlation between MYC expression and infiltration of antitumor immune cells (CD8⁺ T cells and NK cells) (fig. S6E). Furthermore, MCPcounter analysis of the TCGA-LUAD cohort revealed that, relative to the MYC-low group, MYC-high tumors displayed markedly reduced enrichment scores for CD8⁺ T cells and cytotoxic lymphocytes (fig. S6F). These findings suggested the MYC’s role in fostering immunosuppressive microenvironments. Flow cytometry analysis showed that CAFs from Igf2^cKO mice bearing LLC tumors displayed reduced c-Myc expression (Fig. 5C). Immunofluorescence confirmed diminished c-Myc nuclear translocation in Igf2^−/− CAFs relative to WT controls (Fig. 5D). Consistently, pharmacological inhibition of IGF2 using Chromceptin down-regulated c-Myc in a dose-dependent manner, establishing IGF2 as a positive regulator of c-Myc (Fig. 5, E and F).

Fig. 5. IGF2 blockade in CAFs enhances intrinsic MHC-I signaling by suppressing c-Myc activity.

(A) GSEA of signaling pathways enriched in differentially expressed genes between WT and Igf2^−/− CAFs. (B) UpSet plot of overlapping transcription factors identified from knockTF, ENCODE, hTFtarget, ChIP-Atlas, and GTRD. (C) c-Myc expression in CAFs from WT and Igf2^cKO mice. (D and E) Immunofluorescence analysis of c-Myc expression and subcellular localization in WT and Igf2^−/− CAFs and in WT CAFs treated with vehicle or Chromceptin. Scale bars, 50 μm. (F) c-Myc expression in CAFs treated with increasing concentrations of Chromceptin. (G to J) MHC-I and B2M expression in CAFs after Myc knockdown or 10058-F4 treatment. (K and L) MHC-I and B2M expression in WT and Igf2^−/− CAFs with or without Myc inhibition. (M and N) Frequencies of IFN-γ⁺ CD8⁺ T cells cocultured with WT or Igf2^−/− CAFs or with CAFs treated with 10058-F4. Data are mean ± SEM; statistical significance was determined by unpaired two-tailed t test or one-way ANOVA, as appropriate.

To further investigate the functional consequences of MYC suppression in CAFs, we used complementary genetic and pharmacological approaches. Our results demonstrated that MYC blockade substantially up-regulated the expression of B2M and MHC-I, along with mRNA levels of antigen presentation machinery components, compared with control groups (Fig. 5, G and H, and fig. S6G). Consistent with these findings, pharmacological inhibition of MYC recapitulated these effects in a dose-dependent manner (Fig. 5, I and J, and fig. S6H). Genetic ablation of IGF2 markedly enhanced MHC-I and B2M expression compared to control conditions, whereas dual inhibition of IGF2 and MYC did not yield additive effects on these antigen presentation markers in CAFs (Fig. 5, K and L). Notably, this modulation of antigen presentation closely paralleled the expression changes in cytotoxic effector molecules observed in CD8⁺ T cells cocultured with the corresponding CAFs (Fig. 5, M and N). These findings suggest functional cross-talk within the IGF2-MYC regulatory axis, possibly fine-tuning immune recognition mechanisms. Collectively, our findings delineate a cell-intrinsic pathway wherein IGF2 ablation in CAFs attenuates MYC activity, thereby derepressing antigen presentation molecules and potentiating CD8⁺ T cell–mediated antitumor immunity.

Inhibition of IGF2-MYC signaling up-regulates STAT1-dependent MHC-I expression in CAFs by suppressing DNMT1

To further investigate how the IGF2/MYC axis regulates the expression of antigen-presenting molecules in CAFs, we performed transcription factor enrichment analysis on RNA-seq data from both WT and Igf2^−/− CAFs. The results revealed that IGF2 knockout substantially enhanced STAT signaling in CAFs compared to WT CAFs (Fig. 6A). Flow cytometry analysis demonstrated that MYC inhibition markedly up-regulated STAT1 expression, and treatment with rIGF2 did not suppress elevated STAT1 levels in siMyc CAFs or 10058-F4-treated CAFs, relative to siRNA negative control (siNC) or vehicle-treated CAFs (Fig. 6, B and C). Moreover, treatment with rIGF2 protein failed to suppress the elevated expression of phosphorylated STAT1 in siMyc-treated CAFs (fig. S7A). These findings confirm that the IGF2-MYC axis regulates both the expression and activation of STAT1. Analysis of TCGA-LUAD and TCGA-LUSC cohorts indicated that tumors with high STAT1 expression exhibited notably elevated levels of antigen-presenting molecules (fig. S7, B and C), consistent with previous reports that STAT1 transcriptionally regulates antigen presentation–related genes (25). Compared to controls, inhibition of IGF2 or MYC, as well as STAT1 overexpression, enhanced the expression of antigen-presenting molecules B2M and MHC-I. Conversely, STAT1 knockdown in Igf2^−/− or siMyc CAFs reduced B2M and MHC-I expression (Fig. 6, D and E, and fig. S7, D and E). When CAFs under these conditions were cocultured with T cells, the observed changes in antigen-presenting molecules corresponded to variations in CD8⁺ T cell cytotoxic molecule expression (Fig. 6F and fig. S7, F and G), suggesting that suppression of the IGF2-MYC axis enhances CD8⁺ T cell antitumor function via STAT1-mediated antigen presentation.

Fig. 6. Blocking IGF2/MYC axis attenuates DNMT1-dependent STAT1 methylation to enhance antigen presentation in CAFs.

(A) Radar plot of enriched transcription factor signaling pathways based on differentially expressed genes between WT and Igf2^−/− CAFs. (B and C) STAT1 expression in siNC- or siMyc-treated CAFs and in vehicle- or 10058-F4–treated CAFs exposed to rIGF2. (D and E) B2M and MHC-I expression in WT or Igf2^−/− CAFs with Stat1 overexpression or knockdown. (F) Frequencies of IFN-γ⁺ and TNF-α⁺ CD8⁺ T cells cocultured with WT or Igf2^−/− CAFs with Stat1 knockdown or overexpression. (G) Correlation between MYC and DNMT1 expression in TCGA-LUSC. (H) ChIP-qPCR analysis of MYC binding at the DNMT1 promoter in WT CAFs. (I and J) DNMT1 expression in siNC- or siMyc-treated CAFs measured by RT-qPCR and immunofluorescence. (K) H-2k and B2M mRNA expression in siNC- or siDnmt1-treated CAFs. (L and M) B2M and MHC-I expression in WT or Igf2^−/− CAFs with DNMT1 knockdown. (N and O) Frequencies of IFN-γ⁺ CD8⁺ T cells cocultured with WT or Igf2^−/− CAFs, or with siMyc-treated CAFs, with DNMT1 knockdown. (P) ChIP-qPCR analysis of DNMT1 binding at the STAT1 5′ untranslated region in WT CAFs. (Q) DNA methylation levels at the STAT1 5′ untranslated region in siNC- or siDnmt1-treated CAFs. (R) Immunoblot analysis of DNMT1 and STAT1 in siNC- or siDnmt1-treated CAFs. (S and T) MHC-I expression in CAFs and IFN-γ⁺ CD8⁺ T cells in coculture after DNMT1 knockdown combined with Stat1 overexpression or knockdown. Data are mean ± SEM; statistical significance was determined by unpaired two-tailed t test or one-way ANOVA, as appropriate. MW, molecular weight. OE, overexpression; a.u., arbitrary units; AU, arbitrary unit.

To predict transcriptional targets of MYC, we performed integrative bioinformatic analyses leveraging multiple authoritative databases [ENCODE, hTFtarget, ChIP_Atlas, GTRD, and ChIP Enrichment Analysis (CHEA)]. Integrated analysis identified 5038 potential downstream targets. By combining these with differentially expressed genes from scRNA-seq and bulk RNA-seq of WT versus Igf2^−/− CAFs, we identified eight candidate molecules (GDF15, PPP1R15A, LCN2, RND1, DNMT1, TRAF1, IFRD1, and IL11) that may act downstream of IGF2-MYC signaling (fig. S7H). Among these, DNMT1, a classical DNA methyltransferase frequently overexpressed in cancers, can suppress gene expression by promoting methylation (26). Analysis of TCGA-LUSC and TCGA-LUAD datasets revealed a notable positive correlation between IGF2 or MYC expression and DNMT1 levels (Fig. 6G and fig. S7I). ChIP assays confirmed MYC binding to the DNMT1 promoter, and quantitative reverse transcription polymerase chain reaction (RT-qPCR) analysis showed that MYC inhibition substantially reduced DNMT1 transcription, which was further validated by immunofluorescence (Fig. 6, H to J). IGF2 knockout in CAFs suppressed DNMT1 expression, while additional MYC knockdown in Igf2^−/− CAFs produced no additive effect, supporting the notion that IGF2 and MYC jointly promote DNMT1 expression (fig. S7J). Notably, DNMT1 knockdown markedly suppressed the transcription of antigen-presenting molecules H-2K and B2M (Fig. 6K). Flow cytometry revealed that IGF2 knockout, MYC inhibition, or DNMT1 knockdown similarly up-regulated B2M and MHC-I expression. However, DNMT1 knockdown in Igf2^−/− or siMyc CAFs did not further enhance this effect (Fig. 6, L and M, and fig. S7, K and L). Coculture of T cells with CAFs under these conditions showed parallel changes in CD8⁺ T cell cytotoxic molecule expression (Fig. 6, N and O, and fig. S7, M and N), indicating that IGF2/MYC signaling suppresses antigen presentation and CD8⁺ T cell function via DNMT1.

In addition, ChIP analysis indicated that DNMT1 binds to STAT1, and methylated DNA immunoprecipitation assays showed that DNMT1 knockdown substantially reduced STAT1 methylation (Fig. 6, P and Q). Consequently, DNMT1 inhibition enhanced STAT1 expression (Fig. 6R). Flow cytometry data indicated that either DNMT1 knockdown or STAT1 overexpression up-regulated MHC-I to comparable levels, while STAT1 knockdown in siDnmt1 CAFs markedly reduced MHC-I expression (Fig. 6S). These alterations in antigen presentation were mirrored by functional changes in CD8⁺ T cells in coculture experiments (Fig. 6T). In summary, these findings demonstrate that inhibiting the IGF2/MYC axis attenuates DNMT1-mediated repression of STAT1 through DNA methylation, thereby up-regulating antigen-presenting molecule expression and enhancing CD8⁺ T cell activation and antitumor activity.

IGF2 blockade synergized with ICB therapy

To investigate the functional impact of IGF2 deletion in CAFs on immunotherapy efficacy, we performed integrated analyses using both clinical immuno-oncology cohorts and experimental animal models. In a patient with LUAD undergoing immunotherapy, the scRNA-seq of three tumor tissues identified 11 major cell populations, including B cells, CD4 Tconv cells, CD8 T cells, dendritic cells, endothelial cells, epithelial cells, fibroblasts, mast cells, monocytes/macrophages, NK cells, and T_reg cells (Fig. 7A). Notably, comparative analysis revealed markedly elevated IGF2 expression specifically in CAFs from immunotherapy nonresponsive compared to responsive tumors (Fig. 7B). Moreover, analysis of an additional immunotherapy cohort showed that CAF-derived IGF2 expression was notably higher in tumor tissues from patients with stable disease than in those with partial response (fig. S8, A and B). Further analysis of TCGA-LUAD and TCGA-LUSC datasets revealed an inverse correlation between IGF2 expression and immune checkpoint molecules, including CD274, cytotoxic T-lymphocyte antigen-4 (CTLA-4), and programmed cell death protein 1 (PDCD1) (Fig. 7C and fig. S8, C and D). These observations led us to hypothesize that IGF2 suppression might potentiate immunotherapy by modulating checkpoint molecule expression.

Fig. 7. Synergistic effects of IGF2 blockade and ICB therapy.

(A) UMAP of cellular heterogeneity in LUAD tumors from GSE146100. (B) IGF2 expression across stromal and immune cell clusters in ICB-responsive and nonresponsive LUAD nodules. NR, nonresponder; R, responder. Data represent three nodules from one pembrolizumab-treated patient, including two nonresponding nodules and one responding nodule. (C) Correlation of IGF2 with CD274 and CTLA4 in TCGA-LUAD. (D) Schematic of combined IGF2 blockade and anti–PD-1 treatment in the LLC tumor model. (E to G) Representative tumors and tumor weights, growth curves, and survival of WT and Igf2^cKO mice treated with IgG or anti–PD-1. (H to J) Flow cytometry analysis of B2M and MHC-I expression in CAFs and frequencies of IFN-γ⁺ and TNF-α⁺ CD8⁺ T cells in LLC tumors from the indicated groups. Data are mean ± SEM; statistical significance was determined by one-way ANOVA, two-way ANOVA, or log-rank test, as appropriate. ns, not significant.

To test this hypothesis, we evaluated the therapeutic efficacy of anti–PD-1 treatment in WT, Igf2^cKO, Chromceptin, or combine-treated mice bearing LLC tumors (Fig. 7D). The results demonstrated that while either IGF2 deletion in CAFs or anti–PD-1 monotherapy substantially inhibited tumor growth, reduced tumor burden, and prolonged survival compared to controls, the combination treatment showed superior therapeutic outcomes (Fig. 7, E to G). Detailed profiling of the tumor immune microenvironment demonstrated that both IGF2 deletion in CAFs and anti–PD-1 therapy independently augmented multiple aspects of antitumor immunity, including improved antigen presentation capacity by CAFs, increased expression of cytotoxic effector molecules (IFN-γ, TNF-α, GZMB, and PRF1) in CD8⁺ T cells, enhanced proliferation and tumor infiltration of CD8⁺ T cells, and elevated recruitment of CD4⁺ T cells to the tumor site (Fig. 7, H to J, and fig. S8, E to H). Conversely, these treatments reduced the abundance of immunosuppressive populations including myeloid cells, tumor-associated macrophages (TAMs), and T_reg cells (fig. S8, I to K). The combined intervention achieved the most pronounced immunomodulatory effects across all measured parameters.

To further explore the potential role of the IGF2 inhibitor Chromceptin in anti–PD-1 therapy, we evaluated its effects on tumor progression and immune modulation in the LLC model. Compared with the control group, monotherapy with anti–PD-1 only marginally suppressed LLC tumor growth and moderately reduced tumor burden, resulting in a limited extension of overall survival. Notably, this therapeutic effect was markedly enhanced when anti–PD-1 was combined with Chromceptin (fig. S9, A to C). Subsequent analysis of antigen-presenting molecule expression in CAFs revealed that both anti–PD-1 and Chromceptin monotherapy notably up-regulated the expression of B2M and MHC-I compared to untreated controls (fig. S9D). Consistent with this observation, CD8⁺ T cells exhibited elevated cytotoxic molecule expression after either treatment (fig. S9E). Combinatorial therapy further amplified these antitumor effects, demonstrating a synergistic enhancement of immune activation in the TME. Together, these findings provide compelling evidence that genic ablation and pharmacological inhibition of IGF2 in CAFs synergize with anti–PD-1 therapy to enhance antitumor immune responses, establishing the IGF2 pathway as a promising therapeutic target to overcome immunotherapy resistance in lung cancer.

High IGF2 expression predicts immunotherapy resistance in patients with NSCLC

To further investigate the correlation between IGF2 expression and immunotherapy efficacy in patients with lung cancer, we conducted analyses using clinical lung cancer samples and immunotherapy cohorts. In our in-house cohort, we performed multiplex immunofluorescence staining on 80 human LUAD specimens to assess the infiltration of IGF2⁺ CAFs and GZMB⁺ CD8⁺ T cells. We found that, compared with tumors with high IGF2⁺ CAF infiltration, tumors with low IGF2⁺ CAF infiltration exhibited a markedly higher proportion of GZMB⁺ CD8⁺ T cell infiltration (Fig. 8, A and B). Moreover, immunohistochemical analysis revealed that lung cancer tissues with low IGF2 expression exhibited increased GZMB expression when compared with high IGF2-expressing counterparts (fig. S10, A and B). Building on these findings, we next stratified patients in the TCGA-LUAD and TCGA-LUSC cohorts into IGF2-high and IGF2-low groups based on the median IGF2 expression level and then assessed the impact of IGF2 on immune signatures within the tumor microenvironment (fig. S10, C and D). Tumors with low IGF2 expression demonstrated elevated antitumor immune signatures, including IFNG and CD8 markers, while showing decreased protumor signatures such as T cell dysfunction scores in the TCGA-LUAD cohort (Fig. 8C). This pattern was consistently replicated in the TCGA-LUSC cohort (Fig. 8D), mirroring our experimental observations of IGF2-mediated regulation of immune cell infiltration and CD8⁺ T cell function in murine tumor models (Figs. 2, D, E, J, and P, and 3C). Furthermore, deconvolution analysis of TCGA-LUAD and TCGA-LUSC cohorts identified IGF2⁺ CAF-enriched tumors. When patients were grouped by median IGF2⁺ CAF infiltration levels, those with high IGF2⁺ CAF infiltration exhibited notably shorter overall survival compared to controls (Fig. 8, E and F).

Fig. 8. High IGF2 expression correlates with ICB resistance in patients with cancer.

(A and B) Representative multiplex immunofluorescence images showing the spatial relationship between IGF2⁺ CAFs and GZMB⁺ CD8⁺ T cells in LUAD tissues and the corresponding quantification. Scale bars, 200 μm. (C and D) Tumor microenvironment profiling in TCGA-LUAD and TCGA-LUSC, including IFNG score, CD8⁺ T cell score, and T cell dysfunction score. (E and F) Kaplan-Meier survival analysis of patients with LUAD and LUSC stratified by IGF2⁺ CAF levels. (G) Overall survival and clinical benefit in patients with NSCLC treated with anti–PD-1/PD-L1 in GSE135222, stratified by IFS enrichment. NDB, no durable benefit; DCB, durable clinical benefit. (H) Comparison of IFS scores between responders and nonresponders in the NSCLC immunotherapy cohort. (I and J) Predicted immunotherapy response rates in IFS-high and IFS-low tumors in TCGA-LUAD and TCGA-LUSC using TIDE. Data are mean ± SEM; statistical significance was determined by unpaired two-tailed t test or log-rank test, as appropriate.

To directly evaluate clinical relevance, we analyzed immunotherapy cohorts from the tumor immune dysfunction and exclusion (TIDE) database, which revealed that patients with elevated IGF2 expression exhibited poorer survival outcomes after immune checkpoint blockade (ICB) treatment relative to controls (fig. S10, E and F). Given that TME alterations represent transcriptome-wide changes rather than isolated single-gene effects (27), we expanded our investigation to assess the IGF2-positive fibroblast signature (IFS). In the TCGA-LUAD cohort, tumors with low IFS scores showed markedly increased expression of T cell function–related markers compared to control groups (fig. S10G). Further analysis using gene set variation analysis (GSVA) to evaluate LTME scores in the TCGA-LUAD sequencing data demonstrated that tumors with reduced IFS scores displayed elevated antitumor activity scores, including IFNG and CTL scores, alongside diminished protumorigenic scores such as TAM and IL13 scores (fig. S10H). These findings collectively suggest a positive correlation between IFS and an immunosuppressive TME. Subsequently, we investigated the impact of IFS on immunotherapy responsiveness using additional immunotherapy-treated cohorts. Consistent with our earlier findings, analysis of the ICB-treated NSCLC cohorts revealed improved survival and immunotherapy response benefits in patients with low IFS scores (Fig. 8, G and H). Last, TIDE algorithm analysis of TCGA-LUAD and TCGA-LUSC transcriptional profiles indicated that patients with lung cancer with low tumor IFS scores had notably higher proportions of immunotherapy responders (Fig. 8, I and J). Together, these results establish a signature from IGF2-expressing CAFs as both a predictive biomarker for immunotherapy response and a potential therapeutic target in NSCLC.

DISCUSSION

Our findings establish IGF2-expressing CAFs as pivotal drivers of CD8⁺ T cell dysfunction within the LTME. Through comprehensive cross-species multiomic analysis of human and murine systems, we identify IGF2 as a master regulator of CAF-mediated immunosuppression. Specifically, IGF2⁺ CAFs impair CTL infiltration and function, whereas genetic ablation of IGF2 restores CD8⁺ T cell effector activity by rescuing MHC-I–dependent antigen presentation. Molecular interrogation reveals that IGF2-activated MYC signaling represses MHC-I antigen presentation; this occurs through DNMT1-dependent methylation of STAT1, a mechanism that intrinsically constrains CD8⁺ T cell activation. Furthermore, IGF2 inhibition synergizes robustly with anti–PD-1 immunotherapy, providing a rationally designed combinatorial strategy to counteract adaptive immune resistance in NSCLC.

Previous studies have largely emphasized CAF-derived cytokines such as CXCL12 and TGF-β that block CD8⁺ T cell trafficking, as well as checkpoint ligands like PD-L1 that directly attenuate TCR signaling (24). Metabolic reprogramming by CAFs is a well-documented barrier; lactate, kynurenine, prostaglandin E2, and adenosine impose energetic stress and exhaustion on infiltrating T cells (28–30). However, the precise endogenous signaling molecules in CAFs that mediate CD8⁺ T cell dysfunction during tumor immune evasion in LTME remain poorly understood. We systematically integrated scRNA-seq data from 84 NSCLC specimens and subsequently stratified these samples based on CD8⁺ T cell functional states, thereby enabling comprehensive evaluation of dynamic interactions between CAF subsets and CD8⁺ T cells. Our analyses revealed notable up-regulation of endogenous IGF2 expression specifically in CAFs within the LTME exhibiting CD8⁺ T cell dysfunction, which showed strong correlation with poor clinical outcomes. The mechanistic drivers underlying this CAF-specific IGF2 overexpression remained elusive, prompting further investigation that uncovered a distinctive hierarchical tumor-stroma-immune regulatory cascade wherein tumor cell–derived TGF-β, rather than autonomous CAF mechanisms, served as the predominant inducer of IGF2 expression in CAFs. This tumor cell–instructed IGF2 induction establishes a fundamental molecular link between TGF-β signaling in LTME and CAF-mediated immunosuppressive functions. Collectively, these findings illuminate a previously unidentified tumor cell–CAF–CD8⁺ T cell signaling axis that provides an advanced conceptual framework for understanding how neoplasms orchestrate immune evasion through hierarchical modulation of stromal cell functions.

The antigen presentation system principally relies on MHC-I and MHC-II molecules, which exhibit substantial differences in terms of structural composition, functional characteristics, and cellular distribution patterns (31). Specifically, MHC-I molecules predominantly present endogenous antigens to activate CD8⁺ T cells, whereas MHC-II molecules primarily facilitate the presentation of exogenous antigens for CD4⁺ T cell activation (32). Within the TME, malignant cells typically express MHC-I molecules to enable CD8⁺ T cell–mediated recognition and cytotoxic responses; however, as an immune evasion strategy, tumor cells frequently down-regulate or completely lose MHC-I expression (13). In contrast, professional antigen-presenting cells such as dendritic cells mainly use MHC-II molecules to activate CD4⁺ T cells, thereby orchestrating adaptive immune responses (31). Emerging evidence demonstrates that various nonmalignant cell populations including macrophages and endothelial cells can also express MHC-I molecules and may consequently participate in CD8⁺ T cell activation through this mechanism (16). Nevertheless, research on CAFs has historically focused on their MHC-II–mediated antigen presentation capabilities, leaving the potential involvement of MHC-I molecules in CAF-mediated tumor immune regulation largely unexplored. In this study, through integrated transcriptomic and scRNA-seq analyses, we identified that IGF2 can notably down-regulate MHC-I signaling in CAFs, subsequently impairing CD8⁺ T cell activation and antitumor functions while promoting tumor progression. This groundbreaking finding challenges the conventional paradigm that CAFs exclusively mediate antigen presentation through MHC-II molecules, thereby providing mechanistic insights into how nontumor cells, particularly CAFs, contribute to tumor immune evasion via MHC-I deficiency.

The expression of MHC-I molecules in tumor cells is precisely regulated by diverse signaling molecules and pathways, with IFN-γ and its downstream Janus kinase–STAT signaling cascade representing the most well-characterized activation mechanism that potently up-regulates MHC-I expression (33). Furthermore, activation of the nuclear factor κB pathway and ataxia telangiectasia mutated (ATM)/ATM and Rad3-related (ATR)-mediated DNA damage response pathways, as well as epigenetic modifications including histone demethylase Jumonji C domain-containing (JMJD) regulation, has also been implicated in positively modulating MHC-I expression (34). Conversely, several oncogenic signaling pathways prominently suppress MHC-I expression, encompassing aberrant activation of epidermal growth factor receptor–RAS–mitogen-activated protein kinase signaling, hyperactivated β-catenin/Wnt pathways, TGF-β/Smad transduction, and excessive expression of immune checkpoint molecules such as PD-L1 (35, 36). Notably, how endogenous IGF2 signaling regulates MHC-I expression specifically in CAFs remains poorly understood. Our study uncovers a previously unrecognized epigenetic mechanism wherein IGF2-mediated activation of the MYC/DNMT1 pathway in CAFs suppresses MHC-I antigen presentation through STAT1 methylation. Notably, this finding contrasts with our earlier observations in breast and colorectal cancers, where IGF2 primarily promoted immune exclusion and T cell dysfunction via CAF-derived CXCL12 and PD-L1 up-regulation, respectively (24). In this study, we found that loss of IGF2 in CAFs did not substantially reduce PD-L1 expression in lung tumors. This finding suggests that, in lung cancer, IGF2 in CAFs suppresses CD8⁺ T cell cytotoxicity primarily by disrupting antigen presentation machinery rather than by modulating PD-L1 expression. Hence, our results further reveal an organ-specific mode of IGF2-mediated immunosuppression. Although our findings establish that IGF2-deficient CAFs exhibit enhanced surface MHC-I antigen presentation and improved CD8⁺ T cell activation, the intracellular mechanisms governing antigen acquisition, processing, and loading onto MHC-I molecules remain incompletely understood. In particular, whether CAFs acquire tumor antigens through phagocytosis, macropinocytosis, gap junction–mediated transfer, or additional stromal communication mechanisms was not directly investigated in the present study. Likewise, the extent to which CAFs have functional cross-presentation machinery, including proteasomal processing and TAP-dependent peptide loading pathways, and whether these processes operate independently or in coordination with professional antigen-presenting cells such as dendritic cells require further investigation (31, 37). Therefore, our current data primarily support a functional role for the IGF2-MYC-DNMT1-STAT1 axis in suppressing antigen presentation competency in CAFs rather than fully defining the molecular choreography of CAF-mediated antigen processing.

To investigate the functional role of CAF-derived IGF2 within the LTME, we used a fibroblast conditional knockout strategy by crossing S100a4-CreERT mice with Igf2^flox/flox mice to generate inducible fibroblast Igf2-deficient mice. Nevertheless, several limitations of this genetic targeting strategy should be acknowledged. Although our lineage-tracing and scRNA-seq analyses demonstrated that recombination in the LLC model was predominantly enriched in CAF populations with minimal reporter activity detected in macrophage, endothelial, and epithelial compartments, S100a4/FSP1 is not an absolutely CAF-specific marker and has been reported to be expressed in additional stromal and immune cell populations under certain pathological conditions (38, 39). Therefore, while our data strongly support a major CAF-autonomous role for IGF2 in regulating antigen presentation and CD8⁺ T cell dysfunction, we cannot completely exclude potential contributions from minor off-target recombination events. Future studies using other orthogonal fibroblast lineage drivers, such as Col1a2-CreERT or Pdgfra-Cre systems, will be important to further validate the fibroblast-specific functions of IGF2 within the LTME.

Immunotherapy has emerged as a promising antitumor treatment strategy, yet its clinical benefits remain limited to a minority of patients while the majority exhibit poor therapeutic responses or develop resistance, highlighting the urgent need to identify novel targets capable of enhancing immune checkpoint blockade efficacy (40, 41). In this study, we demonstrate that both conditional IGF2 knockout and pharmacological inhibition markedly potentiate the antitumor effects of anti–PD-1 therapy in the LLC lung carcinoma model, as evidenced by reduced tumor burden and prolonged survival, suggesting Chromceptin’s potential as an immunotherapeutic adjuvant to improve ICB outcomes although its efficacy across diverse tumor types requires further validation. Recognizing that tumor progression results from complex gene network alterations rather than single gene mutations, we systematically developed an IFS based on the IGF2/MYC/DNMT1 regulatory axis in CAFs to overcome the limitations of previous single-gene predictive approaches. This signature exhibits strong predictive performance in assessing LTME characteristics, long-term clinical outcomes, and potential ICB responsiveness, offering an innovative stratification approach for patients with NSCLC that reflects tumor-immune-stromal cross-talk dynamics and key resistance-associated epigenetic modifications.

MATERIALS AND METHODS

Cell culture

The murine LLC cell line was obtained from the American Type Culture Collection (CRL-1642), and the TC-1 cell line was obtained from BeNa Culture Collection (BNCC341334). Both cell lines were authenticated using standard procedures. Cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM; 11965092, Gibco) supplemented with 10% fetal bovine serum (FBS; SH30070.03, HyClone) and 1% penicillin-streptomycin solution (15140122, Gibco). Cultures were incubated under controlled humidified conditions (37°C, 5% CO₂) in a cell culture incubator, with medium replacement performed daily to ensure optimal cell viability.

Animals

Six- to 8-week-old male and female C57BL/6J-Igf2^flox/floxS100a4^creERT mice, CD45.1; OT-I mice (Cyagen Biosciences, Suzhou, China), S100a4^creERT mice (Shanghai Model Organisms Center, Shanghai, China), and WT C57BL/6J littermates were maintained under specific pathogen–free conditions with controlled environmental parameters (21° to 23°C, 40 to 60% relative humidity, 12-hour light-dark cycle). Conditional IGF2 knockout mice were generated by crossing Igf2^flox/flox with S100a4^creERT strains, followed by PCR-based genotyping of tail-derived genomic DNA.

For Cre recombinase activation, tamoxifen (T5648, Sigma-Aldrich, dissolved in corn oil) was administered via daily intragastric gavage at 60 mg/kg body weight for 5 consecutive days before tumor inoculation. All animal experiments were conducted in accordance with institutional ethical guidelines for laboratory animal welfare and were approved by the Animal Care and Use Committee of Chongqing Medical University (approval number: IACUC-CQMU-2025-09019), following the Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011).

Tumor models

To establish tumor models, we subcutaneously inoculated 2.0 × 10⁵ LLC cells or TC-1 cells suspended in 100 μl of phosphate-buffered saline (PBS) into the right hind flank of tamoxifen-pretreated Igf2^flox/flox control mice and Igf2^flox/floxS100a4^CreERT conditional knockout mice (n = 5 to 8 per group), with tumor growth kinetics monitored through serial measurements using digital calipers from day 5 to 7 postinoculation (volume = 0.5 × length × width²). As previously described (42), a murine orthotopic lung tumor model was established by intravenous injection of 2.0 × 10⁵ LLC cells via the tail vein. Lung tissues were subsequently subjected to hematoxylin and eosin (H&E) staining, and the proportion of tumor lesion area relative to the total lung tissue area was quantified. In the LLC/CAF coimplantation model, LLC cells and CAFs were mixed at a 1:1 ratio and subcutaneously inoculated into mice. When the tumors reached ∼2 mm³ in volume, 5.0 × 10⁶ CD45.1⁺ OT-I cells were adoptively transferred into the mice via tail vein injection. For therapeutic evaluation, mice received either immunoglobulin G (IgG) isotype control or anti–PD-1 blocking antibody (10 mg/kg, ip every 3 days), while CD8⁺ T cell–depleted cohorts were pretreated with anti-CD8α antibody (clone 53-6.7, 10 mg/kg ip 3 days before inoculation, followed by weekly maintenance doses), with depletion efficiency (>95%) confirmed by flow cytometry analysis of peripheral blood mononuclear cells (CD45⁺ live cells assessed for CD4/CD8 expression). Animals were humanely euthanized by gradual-fill CO₂ asphyxiation (20% chamber volume/min) followed by cervical dislocation upon reaching predefined endpoints (tumor volume ≥ 2000 mm³ or ulceration), with all procedures conducted in strict compliance with institutional animal care guidelines.

Human specimens

Fresh LUAD tumor specimens were obtained from treatment-naïve patients during diagnostic procedures. For cellular composition analysis, tissue samples underwent enzymatic digestion with collagenase to generate single-cell suspensions, enabling characterization of IGF2 expression patterns across TME subpopulations. Parallel specimens were fixed in paraformaldehyde and embedded in paraffin blocks for histological examination. These preparations facilitated subsequent investigations of (i) spatial coexpression patterns between IGF2 and GZMB through immunohistochemistry and (ii) cellular colocalization of IGF2 with α-SMA via multiplex immunofluorescence analysis. The study protocol involving human specimens and associated clinical data was approved by the Institutional Review Board of the First Affiliated Hospital of Chongqing Medical University (approval number: 2025-161-01). Written informed consent was obtained from all participating subjects or their legally authorized representatives in accordance with institutional guidelines and the Declaration of Helsinki principles.

RNA-seq analysis

Total RNA was extracted from WT and Igf2^−/− CAFs using TRIzol Reagent (15596026CN, Thermo Fisher Scientific), with RNA integrity verified using an Agilent 2100 Bioanalyzer (Agilent Technologies), where only samples with RNA integrity number ≥ 8.0 were selected for downstream analysis. Strand-specific RNA sequencing libraries were prepared by Cosmos Wisdom Genomics (Zhejiang, China) using the NEBNext Ultra II Directional RNA Library Prep Kit (New England Biolabs, Ipswich, MA, USA), followed by 150-bp paired-end sequencing on an Illumina NovaSeq 6000 platform (Illumina, San Diego, CA, USA).

Raw sequencing data in FASTQ format underwent quality control assessment using FastQC v0.12.1, followed by alignment to the GRCm38 reference genome (mm10) using STAR v2.7.10b with default parameters. Gene-level read counts were quantified using HTSeq v2.0.2 in union-counting mode. Transcript abundance was normalized as transcripts per million (TPM) values implemented through the tximport R package with transcript length adjustment. Differential expression analysis was performed using DESeq2 v1.48.1 (Bioconductor) applying Wald test statistics, where genes with absolute log₂ fold change > 1, adjusted P value < 0.05 (Benjamini-Hochberg correction), and baseline expression (mean log₂TPM) ≥ 1 in at least one comparison group were considered notably differentially expressed. The resulting differentially expressed genes were subjected to functional enrichment analysis of signaling pathways by GSEA, while the antigen presentation–related signature score was assessed using GSVA.

Primary CAF isolation and culture

Primary CAFs were isolated from human LUAD tissues and mouse tumors using established protocols (43). Briefly, tissues were washed with cold sterile PBS, minced into small fragments, and enzymatically digested in Advanced DMEM/F-12 medium supplemented with type IV collagenase (500 U/ml) and hyaluronidase (100 U/ml) for 1 hour at 37°C under continuous agitation in a thermoshaker. The resulting cellular suspension was passed through a 70-μm cell strainer to obtain single-cell suspensions. The harvested cells were then cultured in Advanced DMEM/F-12 medium supplemented with 10% FBS in 100-mm tissue culture plates. Murine tumor–derived CAFs were isolated using the same methodology. Fibroblast identity was confirmed at the protein level by assessing surface marker expression via Western blotting and flow cytometry analysis.

Isolation, activation, and expansion of primary T cells

OT-I T cells were isolated from the spleens of 6- to 12-week-old male or female CD45.1; OT-I mice. Cells were cultured in RPMI 1640 medium (Gibco) supplemented with 10% FBS (Gibco), 0.1 mM non-essential amino acids (Gibco), 1% GlutaMAX (Gibco), 10 mM Hepes (Gibco), 1 mM sodium pyruvate (Gibco), 50 μM β-mercaptoethanol (Gibco), 1% penicillin-streptomycin (Gibco), and recombinant mouse IL-2 (20 U/ml; 212-12-20UG, PeproTech), as indicated. For activation, OT-I T cells were stimulated with either OVA257-264 peptide (1 μg/ml; 13-5743-82, Invitrogen) or anti-mouse CD3 (2 μg/ml; 100314, BioLegend) plus anti-mouse CD28 (1 μg/ml; 122004, BioLegend) monoclonal antibodies. Cells were maintained at 37°C in a humidified incubator with 5% CO₂ for 48 hours.

Cell isolation

Freshly excised LLC tumor specimens were meticulously minced into uniform 1- to 2-mm³ fragments using sterile iris scissors under aseptic conditions. Tissue fragments were subjected to enzymatic digestion in DMEM (11965092, Gibco) supplemented with Liberase TH (2 mg/ml; 590706, Roche) and incubated for 45 min at 37°C under continuous orbital agitation (100 rpm) to ensure homogeneous dissociation. The resulting cell suspension was sequentially filtered through a 70-μm nylon mesh strainer (BS-70-XBS, Biosharp) to remove undigested debris, followed by erythrocyte depletion using ammonium-chloride-potassium (ACK) lysing buffer (A1049201, Gibco). After three washing cycles in PBS (14190144, Gibco) with centrifugation at 400g for 5 min (4°C), the purified single-cell suspensions were resuspended in cold staining buffer for subsequent flow cytometry analysis.

Flow cytometry analysis

Cell viability discrimination was systematically performed through sequential staining with Fixable Viability Dye eFluor 450 (65-0863-14, Invitrogen) followed by Fc receptor blockade using purified anti-CD16/32 antibody (clone 93, 101301, BioLegend) for 15 min at 4°C before surface marker staining to minimize nonspecific antibody binding. Following three washing cycles with ice-cold PBS containing 2% FBS, cell surface antigen detection was achieved through incubation with specific fluorescently conjugated monoclonal antibodies for 30 min at 4°C in the dark. Subsequent intracellular analysis required distinct processing protocols: transcription factor detection used the Foxp3/Transcription Factor Staining Buffer Set (00-5523-00, eBioscience) for nuclear membrane permeabilization and staining with anti-FOXP3 antibody (clone FJK-16s, 407-5773-82, eBioscience), while cytokine detection necessitated prior 4-hour ex vivo activation with Cell Stimulation Cocktail (00-4975-93, eBioscience) containing phorbol 12-myristate 13-acetate/ionomycin and protein transport inhibitors, followed by sequential fixation, permeabilization, and intracellular staining with anti–IFN-γ (clone XMG1.2, 505808, BioLegend), anti-PRF1 (clone B-D48, 308103, BioLegend), anti-GZMB (clone GB11, 515403, BioLegend), and anti–TNF-α (clone MP6-XT22, 506322, BioLegend). Flow cytometry antibody information is listed in table S1. All stained samples were immediately acquired on a BD FACSCanto II flow cytometer (BD Biosciences) within 24 hours of preparation, with instrument compensation and data acquisition performed using BD FACSDiva software (v8.0.1), followed by rigorous gating strategy implementation and quantitative analysis using FlowJo software (v10.5.3) with Boolean combinations of fluorescence parameters for comprehensive immune subset characterization.

scRNA-seq and bioinformatic analysis of LLC tumors

Single-cell suspensions were prepared from freshly resected LLC tumors obtained from 8-week-old female Igf2^cKO mice and WT littermate controls (n = 3 mice per genotype). To reduce biological variability, tumors from each genotype were pooled at equimolar ratios and processed immediately (within 30 min postresection) to preserve cellular integrity. Tissue specimens were first mechanically dissociated into 1- to 2-mm³ fragments using sterile iris scissors, followed by enzymatic digestion in DMEM medium containing Liberase TH (2 mg/ml; Roche, 590706) at 37°C for 45 min with continuous shaking (100 rpm). The resultant cell suspension was sequentially filtered through 70-μm cell strainers to remove undigested tissue aggregates and subjected to three washing steps in PBS supplemented with 0.05% bovine serum albumin (Sigma-Aldrich, A9418). Cell viability was determined by Trypan blue exclusion using a Countess II FL automated cell counter (Thermo Fisher Scientific), with only samples demonstrating >85% viability proceeding to subsequent analysis. The scRNA-seq libraries were constructed using the 10X Genomics Chromium Next GEM Single Cell 3′ v3.1 kit (1000121) according to the manufacturer’s instructions and sequenced on an Illumina NovaSeq 6000 platform with 150-bp paired-end reads.

Bioinformatic analysis

Raw sequencing data were processed using Cell Ranger v7.1.0 for demultiplexing and alignment to the mm10 reference genome (GENCODE vM25 annotation). Quality control and downstream analyses were performed in Seurat v5.0.1 with the following filtering parameters: Cells with <200 or >8000 detected genes were excluded, along with those exhibiting mitochondrial gene content >10% or hemoglobin gene expression >5%.

Dimensional reduction was performed using principal components analysis on the top 2000 highly variable genes, followed by batch effect correction using Harmony (θ = 2). The processed data were then visualized via Uniform Manifold Approximation and Projection (UMAP) projection and clustered using the Leiden algorithm (resolution = 0.8). Differentially expressed genes across clusters were identified using the Wilcoxon rank-sum test (|log₂ fold change| > 0.25, adjusted P value < 0.05) to facilitate cell type annotation. To investigate intercellular communication patterns, we used the CellChat package to analyze ligand-receptor interactions between distinct cell populations. Cellular differentiation states were assessed using CytoTRACE, an algorithm that predicts differentiation potential from single-cell transcriptomic data. For CAF subpopulations, GSEA was performed to identify markedly enriched signaling pathways.

For tumor microenvironment scoring analyses, patients were stratified into IGF2-high and IGF2-low groups using a cohort-specific median expression cutoff. In the LUSC cohort, the median IGF2 expression value was 0.205, whereas in the LUAD cohort, the median was 0.192. Samples with IGF2 expression values greater than the respective cohort median were assigned to the IGF2-high group, and those with values less than or equal to the median were classified as IGF2 low. The median was calculated after filtering for samples with complete expression data.

Immunohistochemistry and immunofluorescence staining

Immunohistochemistry (IHC) staining was performed on 4-μm formalin-fixed paraffin-embedded tumor sections by sequential deparaffinization in xylene and rehydration through graded ethanol, with antigen retrieval achieved through pressure cooking in citrate buffer (pH 6.0, 120°C, 15 min) before quenching endogenous peroxidase activity with 3% H₂O₂; sections were then blocked with 5% normal goat serum before overnight incubation at 4°C with primary antibodies targeting GZMB (ab255598, Abcam) and IGF2 (MA5-17096, Thermo Fisher Scientific), followed by detection using biotinylated secondary antibodies, avidin-biotin-peroxidase complexes, and 3,3′-diaminobenzidine (ab64238, Abcam) with hematoxylin counterstaining. For dual immunofluorescence, sections were permeabilized with 0.1% Triton X-100, blocked with 5% donkey serum, and incubated with anti-IGF2 and anti–α-SMA (ab7817, Abcam) antibodies overnight at 4°C, subsequently probed with DyLight 488/594–conjugated secondary antibodies (1:500) and counterstained with 4′,6-diamidino-2-phenylindole (DAPI) before confocal microscopy imaging (Leica software v4.7.0), while quantitative assessment included IHC H-scoring and ImageJ-based (v1.8.0.112) fluorescence intensity analysis with background correction.

In situ multiplex immunofluorescence staining

Paraffin-embedded LUAD tissue sections were processed using the OPAL multiplex immunofluorescence staining system (PerkinElmer). Sections were incubated at 62°C for 3 hours, followed by automated deparaffinization and rehydration on a Leica Bond RX stainer. Staining was performed through multiple sequential cycles of primary antibody incubation and tyramide signal amplification according to the manufacturer’s recommended antibody dilutions. Between each primary antibody cycle, signal amplification was achieved using an Opal Polymer horseradish peroxidase (HRP)–conjugated secondary antibody in combination with tyramide-based fluorophore deposition. The fluorophores applied sequentially were Opal 480, Opal 520, Opal 570, Opal 620, and Opal 690, with DAPI used for nuclear counterstaining. Protein expression was quantified on the basis of either staining intensity or the proportion of positively stained cells. Staining intensity was scored as follows: 0, negative; 1, weak; 2, moderate; and 3, strong. The proportion of positive cells was semiquantitatively evaluated across multiple fields. All scoring was independently performed by two blinded observers. Multispectral images were acquired using the PerkinElmer Vectra imaging system.

Western blotting

Western blot analysis was performed according to previously established protocols (44). Briefly, cells were lysed in radioimmunoprecipitation assay buffer supplemented with protease inhibitors, and protein concentrations were quantified using a bicinchoninic acid assay. Equal amounts of protein were separated by 10% SDS–polyacrylamide gel electrophoresis and subsequently transferred electrophoretically onto polyvinylidene fluoride membranes. Following transfer, membranes were blocked with 5% nonfat dry milk prepared in tris-buffered saline containing 0.1% Tween 20. Primary antibodies were then applied overnight at 4°C, including anti-IGF2 (1:1000; MA5-17096, Thermo Fisher Scientific), anti–MHC class I (1:1000; 76828, Cell Signaling Technology), Phospho-STAT1 antibody (1:1000; ET1611-20, HUABIO), glyceraldehyde phosphate dehydrogenase (GAPDH) antibody (1:1000; 10494-1-AP, Proteintech), and anti–β-actin (1:5000; 4970, Cell Signaling Technology) as a loading control. After extensive washing, membranes were incubated for 1 hour at room temperature with HRP-conjugated secondary antibodies (1:5000; 7074 and 7076, Cell Signaling Technology). Protein bands were visualized using enhanced chemiluminescence substrate (34579, Thermo Fisher Scientific) and imaged on a ChemiDoc Imaging System (Bio-Rad). Densitometric analysis was conducted using Image Lab software (v6.0, Bio-Rad), with band intensities normalized to β-actin.

Quantitative reverse transcription polymerase chain reaction

Total RNA was extracted from CAFs using TRIzol reagent according to the manufacturer’s instructions. RNA concentration and purity were assessed using a NanoDrop spectrophotometer. cDNA was synthesized from total RNA using a reverse transcription kit, and quantitative real-time PCR was performed using SYBR Green qPCR Master Mix on a real-time PCR detection system. Relative gene expression was calculated using the 2^-ΔΔCt method and normalized to GAPDH. Primer sequences used for RT-qPCR are listed in table S2.

Chromatin immunoprecipitation–qPCR assay

The chromatin immunoprecipitation (ChIP) assay was performed using the SimpleChIP Enzymatic Chromatin IP Kit (9002S, Cell Signaling Technology) in accordance with the manufacturer’s protocol. Briefly, cells were cultured to a density of approximately 1 × 10⁷ and cross-linked with 1% formaldehyde. After sample collection, chromatin was digested using micrococcal nuclease. Subsequently, the nuclear membrane was disrupted by several rounds of sonication. The resulting DNA fragments were verified to range between 150 and 900 bp. ChIP was then carried out using control IgG (98136-1-RR, Proteintech), anti-MYC (80845-1-RR, Proteintech), or anti-DNMT1 (24206-1-AP, Proteintech) primary antibodies. Following washing and reverse cross-linking, the eluted DNA was quantified by qPCR. The primer sequences used were as follows: for the DNMT1 promoter, forward 5′-ACACAGCTAGGGTGTTTTGGT-3′ and reverse 5′-CGTGTGGTGTAAGATGGCCT-3′; and for the STAT1 promoter, forward 5′-TGCCCGGTACTTTCGGTTTT-3′ and reverse 5′-AGCTCGAGTCTGGGCAAATC-3′.

Gene knockdown and overexpression

Lentiviral vectors encoding short hairpin RNAs (shRNAs) targeting IGF2 and small interfering RNAs (siRNAs) targeting MYC, STAT1, and DNMT1, as well as a STAT1 overexpression plasmid, were purchased from GeneCreate (Wuhan, China). Sequences of siRNAs and overexpression constructs are listed in table S3. Plasmids encoding OVA, EGFP, and ZsGreen were also obtained from GeneCreate. Lentiviral particles were produced in human embryonic kidney (HEK) 293T cells by cotransfecting the transfer vector with the packaging plasmid psPAX2 and the envelope plasmid pMD2.G at a ratio of 3:2:1 using PolyJet DNA transfection reagent (SL100688, SignaGen), according to the manufacturer’s instructions. Briefly, HEK293T cells were transfected with 5 μg of pLKO.1-puro shRNA lentiviral expression vector together with the corresponding packaging plasmids and then incubated at 37°C in a humidified atmosphere containing 5% CO₂. The culture medium was replaced with fresh DMEM after transfection, and viral supernatants were collected 48 hours later and passed through a filter. For lentiviral transduction, CAFs were cultured in RPMI 1640 medium supplemented with polybrene (TR-1003-G, Merck) and incubated with viral supernatants for 72 hours, followed by puromycin selection (2 μg/ml) for 7 days to establish stable knockdown cell lines. Knockdown efficiency was subsequently confirmed by RT-qPCR using PrimeScript SYBR Green Master Mix (RR014A, Takara). In some experiments, OVA or ZsGreen was introduced into LLC cells to generate corresponding stable cell lines. In other experiments, OVA and EGFP were introduced into CAFs to generate engineered cells for subsequent functional assays.

Methylated DNA immunoprecipitation

The methylated DNA immunoprecipitation assay was performed following established methodologies as previously described (45). Genomic DNA was first extracted and purified from cultured cells. The purified DNA was subsequently fragmented into segments ranging from 200 to 1000 bp via ultrasonication. These DNA fragments were then denatured at 95°C to generate single-stranded DNA, which was incubated with a 5-methylcytosine antibody (HA601350, HUABIO) to form DNA–5-mC immunocomplexes. The complexes were captured using magnetic beads, and the immunoprecipitated DNA was lastly purified by phenol/chloroform extraction and analyzed by quantitative PCR. Of note, the primer sequences used were as follows: for the STAT1, forward 5′-GTCATCCCGCAGAGAGAACG-3′ and reverse 5′-AGGGTATGGAGCAGAGCTGA-3′.

Statistics

Statistical analyses were conducted using GraphPad Prism 9.5.0 and R 4.3.2, with intergroup comparisons of continuous variables performed using two-sided unpaired Student’s t tests unless specifically indicated; multifactorial comparisons involving three or more groups were assessed by one-way analysis of variance (ANOVA), while tumor growth kinetics were evaluated by two-way ANOVA, each followed by appropriate post hoc tests where applicable, with survival analyses performed using log-rank (Mantel-Cox) testing. A P value threshold of <0.05 defined statistical significance, with exact P values provided in all relevant figure annotations, where data are represented either as mean ± SEM or median with interquartile range as detailed in figure legends. The experimental design incorporated multiple comparisons corrections when appropriate, with all experiments replicated in at least two independent biological iterations.

Supplementary Materials

This PDF file includes:

Figs. S1 to S10

Tables S1 to S3