Targeting tumor-specific T cells with LAG3-directed interleukin-2 prevents T-cell exhaustion and reinvigorates antitumor immunity Authors: Xiaohong Yu, Huiping Liao, Jiaoyun Lv, Yu Sun, Ruiqi Zhang, Yiwei Chen, Yifan Lin, Lulu Liu, Shijie Li, Hui Tang, Panpan Jia, Bin Shao, Zaopeng Yang, Yang-Xin Fu, Zhenhua Ren Published: 2026 ABSTRACT AbstractLAG3 is a critical inhibitory receptor that is highly enriched on exhausted T cells within the tumor microenvironment (TME), where it acts as a key driver of T-cell exhaustion—an archetypal barrier to robust antitumor immunity. In a colon cancer model, LAG3+CD8+ tumor-infiltrating lymphocytes (TILs) constitute the predominant type of tumor-specific T cells but exhibit defective IL2 signaling. To address whether exogenous IL2 replenishment unpins their dysfunction, we engineered LAG3-LaIL2 (low-affinity IL2), a fusion protein that selectively delivers IL2 to LAG3+CD8+ TILs. LAG3-LaIL2 expanded pre-exhausted tumor-specific CD8+ T cells, reprogrammed their exhaustion trajectory toward an intermediate effector state, and prevented terminal exhaustion, leading to tumor regression and prolonged survival in mice. Mechanistically, LAG3-LaIL2 restored IL2R-JAK3-STAT5 signaling by upregulating the high-affinity IL2 receptor subunit CD122, thereby restoring TIL functionality. Furthermore, LAG3-LaIL2 amplified tumor-specific effector and memory T cells in draining lymph nodes, enabling systemic antitumor immunity against distal tumors and preventing tumor recurrence. Collectively, our findings identify LAG3-LaIL2 as a precision immunotherapy that specifically targets exhausted TILs while restricting IL2 exposure to nontarget cells, thereby enhancing both the efficacy and safety of this approach. This approach provides a translatable strategy to overcome T-cell exhaustion in solid tumors and represents a promising avenue to improve clinical outcomes in cancer patients.Subject terms: Tumour immunology, Immunotherapy Introduction Tumor-reactive CD8+ T cells have the inherent capacity to eradicate cancer cells, as evidenced by the clinical success of immune checkpoint blockade (ICB) therapies targeting pathways such as the PD-1/PD-L1 pathway.1 However, tumors frequently evade immune elimination via compensatory upregulation of alternative inhibitory receptors, including LAG3 and TIM3, which drive T-cell dysfunction and exhaustion.2–4 Lymphocyte activation gene-3 (LAG3; CD223), a coinhibitory receptor enriched in exhausted and dysfunctional tumor-infiltrating CD8+ T cells, synergizes with PD-1 to suppress T-cell receptor (TCR) signaling and cytokine production, thereby establishing a resilient immunosuppressive axis in the TME.3,5–7 Notably, dual anti-PD-1/anti-LAG3 blockade therapy (Opdualag®) was approved by the FDA in 2022 for unresectable or metastatic melanoma, demonstrating that LAG3 inhibition can enhance antitumor immunity when combined with PD-1 blockade.8–10 This regulatory milestone demonstrates the therapeutic potential of cotargeting nonredundant inhibitory pathways, addressing the immune evasion mechanisms that often constrain responses to single-agent checkpoint therapies. While dual PD-1/LAG3 blockade has shown promise for overcoming resistance to single-agent ICBs, many patients still fail to respond because of the epigenetic and functional heterogeneity of exhausted T-cell states.11 This highlights the urgent need for novel immunomodulatory strategies to dismantle treatment resistance mechanisms. Exhausted CD8+ T cells within the TME exhibit significant heterogeneity, encompassing early dysfunctional subsets with plastic epigenetic profiles amenable to reprogramming and late-stage exhausted populations characterized by stable fixed epigenetic states that confer functional irreversibility.12 A critical subset of “pre-exhausted” LAG3+CD8+ T cells retains partial functionality and reprogramming potential, offering a therapeutic window to restore antitumor immunity. IL2, a potent T-cell growth factor, could theoretically rejuvenate these cells by activating the IL2R-JAK3-STAT5 axis, which is often impaired in exhausted T cells. One promising approach involves stimulating early dysfunctional LAG3+ pre-exhausted T cells with IL2. IL2 exerts its primary biological effects through CD8+ T cells and NK lymphocytes.13 However, the therapeutic application of IL2 is hindered by its short half-life in vivo and its propensity to promote the differentiation trajectory, immunosuppressive potential, and homeostasis of regulatory T cells (Tregs), which directly counteract the cytotoxic antitumor potential of effector T cells.14–16 Furthermore, the substantial doses required for therapeutic efficacy pose a risk of lethal toxicity.13 To address these limitations, strategies that selectively deliver IL2 to effector cells within tumor tissues while concomitantly minimizing unintended systemic toxicity are urgently needed. Current therapeutic approaches typically rely on targeting tumor antigens to deliver IL2 to the tumor microenvironment, but these methods risk off-target effects on tumor-specific CD8+ T cells.17–19 To overcome these challenges, we propose two key strategies: targeted delivery of IL2 to LAG3+CD8+ T cells within the TME rather than tumor cell targeting and the use of a low-affinity IL2 variant (LaIL2)20 with reduced binding to both IL2Rα and IL2Rβ to allow better targeting of LAG3+ dysfunctional tumor-infiltrating lymphocytes (TILs), leading to better tumor control. In this study, we designed a T-cell-specific fusion protein (LaIL2 linked to an anti-LAG3 antibody, LAG3-LaIL2) designed to enhance efficacy while minimizing toxicity. Our work establishes LAG3-LaIL2 as a precision immunotherapy that synergizes the antitumor potency of IL2 with the spatial specificity of LAG3 targeting, addressing key limitations of current ICB and cytokine-based therapies. These findings emphasize the therapeutic potential of reprogramming pre-exhausted T-cell states to overcome resistance in solid tumors. Our study highlights the potential of LAG3-LaIL2 as a novel immunotherapeutic strategy with improved efficacy and reduced toxicity. Results Tumor-specific T cells are predominantly LAG3+CD8+ T cells but become exhausted and exhibit weakened IL2 signaling in advanced cancers To map the location of tumor-specific CD8+ T cells, we analyzed their distribution across the spleen, draining lymph nodes (dLNs), and tumor tissues in MC38 tumor-bearing mice. Compared with those in the spleen and dLN, we observed that tumor-specific CD8+ T cells21,22 were predominantly localized in the TME, with the majority of these antigen-specific cells exhibiting LAG3 coexpression. (Fig. 1a, b and Supplementary Fig. 1a, b). Notably, tumor-specific CD8+ T cells were markedly enriched in the LAG3+CD8+ population relative to LAG3−CD8+ T cells. (Fig. 1c and Supplementary Fig. 1c). LAG3 is preferentially expressed on tumor-reactive T cells and operates as a context-dependent marker, particularly in conditions of chronic antigen exposure, such as the TME, for identifying antigen-specific T cells. It thus represents a promising target for delivering immunomodulatory agents, including cytokines, to intratumoral T cells. Tumor-specific T cells are predominantly LAG3+CD8+ T cells but become exhausted and exhibit weakened IL2 signaling in advanced cancers. a Proportion of tetramer+ cells in the spleen, dLN and tumor tissues of MC38 murine tumor models (n = 5/group). b Proportion of LAG3+ and LAG3- subsets among tetramer+CD8+ T cells from MC38 murine tumors (n = 5/group). c Frequencies of tetramer+ cells within LAG3-CD8+ and LAG3+CD8+ T-cell populations in tumors from MC38 murine tumor models (n = 5/group). d LAG3 expression levels in T-cell clusters from T10 and T20 MC38 tumors. e Differential expression of IL2 receptor-related genes and Stat5a in effector and exhausted CD8+ T-cell clusters between T10 and T20 MC38 tumors. The data are displayed as the mean ± SEM and represent the results from two to three separate experimental repetitions. Statistical significance was evaluated through one-way ANOVA followed by Tukey’s test for multiple comparisons (a–c). The normal distribution of the data was confirmed via the Shapiro‒Wilk test. ***P < 0.001, ****P < 0.0001 To investigate the temporal dynamics of LAG3+CD8+ T-cell states during tumor progression, we isolated TILs from MC38 tumor-bearing mice at early (day 10 postinoculation, T10) and advanced (day 20 postinoculation, T20) stages. Single-cell transcriptomic profiling revealed an expansion of the LAG3+CD8+ T-cell subset as tumors progressed (Fig. 1d). Notably, compared with those at T10, effector and exhausted CD8+ T cells at T20 presented significantly reduced expression of IL2 signaling-associated genes (Il2ra, Il2rb and stat5a) (Fig. 1e and Supplementary Fig. 1d), suggesting that progressive dysfunction in LAG3+CD8+ T cells is accompanied by impaired IL2 responsiveness. To confirm our findings at the protein level, we assessed the surface expression of the IL-2 receptor subunits and LAG3 on CD8+ T cells isolated from MC38 tumors at both the early and late stages. At the late stage, all the IL-2 receptor subunits were significantly downregulated in LAG3+CD8+ T cells (Supplementary Fig. 1e–g). LAG3-LaIL2 mediates the cis-delivery of LaIL2 to intratumoral LAG3+CD8+ T cells To assess the suitability of LAG3 as a therapeutic target for intratumoral CD8+ T cells, we systematically profiled LAG3 expression across CD4+ T helper cells, Tregs, and CD8+ T cells in distinct anatomical compartments—including the spleen, dLN and tumor tissue—using the MC38 murine tumor model. Strikingly, compared with other immune cell subsets, tumor-infiltrating CD8+ T cells exhibited markedly elevated LAG3 expression within the TME (Fig. 2a and Supplementary Fig. 2a, b). Moreover, LAG3 expression on CD8+ T cells in tumor tissue was significantly greater than that in secondary lymphoid organs (spleen and dLN), as indicated by both an increased frequency of LAG3+CD8+ T cells (Supplementary Fig. 2a, b) and increased expression intensity (Fig. 2a). These results highlight the selective enrichment of LAG3+CD8+ T cells within the tumor milieu. LAG3-LaIL2 mediates the cis-delivery of LaIL2 to intratumoral LAG3+CD8+ T cells. a LAG3 expression in CD4+ T helper (Th) cells, Tregs and CD8+ T cells within the spleen, dLN and tumor tissues of MC38 murine tumor model mice (n = 5/group). b Schematic illustration of the anti-LAG3×LaIL2 heterodimer (LAG3×LaIL2) and the LAG3-InFc-LaIL2 homodimer. Binding affinities of InFc-WTIL2, InFc-LaIL2, anti-LAG3-InFc, LAG3×LaIL2, and LAG3-InFc-LaIL2 for Tregs (c) and CD8+ T cells (d) in the dLNs of MC38 murine tumor models (n = 3/group). e–g Binding affinities of InFc-WTIL2, InFc-LaIL2, anti-LAG3-InFc, LAG3×LaIL2, and LAG3-InFc-LaIL2 for LAG3-CD8+ T cells and LAG3+CD8+ T cells in the tumors of MC38 murine tumor models (n = 3/group). The data are displayed as the means ± SEMs and represent the results from two to three separate experimental repetitions. The statistical significance was evaluated through two-way (a, g) or one-way (c–f) ANOVA with Tukey’s procedure for multiple comparisons. The normal distribution of the data was confirmed via the Shapiro‒Wilk test. ***P < 0.001, ****P < 0.0001 To selectively activate intratumoral LAG3+CD8+ T cells via IL2, we engineered LAG3-LaIL2 fusion proteins designed to preferentially deliver IL2 agonists to LAG3+ tumor-reactive T cells. This strategy enables concurrent blockade of the LAG3 inhibitory pathway and activation of IL2 signaling within the same cell through cis-targeted delivery. To enable a broad investigation of drug efficacy and mechanisms of action in both murine and humanized mouse models, we used a single-chain variable fragment (scFv) from an anti–LAG3 antibody (clone S119) that recognizes both human and murine LAG3. To reduce IL-2 binding to Tregs while selectively engaging CD8+ T cells, our laboratory previously engineered a low-affinity IL-2 variant (LaIL-2) containing R38L and F42A mutations, which decreased its affinity for both IL-2Rα and IL-2Rβ.14,23 To this end, we generated and evaluated two formats of LAG3-LaIL2 fusion proteins: a homodimeric construct (LAG3-InFc-LaIL2) and a heterodimeric version (LAG3×LaIL2) (Fig. 2b). For comparative analysis, anti-LAG3-InFc and InFc-WTIL2 were included as positive controls for T-cell targeting. Notably, LAG3-LaIL2 (LAG3-InFc-LaIL2 and LAG3×LaIL2) exhibited minimal binding to peripheral Tregs and CD8+ T cells (Fig. 2c, d, Supplementary Figs. 2c, d and 3a, b). Importantly, among the CD8+ T-cell subsets, the LAG3-LaIL2 construct preferentially bound to intratumoral LAG3+CD8+ T cells, with negligible binding observed in LAG3-CD8+ T cells (Fig. 2e–g and Supplementary Fig. 3b). Moreover, LAG3-LaIL2 demonstrated enhanced specificity for LAG3+CD8+ T cells relative to untargeted InFc-LaIL2 (Fig. 2f, g and Supplementary Fig. 3b). Collectively, these findings indicate that LAG3-LaIL2 effectively and selectively targets intratumoral LAG3+CD8+ T cells, engaging tumor-specific cytotoxic T cells while sparing peripheral Tregs and LAG3-CD8+ T cells. LAG3-InFc-LaIL2 reprograms exhausted T cells by expanding pre-exhausted and intermediate exhausted CD8+ T cells in the tumor microenvironment To elucidate the mechanisms by which LAG3-InFc-LaIL2 modulates the TME and delineate the molecular distinctions between LAG3-InFc-LaIL2 monotherapy and anti-LAG3 + InFc-LaIL2 combination therapy, we assessed therapeutic responses during the early treatment phase. LAG3-InFc-LaIL2 treatment led to a significant increase in both the frequency and absolute number of intratumoral CD3+ T cells (Fig. 3a, Supplementary Figs. 4a–c and 5a). LAG3-InFc-LaIL2 reprograms exhausted T cells by expanding pre-exhausted and intermediate exhausted CD8 + T cells in the tumor microenvironment. a MC38 murine tumor models (n = 5/group) were administered equimolar doses of InFc-LaIL2 plus anti-LAG3-InFc or LAG3-InFc-LaIL2 on days 11 and 14. Five days after the initial therapy, TILs were analyzed for the percentage of CD3+ T cells. b MC38 murine tumor models (n = 5/group) were generated as described in (a). CD3+ T cells were isolated from tumors and processed for scRNA-seq. t-SNE plots depict Seurat-mediated clustering without supervision and the distribution of 16 T-cell clusters. c t-SNE visualization of scRNA-seq clusters of CD3+ T cells from the PBS, LAG3 + LaIL2, and LAG3-InFc-LaIL2 treatment groups. d Bar plot displaying cluster proportions as percentages of cells aligned with the t-SNE plot in (b). e, f MC38 murine tumor models (n = 5/group) were generated as described in (a). The proportion of CD8+ T cells among CD45+ cells (e) and the CD8+/Treg (f) ratio were assessed in TILs. The data are displayed as the means ± SEMs and represent the results from two to three separate experimental repetitions. The statistical significance was evaluated through one-way ANOVA followed by Tukey’s test for multiple comparisons (a, e–f). The normal distribution of the data was confirmed via the Shapiro‒Wilk test. ***P < 0.001, ****P < 0.0001 To further characterize T-cell changes after LAG3-InFc-LaIL2 treatment, we conducted unbiased single-cell RNA sequencing (scRNA-seq) on tumor-isolated CD3+ T cells. Unsupervised clustering identified sixteen transcriptionally distinct T-cell subsets, which were annotated on the basis of canonical gene expression signatures (Fig. 3b and Supplementary Fig. 5b). Pre-exhausted CD8+ T cells (Tex-pre) expressed elevated levels of CD44, Pdcd1, Lag3, Mki67, and Ifng, indicating a proliferative and functional state. Intermediately exhausted effector CD8+ T cells (Tex-int) presented enriched effector-like profiles but lacked Ki67 expression. In contrast, terminally exhausted CD8+ T cells (Tex-term) were characterized by the upregulation of Pdcd1, Lag3, and Havcr2 (genes encoding TIM3), with minimal expression of Ki67 (Supplementary Fig. 5c). Compared with the PBS- or anti-LAG3 plus Fc-LaIL2-treated groups, LAG3-InFc-LaIL2 monotherapy resulted in a marked expansion of Tex-pre and Tex-int populations while reducing the frequency of Tex-term (Fig. 3c, d and Supplementary Fig. 5c). Flow cytometric analysis of TILs conducted five days after treatment further corroborated these findings, revealing increased CD8+ T-cell infiltration and a higher CD8 + /Treg ratio in the LAG3-InFc-LaIL2 group (Fig. 3e, f and Supplementary Fig. 5d, e). Collectively, these results demonstrate that LAG3-InFc-LaIL2 selectively maintained pre-exhausted and intermediately exhausted CD8+ T cells while diminishing terminally exhausted populations, thereby reshaping the TME. LAG3-InFc-LaIL2 treatment prevents terminal exhaustion of intratumoral T cells through the IL2R-JAK3-STAT5 pathway Persistent antigen stimulation within the TME drives T-cell exhaustion, which is characterized by impaired effector activity. To assess whether LAG3-InFc-LaIL2 can reprogram the effector capacity of intermediately exhausted CD8+ T cells (Tex-int), we conducted pathway enrichment analyses. Compared with terminally exhausted T cells (Tex-term), Tex-int subsets presented significant enrichment of genes associated with the T-cell receptor (TCR) signaling pathway, indicating that LAG3-InFc-LaIL2 enhances the tumor-specific T-cell response (Fig. 4a). Correspondingly, a marked increase in tumor-specific CD8+ T cells21,22 was observed following LAG3-InFc-LaIL2 treatment (Fig. 4b, Supplementary Fig. 6a, b). LAG3-InFc-LaIL2 treatment induced significant enrichment of the IFN-γ response hallmark in Tex-int, concomitant with expansion of IFN-γ-producing tumor-specific T cells within the TME (Fig. 4c-e and Supplementary Fig. 6c, d). LAG3-InFc-LaIL2 treatment prevents terminal exhaustion of intratumoral T cells through the IL2R-JAK3-STAT5 pathway. a KEGG analysis of T-cell receptor (TCR) signaling in the CD8+ Tex-int cluster compared with that in the CD8+ Tex-term cluster. b MC38 murine tumor models (n = 6/group) were treated with equimolar doses of InFc-LaIL2 plus anti-LAG3-InFc or LAG3-InFc-LaIL2 on days 11 and 14. Five days after the initial treatment, TILs were analyzed for the proportion of tetramer+CD8+ T cells. c GSEA of IFN-γ signaling in the CD8+ Tex-int cluster compared with that in the CD8+ Tex-term cluster. d Tumor-sorted CD45+ TILs were restimulated with MC38 cells subjected to irradiation for 48 hours. IFN-γ-secreting cells were quantified via an ELISPOT assay (n = 5/group). e CD8+ T cells isolated from tumor tissues were restimulated with ionomycin and PMA for 4 hours with brefeldin A (BFA) supplementation, followed by flow cytometric quantification of IFN-γ-secreting cells (n = 7/group). f Ingenuity pathway analysis (IPA) was used to identify genes whose expression varied and was enriched in the Tex-int cluster of the LAG3-InFc-LaIL2 treatment group compared with the Tex-term cluster in the PBS treatment group. g MC38 murine tumor models (n = 4/group) were treated as described in (b). TILs were analyzed for the proportion of GZMA+ CD8+ T cells. h Tumor-isolated CD8+ T cells (2.5e6) were cocultured with MC38K target cells (5e5) for 72 hours. Cytotoxicity was quantified by LDH released in culture supernatants. GSEA of IL2-STAT5 signaling in the CD8+ Tex-pre cluster compared with that in the CD8+ Tex-int cluster (i) and IL2–STAT5 signaling in the CD8+ Tex-int cluster compared with that in the CD8+ Tex-term cluster (j). k Levels of p-STAT5, STAT5, p-JAK3 and JAK3 proteins in activated PBMCs after stimulation with anti-LAG3-InFc, InFc-LaIL2, or LAG3-InFc-LaIL2 in the presence of anti-CD3. j MC38 tumor-bearing mice (n = 5/group) were treated as described in (b). TILs were analyzed for the proportion of CD122+ cells among CD8+ T cells. The data are displayed as the means ± SEMs and represent the results from two to three separate experimental repetitions. The statistical significance was evaluated through one-way ANOVA followed by Tukey’s test for multiple comparisons (b, d, e, g, h). The normal distribution of the data was confirmed via the Shapiro‒Wilk test. ***P < 0.001, ****P < 0.0001 To further elucidate how LAG3-InFc-LaIL2 modulates Tex-int function, we performed Ingenuity Pathway Analysis (IPA) on differentially expressed genes (DEGs) between Tex-int and Tex-term subsets. The analysis revealed upregulated expression of effector-associated genes, such as GZMA, in the Tex-int cluster of the LAG3-InFc-LaIL2 group (Fig. 4f). In agreement with these findings, LAG3-InFc-LaIL2 treatment significantly expanded the population of GZMA+CD8+ T cells, especially tumor-specific GZMA+CD8+ T cells (Fig. 4g, Supplementary Fig. 6e–h and 7a), and potentiated the tumor-specific cytotoxic capacity of CD8⁺ T cells, as measured by target cell elimination (Fig. 4h), indicating that the observed transcriptional and phenotypic reprogramming underlies enhanced antitumor effector function. Moreover, LAG3-InFc-LaIL2 impeded the differentiation trajectory from Tex-pre and Tex-int toward terminal exhaustion, promoting the transition from Tex-pre to Tex-int while limiting progression to the Tex-term state (Supplementary Fig. 7b). Notably, Tex-pre cells express a high level of LAG3, enabling them to respond directly to LAG3-targeted IL2 agonism. This intervention resulted in the expansion of Tex-pre and Tex-int populations and a corresponding reduction in Tex-term cells. Emerging evidence indicates that constitutive Stat5a signaling can antagonize TOX-driven epigenetic exhaustion and reprogram CD8+ T cells into durable effectors with increased polyfunctionality and antitumor efficacy.24 LAG3-InFc-LaIL2 treatment induced a significant reduction in the frequency of TOX-high cells within the terminally exhausted PD-1⁺TIM-3⁺CD8+ T-cell subset (Supplementary Fig. 7c) and increased the activity of the IL2-STAT5 pathway in both the Tex-pre (vs. Tex-int) and Tex-int (vs. Tex-term) populations (Fig. 4i, j). Consistent with these in vivo single-cell transcriptomic findings, LAG3-InFc-LaIL2 also induced robust phosphorylation of STAT5 and JAK3 in activated PBMCs stimulated with anti-CD3, outperforming either anti-LAG3 or InFc-LaIL2 monotherapies (Fig. 4k), and promoted robust STAT5 phosphorylation in anti-CD3-stimulated mouse TILs (Supplementary Fig. 8a, b). Moreover, treatment with LAG3-InFc-LaIL2 resulted in marked upregulation of CD122 on intratumor CD8+ T cells (Supplementary Fig. 8c, d). Given that IL-2 signaling is known to further induce CD122 expression, these data suggest that LAG3-InFc-LaIL2 may initiate a positive feedback loop that amplifies its own activity, thereby enhancing responses in effector CD8+ T cells. In conclusion, LAG3-InFc-LaIL2 alleviates CD8+ T-cell exhaustion and prevents terminal differentiation through IL2Rβ-JAK3-STAT5 signaling, thereby sustaining an effector-like phenotype in Tex-pre and Tex-int subsets. LAG3-InFc-LaIL2 shows superior efficacy without toxicity To evaluate the therapeutic efficacy and potential toxicity of LAG3-targeted therapies, mice bearing MC38 tumors were intraperitoneally administered equimolar doses of LAG3-InFc-LaIL2, LAG3×LaIL2, or a combination of anti-LAG3 and InFc-LaIL2. LAG3-InFc-LaIL2 demonstrated significantly greater potency in controlling MC38 tumor growth than LAG3×LaIL2 at equimolar doses with minimal toxicity (Fig. 5a, b). In contrast, equivalent doses of combination treatments with anti-LAG3 and InFc-LaIL2 had negligible effects (Fig. 5a, b and Supplementary Figs. 9a–d and 10a). Notably, the antitumor efficacy of LAG3-InFc-LaIL2 was not limited to the MC38 model. In the poorly immunogenic B16-F10 tumor model, LAG3-InFc-LaIL2 exhibited superior antitumor activity compared with anti-LAG3 plus InFc-LaIL2 (Fig. 5c and Supplementary Fig. 8e). Similarly, in the CT26 colon tumor model, LAG3-InFc-LaIL2 demonstrated better tumor control than LAG3-InFc-WTIL2 and mixed treatments did (Fig. 5d, Supplementary Fig. 9f–I and 10b). Serum analysis revealed a greater increase in inflammatory cytokines, such as IFN-γ, following LAG3-InFc-WTIL2 administration than following LAG3-InFc-LaIL2 administration (Fig. 5e). Notably, LAG3-InFc-LaIL2-treated mice presented no toxicity, which was correlated with lower serum IFN-γ levels and better safety profiles (Fig. 5e and Supplementary Fig. 9j). LAG3-InFc-LaIL2 shows superior efficacy without toxicity. a, b MC38 murine tumor models (n = 5/group) were administered equimolar doses of InFc-LaIL2 plus anti-LAG3-InFc, LAG3×LaIL2 or LAG3-InFc-LaIL2 on days 10 and 14. Tumor progression was assessed twice per week (a). The mice were subjected to two weekly body weight monitoring (b). c B16F10 murine tumor models (n = 5/group) were administered equal molar doses of InFc-LaIL2 plus anti-LAG3-InFc or LAG3-InFc-LaIL2 on days 8, 11 and 14. Tumor progression was assessed twice per week. d CT26 murine tumor models (n = 5/group) were treated with equal molar amounts of InFc-LaIL2 plus LAG3-InFc, LAG3-InFc-WTIL2 or LAG3-InFc-LaIL2 on days 8 and 11. Tumor progression was assessed twice per week (d). Serum interferon-γ levels were quantified via CBA one day after treatment (e). f CFPAC1 tumor-bearing mice (n = 5/group) were treated with equal molar amounts of InFc-LaIL2 plus anti-LAG3-InFc or LAG3-InFc-LaIL2 on days 17 and 20. Tumor progression was assessed twice per week. The data are displayed as the means ± SEMs and represent the results from 2 to 3 separate experimental repetitions. The P value was determined via two-way ANOVA (a, c, d, f). The statistical significance was evaluated through one-way ANOVA followed by Tukey’s test for multiple comparisons (e). The normal distribution of the data was confirmed via the Shapiro‒Wilk test. ***P < 0.001, ****P < 0.0001 To further assess antitumor efficacy in a clinically relevant context, a humanized mouse model was generated via the reconstitution of NSG-SGM3 mice with human PBMCs. These mice were implanted with human pancreatic cancer CFPAC1 cells and treated with LAG3-InFc-LaIL2 or anti-LAG3 plus LaIL2. Remarkably, LAG3-InFc-LaIL2 treatment resulted in complete tumor eradication (Fig. 5f). These findings validate the therapeutic potential of LAG3-InFc-LaIL2-based immunotherapy as a viable strategy for the treatment of human cancers. LAG3-InFc-LaIL2 drives local and systemic antitumor immunity via intratumoral CD8+ T cells To investigate the cellular mechanisms underlying the antitumor activity of LAG3-InFc-LaIL2, we assessed the involvement of various immune cell populations, given the broad effects of IL-2R on both NK and CD4+ T cells. We depleted NK cells in wild-type mice via an anti-NK1.1 antibody, which did not affect the antitumor efficacy of LAG3-InFc-LaIL2, indicating that NK cells are not essential for the therapeutic response (Supplementary Fig. 11a). Similarly, depletion of CD4+ T cells with an anti-CD4 antibody revealed that these cells are also not required for treatment efficacy (Supplementary Fig. 11b, c). Single-cell sequencing analysis revealed that LAG3-InFc-LaIL2 increases the proportion of pre-exhausted and inter-exhausted CD8+ T cells relative to terminally exhausted CD8+ T cells, reprogramming CD8+ T cells toward durable, functional effector states within the TME. To further elucidate the underlying mechanisms, we assessed the contribution of CD8+ T cells and lymph node egress in MC38-bearing C57BL/6 mice treated with LAG3-InFc-LaIL2. Notably, the antitumor efficacy of LAG3-InFc-LaIL2 was completely abolished upon CD8+ T-cell depletion (anti-CD8; Fig. 6a) and was only marginally reduced following the administration of FTY720, an inhibitor of lymphocyte egress from secondary lymphoid organs (Fig. 6b and Supplementary Fig. 11d, e). These results indicate that the therapeutic benefit of LAG3-InFc-LaIL2 is dependent primarily on CD8+ T cells, with minimal reliance on newly recruited lymph node-derived T cells. To define the requirement for LAG3 expression on CD8+ T cells in the antitumor response to LAG3-InFc-LaIL2, we employed an adoptive transfer approach. Rag1−/−mice bearing MC38 tumors were reconstituted with CD8+ T cells from either wild-type or Lag3−/− donors. Treatment with LAG3-InFc-LaIL2 inhibited tumor growth in mice receiving wild-type CD8+ T cells. In contrast, the therapeutic benefit was significantly compromised in hosts reconstituted with Lag3−/− CD8+ T cells (Supplementary Fig. 11f). These results establish that the antitumor efficacy of LAG3-InFc-LaIL2 is strictly dependent on LAG3 expression on CD8+ T cells, providing direct evidence for a LAG3-specific mechanism of action and underscoring its precise in vivo targeting ability. LAG3-InFc-LaIL2 drives local and systemic antitumor immunity via intratumoral CD8+ T cells. a, b MC38 murine tumor models (n = 5/group) were treated with LAG3-InFc-LaIL2 on days 12 and 15. Anti-CD8 antibody (200 μg/mouse) was administered twice weekly beginning on day 11 (a). The FTY720 treatment was started on day 11 and was given at 2-day intervals until the end of the experimental period. (b). Tumor growth was measured twice weekly to track progression over time. c Schematic diagram of MC38 tumor fragment transplantation from C57BL/6 mice to subcutaneous Rag1−/− mice. d, e Rag1−/− mice were subcutaneously implanted with MC38 tumor fragments from WT C57BL/c mice in the right flank (70–80 mm³, treated tumor) or left flank (5 × 10⁵ cells, untreated tumor) on day 0. The right tumor was intratumorally (i.t.) treated with LAG3-InFc-LaIL2 on days 0 and 3. The growth of treated (d) and untreated (e) tumors was monitored twice weekly. f Peripheral blood samples were analyzed for the number of tetramer+CD8+ T cells 15 days post-treatment (n = 5/group). The data are displayed as the means ± SEMs and represent the results from 2 to 3 separate experimental repetitions. The P value was determined by unpaired two-tailed t tests (f). The statistical significance was evaluated through two-way ANOVA (a, b). The normal distribution of the data was confirmed via the Shapiro‒Wilk test. ***P < 0.001, ****P < 0.0001 To evaluate whether local administration of LAG3-InFc-LaIL2 could elicit systemic antitumor immunity, we employed a dual-tumor model. MC38 tumor fragments derived from wild-type C57BL/6 mice were transplanted into Rag1−/− mice, which was accompanied by contralateral subcutaneous implantation of MC38 cells (Fig. 6c). LAG3-InFc-LaIL2 treatment not only suppressed primary tumor growth but also significantly inhibited the progression of distal tumors (Fig. 6d, e and Supplementary Fig. 12a–d). Strikingly, tumor-specific CD8+ T cells were detected in the peripheral blood, indicating their migration from the primary TME and incorporation into the systemic T-cell pool (Fig. 6f and Supplementary Fig. 12e). These results further support the idea that effective tumor control can be achieved through the expansion and functional restoration of resident intratumoral CD8+ T cells without a primary reliance on peripheral CD8+ T-cell recruitment. LAG3-InFc-LaIL2 expands effector and memory tumor-specific CD8+ T cells to control distant tumors and prevent tumor recurrence Emerging evidence from recent studies substantiates the concept that neoadjuvant therapy, when administered prior to surgical intervention, can elicit systemic immune responses that target metastatic lesions.25–27 Following LAG3-InFc-LaIL2 treatment, a significant accumulation of tumor-specific CD8+ T cells was observed in the dLN, in contrast to other therapeutic modalities (Fig. 7a, Supplementary Figs. 13a and 14a). These cells manifested a CD44+CD62L- effector phenotype and expressed GZMA+ as a definitive effector marker (Fig. 7b, c, Supplementary Figs. 13b, c and 14b, c). Furthermore, the expression of IL2Rβ in the majority of CD8+ T cells in the dLN was upregulated following LAG3-InFc-LaIL2 treatment (Fig. 7d and Supplementary Fig. 14d). To investigate the function of these cells, after treatment, dLN cells were adoptively transferred into Rag1−/− mice, which subsequently received subcutaneous inoculation with MC38 tumor cells (Fig. 7e). Notably, compared with other treatment groups, dLN cells derived from LAG3-InFc-LaIL2–treated mice demonstrated superior efficacy in preventing tumor recurrence (Fig. 7f). These findings suggest that tumor-specific effector CD8+ T cells within the dLN, potentially originating from the TME, may play a pivotal role in preventing recurrence and controlling distal tumor growth. LAG3-InFc-LaIL2 expands effector and memory tumor-specific CD8+ T cells to control distant tumors and prevent tumor recurrence. a, b MC38 murine tumor models (n = 6/group) were treated with equal molar amounts of InFc-LaIL2 plus anti-LAG3-InFc or LAG3-InFc-LaIL2 on days 10 and 13. Five days after the first treatment, total dLN cells were analyzed for the proportion of tetramer+CD8+ T cells (a) and the proportion of CD62L-CD44+ Tetramer+CD8+ T cells (b). c MC38 murine tumor models (n = 4/group) were generated as described in (a). Five days after the first treatment, total dLN cells were analyzed for the proportion of GZMA+ cells among the tetramer+CD8+ T cells. d MC38 tumor-bearing mice (n = 5/group) were treated as described in (a). Five days following the initial treatment, the dLNs were analyzed to determine the proportion of CD122+ cells among the CD8+ T cells. e, f Rag1−/− mice (n = 5/group) were transferred with 1e6 dLN cells from MC38 murine tumor models five days after treatment with equal molar amounts of InFc-LaIL2 plus anti-LAG3-InFc or LAG3-InFc-LaIL2 and then implanted s.c. with 1e6 MC38 cells. Subcutaneous MC38 tumor growth curves in Rag1−/− mice were monitored twice a week. g MC38 tumor-free mice following LAG3-InFc-LaIL2 or naïve control mice (n = 3/group) received subcutaneous injections of 5 × 106 MC38 cells at a secondary site. Tumor progression was assessed twice weekly. The data are displayed as the means ± SEMs and represent the results from two to three separate experimental repetitions. The statistical significance was evaluated through one-way ANOVA followed by Tukey’s test for multiple comparisons (a–d). The normal distribution of the data was confirmed via the Shapiro‒Wilk test. ***P < 0.001, ****P < 0.0001 Five weeks post-tumor regression, the cured mice were rechallenged with MC38 cells. All LAG3-InFc-LaIL2-treated mice exhibited spontaneous tumor rejection, indicating the establishment of durable immunological memory (Fig. 7g). Overall, LAG3-InFc-LaIL2 treatment induced robust antitumor immunity at the primary site, which subsequently disseminated systemically via the dLN to exert control over distant tumors and prevent recurrence (Fig. 8). A proposed model for LAG3-InFc-LaIL2 preventing T cells from terminal exhaustion and reinvigorating antitumor immunity. This schematic model was illustrated using Adobe Illustrator Consequently, targeting the LAG3 pathway potentiates local immune activation, augmenting effector T-cell responses and mitigating immune checkpoint-mediated suppression of antitumor activity. The localized delivery of engineered IL2 synergistically amplifies this effect, fostering robust T-cell expansion and trafficking. This coordinated immune modulation influences both the TME and systemic immunity, thereby providing a multifaceted therapeutic strategy to overcome immune exhaustion and resistance across local and peripheral immune compartments. Discussion Moreover, Tex-pre- and Tex-int-specific CD8+ T cells are critical for immune responses in chronic infections and cancer, and their tumor-associated pool size is pivotal to the success of cancer immunotherapies.28,29 ICB rescues Tex-int CD8+ T cells, which are endowed with chromatin plasticity, but not terminal exhausted CD8+ T cells, whose chromatin accessibility is fixed.30,31 During tumor progression, immune-inhibitory receptors such as LAG3 are upregulated to suppress cell-autonomous IL2 synthesis in exhausted CD8+ T cells,6,32 leading to widespread LAG3 expression and impaired IL2-STAT5 signaling in most CD8+ T cells after chronic tumor antigen exposure. In the MC38 model, most tetramer+ tumor-specific CD8+ T cells were found within the LAG3+ compartment, suggesting that LAG3 targeting covered most of the tumor-specific T cells. To overcome dysfunction in LAG3+CD8+ T cells, we designed LAG3-LaIL2, an antibody–cytokine fusion that delivers avidity-driven, LAG3-restricted IL-2Rβ stimulation preferentially to intratumoral T cells. This approach mitigated the inherent bias of IL-2 toward peripheral CD8+ T cells and Tregs. LAG3-LaIL2 maintained a cytotoxic effector-like phenotype in Tex-pre cells, promoted their differentiation into Tex-int states, and may impede terminal exhaustion (Tex-term) within the tumor microenvironment. This fusion enhanced pSTAT5 signaling, proliferation, and cytotoxicity in the Tex-pre and Tex-int subsets, yielding robust antitumor responses without systemic toxicity in both immunologically hot and cold tumors. Furthermore, LAG3-LaIL2 confers control of distal tumors and prevents recurrence by expanding tumor-specific effector and memory CD8+ T cells in draining lymph nodes (Fig. 8). Chronic TCR signaling and TME-derived metabolic cues trigger T-cell exhaustion, which elicits TCF1 reduction, TOX activation and subsequent EMOS engagement to promote terminal differentiation.33–40 Exhaustion involves transcriptomic changes and upregulated coinhibitory receptors: PD-1 is broadly expressed across stem-like, pre-exhausted, exhausted, and terminally exhausted CD8+ T cells in the TME, whereas LAG3 is restricted to Tex-pre- and Tex-term cells.41 These findings indicate that LAG3 blockade is effective in targeting tumor-specific T cells. PD-1 blockade can relieve the inhibition of Tex cells and is effective in some cancers but does not fully reprogram Tex cells epigenetically.1,42–45 Unlike PD-1, LAG3 sustains TOX expression, modulates the CD94/NKG2-Qa-1b axis, and it cooperates with PD-1 to exacerbate T-cell exhaustion and constrain IFN-γ-mediated antitumor immunity.6,46 In an advanced melanoma trial, combining Relatlimab (anti-LAG3) and nivolumab (anti-PD-1) improved CD8+ T-cell activation, changed differentiation, and increased cytotoxicity, despite their exhaustion features.8 Unlike PD-1, LAG3 mediates immunosuppression via a distinct mechanism whereby ligand-induced, nondegradative polyubiquitination by c-Cbl and Cbl-b triggers a conformational change that enables inhibitory signaling—a process essential for its immune checkpoint function and a potential biomarker for immunotherapy.47 In primary human NSCLC T cells, coblockade of PD-1 and LAG3 suppressed CBL-B expression, whereas a bispecific inhibitor additionally targeted C-CBL, effectively reversing T-cell dysfunction.48 Given that anti-LAG3 (alone or with anti-PD-1) enhances tumor CD8+ T-cell cytotoxicity/function (despite an exhausted phenotype) clinically, we designed LAG3-LaIL2 to block LAG3-mediated immunosuppression while activating IL2/STAT5 signaling to prevent Tex-term progression. TME-resident exhausted CD8+ T cells are heterogeneous and include exhausted progenitors (Tprog-ex) and Tex-term subsets.34,35 PD-1 blockade targets Tex-pro cells to drive transient Tex-int expansion but fails to fully restore effector function or sustain responses due to a rigid, TOX-mediated epigenetic program.49–51 Similarly, LAG3-LaIL2 promotes Tex-pre to-Tex-int differentiation and may prevent Tex-int-to-Tex-term transition, increasing Tex-pre/Tex-int abundance and reducing Tex-term populations in the TME. Critically, these Tex-pre/Tex-int cells retain effector features and robust IL2-STAT5 signaling—boosting Tex-int formation and antagonizing TOX. As STAT5 activity decreases with tumor progression, IL2-driven STAT5 upregulation may reverse T-cell dysfunction; consistent with this, STAT5 signaling maintains progenitor CD8+ T cells, promotes Tex-int differentiation, and blocks Tex-term progression.24 LAG3-LaIL2 thus broadly reactivates the intratumoral T-cell landscape and revives dysfunctional T cells for antigen-dependent immune reactivation. Future studies should clarify whether these gene expression changes quantitatively/qualitatively modify existing exhaustion lineage subsets or generate novel differentiation states. Our data revealed that LAG3-LaIL2 expands preexisting tumor-specific CD8+ T cells in solid tumors; in Rag1−/− mice transplanted with wild-type tumor fragments, these cells control primary and distal tumors. Unlike prior evidence highlighting the role of systemic CD8+ T cells in ICB responses,52–55 we found that tumor-specific CD8+ T cells migrated from treated primary tumors to peripheral blood and then to distal tumors in Rag1−/− mice. We also observed increased numbers of effector/memory tumor-specific CD8+ T cells in dLNs 5 days after therapy. Tumor-draining lymph nodes (TDLNs) are critical for radiotherapy (RT)- and ICB-mediated antitumor responses56–58 and serve as reservoirs for stem-like tumor-specific CD8+ T cells.56 Tumor-specific T cells migrate bidirectionally between the TME and dLNs.59–61 TdLN-Ttsm cells exhibit superior antitumor efficacy after adoptive transfer and respond to PD-1/PD-L1 blockade.53 Consistent with these findings, the transfer of dLN cells from LAG3-LaIL2-treated MC38 tumor-bearing mice to Rag1−/− mice inhibited tumor growth. However, the mechanisms underlying the ability of T-cell “education” to prevent distant tumor control/recurrence and the synergy between LAG3 blockade and exogenous IL2 signaling require further investigation. Unlike other IL-2 strategies, LAG3-LaIL2 enables precise tumor targeting and ameliorates immunosuppression within the TME. Initial immuno-cytokines with native IL-2 induce systemic toxicity via high-affinity receptor binding, whereas more recent low-affinity muteins (e.g., FAP–IL-2v, NHS–IL-2LT) still depend on variable tumor-associated antigens and lack inhibitory pathway disruption.62,63 In contrast, LAG3-LaIL2 uses LAG3—which is highly expressed on exhausted T cells—as both a target site and functional block, enabling TAA-independent IL-2 delivery while countering LAG3-mediated suppression of T-cell function and IL-2 secretion. Furthermore, unlike CD122-selective agonists, which expand peripheral T cells but fail to address exhaustion64, or PD-1/IL-2 fusion proteins, whose targeting effects may be compromised by prior PD-1 treatment65, our construct targets LAG3—an independent pathway that works synergistically with PD-1. This leads to intratumoral T-cell reactivation and long-lasting memory responses in lymph nodes, resulting in broader applicability and sustained antitumor immunity. In summary, we developed the novel LAG3-targeted fusion protein LAG3-InFc-LaIL2, which selectively delivers IL2 to tumor-infiltrating CD8+ T cells. This agent enhances TIL expansion and cytotoxicity, blocks Tex-int-to-Tex-term differentiation, and exerts potent antitumor efficacy in mouse and humanized models while mitigating adverse effects—with key advantages over next-generation IL-2 agonists (TME specificity) and anti-PD-1 + IL-2 fusions (independent of PD-L1 and dual targeting of the LAG3/IL2 pathways). Importantly, it increases the number of effector/memory tumor-specific CD8+ T cells in dLNs, potentially enhancing recurrence and controlling metastasis. With dual PD-1/LAG3 blockade already approved, LAG3-InFc-LaIL2 offers a novel strategy for synergistic pairing with anti-PD-1 therapy. Our findings demonstrate that LAG3-InFc-LaIL2 overcomes key limitations of current cytokine therapies, supporting robust clinical evaluation in cancer patients. Materials and methods Ethics approval Every procedure involving laboratory mice was examined and approved by the Institutional Animal Care and Use Committee of the Chinese Institute for Brain Research (CIBR-IACUC-048). Mice We procured 6–8-week-old female C57BL/6 N, BALB/c, Rag1−/− and NSG-SGM3 mice from Beijing Vital River Laboratory Animal Technology Co., Ltd., which were maintained under specific pathogen-free conditions. Every animal experiment was conducted in accordance with the operational guidelines set by the Chinese Institute for Brain Research, Beijing. Cell lines and reagents We obtained MC38, B16F10, CT26 and CFPAC1 cell lines from the American Type Culture Collection. These cells were cultured in DMEM supplemented with 10% heat-inactivated FBS, 100 U ml-1 penicillin and 100 μg ml-1 streptomycin under 5% CO2 incubation conditions and subjected to a routine mycoplasma contamination test. Bio X Cell was supplied with anti-CD8 (53-5.8), and FTY720 was acquired from Selleckchem. Production of the LAG3-LaIL2 protein Transient transfection of the corresponding plasmids into FreeStyle 293-F cells was performed to generate LAG3-InFc-LaIL2 and other proteins of interest. The fusion protein-containing supernatant was subsequently purified via protein A affinity chromatography. The purity of the proteins was checked via SDS–PAGE. Tumor growth and treatment Murine right flanks were subcutaneously inoculated with tumor cells at the following doses: 1 × 106 MC38, 5 × 105 B16F10, 1 × 106 CT26 and 1 × 106 CFPAC1 tumor cells. Following tumor formation, the mice were randomly distributed into different experimental groups. The mice were intraperitoneally injected with 30 μg of LAG3-LaIL2, 30 μg of LAG3-InFc-WTIL2, 30 μg of LAG3-InFc-LaIL2 or anti-LAG3 (23 μg) + InFc-LaIL2 (18.24 μg) at the indicated timepoints. For the CT26 model, BALB/c mice were used. CD8+ T-cell depletion was achieved via i.p. administration of 200 μg of antibody 1 day before treatment commenced, with injections repeated every 3 days for 3 weeks.20 Thirty micrograms of FTY720 was i.p. administered 1 day before treatment commenced, followed by the administration of a 10 μg dose every other day for two weeks.20 The tumor dimensions were assessed by measuring the length (a), width (b) and height (h), with the volume calculated via the formula abh/2. The mice were sacrificed as part of the survival curve assessment if any of the following criteria were met: tumor length/width/height >2 cm, tumor volume >1500 mm³, or ≥ 20% reduction in body weight from treatment-onset levels. LAG3-LaIL2 was administered via i.p. injection in all relevant in vivo experiments, except for the intratumoral (i.t.) administration specified in Fig. 6c–f and Supplementary Fig. 12. Flow cytometry analysis To prevent nonspecific binding, we subjected single-cell suspensions to 15 minutes of pretreatment with the anti-FcgIII/II receptor (clone 2.4G2). This was followed by a 30-min incubation with fluorochrome-labeled antibodies under refrigeration in the dark. Fixable viability Dye eFluor 506 or eFluor 780 was used to exclude dead cells. Intracellular staining of Foxp3+, Ki-67+, and Granzyme A+ cells was performed via a true-nuclear transcription factor buffer (BioLegend) according to the manufacturer’s guidelines.20 The data were gathered via a CytoFLEX flow cytometer from Beckman Coulter and analyzed via CytExpert from Beckman Coulter or FlowJo from Tree Star. Adoptive transfer of total dLN cells Five days after LAG3-InFc-LaIL2 treatment, dLNs were distributed within individual cell suspensions. Following thorough washing with PBS, the samples were passed through a 70-μM filter to ensure uniformity. Rag1−/− mice were treated by infusion of purified cells (2 × 106). The following day, 1 × 106 MC38 cells were s.c. injected. Establishment of a C57-derived allograft model of the MC38 cell line in Rag1−/− mice MC38 tumor-bearing C57BL/6 J mice were euthanized under anesthesia via intravenous administration of sodium pentobarbital at a dosage of 10 mg/kg body weight. Once the tumor volume reached approximately 150–160 mm3. The tumors were harvested and immersed in Matrigel immediately after the tumor-bearing mice were euthanized. For the tissue fragments, the chunks of the excised tumors were mechanically cut into 70–80 mm3 pieces with a scalpel. Tumor fragments were surgically implanted beneath the skin on the right flank of Rag1−/− mice, which were then treated with 30 μg of LAG3-InFc-LaIL2 or anti-LAG3 (23 μg) + InFc-LaIL2 (18.24 μg) at the indicated timepoints by intratumoral administration. Subcutaneous implantation of 5 × 105 MC38 cells was performed on the left flank of Rag1−/− mice. Establishing a LAG3KO CD8 model of the MC38 cell line in Rag1−/− mice Rag1−/− mice received subcutaneous injections of 1 × 106 MC38 tumor cells. Six days later, the spleens of both wild-type and LAG3 knockout mice were harvested to prepare single-cell suspensions. From these preparations, CD8+ T cells were then isolated via a negative selection method with a magnetic bead-based separation kit. CD8+ T cells (1 × 10⁶ cells/mouse) were transferred intravenously into Rag1−/− tumor hosts during concurrent treatment. The mice were distributed equally into 4 groups: Group 1 (WT CD8+ T cells + PBS), Group 2 (WT CD8+ T cells + 30 μg LAG3-InFc-LaIL2 (i.p.)), Group 3 (LAG3-KO CD8+ T cells + PBS), and Group 4 (LAG3-KO CD8+ T cells + 30 μg LAG3-InFc-LaIL2 (i.p.)). A total of two treatments were administered, with a 3-day interval between them. Quantification of IFN-γ–secreting T cells via ELISPOT Tumor tissues were resuspended in RPMI 1640 medium supplemented with 10% FBS, along with 2 mmol/L l-glutamine, 100 U/mL penicillin, and 100 mg/mL streptomycin to create an optimal culture environment. For the subsequent assay, 3 × 10e5 dLN cells were utilized. MC38 cells were exposed to irradiation at a single 60 Gy dose.20 A 1:4 ratio of MC38 cells to immune cells was used for the coculture experiments. Following a 48-hour incubation period, we identified elevated levels of IFN-γ via the BD Biosciences IFN-γ ELISPOT Assay Kit, which adheres strictly to the manufacturer’s guidelines.20 We then counted the visualized cytokine spots via an ImmunoSpot S6 Analyzer from Cellular Technology Limited. LDH cytotoxicity assay CD8+ T cells were isolated from tumors from LAG3-InFc-LaIL2- or control-treated mice via magnetic bead purification (BioLegend, 480035). To assess CTL-mediated cytotoxicity, CD8+ T cells (2.5e6, effector) and MC38 tumor cells (5e5, target) were mixed at a 5:1 ratio for a period of 72 h. LDH release in the supernatant was measured via an LDH kit (C0016, Beyotime). Detection of intratumor IFN-γ+ CD8+ T cells CD8+ T cells that infiltrated tumors were purified from LAG3-InFc-LaIL2- or control-treated mice via magnetic bead purification (BioLegend, 480035) and resuspended in RPMI 1640 medium supplemented with 10% FBS. The medium was further enriched with phorbol 12-myristate 13-acetate (PMA, Sigma‒Aldrich, Cat. No. P8139; final 50 ng/mL), ionomycin (Sigma‒Aldrich, Cat. No. I0634; final 1 μg/mL) and brefeldin A (BFA, eBioscience, Cat. No. 00--4506--51; final 10 μg/mL), which were maintained at 37 °C in 5% CO2 for 4 h; after harvesting, the cells were stained with surface flow cytometry antibodies at 4 °C for 30 min before being rinsed with PBS. The cells were then subjected to fixation buffer (BD 554655) at room temperature for 20 min, permeabilized with permeabilization buffer (BD 554723) for 10 min, and then stained with an AF700-conjugated anti-mouse IFN-γ antibody (Thermo Fisher 56731182) at 4 °C for 30 min. The study utilized a Beckman Coulter CytoFLEX flow cytometer for data collection, followed by data analysis via CytExpert from Beckman Coulter or FlowJo software from Tree Star. Detection of intratumor p-STAT5+ CD8+ T cells TILs were extracted from MC38K tumors via a mouse CD90.2-positive selection magnetic bead kit (STEMCELL Technologies, 18951). To eliminate nonspecific binding, single-cell suspensions were first treated with an anti-FcγIII/II receptor antibody (clone 2.4G2) for 15 min. The cells were then stained with conjugated antibodies for 30 min while they were kept cool at 4 °C in the dark. Following staining, the cells were treated with either 1 μg/mL LAG3-InFc-LaIL2 or other proteins for 15 min. The cells were fixed with an equal volume of 4% paraformaldehyde (PFA) for 1 h on ice, followed by two washes with prechilled 90% methanol at 1500 rpm. Permeabilization was performed with 90% methanol for 1 h. After two additional washes with FACS buffer, the samples were incubated with an anti-p-STAT5 antibody for 30 min. The grand finale involved analyzing the stained cells via a CytoFLEX flow cytometer. Toxicity evaluation Tumor-bearing mice were treated with various fusion proteins. The initial body weight was recorded at the start of the treatment, and subsequent measurements were taken to determine changes. The relative weight change was calculated via the following formula: (measured body weight/initial body weight) × 100%. Blood samples were collected at multiple time intervals posttreatment, and the serum was isolated. Cytokine titers in the serum were assessed with a cytometric bead array from BD Biosciences. Western blot A total of 5e7 PBMCs were activated for 72 h with a 5 μg/mL cocktail of anti-CD3 and anti-CD28 antibodies and then cultured in serum-free medium for an hour. The mixture was then stimulated for 0, 5, 15 or 60 min with 0.01 μg/mL LAG3-InFc, LaIL2-InFc or LAG3-InFc-LaIL2 with 5 μg/mL CD3, after which the cells were lysed with RIPA lysis buffer supplemented with both phosphatase and protease inhibitors to preserve protein integrity. The protein content was measured via the bicinchoninic acid method by Pierce. Western blot analysis was then performed on these protein samples, with the transfer membrane being incubated with p-STAT5, STAT5, p-JAK3, JAK3 and β-actin antibodies (dilution 1:10,00) and subsequently exposed to the corresponding peroxidase-conjugated secondary antibodies (dilution 1:5000). Chemiluminescent signals were captured via a ChemiDoc MP from Bio-Rad Laboratories, with band intensities quantified via Image Lab Software from the same company. Tumor digestion The tumor samples were minced into small fragments and then subjected to enzymatic breakdown using a solution containing 1 mg/ml collagenase I (Sigma) and 0.5 mg/ml DNase I (Roche) at 37 °C for 20 min. A 70 μm strainer was used to sieve the tumor cell suspensions, thus removing large undigested tumor particles. The filtered cells underwent two wash cycles with PBS containing 2 mM EDTA. Data collection and preprocessing for single-cell RNA sequencing MC38 tumors were isolated on day 15 after 5 days of treatment. The tumor tissues were processed via a MACS tumor dissociation kit (Miltenyi Biotec) per the manufacturer’s guidelines. Following mechanical dissociation, the cells were stained with anti-CD3 antibodies and sorted via BD FACS Aria. scRNA-seq was conducted via the Chromium Next GEM Single Cell 3’ Kit (10x Genomics). All subsequent steps were carried out in accordance with the manufacturer’s protocol. The resulting libraries were purified and sequenced on a NextSeq sequencing system (Illumina). The raw sequencing data were demultiplexed and processed via the Cell Ranger software suite (version 2.1.1). The raw Fastq files were initially aligned to the mouse reference genome via the Cell Ranger single-cell software suite (version 7.1.0). Following alignment, a unique molecular identifier (UMI) matrix was generated through Seurat v4.1.0 analysis. Initial quality control filtering removed low-quality cells characterized by fewer than 200 detected genes or mitochondrial UMI counts exceeding 10% of the total UMIs. The count data were subsequently normalized via the normalization data algorithm, and a log-transformed normalized matrix was prepared for further analyses. To identify highly variable features, the Find Variable Features function was applied to the dataset. Dimensionality reduction was performed by projecting the data onto the first 30 principal components (PCs) via principal component analysis (PCA). A UMAP representation of the data and neighbor relationships was then constructed via the Run UMAP and Find Neighbor functions, respectively. Clustering was performed via shared nearest neighbor algorithms, and t-distributed stochastic neighbor embedding (t-SNE) plots were generated to visualize cluster separation across selected PCA dimensions. Marker gene identification was conducted via the Find All Markers or Find_markers functions with thresholds of at least 25% expression in a cluster (min.pct = 0.25) and a minimum log-fold change threshold of 0.5 (logfc.threshold = 0.5). The expression of marker genes was weighted according to their predicted loss probability, computed from nearest neighbors, and the normalized sum of expression across all genes in the gene set. All the scRNA-seq data analyses were conducted via the R programming language (version 4.2.2) to facilitate downstream interpretation and integration of the scRNA-seq data. Statistics and reproducibility The number of experimental animals and technical/biological replicates is specified in each figure legend. Prior to treatment initiation, the animals were stratified randomly by tumor volume and then randomly assigned to experimental groups. For in vitro assays, cell culture wells were randomized across all groups, and equal analytical weighting was applied to all the cellular samples. The data were thoroughly analyzed via GraphPad Prism 10 statistical software, and the results are presented as the means ± standard errors of the means (SEMs). P values were determined by two-way analysis of variance (ANOVA) for tumor growth, the log-rank test for survival, Student’s t test or one-way ANOVA for other analyses. P values less than 0.05 were considered statistically significant. For the scRNA-seq data, multiple comparison p values were adjusted via the Benjamini–Hochberg method implemented through the p.adjust function. Supplementary information Supplementary Information Acknowledgements The authors would like to thank all members of the Fu laboratory for their insightful feedback and the faculty at the animal facility of the Chinese Institute for Brain Research. This work was supported by the National Science and Technology Major Project (SQ2026AAA161214; 2023ZD0520200, to Z.R.), grants from the National Natural Science Foundation of China (82250710684, to Y.-X.F.) and Changping Laboratory (2021C-02-04). Author contributions X.Y. and Z.R. designed the experiments, analyzed the data, and wrote the manuscript. X.Y., H.L., P.J., B.S., Z.Y. and Z.R. developed the study methodology. X.Y., J.L., Y.S., R.Z., Y.C., Y.L., L.L., and S.L. performed the experiments. H.T. provided reagents. Y.-X. F. and Z.R. supervised the project. All the authors have read and approved the article. Data availability Single-cell RNA-seq data newly generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE318979. In addition, the raw sequencing data reported in this paper are available in the Genome Sequence Archive at the National Genomics Data Center (GSA: CRA038462). 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: Xiaohong Yu, Huiping Liao Contributor Information Yang-Xin Fu, Email: yangxinfu@mail.tsinghua.edu.cn. Zhenhua Ren, Email: renzhenhua@cpl.ac.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41392-026-02667-8. References Associated Data Supplementary Materials Supplementary Information Data Availability Statement Single-cell RNA-seq data newly generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE318979. In addition, the raw sequencing data reported in this paper are available in the Genome Sequence Archive at the National Genomics Data Center (GSA: CRA038462).