Universal STING mimic boosts antitumour immunity via preferential activation of tumour control signalling pathways

Abstract

The efficacy of STING (stimulator of interferon genes) agonists is due to various factors, primarily inefficient intracellular delivery, low/lack of endogenous STING expression in many tumours, and a complex balance between tumour control and progression. Here we report a universal STING mimic (uniSTING) based on a polymeric architecture. UniSTING activates STING signalling in a range of mouse and human cell types, independent of endogenous STING expression, and selectively stimulates tumour control IRF3/IFN-I pathways, but not tumour progression NF-κB pathways. Intratumoural or systemic injection of uniSTING-mRNA via lipid nanoparticles (LNPs) results in potent antitumour efficacy across established and advanced metastatic tumour models, including triple-negative breast cancer, lung cancer, melanoma and orthotopic/metastatic liver malignancies. Furthermore, uniSTING displays an effective antitumour response superior to 2′3′-cGAMP and ADU-S100. By favouring IRF3/IFN-I activity over the proinflammatory NF-κB signalling pathway, uniSTING promotes dendritic cell maturation and antigen-specific CD8⁺ T-cell responses. Extracellular vesicles released from uniSTING-treated tumour cells further sensitize dendritic cells via exosome-containing miRNAs that reduced the immunosuppressive Wnt2b, and a combination of LNP-uniSTING-mRNA with α-Wnt2b antibodies synergistically inhibits tumour growth and prolongs animal survival. Collectively, these results demonstrate the LNP-mediated delivery of uniSTING-mRNA as a strategy to overcome the current STING therapeutic barriers, particularly for the treatment of multiple cancer types in which STING is downregulated or absent.

DNA engagement of the mammalian cyclic GMP-AMP synthase (cGAS) generates 2′3′-cGAMP, which binds to the homodimer of the adaptor protein STING (stimulator of interferon genes) on the endoplasmic reticulum (ER) membrane, forming a polymeric STING architecture that is essential to activate the downstream signalling pathways^1,2. The activated STING causes the initiation of transcriptional cascades primarily involving interferon (IFN) regulatory factor 3 (IRF3) and nuclear factor (NF-κB), thereby upregulating the transcription of type I IFNs, proinflammatory cytokines and chemokines. Recent studies have demonstrated that STING activation and subsequent IFN-I secretion play a critical role in anticancer immunity^3,4. The therapeutic potential of cyclic dinucleotide (CDN)-based STING agonists is being examined in several clinical trials. However, the poor stability and dose-dependent toxicity of these compounds make their clinical applications a major undertaking^5–7. Recently, orally available non-nucleotide STING agonists with improved antitumour efficacy and broadened indications have been reported^8,9. Despite their therapeutic potential, all the CDN- and non-nucleotide-based STING agonists are optimal when the tumour also expresses STING. Several studies have shown that STING signalling pathways in tumour cells are suppressed due to the epigenetic silencing or missense mutations of either STING or its upstream regulator cGAS^10,11. Despite the recent report of constitutively active mutations of STING, such as V155M, its effectiveness and safety profiles have not yet been investigated in clinical trials¹². Moreover, the NF-κB/IL6/STAT3 axis downstream of cGAS/STING activation promotes cancer cell progression, reflecting the pleiotropic effects of STING activation¹³. Previously, a recombinant STING protein lacking a transmembrane domain (STINGΔTM) was used as a carrier for delivering cGAMP¹⁴. However, this approach depended on the in vitro addition of cGAMP to preassemble the cGAMP-STINGΔTM ribonucleoprotein complex before the challenging intracellular delivery of this large protein complex. All the STING agonists documented thus far stimulate both the IRF3 and NF-κB pathways. The antitumor response and activation of IRF3/IFN-I and NF-κB among the representative STING agonists^12,14–16 are summarized in Supplementary Table 1. Thus, there is an urgent unmet need to develop novel STING stimulators that can induce the STING-mediated immune response regardless of the presence of endogenous STING. This report used a universal STING activation strategy in a wide range of cancer types without promoting cancer growth.

STING polymerization and trafficking are crucial for its downstream immunological outcomes^15,17,18. The transmembrane domain of STING undergoes a conformational change upon binding to cGAMP, forming a side-by-side tetramer that could grow further into a larger oligomer¹⁶. By bringing multiple TANK binding kinase 1 (TBK1) dimers together for trans-autophosphorylation of the activation loop, this tetramer-based higher-order architecture of STING provides an efficient scaffold for phosphorylation of the C-terminal tail of STING, which in turn recruits and activates IRF3 for the upregulation of IFN-I^19,20. Although the role of the transmembrane domain of STING in its activation is still elusive, strong evidence indicates that the tetramer-based higher-order architecture of STING plays a central role in initiating the downstream signalling responses. More importantly, STING trafficking could shape the downstream transcriptional cascades. NF-κB signalling cascade occurs when STING exits the ER, whereas IRF3 activation requires STING’s presence in the late endosomes^21,22.

To address the deficiency of the functional STING signal in certain tumours, we report an effective universal STING mimic (uniSTING), designed by genetically fusing a highly stable tetramerization motif with the non-membrane-bound domain of STING. UniSTING can self-assemble into a tetrameric STING and further nucleate into polymeric STING with a higher-order architecture independent of the endomembrane system. UniSTING imitates the cGAMP-induced oligomerization of endogenous STING, leading to constitutive activation of STING downstream pathways that favour the IRF3/IFN-I cascade over proinflammatory NF-κB signalling (Fig. 1). Taking advantage of lipid nanoparticle (LNP)-mRNA, which allows for efficient intracellular delivery and substantial enhancement of innate responses²³, we encapsulated the mRNA encoding the mouse version of uniSTING fusion protein into LNPs for delivery into the tumour microenvironment (TME). UniSTING induced phosphorylation of TBK1 and IRF3 and subsequent robust production of IFN-β and ISG in murine dendritic cells (DCs) and tumour cells, with little activation of the NF-κB pathway. This activates mature DCs to elicit an enhanced T-cell response. Importantly, the human version of uniSTING (huSTING) also showed comparable effects on activating STING pathways in STING-deficient human ovarian cancer cells or monocytes. Robust activation of STING signalling was also observed in STING-deficient bone-marrow-derived DCs (BMDCs), indicating that uniSTING works independently of endogenous STING expression. Accordingly, LNP-uniSTING-mRNA treatment inhibited tumour growth and improved survival in both wild-type (WT) and STING^−/− mice. Mechanistically, extracellular vesicles (EVs) promoted tumour cell–DC interaction following uniSTING treatment. An unbiased assessment of microRNA (miRNA) profiles revealed that EVs released from uniSTING-treated tumour cells contained markedly elevated miRNAs, including miR-130–3p, miR-15b-5p and miR-16–3p. The expression of immunosuppressive protein Wnt2b in DCs was reduced by these miRNAs, further improving DC function. Consistent with these findings, a combination of LNP-uniSTING-mRNA with anti-Wnt2b (α-Wnt2b) antibody was superior in inhibiting tumour growth and lengthening the survival than either alone.

Fig. 1 |. LNP-uniSTING-mRNA induced constitutive STING activation and EV-mediated crosstalk between tumour cells and DCs.

LNP-uniSTING-mRNA treatment resulted in the expression of a universal STING mimic in tumour cells and DC cells, which self-assembled into a tetrameric subunit followed by the formation of a higher-order STING architecture for efficient downstream phosphorylation of IRF3 and subsequent release of type I IFNs and ISG cytokines. EVs released by uniSTING-treated tumour cells further sensitized DCs’ function in the TME by the secretion of miRNAs, including miR-130–3p, miR-15b-5p and miR-16–3p, which targeted Wnt2b and reduced immunosuppressive signalling molecules. P, phosphorylation. Illustration created with BioRender.com.

UniSTING activates STING independently of its endogenous level

To overcome the lack of functional STING signal in TME and undesired toxicity after systemic administration, we developed a universal STING mimic by genetically fusing a thermostable tetramerization motif (52 residues)²⁴ with STING’s C-terminal cytoplasmic domain (residues 138–378 of murine STING). Four copies of the resulting fusion protein self-assembled into a tetrameric structure independent of membrane association, mimicking the side-by-side tetrameric STING on the membrane. Further oligomerization based on this tetrameric core resulted in polymeric STING with higher-order architecture for downstream signalling activation (Fig. 2a). UniSTING fusion protein was expressed in ExpiCHO cells (37.5 kDa, monomer in SDS–PAGE), purified based on the engineered N-terminal Flag tag and subsequently confirmed by fast protein liquid chromatography analysis. UniSTING existed predominantly as an octamer with an estimated molecular weight of ∼300 kDa under non-denaturing conditions (Supplementary Fig. 1a,b) with good stability (T_m ≈ 52.7 °C, Supplementary Fig. 1c). TBK1 and IRF3 colocalized with uniSTING in the cytosol when coexpressed alongside the uniSTING in HEK293T cells. IRF3 was additionally observed in the nucleus, indicating its activated state (Fig. 2b). Phosphorylated TBK1 and IRF3 were coimmunoprecipitated with uniSTING in DC2.4 cells (Fig. 2c), confirming uniSTING’s binding with TBK1 and IRF3. Furthermore, gene set enrichment analysis (GSEA) of RNA sequencing (RNA-seq) data showed that IFNα and IFNγ response genes were markedly enriched in uniSTING-mRNA-treated DC2.4 cells compared with the EGFP-mRNA (mock) group (Fig. 2d and Supplementary Fig. 2a). Endogenous STING trafficks through endomembranes¹⁵. It was reported that a proinflammatory NF-κB signal occurs when STING exits the ER, while STING’s presence in late endosome is sufficient for IRF3 transcriptional activity²². UniSTING was designed to form a tetrameric structure independent of membrane association. To examine uniSTING’s effect on downstream pathways in an unbiased fashion, we performed RNA-seq on 4T1 cells. Specifically, 235 DEGs were elevated in uniSTING-treated 4T1 cells, with 80% of them connected to pIRF3 binding and only 5% related to p-p65 binding, emphasizing the highly selective activation of STING downstream pathways by uniSTING, which favours IRF3/IFN-I activity over the proinflammatory but also protumourigenic NF-κB signalling (Fig. 2e). STING signallings that are reported to be IRF3-dependent^25–27 and NF-κB-dependent^28–30 were displayed (Fig. 2f). IRF3-dependent signallings were extensively upregulated by uniSTING treatment, whereas NF-κB target signallings were unaffected. Both uniSTING and 2′3′-cGAMP treatments elevated Ifnb1 and Cxcl10 transcripts (Fig. 2g). While 2′3′-cGAMP treatment increased the production of Il6 and Tnf, uniSTING treatment failed to do so (Fig. 2g). Furthermore, treatment with uniSTING-mRNA increased STING downstream pTBK1, pIRF3 and IFNβ secretion in a dendritic cell line (DC2.4) and two murine TNBC cell lines (4T1 and E0771), but not when a monomeric version of STING (mSTING, which lacked the tetramerization motif) was used (Fig. 2h and Supplementary Fig. 2b). The human version of uniSTING (huSTING) activated STING signalling in STING-deficient human ovarian cancer cell line ES-2 and human macrophage THP-1 cells (Fig. 2h,i and Supplementary Fig. 2c). Moreover, while NF-κB phosphorylation was significantly increased by 2′3′-cGAMP in DC2.4 and 4T1, it was not obviously increased by uniSTING in different types of cell lines (Fig. 2h). These data in both murine and human cells indicate that uniSTING functions as a universal activator with the specific preference for IRF3/IFN-I over NF-κB signallings.

Fig. 2 |. Characterization of tetramer-based uniSTING as a universal STING agonist independent of cGAMP or endogenous STING.

a, Schema for uniSTING construction by genetically fusing an N-terminal Flag tag and a 52-residue tetramerization motif (green) with the C-terminal cytoplasmic domain of STING (red). b, Immunofluorescence staining revealed TBK1-GFP and IRF3-HA (purple) colocalized with uniSTING-Flag (red) in the cytosol of HEK293T cells. DAPI staining was used to show the nucleus (blue). Scale bars, 10 μm (left); 2 μm (right). A representative of three repeated experiments is shown. c, Coimmunoprecipitation of uniSTING with pTBK1, pIRF3 or Flag in cell lysates from DC2.4 cells. A representative of three repeated experiment is shown. d, Top seven enriched Gene Ontology pathways in uniSTING-treated DC2.4 cells. e, Venn diagram and volcano plot of the percentages of DEGs that were associated with either pIRF3 binding (red) or p-p65 binding (cyan) in uniSTING-treated 4T1 versus the mock group. FC, fold change. f, Selected IRF3-dependent or NF-κB-dependent pathways in 4T1 tumour cells (P < 0.01). IKK, inhibitory-κB kinase; IL-2, interleukin-2; TNF, tumour necrosis factor. g, Transcripts of Ifnb1, Cxcl10, Il6 and Tnf in 4T1 tumour cells 24 h after indicated treatments. n = 6 biologically independent samples. h, Immunoblot analysis of STING signalling in DC2.4 cells, 4T1 or ES-2 tumour cells treated with PBS, mock mRNA (1 μg ml⁻¹), mSTING mRNA (1 μg ml⁻¹), uniSTING mRNA (1 μg ml⁻¹), or 2′3′-cGAMP (5 or 10 μg ml⁻¹). i, ELISA revealed IFNβ production in murine DCs, 4T1 or ES-2 tumour cells 24 h after indicated treatments (mSTING, monomeric STING mRNA; mock, EGFP-mRNA; uniSTING, universal polymeric STING mRNA; 1 μg mRNA or 5 μg/10 μg 2′3′-cGAMP per ml). n = 5 biologically independent samples. j, Fluorescence-activated cell sorting gating strategy of BMDCs (CD11c⁺F4/80⁻ cells gated on CD45⁺CD11b⁻). mRNA expression of Ifnb1 and Cxcl10 in BMDCs derived from WT, Tmem173^−/−, Irf3^−/− and Ifnαr1^−/− mice 24 h after PBS, mock mRNA (1 μg ml⁻¹), uniSTING mRNA (1 μg ml⁻¹), or 2′3′-cGAMP (10 μg ml⁻¹) treatment. n = 5 biologically independent samples. Significant differences were assessed by unpaired two-tailed Student’s t-test (d–f), a one-way ANOVA and Tukey’s multiple-comparisons test (g,i) and two-way ANOVA with multiple comparisons (j). Results are presented as mean ± s.d.

We next delineated uniSTING’s reliance on the cGAS/STING pathway. uniSTING downstream signalling was resistant to a cGAS inhibitor, Ru.521 (Supplementary Fig. 3). Both uniSTING and 2′3′-cGAMP elevated transcripts of Ifnb1 and Cxcl10 in BMDCs. This was abolished in Irf3^−/− and Ifnαr1^−/− BMDCs (Fig. 2j). Importantly, uniSTING but not 2′3′-cGAMP retained its ability to activate IFNβ and CXCL10 in STING-deleted (Tmem173^−/−) BMDCs (Fig. 2j). Collectively, these data indicate that in the absence of endogenous STING, uniSTING selectively strengthened the IRF3-dependent IFN-I release for antitumour immunity without promoting side effects from the multifunctional NF-κB signalling that has been documented to promote tumour progression.

uniSTING delivered by LNP-mRNA inhibits established tumours

Endogenous STING expression is significantly downregulated in human breast tumour compared with nontumour counterparts (Supplementary Fig. 4a). A low level of STING protein in human breast cancer correlates with a poor prognosis (Supplementary Fig. 4b). Among all breast cancer subtypes, triple-negative breast cancer (TNBC) has the poorest outcomes³¹. We first explored the therapeutic potential of uniSTING mRNA in TNBC models. To achieve efficient intracellular delivery of uniSTING, mRNA encoding uniSTING was encapsulated into LNPs (LNP-uniSTING-mRNA) based on ionizable lipid-COATSOME SS-OP^32,33. SS-OP contains two sensing motifs: a tertiary amine that is responsive for endosomal membrane destabilization, and a disulfide bond that can be cleaved by reduced glutathione in a reductive environment (cytoplasm) for spontaneous particle collapse³⁴. The resulting LNP-mRNAs represent surface-neutral particles with a median diameter of ∼100 nm (Fig. 3a and Supplementary Fig. 5a). We confirmed an elevated luciferase activity in tumour cells and DC2.4 cells when luciferase mRNA was delivered via SS-OP LNPs with redox sensitivity (Supplementary Fig. 5b). LNP-mCherry-mRNA was intratumourally administered into established 4T1-GFP tumours. Both tumour cells and CD11c⁺ cells showed pronounced uptake and translation of the LNP-delivered mRNA compared with that of the other cell types, while cancer cells displayed the highest mCherry expression in line with the in vitro assay (Fig. 3b,c and Supplementary Fig. 5c). Additionally, SS-OP LNPs showed 6-fold higher luciferase activity in the tumour site 24 h after dosing compared with that of clinically approved MC3 LNPs³⁵. These results indicated that SS-OP-based LNPs functioned as an efficient delivery system (Fig. 3d). Furthermore, uniSTING was enriched in tumours (Fig. 3e), peaking after 4 h with declining yet detectable signals until day 7 through intratumoural administration (Fig. 3f). UniSTING protein was detectable in the liver from 2 h to 72 h postinjection (Supplementary Fig. 5d). Additionally, no discernible reduction in body weight was noticed despite the intravenous injection of uniSTING (Supplementary Fig. 5e). Serum alanine transaminase, aspartate aminotransferase, blood urea nitrogen and creatinine levels were maintained in the normal range³⁶ 24 h and 48 h after intravenous treatments with LNP-uniSTING-mRNA, LNP-mock-mRNA or PBS (Supplementary Fig. 5f–g), suggesting the safety of uniSTING for systemic administration. LNP-mediated mRNA delivery therefore addresses the poor stability of current STING agonists.

Fig. 3 |. Cytosolic delivery of uniSTING-mRNA based on LNPs intratumourally inhibits tumour growth.

a, Schematic of mRNA-loaded SS-OP LNPs and cryogenic electron microscopy image of LNP-uniSTING-mRNA. Scale bar, 100 nm. Each experiment was repeated three times. b, LNP-mCherry-mRNA expression in tumour cells (4T1-GFP), DCs (CD45⁺CD11c⁺), T cells (CD45⁺CD3⁺) and macrophages (CD45⁺CD11c⁻CD11b⁺F4/80⁺) following intratumoural injection (i.t.) (mCherry mRNA, 0.5 mg kg⁻¹), quantified by flow cytometry. n = 6 mice per group. c, Representative confocal microscopy images of mCherry⁺ cells in 4T1-GFP tumour 6 h postinjection (n = 6). 40× magnification (scale bar, 10 μm) and its regional magnification (scale bar, 2 μm) are shown. Immunofluorescence staining with anti-CD11c antibody (white) and DAPI (blue). Each experiment was repeated three times. d, In vivo transfection efficiency of SS-OP LNPs and MC3 LNPs measured by IVIS by intratumoural administration of LNP-luciferase-mRNA (mRNA, 0.5 mg kg⁻¹) in 4T1 models (n = 3). e, In vivo uniSTING protein in tumour tissues, normal organs and serum 6 h following i.t. injection of LNP-uniSTING-mRNA (mRNA, 1 mg kg⁻¹) (n = 8). f, Time course of uniSTING expression in tumours (n = 8). g, Treatment scheme and Kaplan–Meier survival of 4T1-Luc2 tumour-bearing mice treated with the indicated formulations (for PBS, LNP-mock-mRNA, LNP-uniSTING-mRNA, 2′3′-cGAMP, n = 10, 10, 12 and 8, respectively). h, 4T1-Luc2 tumour weight following the indicated treatments (n = 8). i, Spider plots of 4T1-Luc2 growth curves, measured by bioluminescence signals (n = 10). j, Treatment scheme and Kaplan–Meier survival of LLC1 tumour-bearing mice with the indicated formulations (n = 7). k, LLC1 tumour weight following the indicated treatments (n = 8). l, Spider plots of LLC1 tumour growth curves (for PBS, LNP-mock-mRNA, LNP-uniSTING-mRNA and 2′3′-cGAMP, n = 7, 7, 7 and 8, respectively). m, Treatment scheme and Kaplan–Meier survival of B16-F10 tumour-bearing mice treated with the indicated formulations (n = 8). n, Spider plots of individual tumour growth curves (n = 6). Log-rank (Mantel–Cox) test. o, Cured mice were reinoculated with 10⁵ B16-F10 cells. Rechallenging tumour growth and Kaplan–Meier survival curves are shown (n = 5). Significance assessed with one-way ANOVA and Tukey’s multiple comparisons test for tumour growth and log-rank (Mantel–Cox) test for survival. p, Treatment scheme and Kaplan–Meier survival of E0771 tumour-bearing mice (WT, Tmem173^−/− and Ifnar^−/− mice) treated with the indicated formulations (for PBS and LNP-uniSTING-mRNA in WT, Tmem173^−/− and Ifnαr^−/− mice, n = 7, 8, 5, 7, 5 and 6, respectively). q, E0771 tumour weight following the indicated treatments (for PBS and LNP-uniSTING-mRNA in WT, Tmem173^−/− and Ifnαr^−/− mice, n = 6, 8, 7, 8, 5 and 6, respectively). Significant differences were assessed using a one-way ANOVA and Tukey’s multiple-comparisons test (b,d,e,f,h,k), log-rank (Mantel–Cox) test (g,j,m,p), and two-way ANOVA with multiple comparisons (q). Results are presented as mean ± s.d.

We next investigated the antitumour effect of LNP-uniSTING-mRNA in vivo. Intratumoural injection of LNP-uniSTING-mRNA efficiently inhibited 4T1-Luc2 breast tumour growth and prolonged survival compared with soluble 2′3′-cGAMP or LNP-mock-mRNA (Fig. 3g–i and Supplementary Figs. 6 and 7). A long-term monitoring experiment demonstrated that uniSTING was superior to mSTING in terms of prolonged survival (Supplementary Fig. 6d). Similar results were obtained in Lewis lung carcinoma (LLC1), B16-F10 melanoma tumour and E0771 breast cancer (Fig. 3j–n and Supplementary Fig. 8). UniSTING eradicated 50% of B16-F10 tumours and effectively prevented tumour rechallenge (Fig. 3m–o). Consistent with in vitro assays, the antitumour efficacy of uniSTING displayed an independence from the endogenous STING expression based on the observed antitumour activity in Tmem173^−/− mice. UniSTING failed to inhibit tumour growth in Ifnαr^−/− mice, confirming its mechanism-of-action is via IFNs (Fig. 3p,q and Supplementary Fig. 9). Hence, LNP-uniSTING-mRNA robustly boosted antitumour efficacy in multiple tumour models independently of endogenous STING. To seek the treatment potential that is feasible for the majority of solid tumours, we next examined the therapeutic efficacy of LNP-uniSTING-mRNA through systemic injection in advanced tumours, such as the 4T1-Luc2 liver metastatic tumour model and the orthotopic Hepa1–6 liver cancer model. Intravenous administration of uniSTING drastically inhibited 4T1 liver metastatic tumour growth and eradicated metastatic tumours in 40% of mice compared with the 2′3′-cGAMP-treated group which showed negligible effect (Fig. 4a–d). Furthermore, the orthotopic Hepa1–6 liver cancer had a much lower tumour burden and prolonged survival after intravenous injection of LNP-uniSTING-mRNA than controls (Fig. 4e–g). Thus, systemic delivery of LNP-uniSTING-mRNA effectively restrained advanced orthotopic/metastatic tumours.

Fig. 4 |. Systemic uniSTING treatment exerts potent antitumour effects on orthotopic/metastatic tumours.

a, Treatment scheme for the 4T1-Luc2 liver metastatic tumour model with the indicated formulations. b, Spider plots of individual tumour growth curves measured by bioluminescence intensity. n = 5 in each group. c, In vivo bioluminescence imaging of mice bearing liver metastatic tumours on days 1 and 10 following treatment. n = 5 biologically independent samples. d, Kaplan–Meier survival curves of mice treated with indicated formulations. n = 8 biologically independent samples. e, Treatment scheme for orthotopic Hepa1–6 HCC tumour-bearing mice with the indicated formulations. f, Average tumour weight in HCC tumour models after indicated treatments. n = 7 biologically independent sample. g, Kaplan–Meier survival curves of mice treated with indicated formulations. n = 8 biologically independent samples. For survival studies, 5 × 10⁷ of bioluminescence intensity was used as the endpoint criteria in the 4T1 liver metastatic tumour model and a 30% weight loss was used as the endpoint criteria in the HCC tumour model. Each line represents one survival curve for each group; log-rank (Mantel–Cox) test. Significant differences were assessed using a one-way ANOVA and Tukey’s multiple-comparisons test. Results are presented as mean ± s.d.

Through immune cell profiling in 4T1-Luc2 tumours, we observed that the overall numbers of CD45⁺ leukocytes or DCs were unchanged among groups (Supplementary Fig. 10a). However, expression of the costimulatory molecule CD86 in DCs was substantially increased by LNP-uniSTING-mRNA (Supplementary Fig. 10a). Similar results were observed in the E0771 tumour model (Supplementary Fig. 11). A notable increase in tumoural cytotoxic CD8⁺ T and NK1.1⁺ cells was observed in the uniSTING-treated group compared with that of control groups, while the change in CD4⁺ T cells was minor (Supplementary Fig. 10a), demonstrating that uniSTING drove immune cell activation^37,38. UniSTING significantly raised the frequency of intratumoural IFNγ⁺ or GzmB⁺ cytotoxic CD8⁺ T cells (Supplementary Fig. 10b). Importantly, compared with the control groups, uniSTING increased tumoural Ifnb1 and Il12a transcripts as expected, but not that of Tnf and Il6 (Supplementary Fig. 10c), consistent with the in vitro findings (Fig. 2g). LNP-uniSTING-mRNA also enhanced the CD8⁺ T-cell frequency in PBMCs in a vaccine study, as illustrated in Supplementary Fig. 10d. This was in contrast to the group that received ADU-S100, a STING agonist used in clinical trials (Supplementary Fig. 10e). Additionally, uniSTING treatment decreased the frequency of CD62L⁺CD44⁻ naive CD8⁺ T cells without a notable alteration of the CD62L⁺CD44⁺ central memory subpopulation (Supplementary Fig. 10f). Importantly, the frequency of CD62L⁻CD44⁺ effect memory CD8⁺ T cells was dramatically increased in the uniSTING group compared with the other groups (Supplementary Fig. 10f, g). Stem-like CD44⁺CD27⁺TCF1⁺CD8⁺ T cells can proliferate, regenerate and differentiate into effector T cells³⁹. We observed a surge in intratumoural CD44⁺CD27⁺TCF1⁺CD8⁺ T cells by injection of LNP-uniSTING-mRNA or ADU-S100, with a higher frequency in the former group (Supplementary Fig. 10h). These findings showed that LNP-uniSTING-mRNA promoted DC maturation and robust CD8⁺ T-cell responses.

UniSTING triggers intracellular communication by exosomal miRNAs

To explore the underlying mechanism of uniSTING’s effect on antitumour immunity, we incubated the DC line DC2.4 with conditioned medium (CM) from uniSTING-mRNA (CM_uniSTING) or mock-mRNA (CM_mock)-treated tumour cells. The levels of IFNβ and CXCL10 mRNA were significantly increased in the CM_uniSTING-treated DC2.4 cells compared with those incubated with CM_mock (Supplementary Fig. 12a), indicating that CM from uniSTING-treated tumour cells contributed to DC activation. Cells communicate with other cells nearby or far away by releasing EVs⁴⁰. We thereby isolated EVs from mock- or uniSTING-treated tumour cell CM (EXO_mock or EXO_uniSTING). The average diameter of obtained EVs was ∼130 nm (Supplementary Fig. 13a), consistent with the reported size of EVs⁴¹. The presence of the canonical exosomal proteins CD9 and CD81 of EVs suggested that these are of an endosomal origin (Supplementary Fig. 13a). The resulting EVs or the supernatants were separately added into DC2.4 cells. Remarkably, DC2.4 cells exposed to EXO_uniSTING exhibited a sharp rise in Ifnb1 and Cxcl10 mRNA compared with cells exposed to EXO_mock, suggesting an increased IFN-I signalling cascade in the EXO_uniSTING-treated group (Supplementary Fig. 13b,c). Intratumoural administration of EXO_uniSTING markedly retarded 4T1-Luc2-tumour growth when compared with EXO_mock-treated group, consistent with an increased release of Ifnb1 and reduced expression of immunosuppressive cytokines in the TME (Supplementary Fig. 13d). An miRNA profiling assay was applied to examine which exosomal miRNAs from EXO_uniSTING are responsible for the crosstalk between tumour cells and DCs. Scatter plots with boundaries defined as ±3-fold changes revealed that 11 miRNAs were significantly changed between EXO_mock and EXO_uniSTING. The role of the top five miRNAs (miR-130b-3p, miR-130a-3p, miR-19a-3p, miR-16–5p and miR-15b-5p) that were upregulated in EXO_uniSTING was evaluated (Supplementary Fig. 13e,f). Target miRNA levels in the EXO_uniSTING were not altered by RNase treatment (Supplementary Fig. 13g), indicating that the miRNAs were contained in EVs rather than on the outer surface. We further synthesized related miRNAs, including miR-130b-3p, miR-130a-3p and miR-19a-3p mimics that have been implicated in the immune response modulation^42–44, and mixed them as a pool to treat DCs. Compared with the control group, the synthetic miRNA pool in DC2.4 cells increased Ifnb1 and Cxcl10 and downregulated immunosuppressive molecules such as Tgfb, Il10 and Mrc1 (Supplementary Figs. 12b and 13h). Notably, miR-130a, miR-19a and miR-16, three of the top five elevated miRNAs in EXO_uniSTING, exhibited a favourable correlation with the prognosis of human breast cancer (Supplementary Fig. 11c). These data suggested that uniSTING-treated tumour cells modulated the immunological responses of DCs via exosomal miRNAs.

To delineate the genes in DCs that were regulated by the miRNAs in EXO_uniSTING, we conducted RNA-seq of EXO_uniSTING- or EXO_mock-treated DC2.4 cells. EXO_uniSTING regulated the expression of a cluster of genes involved in IFN signalling responses, including Ifit2, Ifit3, Isg15 and Usp18⁴¹ (Fig. 5a). GSEA showed that the IFNα response genes were markedly enriched in EXO_uniSTING-treated DC2.4 cells (Fig. 5a). Notably, EXO_uniSTING-treated DC2.4 cells drastically downregulated Wnt signalling-associated genes, especially Wnt2b and Snail^45,46(Fig. 5b). Wnt signalling activation induces tolerogenic regulatory DCs, increases Treg infiltration and impairs CD8⁺ effector T-cell differentiation⁴⁷. EXO_uniSTING treatment was confirmed to reduce Wnt2b transcripts (Fig. 5c). Similar outcomes were also observed in treatment with EVs from E0771 cells and BMDCs (Supplementary Fig. 14a–c). Wnt2b siRNA or α-Wnt2b neutralizing antibody upregulated Ifnb1, Cxcl9 and Cxcl10 mRNA levels but downregulated immunosuppressive Tgfb and Il10 in DC2.4 cells (Fig. 5d and Supplementary Fig. 14d,e), suggesting that Wnt2b serves as a regulatory signal for IFN-I response. Moreover, the Wnt2b level was linked to a poor prognosis of breast cancer in humans (Supplementary Fig. 14f). We next assessed whether the Wnt2b downregulation in DCs is mediated by miRNAs secreted from EXO_uniSTING. The candidate miRNAs targets were predicted by using a sophisticated database⁴⁸. Interestingly, four of the top fourteen Wnt2b-linked miRNAs (miR-130b-3p, miR-130a-3p, miR-16–5p and miR-15b-5p) were found in EXO_uniSTING, statistically demonstrating the impact of these miRNA–target interactions (Fig. 5e). Treatment with the corresponding synthetic miRNA pool decreased Wnt2b expression in DC2.4 cells as expected (Fig. 5f). EXO_uniSTING co-treatment with the synthetic miRNA pool could further repress Wnt2b expression. We next applied miRNA hairpin inhibitors (miRi) to knockout miR-130a-3p and miR-130b-3p in tumour cells. Transfection of miRi Pool resulted in at least a 95% downregulation of miR-130a-3p and miR-130b-3p in 4T1 cells, compared with that of the negative-control miRNA hairpin inhibitor (miRiCtr) (Fig. 5g). EVs derived from 4T1 cells treated with miRi Pool + uniSTING failed to downregulate Wnt2b relative to those derived from cells treated with miRiCtr + uniSTING (Fig. 5h,i). Importantly, miRi Pool + uniSTING treatment did not prime the immune response as assessed by Ifnb1, Cxcl10, Cd80 and Cd86 expression, while miRiCtr + uniSTING showed a remarkable increase in these genes (Fig. 5j). Thus, exosome-mediated crosstalk between DC cells and uniSTING-treated tumour cells was mainly attributed to miRNAs, which targeted the critical regulatory Wnt2b signalling and improved the function of DCs.

Fig. 5 |. Exosomal miRNAs derived from uniSTING-treated tumour cells potentiate DC function by blocking Wnt2b signalling.

a, Left: heatmap of DEGs in EXO_mock- or EXO_uniSTING-treated DC2.4 cells (n = 2 per group). Upregulated (red) or downregulated (blue) DEGs are shown. Right: GSEA enrichment of the IFNα response signatures in samples shown in the left panel. b, Volcano plot showing fold changes of DEGs between EXO_mock-treated and EXO_uniSTING-treated DC2.4 cells (log₂ FC, >1.5; adjusted P < 0.05). c, Wnt2b mRNA expression in DC2.4 cells pretreated with PBS, mock or uniSTING-mRNA, upon the addition of EXO_mock or EXO_uniSTING. n = 6 per group. d, Transcripts of Wnt2b, Ifnb1, Cxcl9, Cxcl10, Tgfb and Il10 in DC2.4 cells 48 h after treatment with siRNA against Wnt2b or siRNA negative control, α-Wnt2b antibody and PBS, respectively. n = 6 per group. e, Integrated scores predicting the association between miRNAs with Wnt2b mRNA. Higher scores indicate a stronger association. f, Wnt2b expression in DC2.4 cells after treatment with synthetic miRNA pool (mixture of 25 nM miR-130b-3p, miR-130a-3p and miR-19a-3p mimics) or control miR mimic, upon the addition of EXO_mock or EXO_uniSTING. n = 5 per group. g, Relative miR-130a-3p and miR-130b-3p expression in mock/uniSTING-treated tumour cells transfected with inhibitory miRiCtr or miRi Pool (miRi 130a-3p, miRi 130b-3p). n = 6 per group. h,i, Wnt2b expression in DC2.4 cells after treatment with EVs collected via ultracentrifugation of the CM from 4T1 tumour cells treated with mock + miRiCtr, mock + miRi Pool, uniSTING + miRiCtr or uniSTING + miRi Pool, analysed by western blotting (h) or qPCR (i). Non-treated DC2.4 cells were used as control. n = 6 per group. j, Transcripts of Ifnb1, Cxcl10, Cd80 and Cd86 in DC2.4 cells after treatment with EVs collected via ultracentrifugation of the CM from 4T1 tumour cells treated with mock + miRiCtr, mock + miRi Pool, uniSTING + miRiCtr and uniSTING + miRi Pool. Non-treated DC2.4 cells were used as control. n = 5 per group. Significant differences were assessed using a one-way ANOVA and Tukey’s multiple-comparisons test and two-way ANOVA with multiple comparisons. Results are presented as mean ± s.d.

anti-Wnt2b enhances in vivo antitumour activity of uniSTING

We next investigated whether Wnt2b blockade enhanced uniSTING’s antitumour activity. Dual treatment of LNP-uniSTING-mRNA with α-Wnt2b greatly augmented antitumour efficacy as evidenced by reduced 4T1-Luc2 tumour weight and prolonged survival compared with monotherapies or the soluble 2′3′-cGAMP plus α-Wnt2b group (Fig. 6a–c). A similar effect was observed in the E0771 breast cancer model and the LLC1 lung cancer model (Fig. 6d,e and Supplementary Fig. 15). α-Wnt2b with LNP-uniSTING-mRNA but not with ADU-S100 gained maximum antitumour benefit (Fig. 6d,e). Along with strengthening cytotoxic CD8⁺ T-cell responses (Fig. 5a,b,e) and promoting memory T-cell differentiation (Fig. 5f–h), LNP-uniSTING-mRNA together with Wnt2b blockade further increased the frequency of intratumoural CD45⁺CD8⁺ T cells and its CD27⁺TCF1⁺ memory subpopulation (Fig. 6f,g). When paired with LNP-uniSTING-mRNA but not with ADU-S100, more cytotoxic GzmB⁺CD8⁺ T cells were detected in the group treated with α-Wnt2b (Fig. 6h). These findings show that α-Wnt2b synergized with LNP-uniSTING-mRNA to elicit antitumour immunity.

Fig. 6 |. α-Wnt2b antibody enhances in vivo antitumour activity of STING activation.

a, Treatment scheme for 4T1-Luc2 tumour-bearing mice with the indicated formulations. b, Average tumour weight in 4T1-Luc2 tumour models after indicated treatments. n = 10 biologically independent sample. c, Kaplan–Meier survival curves of 4T1-Luc2 tumour-bearing mice treated with indicated formulations. For PBS, LNP-mock-mRNA, α-Wnt2b, LNP-uniSTING-mRNA, LNP-uniSTING-mRNA + α-Wnt2b, 2′3′-cGAMP and 2′3′-cGAMP + α-Wnt2b, n = 10, 10, 10, 10, 9, 7 and 11 biologically independent samples, respectively. Each line represents one survival curve for each group; log-rank (Mantel–Cox) test. d, Treatment scheme for LLC1 tumour-bearing mice with the indicated formulations. e, Kaplan–Meier survival curves of LLC1 tumour-bearing mice treated with indicated formulations. n = 5 biologically independent sample. In all the survival analyses, an 800 mm³ tumour volume was used as the endpoint criterion. Each line represents one survival curve for each group; log-rank (Mantel–Cox) test. f, Impact of indicated treatments on CD8⁺ frequency in CD45⁺ cells. For PBS, LNP-mock-mRNA, α-Wnt2b, LNP-uniSTING-mRNA, LNP-uniSTING-mRNA + α-Wnt2b, ADU-S100 and ADU-S100 + α-Wnt2b, n = 3, 3, 3, 3, 5, 3 and 3 biologically independent samples, respectively. g, The percentage of CD27⁺TCF1⁺ memory CD8⁺ T cells in lung tumours. For PBS, LNP-mock-mRNA, α-Wnt2b, LNP-uniSTING-mRNA, LNP-uniSTING-mRNA + α-Wnt2b, ADU-S100 and ADU-S100 + α-Wnt2b, n = 3, 3, 3, 3, 5, 5 and 3 biologically independent samples, respectively. h, The percentage of GzmB⁺ cells in CD8⁺ T cells. For PBS, LNP-mock-mRNA, α-Wnt2b, LNP-uniSTING-mRNA, LNP-uniSTING-mRNA + α-Wnt2b, ADU-S100 and ADU-S100 + α-Wnt2b, n = 3, 3, 3, 3, 5, 3 and 3 biologically independent samples, respectively. Gating strategies for flow cytometry are shown in Supplementary Fig. 18. Significant differences were assessed using a one-way ANOVA and Tukey’s multiple-comparisons test (b,f–h). Results are presented as mean ± s.d.

Conclusions

The clinical potential of STING agonists is currently constrained by issues including poor stability, deliverability, undesired side-effect toxicity following systemic delivery, and downregulation or loss of endogenous STING in the TME. To address these challenges, we developed uniSTING based on a non-membrane-associated tetrameric subunit as a universal, constitutively active STING mimic for the activation of STING signalling pathways, favouring IRF3/IFN-I activity over proinflammatory NF-κB signalling in a manner independent of the endogenous STING. Furthermore, through LNP-mediated delivery of mRNA encoding designed STING mimic, we demonstrated the continuous expression of uniSTING in tumour cells and DCs for up to 7 days. Importantly, treatment with LNP-uniSTING-mRNA greatly reduced the tumour burden in multiple poorly immunogenic murine tumours, including TNBCs, LLC1, B16-F10 and advanced orthotopic/metastatic liver tumour models, which showed negligible response when treated with 2′3′-cGAMP. Mechanistic studies validated that uniSTING-induced activation of STING signalling remained in both Tmem173^−/− BMDCs in vitro and tumour-bearing Tmem173^−/− mice, mirroring the downregulation or absence of STING in the TME. Surprisingly, uniSTING selectively induced IRF3-dependent signalling, fostering the specific CD8⁺ T-cell antitumour response but not the undesired proinflammatory NF-κB signalling. LNP-uniSTING-mRNA potently boosted the activation of DCs, increased the frequency of tumoural CD8⁺ T cells and promoted the differentiation of the T memory subpopulation and the cytotoxic CD8⁺ T-cell antitumour response. Importantly, EVs from uniSTING-treated tumour cells mediated crosstalk between tumour cells and DCs. Furthermore, Wnt2b-signalling-mediated negative feedback was revealed in DCs in response to STING activation, which can be relieved by exosomal miRNAs derived from uniSTING-treated tumour cells. A combination of LNP-uniSTING-mRNA with α-Wnt2b antibody synergistically inhibited tumour growth and prolonged survival. Mechanically, dual therapy of LNP-uniSTING-mRNA with α-Wnt2b further replenished tumoural CD8⁺ T cells and their memory subpopulation, and increased the percentage of cytotoxic GzmB⁺ CD8⁺ T-cells. Taken together, LNP-uniSTING-mRNA triggered robust antitumour immunity across multiple advanced cancer types and addressed current limits of STING agonists, rendering it as an attractive candidate for translational applications, especially for the treatment of intractable STING-deficient tumour types.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41565–024-01624–2.

Methods

Materials

COATSOME SS-OP, 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and 1,2-dimyristoyl-rac-glycero-3-methylpolyoxyethylene (DMG-PEG2000) were purchased from NOF. Cholesterol was purchased from Avanti Polar Lipids. 2′3′-cGAMP was purchased from Invivogen. Luciferase Assay System was purchased from Promega. IVISbrite D-Luciferin potassium salt bioluminescent substrate was purchased from PerkinElmer. EndFree Plasmid Maxi Kit was purchased from Qiagen. All other reagents and chemicals were obtained from Sigma-Aldrich, Life Technologies and TriLink BioTechnologies unless otherwise stated. Antibodies used for western blotting, flow cytometry and immunofluorescence staining are listed in Supplementary Table 1. Primers used for quantitative polymerase chain reaction (qPCR) are listed in Supplementary Table 2.

Cell culture

4T1, 4T1-Luc2, B16-F10, Hepa1–6, ES-2, THP-1, LLC1 and HEK293T cell lines were obtained from ATCC. DC2.4 cell line was obtained from MilliporeSigma. ExpiCHO cell line was obtained from ThermoFisher Scientific. E0771 cell line was obtained from CH3 Biosystems. 4T1 parental cells were stably transfected with a vector carrying green fluorescent protein (GFP) and the puromycin resistance gene. 4T1-Luc2 cells were cultured in RPMI 1640 (Gibco) supplemented with 10% fetal bovine serum (FBS, Gibco) and 8 μg ml⁻¹ blasticidin. 4T1-GFP cells were cultured in RPMI 1640 supplemented with 10% FBS and 1 μg ml⁻¹ puromycin. 4T1 WT cells were cultured in RPMI 1640 supplemented with 10% FBS. DC2.4 cells were cultured with RPMI 1640 supplemented with 10% FBS, 1× non-essential amino acids (Gibco), 1× HEPES buffer (Gibco) and 0.0054× β-mercaptoethanol (Gibco). HEK293T cells were cultured in high-glucose Dulbecco’s modified Eagle’s medium (DMEM, Gibco) supplemented with 10% FBS. E0771, LLC1, Hepa1–6 and B16-F10 cells were cultured in DMEM (Gibco) supplemented with 10% FBS. ES-2 cells were cultured in McCoy’s 5a medium with 10% FBS. THP-1 cells were cultured with RPMI 1640 supplemented with 10% FBS and 0.05 mM 2-mercaptoethanol. All the adherent cell lines were cultured at 37 °C and 5% CO₂ in a humidified atmosphere. ExpiCHO cells were cultured in ExpiCHO Expression Medium at 37 °C and 8% CO₂ in a humidified atmosphere on an orbital shaker platform (125 ± 5 r.p.m.).

Primary BMDCs were prepared from WT, Tmem173^−/−, Irf3^−/− and Ifnαr^−/− C57BL/6 mice. Briefly, femurs and tibias were collected, and bone marrow was flushed out with ice-cold RPMI 1640 using a sterile syringe and a 21G needle. After red blood cell lysis, bone marrow cells were resuspended in RPMI 1640 medium containing 10% heat-inactivated FBS, murine recombinant granulocyte–macrophage colony-stimulating factor (GM-CSF, 20 ng ml⁻¹, BioLegend) and 1% antibiotic–antimycotic (Gibco) and distributed into 100 mm Petri dishes at a density of 3 × 10⁶ cells per dish for 6 days. Every 2 days, cells were resupplemented with fresh complete medium. On day 6, adherent cells were discarded and CD11c⁺F4/80⁻ cells were sorted.

Experimental mouse models

All animals were maintained at the animal facilities at the University of North Carolina at Chapel Hill under specific pathogen-free conditions. Animal experiments were carried out in accordance with protocols (ethical approval numbers 17–075.0 and 18–071.0) approved by the University of North Carolina’s Institutional Animal Care and Use Committee. Six-week-old female WT C57BL/6J and BALB/cJ mice were purchased from Jackson Laboratory. Tmem173^−/− mice and Ifnαr^−/− mice were from Jackson Laboratory. Irf3^−/− mice were provided by T. Taniguchi (University of Tokyo) and were obtained from J. K. Whitmire (University of North Carolina at Chapel Hill). All knockout strains were maintained on a C57BL/6J background. Mice were held in animal facilities with a 12 h dark/light cycle at noise levels below 50 dBA. The average temperature was held around 20–22 °C and the air humidity was maintained between 45% and 65%.

For E0771 tumour models, 4 × 10⁵ cells in 20 μl PBS were injected into the mammary fat pads of C57BL/6J mice and the tumour size was measured every 2–3 days using a digital caliper. Tumour volume was estimated using the formula: tumour volume = length × width²/2. For 4T1-Luc2 tumour models, 2 × 10⁵ to 1 × 10⁶ cells in 20 μl PBS were injected into the mammary fat pads of BALB/cJ mice. The tumour progression was monitored by an IVIS Kinetics Optical System (PerkinElmer) after intraperitoneal injection of 100 μl of D-luciferin (10 mg ml⁻¹). For LLC1 tumour models, 5 × 10⁵ cells in 25 μl PBS were injected into the right flank of C57BL/6J mice. For B16-F10 tumour models, 1 × 10⁵ cells in 25 μl PBS were injected into the right flank of C57BL/6J mice. For the rechallenge studies, cured mice were inoculated with 1 × 10⁵ B16-F10 cells for tumour rechallenging. Tumour size was directly measured with a digital caliper. For orthotopic HCC tumour models, 5 × 10⁶ Hepa1–6 cells in 30 μl PBS were inoculated at the subcapsular region of the liver’s left lobes in C57BL/6J mice. To establish the hemi-spleen 4T1 liver metastasis model, the spleen was exposed and tied in half using suture. Then, 1 × 10⁶ 4T1-Luc2 tumour cells in 50 μl PBS were inoculated directly into the spleen, followed by resecting the inoculated portion of the spleen. The rest of the spleen was returned before suturing. For all the different mouse models, mice were randomly assigned to treatment groups. The investigators were blinded to the group allocation during the animal experiments. On reaching an average tumour size of ∼60 mm³, mice were treated by intratumoural injection with either PBS, 2′3′-cGAMP or the indicated LNP-mRNA formulations, or intraperitoneal injection with α-Wnt2b antibody. For survival studies, mice were killed when tumours reached 800 or 1000 mm³ or when the mice lost more than 30% of their weight.

In vitro transfection studies

For plasmid transfection, HEK293T cells were cultured until 80% confluence and transfected with the vector-encoding gene of interest through lipofectamine 2000, according to the manufacturer’s instructions. HEK293T cells were collected 48 h after transfection for further analysis. For recombinant expression of uniSTING, ExpiCHO cells were transfected with pCDNA3.4-plasmid-encoding uniSTING using the ExpiFectamine CHO Transfection Kit (ThermoFisher Scientific), according to the manufacturer’s instructions. After 6 days, cells were collected for protein purification. For mRNA transfection, cells were transfected with mRNA either mixed with TransIT-mRNA transfection agent (Mirus Bio) or encapsulated into SS-OP LNPs in DMEM culture medium supplemented with 10% FBS, according to the manufacturer’s instructions or conditions optimized in the Liu lab. After 24 or 48 h, cells or supernatant were collected for further analysis. For siRNA transfection, cells were transfected with 10 nM of ON-TARGETplus mouse Wnt2b siRNA pool or non-targeting Control siRNAs (Dharmacon, Horizon Discovery) using the INTERFERin transfection reagent (Polyplus-transfection) in a DMEM culture medium supplemented with 10% FBS, according to the manufacturer’s instructions. After 24 or 48 h, cells were collected and followed by qPCR analysis. For miRNA transfection, DCs were transfected with 15 nM miRIDIAN mouse miRNA pool (miR-130b-3p mimic, miR-130a-3p mimic and miR-19a-3p mimic) (Dharmacon, Horizon Discovery) or miRIDIAN miRNA mimic negative control using the DharmaFECT 1 transfection reagent (Dharmacon, Horizon Discovery) in DMEM culture medium, according to the manufacturer’s instructions. After 4 h, EVs isolated from tumour cells with different treatments were added into the culture medium, and the cells were incubated at 37 °C and 5% CO₂ in a humidified atmosphere for 24 h followed by qPCR analysis.

uniSTING protein generation and characterization

The constitutively active STING mimic protein was designed by genetically fusing an extremely thermostable tetramerization motif (amino acid sequence GIINETADDIVYRLTVIIDDRYESLKNLITL-RADRLEMIINDNVSTILASIG)²⁴ with the C-terminal cytoplasmic domain of STING (residues 138–378 of murine STING for the mouse version or residues 139–379 of human STING for the human version) via a flexible linker. A Flag tag was engineered at the N terminus to facilitate protein purification and detection. The expression vectors encoding the fusion protein were generated by inserting synthetic DNAs (codon optimized for expression in mammalian cells) into pCDNA3.4 vector. For protein expression, ExpiCHO cells were transfected with pCDNA3.4-plasmid-encoding uniSTING. The cell lysates from 1 × 10⁷ cells were collected 6 days after transfection by using 1 ml of lysis buffer (50 mM Tris–HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA and 1% Triton X-100) with Halt protease inhibitor cocktail (ThermoFisher Scientific). The uniSTING protein was purified via anti-Flag M2 affinity gel (Sigma-Aldrich) according to the manufacturer’s protocol. The purified proteins were analysed on 4–12% SDS–PAGE gel (Invitrogen) with Coomassie G-250 (Bio-Rad) stain. Molecular weight analysis was performed by fast protein liquid chromatography (GE AKTA) with a Superdex 200 10/300 GL gel filtration column (MilliporeSigma). First, 200 μg of each standard (as shown in Supplementary Fig. 1b) or 400 μg of uniSTING protein was dissolved in 200 μl of TBS (50 mM Tris–HCl, 150 mM NaCl, pH 7.4) buffer with 0.02% Tween 80 and filtered through a 0.45 μm filter. The protein samples were then injected into the column and separated at a flow rate of 0.3 ml min⁻¹ and detected at 280 nm. The molecular weight of uniSTING was finally determined from the standard curve. The melting temperature (T_m) of uniSTING was measured by using a Protein Thermal Shift kit (ThermoFisher Scientific) according to the manufacturer’s protocol. Briefly, uniSTING was mixed with different buffers (as shown in Supplementary Fig. 1c) and Protein Thermal Shift dye, and a melting curve experiment was run on a real-time PCR instrument (Analytik Jena, qTOWER³ G). The T_m was calculated from the melting curve based on Protein Thermal Shift software (ThermoFisher Scientific).

mRNA synthesis and LNP preparation

mRNA for studies of mCherry expression in vivo was obtained from TriLink BioTechnologies⁴⁹. The expression vectors encoding uniSTING and mSTING were generated by inserting the corresponding open reading frame into a modified pUC57 vector optimized for in vitro transcription. All plasmids were sequenced to confirm the sequence accuracy. The cDNA used for in vitro transcription was PCR-amplified using a T7 promoter primer and a 3′ primer containing 150 Ts. mRNAs were transcribed in vitro by using MEGAscript kits in the presence of Cleancap AG (TriLink) as a 5′-capping agent according to the manufacturer’s instructions. The resulting mRNAs were purified by using a MEGAclear Transcription Clean-Up Kit (ThermoFisher Scientific) and characterized prior to being encapsulated into LNPs. All mRNAs were stored at −80 °C, and were allowed to thaw on ice before use. The SS-OP LNPs were prepared according to previous reports and the manufacturer’s instructions⁵⁰. In brief, an organic phase was prepared by solubilizing with ethanol a mixture of ionizable SS-OP lipid, DOPC, cholesterol and DMG-PEG2000 at a 52.5/7.5/40/1.5 molar ratio. The mRNA solution was prepared in 20 mM malic acid buffer with 30 mM NaCl (pH 3.0) at aconcentration of 66.7 μg ml⁻¹. To prepare mRNA/LNPs, mRNA solution was rapidly mixed with the lipid ethanol solution at room temperature under vortexing. MES buffer (5.25 ml, pH 5.5, 20 mM) was then added to the mixture. Subsequent buffer exchange was conducted by using Amicon Ultra-15–100 Kcentrifugal units (Merck Millipore). The size, polydispersity index and zeta potentials of LNPs were measured with a Malvern Zetasizer Nano ZS90. Diameters were reported as the volume mean peak average. LNPs were also characterized by using cryogenic electron microscopy. To define the cell types responsible for mRNA uptake and its translation in vivo, 4T1-GFP-tumour-bearing mice were intratumorally administered with LNP-mCherry-mRNA, followed by flow cytometric and confocal microscope analyses for mCherry⁺ cells in tumour tissues.

Western blot analysis

Cells were lysed in radioimmunoprecipitation buffer (Invitrogen). The protein concentration of samples was determined with a BCA Protein Assay Kit (Pierce). First, 10–30 μg of proteins were mixed with NuPage LDS sample buffer (Invitrogen) and NuPage sample reducing agent (Invitrogen) and heated at 95 °C for 5 min. The denatured samples were then loaded and separated on a 4–12% Bis–Tris NuPage gel (Invitrogen). The proteins were then blotted on a polyvinylidene difluoride membrane (Bio-Rad) for 1 h at 20 V. The membrane was blocked with 5% bovine serum albumin (BSA) in TBST for 1 h at room temperature and then incubated with primary antibodies overnight at 4 °C while shaking. The membranes were washed and further incubated with a secondary antibody for 40 min at room temperature and then detected using the Clarity Western ECL Substrate (Bio-Rad). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as the control. The ChemiDoc Imaging System and Image lab 5.1 (Bio-Rad) were used to acquire and analyse the images.

Colocalization and coimmunoprecipitation of uniSTING with TBK1 and IRF3

The interactions between uniSTING and TBK1 or IRF3 were analysed through confocal microscopy imaging and coimmunoprecipitation. For confocal microscopy imaging, the expression vectors encoding Flag-tagged uniSTING, TBK1-GFP and IRF3-HA were generated by inserting synthetic coding DNAs (codons optimized for expression in mammalian cells) into pCDNA3.4 vectors. HEK293T cells were cultured until 80% confluence and transfected with the expression vector encoding a gene of interest. HEK293T cells were rinsed in PBS 48 h after transfection and then placed in 4% paraformaldehyde for 15 min at 4 °C. Cells were then processed through permeabilization and blocking with 5% goat serum at room temperature for 1 h. Primary antibodies were incubated overnight at 4 °C and rinsed with PBS three times, followed by fluorescent secondary antibody staining (37 °C, 1 h). Finally, the cells were mounted with Prolong Diamond Antifade Mountant with 4,6-diamidino-2-phenylindole (DAPI, ThermoFisher Scientific). Immunofluorescence images were taken with a laser scanning confocal microscope (Zeiss LSM 700). Zeiss ZEN2.3 was used to analyse the colocalization of uniSTING, TBK1 and IRF3. For coimmunoprecipitation experiments, Flag-tagged uniSTING was first captured on anti-Flag M2 affinity resin in TBS buffer with protease inhibitor cocktails. Cell lysates from DC2.4 cells were incubated with uniSTING bound to anti-Flag M2 resin for 4 h at 4 °C and washed three times with TBS buffer. The resin was then boiled in SDS-loading buffer and analysed by immunoblotting. The bound Flag fusion protein was also eluted with 0.1 M glycine HCl, pH 3.5, into vials containing 15–25 μl of 1 M Tris, pH 8.0, and analysed by immunoblotting.

Enzyme-linked immunosorbent assay

The expression of Flag-tagged uniSTING protein in the tumour tissue, normal organs and serum was measured after intratumoural injection of LNP-uniSTING-mRNA. Briefly, tissue and plasma samples were collected at different time points after injection. The tissue samples were lysed using RIPA lysis buffer containing protease inhibitor cocktail mix and 5 mM EDTA. The cell lysates were centrifuged and the proteins in the supernatant were collected. A BCA kit was used to measure the protein concentration according to the manufacturer’s protocols. For enzyme-linked immunosorbent assay (ELISA), flat-bottomed 96-well plates (ThermoFisher Scientific) were precoated with anti-Flag-tag monoclonal antibody (ThermoFisher Scientific) at a concentration of 0.5 μg ml⁻¹ per well in 100 mM carbonate buffer (pH 9.6) at 4 °C overnight, and 5% BSA in PBST was then used to block the non-specific binding for 1 h at room temperature. The samples were diluted 50 times in PBST buffer and added to the wells. After incubation for 2 h at room temperature, the plate wells were washed with PBST five times, followed by incubation with the detective antibody (rabbit polyclonal anti-STING antibody, 1:1,000 dilutions, ThermoFisher Scientific) for 1 h at room temperature. Horseradish peroxidase (HRP)-conjugated goat anti-rabbit IgG (CST) was used at a dilution of 1:10,000 in the PBST buffer with 5% BSA for labelling. After adding the HRP substrates, optical densities were determined at a wavelength of 450 nm in the plate reader (Bio-Rad). IFNβ expression in DCs and tumour cells after treatment with either PBS, EGFP-mRNA (mock), uniSTING-mRNA, mSTING-mRNA or 2′3′-cGAMP was also measured by using a mouse IFNβ ELISA kit (R&D systems) according to the manufacturer’s instructions.

Flow cytometry assay

Tumour tissues were harvested and digested in RPMI 1640 medium containing 2% FBS, collagenase type I (200 U ml⁻¹, Invitrogen), collagenase type IV (200 U ml⁻¹, Invitrogen) and DNAase I (100 μg ml⁻¹, Invitrogen) at 37 °C for 1 h to generate single-cell suspensions. The tumour-infiltrating lymphocytes were isolated by density gradient centrifugation and diluted to 1 × 10⁶ cells ml⁻¹ for staining with LIVE/DEAD Fixable Near-IR dye, followed by surface staining with fluorescently conjugated antibodies. For determination of cytokine production, tumour-infiltrating lymphocytes were incubated with phorbol myristate acetate/ionomycin for 4 h. Brefeldin A (Biolegend) was added 30 min after the beginning of stimulation. After stimulation, cell surfaces were stained, and cells were subsequently fixed and permeabilized with a Cytoperm Fixation/Permeabilization Solution Kit (BD) for intracellular staining according to the manufacturer’s protocol. Data were collected with a BD LSRII and a LSRFortessa flow cytometer with BD FACSDIVA software and analysed using FlowJo v.10 software (TreeStar); the gating strategy is shown in Supplementary Figs. 16–18. All antibodies are listed in Supplementary Table 1.

For sorting BMDCs, cells were stained with BV510 anti-mouse CD45, APC anti-mouse CD11c, PE anti-mouse F4/80 and LIVE/DEAD fixable dye staining, and sorted according to the gating strategy displayed in Fig. 2o and Supplementary Figs. 16–18.

qPCR assay

Total RNAs were extracted from cells or tumour tissue samples with an RNeasy Plus Mini Kit (Qiagen) according to the manufacturer’s instructions. cDNA was reverse transcribed with iScript Reverse Transcription Supermix (Bio-Rad) and amplified with TaqMan Gene Expression Master Mix (Applied Biosystems) or iTaq Universal SYBR Green Supermix (Bio-Rad). qPCR was performed using an Applied Biosystems 7500 Fast and 7500 Real-Time PCR system. Data were analysed with 7500 v.2.3 software. All the mouse-specific primers are listed in Supplementary Table 2. Expression was normalized to the glyceraldehyde 3-phosphate dehydrogenase internal control.

EV isolation

Tumour cells were cultured in medium supplemented with 10% EV-depleted FBS. After treatment with mock- or uniSTING-mRNA, EVs were collected by serial centrifugation as reported in the literature⁴¹. Briefly, the supernatants were sequentially centrifuged at 500g for 10 min and 1,000g for 20 min at 4 °C to remove debris and dead cells. The collected supernatant was then ultracentrifuged at 12,000g for 20 min at 4 °C (Optima L-100 XP, Beckman Coulter) to remove apoptotic bodies and shedding vesicles. EVs were pelleted by ultracentrifugation at 100,000g for 70 min at 4 °C. The collected EVs were washed twice with PBS. The biomarkers of EVs were characterized by western blot using CD9 and CD81 antibodies. EVs were also characterized by transmission electron microscopy.

miRNA array analysis

miRNAs were extracted from EVs using an miRNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. The isolated RNAs were converted into cDNA by using the miScript II reverse transcription reaction with HiSpec Buffer (Qiagen). cDNA was used as the template in qPCR with the miScript miRNA PCR Array Mouse Inflammatory Response & Autoimmunity and miScript SYBR Green PCR Kit (Qiagen). qPCR was performed on a Thermal Cycler ABI9700 and QuantStudio 6 Flex Real-Time PCR system using the following thermocycling parameters: 94 °C for 15 min, 40 cycles at 94 °C for 10 s, 55 °C for 30 s and 70 °C for 30 s, followed by a melting curve analysis with QuantStudio software v.1.7.2 (Applied Biosystems). The data were further analysed by using the online miRNA PCR array data analysis tool (www.qiagen.com/us/shop/pcr/primer-sets/miscript-mirna-pcr-arrays/#resources) according to the manufacturer’s instructions. miRNAs with a threshold cycle value >35 were considered as undetermined to minimize the potential noise introduced by measurements below the detection threshold.

RNA deep-sequencing

Total RNA was extracted from DC2.4 cells and 4T1 cells after different treatments by using an RNeasy Plus Mini Kit (Qiagen) according to manufacturer’s instructions. Contaminating genomic DNA was removed by using the specific column from the kit. Library preparation and RNA deep-sequencing were performed by Novogene. Briefly, a library for transcriptome sequencing was generated by using an NEBNext Ultra RNA Library Prep Kit for Illumina (NEB) following the manufacturer’s recommendations, and the index codes were added to attribute sequences to each sample. The library quality was determined in an Agilent Bioanalyzer 2100 system. After removing reads with adapter or of low quality, clean reads were aligned to the mouse GRCm39 reference genome using hisat2 software, and only the uniquely mapped reads were retained for downstream analysis. The gene expression level was quantified using the FeatureCounts function of Subread v.2.0.1. Data were then processed with a pipeline that used the EdgeR Bioconductor package for normalization and differential expression analysis in a paired strategy. The differential expression genes were analysed by the DESeq2 package with default settings using total read counts as input. Fold change and log₂(ratio) values were calculated to represent gene expression differences between conditions. Heatmaps ofgene expression were generated based on the z-score values of the normalized expression matrix from DESeq2 analysis (www.broadinstitute.org/GENE-E/).

GSEA

GSEA 4.2.2 was performed using the Java application available from the Broad Institute (www.broadinstitute.org/gsea/). The gene set database Hallmarks (h.all.v6.1.symbols.gmt) from the Molecular Signatures Database (MSigDB) was used in the analysis. One thousand gene set permutations were performed. An false discovery rate cut-off of <0.05 was used for enriched terms, as is recommended when performing permutations by gene set. R v.3.5.0 was used for analysis.

Statistical analysis

Data are expressed as mean ± s.d. Statistical significance was determined by unpaired two-tailed Student’s t-test when only two value sets were compared or by analysis of variance (ANOVA) comparison between multiple groups. The log-rank Mantel–Cox test was used for survival curves. Exact P values are documented in the figures. All statistical analyses were performed with GraphPad Prism 7.0. No exclusion criteria were incorporated in the design of the experiments for this study.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The MicroRNA Data Integration Portal (mirDIP) was used to identify gene targets for exosomal miRNAs and can be accessed at http://ophid.utoronto.ca/mirDIP. The gene set database Hallmarks (h.all. v6.1.symbols.gmt) from the Molecular Signatures Database (MSigDB) was used in the analysis. All raw sequencing data and associated processed data files that support the findings of this study have been deposited in the Gene Expression Omnibus under accession code GSE253724 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE253724). Source data are available for Figs. 2d–g,i,j, 3b,d–q, 4b,d,f,g, 5a–g,i,j and 6b,c,e–h and Supplementary Figs. 1c, 2a, 6b–d, 8c,d, 9 and 13d–f in the associated source data file. Source data are provided with this paper.