Research ArticleTUMOR IMMUNOLOGY

RIG-I activation is critical for responsiveness to checkpoint blockade

See allHide authors and affiliations

Science Immunology  13 Sep 2019:
Vol. 4, Issue 39, eaau8943
DOI: 10.1126/sciimmunol.aau8943

Priming responses to checkpoint blockade

Although activation of intracellular DNA sensing has been proposed as a means to promote antitumor immunity, molecules that regulate sensing of intracellular RNAs have received considerably less attention in this setting. Here, Heidegger et al. report that expression of RNA sensor RIG-I in tumor cells plays a vital role in promoting responsiveness to anti–CTLA-4 therapy in mouse models of cancer. By engineering melanoma cell lines lacking key molecules involved in DNA and RNA sensing, cell death, and type I interferon signaling, the authors have catalogued the relative importance of these pathways in regulating antitumor immunity and responsiveness to checkpoint blockade. The authors propose that activation of RNA sensing could be used to increase the immunogenicity of poorly immunogenic tumors.

Abstract

Achieving durable clinical responses to immune checkpoint inhibitors remains a challenge. Here, we demonstrate that immunotherapy with anti–CTLA-4 and its combination with anti–PD-1 rely on tumor cell–intrinsic activation of the cytosolic RNA receptor RIG-I. Mechanistically, tumor cell–intrinsic RIG-I signaling induced caspase-3–mediated tumor cell death, cross-presentation of tumor-associated antigen by CD103+ dendritic cells, subsequent expansion of tumor antigen–specific CD8+ T cells, and their accumulation within the tumor tissue. Consistently, therapeutic targeting of RIG-I with 5′– triphosphorylated RNA in both tumor and nonmalignant host cells potently augmented the efficacy of CTLA-4 checkpoint blockade in several preclinical cancer models. In humans, transcriptome analysis of primary melanoma samples revealed a strong association between high expression of DDX58 (the gene encoding RIG-I), T cell receptor and antigen presentation pathway activity, and prolonged overall survival. Moreover, in patients with melanoma treated with anti–CTLA-4 checkpoint blockade, high DDX58 RIG-I transcriptional activity significantly associated with durable clinical responses. Our data thus identify activation of RIG-I signaling in tumors and their microenvironment as a crucial component for checkpoint inhibitor–mediated immunotherapy of cancer.

INTRODUCTION

Targeting immune checkpoints such as cytotoxic T lymphocyte–associated protein 4 (CTLA-4) or programmed cell death protein 1 (PD-1) has been proven to be an effective tumor therapy (1, 2). Dual inhibition of both pathways showed particularly strong clinical responses in different malignancies including melanoma, renal carcinoma, and non–small cell lung cancer (35). However, substantial variation in response between patients exists, and our understanding of the underlying mechanisms remains limited. Several factors have been suggested to predict treatment response to immune checkpoint blockade (ICB) including tumor mutational burden, expression and variety of major histocompatibility complex (MHC) molecules, a T cell–inflamed tumor microenvironment, composition of the gut microbiota, and absence of tumor immunoinhibitory pathways (6). In general, ICB is considered to (re)invigorate tumor antigen–specific T cells (79). Type I interferon (IFN-I) is an important factor for the development of such antitumor T cell responses because it drives intratumoral accumulation of CD8α+ dendritic cells (DCs) and cross-priming of antigen-specific cytotoxic T cells (CTLs) against growing tumors, ultimately leading to tumor regression (10, 11). Accordingly, peritumoral activation of the IFN-I system was associated with intratumoral CTL accumulation and signs of spontaneous regression of primary murine melanomas (12). IFN-I gene signatures in bulk tumor samples have been associated with favorable clinical responses to ICB in patients with melanoma (13).

Regarding IFN-I–inducing pathways in the tumor microenvironment, recent work uncovered an important contribution of cytosolic nucleic acid–sensing receptor systems (14). Although these innate immune receptors are frequently expressed in tumor cells, their role in the development of antitumor immunity has mainly been addressed within host myeloid cells. In particular, the cGAS (cyclic guanosine monophosphate–adenosine monophosphate synthase)/STING [stimulator of interferon genes, also known as transmembrane protein 173 (TMEM173)] pathway in DCs has been associated with tumor rejection during ICB and radiation therapy–induced tumor immune surveillance (15, 16). cGAS/STING is a cytosolic pattern recognition receptor system that recognizes and signals the presence of aberrant cytosolic double-stranded DNA (dsDNA) produced under a variety of conditions including apoptosis, necrosis, and replication of DNA pathogens. In DCs, STING can be activated by DNA originating from dying cancer cells (enhanced by genotoxic stress during irradiation or chemotherapy), thus inducing high levels of IFN-I, which can promote optimal cross-priming of T cells and efficient antitumor immune responses (16).

Another important cytosolic nucleic acid sensor, the RNA receptor RIG-I (retinoic acid inducible gene I, encoded by DDX58) has also been linked to antitumor immune responses (17, 18). RIG-I primarily detects double-stranded 5′–triphosphate RNA (3pRNA) during viral or bacterial infection. Upon ligand binding in myeloid cells, RIG-I recruits the adaptor mitochondrial antiviral signaling protein (MAVS) to induce proinflammatory cytokine and IFN-I production, as well as ASC (apoptosis-associated speck-like protein containing a CARD)–dependent inflammasome activation (18, 19), orchestrating a diverse innate and adaptive immune response. Engagement of the RIG-I pathway in myeloid cells can induce IFN-I–dependent natural killer cell activation and regression of melanoma metastases (18). Both cGAS/STING and RIG-I signaling can trigger various forms of programmed cell death (20). However, in contrast to other cytosolic nucleic acid receptors involved in IFN-I release (i.e., cGAS/STING), targeted activation of RIG-I within tumor cells has been shown to induce an immunogenic variant of programmed cancer cell death (ICD) that results in growth inhibition of a variety of preestablished tumor entities (17, 18, 21). Yet, the molecular mechanisms that drive RIG-I–induced ICD remain incompletely defined but seem to involve proapoptotic BH3 (Bcl-2 homology domain 3)–only proteins and activation of caspase-3 (21).

To what extent tumor-intrinsic nucleic acid receptor signaling affects ICB-mediated antitumor immunity is unknown. Given that the RIG-I/MAVS pathway is an important regulator of IFN-I production and programmed cancer cell death, we hypothesized that it may be involved in shaping the outcome of ICB and could be suitable for combined modality immunotherapeutic approaches.

RESULTS

Anti–CTLA-4–mediated systemic antitumor immunity requires tumor cell–intrinsic RIG-I signaling

We first analyzed the potential impact of RIG-I signaling in tumor cells on the efficiency of anti–CTLA-4 checkpoint blockade in an immunogenic tumor model of disseminated disease. To this end, we used a B16 melanoma cell line expressing the model antigen ovalbumin (OVA) (B16.OVA). RIG-I–deficient (RIG-I−/−) B16.OVA cells were engineered using the CRISPR-Cas9 technology (fig. S1A). Consistent with its role in triggering IFN-I production and cell death in tumor cells, all RIG-I−/− B16.OVA clones failed to produce IFN-I in response to a specific RIG-I ligand (in vitro transcribed 3pRNA) and were resistant to RIG-I–mediated cell death in vitro (fig. S1, B to D). We then used either wild-type (WT) or RIG-I−/− B16.OVA cells in a subcutaneous melanoma model in the presence or absence of anti–CTLA-4 (Fig. 1A). Although systemic anti–CTLA-4 monotherapy delayed the growth of both WT and RIG-I−/− tumors, it was significantly less potent in mice bearing RIG-I−/− tumors, resulting in faster tumor growth and reduced host survival benefit compared with WT controls (Fig. 1, B and C). The cell clones selected for in vivo experiments were carefully evaluated for the same in vitro growth dynamics as the parental WT cell line (fig. S1E). Using an alternative RIG-I−/− B16.OVA clone confirmed the importance of tumor cell–intrinsic RIG-I signaling for successful CTLA-4 inhibition (fig. S1F) and excluded possible clonal effects of the selected cell line. Our data thus demonstrate that anti–CTLA-4 immunotherapy relies on tumor cell–intrinsic RIG-I signaling.

Fig. 1 Anti–CTLA-4–mediated systemic antitumor immunity requires tumor cell–intrinsic RIG-I signaling.

(A) Treatment scheme 1: WT mice were implanted with either WT or RIG-I–deficient (RIG-I−/−) B16.OVA cells. Recipients were injected intraperitoneally with anti–CTLA-4 or isotype control antibodies. (B) Tumor growth and (C) overall survival of mice bearing either WT or RIG-I−/− tumors. Data show survival of n = 35 mice per group that were pooled from five independent experiments. (D) Treatment scheme 2: WT mice were bilaterally challenged with WT or RIG-I−/− B16.OVA cells. Right-sided tumors were induced with more tumor cells to facilitate a faster growth dynamic in comparison with left-sided tumors. Recipients were treated with anti–CTLA-4 as described above. Some mice were additionally injected with the RIG-I ligand 3pRNA into the right-sided (local) tumor. (E and F) Tumor growth of 3pRNA-injected (E) and noninjected (distant) (F) B16.OVA tumors. All tumor growth curves show mean tumor volume ± SEM of n = 12 to 15 individual mice. (G) Overall survival of treated mice bearing WT or RIG-I−/− B16.OVA tumors. All data were pooled from at least three independent experiments. (H and I) WT mice were implanted with WT or RIG-I−/− tumor cells. Recipient mice were injected intraperitoneally with anti–CTLA-4 or isotype control antibodies. (H) Tumor growth of n = 6 to 12 mice per group bearing CT26 colon carcinomas. Data are representative of two independent experiments. (I) Tumor growth in n = 14 to 15 mice per group bearing Panc02 pancreatic carcinomas. Data were pooled from two independent experiments. All tumor growth curves show mean tumor volume ± SEM.

We next analyzed the impact of therapeutically enhanced RIG-I signaling on the systemic immune response to anti–CTLA-4. Here, we used either WT or RIG-I−/− B16.OVA cells in a bilateral flank tumor model in which recipient animals are challenged with two tumors of the same genotype. We injected 3pRNA into the right-sided flank tumors to locally activate RIG-I within the tumor microenvironment, hereby targeting both tumor and immune cells. We then evaluated the “local” (in the injected tumor) and “systemic” (in the contralateral, noninjected, distant tumor) response (Fig. 1D). First, we analyzed mice bearing only WT tumors. Therapeutic targeting of RIG-I by administration of 3pRNA into the right flank tumor resulted in total regression of the injected (local) WT tumor but not of the contralateral, noninjected (“distant”) WT tumor [Fig. 1, E and F (left panels), and individual tumor growth curves in fig. S1G]. However, combining local 3pRNA-mediated RIG-I activation with systemic anti–CTLA-4 also led to growth control of distant tumors and was associated with long-term survival in most of the treated animals bearing WT tumors (Fig. 1, E to G). Next, we analyzed mice bearing RIG-I−/− tumors. In contrast to mice bearing WT tumors, 3pRNA treatment of local RIG-I−/− tumors combined with anti–CTLA-4 treatment failed to prevent rapid outgrowth of distant RIG-I−/− tumors, resulting in reduced host survival [Fig. 1, F (right) and G]. Furthermore, local tumor control induced by 3pRNA alone or in combination with anti–CTLA-4 was marginally but significantly impaired in mice bearing RIG-I−/− tumors (Fig. 1E). All long-term surviving animals (predominantly mice that were challenged with WT tumors) were tumor free and immune to subsequent rechallenge with WT B16.OVA tumor cells (fig. S1H). Our findings were not restricted to melanoma because animals bearing subcutaneous RIG-I−/− colon carcinoma (CT26) or pancreatic tumors (Panc02) showed a markedly impaired response to anti–CTLA-4 with more rapid tumor growth when compared with mice bearing respective WT tumors (Fig. 1, H and I). We therefore concluded that (i) localized tumor–intrinsic RIG-I activation augments systemic potency of anti–CTLA-4 immunotherapy and that (ii) after therapeutic targeting of the RIG-I pathway, differential aspects of immunity seem to be responsible for tumor regression in local versus distant tumors, probably involving both tumor-intrinsic and immune cell–mediated mechanisms.

To prove that the activity of anti–CTLA-4 on the distal tumors depended on local tumor cell–intrinsic RIG-I signaling in the RNA-injected tumor, we established a model in which mice were inoculated with a WT tumor in one flank and a RIG-I−/− tumor in the other flank (fig. S1, I and J). These mice were then injected with 3pRNA into their right-sided (local) tumors (which were either WT or RIG-I−/−) together with systemic anti–CTLA-4. We then evaluated the systemic immune response by measuring the growth of distant (noninjected) tumors (which were of the “opposite” genotype). We found that RIG-I activation in local WT tumors in combination with anti–CTLA-4 eradicated distant tumors independently of the genotype of the distant tumor (WT or RIG-I−/−) (Fig. 1F and fig. S1I). In contrast, 3pRNA injection into local RIG-I−/− tumors in combination with anti–CTLA-4 failed to control distant tumors irrespectively of whether they were WT or RIG-I−/−, resulting in poor recipient survival (fig. S1, I and J). Thus, tumor-intrinsic RIG-I activity is required to trigger the development of systemic antitumor immunity after combined treatment with intratumoral 3pRNA and systemic anti–CTLA-4. Once such a systemic immune response has developed, eradication of established tumors occurs independent of intratumoral RIG-I expression.

Tumor cell–intrinsic RIG-I signaling facilitates cross-presentation of tumor-associated antigen by CD103+ DCs and subsequent priming of CD8+ T cells

Thus far, our data suggested that RIG-I signaling, both intrinsically within tumor cells and within cells of the host microenvironment, promoted anti–CTLA-4 antitumor activity. We therefore investigated the processes underlying these effects. First, we focused on the cellular mechanism mediating the reduction of tumor volume caused by intratumoral 3pRNA injection. Using our established model with bilateral tumors of the same genotype (either WT or RIG-I−/−) (Fig. 1D), we administered 3pRNA into right-sided flank tumors and assessed tumor cell death 24 hours later. We found that 3pRNA application into WT tumors on one side resulted in rapid induction of tumor cell death in the injected but only marginally in distant (opposite flank) WT tumors (Fig. 2A). Such localized tumor cell death was absent after 3pRNA treatment of RIG-I−/− tumors, indicating that tumor cell–intrinsic RIG-I signaling mediates this process.

Fig. 2 Tumor cell–intrinsic RIG-I signaling facilitates cross-presentation of tumor-associated antigen by CD103+ DCs and subsequent priming of CD8+ T cells.

(A) WT mice bearing bilateral WT or RIG-I−/− B16.OVA tumors were treated with a one-sided intratumoral 3pRNA administration or its vehicle in vivo-jetPEI. Frequency of tumor cells undergoing programmed cell death was analyzed by TUNEL 24 hours later. (B to D) WT mice bilaterally inoculated with WT or RIG-I−/− B16.OVA cells were repeatedly treated with anti–CTLA-4 ± intratumoral 3pRNA as described in Fig. 1D. Analyses were performed on day 15 after tumor induction. (B) Cross-presentation of the processed OVA peptide-epitope SIINFEKL in the context of MHC-I on CD103+ DCs in the lymph node draining the 3pRNA-injected tumor (TdLN) determined by flow cytometry. (C) Frequency of H-2Kb–SIINFEKL Tetramer+ CD8+ T cells (carrying an OVA-specific T cell receptor) in peripheral blood. All data give values of individual mice (and group means presented as bars) that were pooled from at least two independent experiments. (D) Frequency of tumor-infiltrating CD8+ T cells analyzed by immunohistochemistry; representative tumor sections after hematoxylin and eosin and anti–CD8 (red, arrows) staining from one of two independent experiments. Magnification, ×20. Scale bars, 50 μm. N, highly necrotic area. (E and F) WT and Batf3-deficient (Batf3−/−) mice were bilaterally inoculated with WT B16.OVA cells and were treated with anti–CTLA-4 ± intratumoral 3pRNA as described in Fig. 1D. (E) Overall survival of tumor-bearing mice and (F) the frequency of H-2Kb–SIINFEKL Tetramer+ CD8+ T cells in peripheral blood on day 15 after tumor induction. All data give values of individual mice + group mean as bar that were pooled from at least two independent experiments.

During tumor cell death, tumor antigens can be released and processed by local DCs to induce a tumor antigen–specific CTL response (22). We therefore analyzed the role of tumor-intrinsic RIG-I signaling for DC activation and cross-priming of CD8+ T cells. In this context, CD103+ migratory DCs are particularly important for cross-priming against both foreign and self, skin-associated antigens (23) and play a unique role for the transport of melanoma-associated antigens to the local draining lymph node and for efficient cross-priming of CD8+ CTLs (24). We thus focused on this specific DC population. We found that tumor-intrinsic RIG-I signaling and associated cell death induction facilitated phagocytosis of tumor cell debris by bone marrow–derived DCs in vitro (fig. S2A). In the bilateral tumor model, we next analyzed processing of the tumor-associated antigen OVA and cross-presentation of its immune-dominant peptide epitope SIINFEKL by CD103+ DCs in the tumor-draining lymph nodes (TdLNs) by flow cytometry. We found that cross-presentation of tumor-associated antigen by CD103+ DCs in WT tumors was significantly enhanced by anti–CTLA-4 monotherapy (Fig. 2B and fig. S2B). This effect was lost in RIG-I−/− tumors, revealing the importance of intrinsic RIG-I signaling. Additional activation of the RIG-I pathway by 3pRNA injection into WT tumors, but not RIG-I−/− tumors, potently enhanced cross-presentation of tumor-associated antigen by CD103+ DCs in TdLNs (Fig. 2B). This is a localized response restricted to lymph nodes in close proximity to local tumors that undergo 3pRNA-induced programmed cell death (fig. S2C). Thus, cross-presentation of tumor-associated antigens by CD103+ DCs promoted by local RIG-I activation is facilitated by anti–CTLA-4 treatment and is dependent on tumor cell–intrinsic RIG-I signaling as well as associated tumor cell death.

Furthermore, using the same bilateral tumor model, we found strong systemic expansion of OVA-specific CTLs after single-agent treatment with either 3pRNA or anti–CTLA-4 and enhanced CTL expansion after combination treatment with both 3pRNA and anti–CTLA-4, all of which required tumor cell–intrinsic RIG-I signaling (Fig. 2C). Together, these data suggested that local activation of tumor cell–intrinsic RIG-I signaling results in spatially restricted cross-presentation of tumor antigens by CD103+ DCs, thereby potentiating systemic CTL antitumor responses. Consistent with this, histological analysis of local and distant tumors (Fig. 2D and fig. S3A) showed that treatment with both anti–CTLA-4 and/or local RIG-I activation resulted in increased accumulation of CD8+ tumor-infiltrating leukocytes (TILs). This accumulation was significantly reduced in RIG-I−/− tumors, suggesting that tumor cell–intrinsic RIG-I signaling is required for this process. TILs activated by tumor-intrinsic RIG-I signaling appear to be particularly active, as gene expression analysis of whole tumor lysates of explanted RIG-I−/− tumors showed reduced expression of proteins involved in T cell lytic function (including IFN-γ, granzyme B, and perforin) and increased expression of inhibitory receptors [including TIM-3 (T-cell immunoglobulin and mucin domain-containing protein 3), LAG3 (lymphocyte-activation gene 3), and PD-1] (fig. S3B). Last, mice deficient for the transcription factor Batf3 (basic leucine zipper transcription factor ATF-like 3), which lack CD8α-like DCs (including CD103+ DCs), failed to induce antitumor responses to both anti–CTLA-4 and local RIG-I activation (Fig. 2, E and F), emphasizing the critical role of this DC subset for antitumor immunity.

To explore how tumor cell–intrinsic RIG-I signaling affects ICB, we performed transcriptomic analyses by next-generation RNA sequencing (RNA-seq) of treated (3pRNA or anti–CTLA-4) and untreated WT as well as RIG-I−/− melanomas. To this end, tumor-bearing WT animals were treated as described above (Fig. 1D). RNA was isolated from bulk tumors either on day 7 or on day 13 after tumor induction and was subjected to gene set enrichment analysis (GSEA). Here, active RIG-I signaling in WT tumors injected with 3pRNA strongly associated with cytokine production (IFN-α and IFN-γ), apoptosis, and inflammatory signaling (fig. S3C, column 1, and single gene level data in table S1). IFN-α and IFN-γ response gene sets were less enriched in RIG-I−/− compared with WT tumors (fig. S3C, column 2), indicating that tumor cell–intrinsic RIG-I signaling contributes to these effects. After anti–CTLA-4 treatment of WT tumors, we observed activation of proinflammatory pathways including interleukin-6 (IL-6)/Janus kinase (JAK)/signal transducer and activator of transcription (STAT) signaling (fig. S3C, column 3). CTLA-4 blockade in mice bearing RIG-I−/− tumors induced less activity of IFN-α and IFN-γ response pathways in comparison with WT tumor–bearing animals (fig. S3C, column 4).

Host IFN-I contributes to efficacy of anti–CTLA-4 therapy

On the basis of our transcriptomic data, we considered that the positive effect of local RIG-I activation in the tumor and its microenvironment on anti–CTLA-4 activity may be mediated by IFN-I signaling. To address this, we induced bilateral WT flank tumors in cohoused WT or IFN-α receptor subunit 1 (IFNaR1)–deficient recipient mice (Fig. 3A). We found that the effects of anti–CTLA-4 monotherapy and the enhancement of these effects by local 3pRNA injection required intact host cell IFN-I signaling as therapy-induced tumor regression (fig. S4A) and associated prolonged overall survival were absent in Ifnar1−/− mice (Fig. 3B). We could also confirm IFN-I production in both injected and distant tumors after local 3pRNA administration (Fig. 3C). However, release of IFN-I (as well as proinflammatory cytokines) was independent of tumor cell–intrinsic RIG-I signaling in vivo (Fig. 3C and fig. S4B). Together, these data suggest that host cell–derived IFN-I—induced by either RIG-I or alternative innate pathways—is a prerequisite for anti–CTLA-4 efficacy. In line with these findings, we found that RIG-I– and anti–CTLA-4–mediated expansion of tumor antigen–specific CD8+ T cells was strictly dependent on host IFNaR1 signaling (Fig. 3D).

Fig. 3 Host IFN-I contributes to efficacy of anti–CTLA-4 therapy.

(A) Treatment scheme 1: WT and IFN-α receptor 1–deficient mice (Ifnar1−/−) were bilaterally inoculated with WT B16.OVA cells and were repeatedly treated with anti–CTLA-4 ± intratumoral 3pRNA as described in Fig. 1D. (B) Overall survival of n = 4 to 8 tumor-bearing animals per group that were pooled from two independent experiments. (C) Mice bearing bilateral WT or RIG-I−/− B16.OVA tumors were treated with a one-sided intratumoral 3pRNA administration or its vehicle in vivo-jetPEI. Concentration of IFN-I within the tumor microenvironment 24 hours later determined by enzyme-linked immunosorbent assay. (D) Frequency of H-2Kb–SIINFEKL Tetramer+ CD8+ T cells in peripheral blood on day 15 after tumor induction in WT and Ifnar1−/− mice bearing WT tumors that were treated as described in (A). (E) WT mice were bilaterally inoculated with WT or IRF3/7-deficient (IRF3/7−/−) B16.OVA cells and were repeatedly treated with anti–CTLA-4 ± 3pRNA as described in Fig. 1D. (F) Overall survival and (G) frequency of H-2Kb–SIINFEKL Tetramer+ CD8+ T cells in peripheral blood on day 15 after tumor induction in treated mice bearing WT or IRF3/7−/− B16.OVA tumors. All data give values of n = 10 individual mice and were pooled from at least two independent experiments. ns, not significant.

Next, we assessed the role of tumor cell–derived IFN-I in this context. To this end, we generated B16.OVA cells deficient for the transcription factors IRF3 (interferon regulatory factor 3) and IRF7 (IRF3/7−/−) using CRISPR-Cas9. These cells cannot produce IFN-I in response to 3pRNA (fig. S4, C and D). Using these cells in the bilateral tumor model (Fig. 3E), we observed that WT recipient mice bearing IRF3/7−/− tumors treated with anti–CTLA-4 alone or in combination with 3pRNA showed similar responses to treatment (Fig. 3F and fig. S4E) and expansion of tumor antigen–specific CD8+ T cells (Fig. 3G) as compared with mice bearing WT tumors. On the basis of these data, we concluded that (i) activation of tumor-intrinsic RIG-I signaling and (ii) IFN-I production in host (nontumor) cells including DCs is a prerequisite for anti–CTLA-4 efficacy, because these mechanisms facilitate cross-presentation and priming of tumor antigen–specific CD8+ T cells.

Caspase-3–mediated tumor cell death is crucial for anti–CTLA-4 efficacy

We next investigated why local tumor cell–intrinsic RIG-I signaling was required for the anti–CTLA4 treatment response. Although IRF3/7−/− melanoma cells failed to produce IFN-I in response to 3pRNA treatment, they were still susceptible to RIG-I–mediated cell death (fig. S4F). To analyze the role of RIG-I–induced tumor cell death in mediating anti–CTLA-4 responses, we generated B16.OVA cell lines deficient in critical components of distinct cell death pathways including apoptosis (caspase-3−/−) and necroptosis [mixed lineage kinase domain–like pseudokinase (MLKL−/−)]. We found that tumor cell–intrinsic RIG-I activation by 3pRNA in vitro induced apoptotic rather than necroptotic cell death (Fig. 4, A and B). This was evidenced by rapid translocation of phosphatidylserine into the outer plasma membrane leaflet (annexin V+ cells), as well as cleavage and thus activation of the apoptosis executioner caspase-3 (caspase-3+ cells) before loss of membrane integrity (Live/Dead marker+ cells) (Fig. 4, A and B). Membrane permeabilization at later time points is a result of secondary necrosis. We did not observe 3pRNA-induced rapid loss of membrane integrity in the absence of caspase-3 activation, which would have been a feature of necroptosis (25). Consistent with this, 3pRNA-induced programmed cell death was abrogated in RIG-I−/− and caspase-3−/− but not MLKL−/− melanoma cells. Furthermore, activation of RIG-I signaling by 3pRNA failed to induce phosphorylation and thus activation of the necroptosis executioner protease MLKL in either B16.OVA melanoma cells or L-929 fibroblasts (Fig. 4C). Note that using a series of known inducers of necroptosis [including the combination of tumor necrosis factor–α (TNF-α), SMAC mimetics, and the pan-caspase inhibitor Z-VAD-FMK], we were unable to see induction of phosphorylated MLKL in B16.OVA melanoma, suggesting that this pathway may be suppressed in this cell line.

Fig. 4 Caspase-3–mediated tumor cell death is crucial for anti–CTLA-4 efficacy.

(A and B) WT, RIG-I−/−, caspase-3–deficient (caspase-3−/−), and MLKL-deficient (MLKL−/−) B16.OVA cells were transfected with 3pRNA in vitro and were harvested at the indicated time points. Cell viability determined by (A) annexin V and Live/Dead or (B) cleaved caspase-3 and Live/Dead staining by flow cytometry. Representative data are from one of three independent experiments. (C) WT B16.OVA and murine L-929 fibroblasts were transfected with 3pRNA, and induction of necroptotic cell death by phosphorylated MLKL (pMLKL) was analyzed by Western blot. The combination of TNF-α, SMAC mimetics, and the pan-caspase inhibitor Z-VAD-FMK was used as a positive control to induce necroptosis. (D) WT mice bearing bilateral WT or RIG-I−/− B16.OVA tumors were treated with a one-sided intratumoral 3pRNA administration or its vehicle in vivo-jetPEI as described for Fig. 2A. Frequency of tumor cells expressing cleaved caspase-3 24 hours later. Data give values of individual mice + group mean as bar that were pooled from two independent experiments. (E) Treatment scheme: WT mice were bilaterally inoculated with WT or caspase-3−/− B16.OVA cells and were treated with anti–CTLA-4 ± intratumoral 3pRNA as described in Fig. 1D. (F) Frequency of H-2Kb–SIINFEKL Tetramer+ CD8+ T cells in peripheral blood on day 15 after tumor induction. Tumor growth of (G) 3pRNA-injected and (H) non-injected (distant) B16.OVA tumors. All tumor growth curves show mean tumor volume ± SEM of n = 10 individual mice. (I) Overall survival of treated mice bearing WT or caspase-3−/− B16.OVA tumors. All figure panels give data of n = 10 individual mice per group that were pooled from two independent experiments.

Using our bilateral flank tumor model, local injection of 3pRNA into WT tumors resulted in rapid, localized cleavage of caspase-3 and subsequent tumor cell death in vivo (Fig. 4D). This effect was impaired in RIG-I−/− flank tumors (Fig. 4D), suggesting that RIG-I–mediated tumor cell death is mediated via caspase-3 in vivo. Next, we used WT or caspase-3−/− melanoma cells in the bilateral flank tumor model (Fig. 4E) and analyzed T cell expansion and response to treatment. We found that the development of tumor antigen–specific T cell responses (Fig. 4F) and subsequent antitumor immunity after single-agent treatment with anti–CTLA-4 or combination of anti–CTLA-4 and 3pRNA were almost completely abolished in mice with caspase-3−/− tumors. Mice bearing caspase-3−/− tumors showed rapid tumor growth (Fig. 4, G and H) and very poor survival (Fig. 4I) in response to therapy. In sum, our data identify caspase-3–mediated tumor cell death and host IFN-I signaling, but not tumor cell–derived IFN-I, as prerequisites for successful anti–CTLA-4 immunotherapy. Tumor cell–intrinsic RIG-I signaling initiates caspase-3–mediated programmed tumor cell death, which facilitates tumor antigen cross-presentation by CD103+ DCs and subsequent expansion of CD8+ T cells. These processes are additionally dependent on host IFN-I, induced by RIG-I signaling in host cells. Targeting the RIG-I pathway in both tumor cells and their microenvironment by intratumoral application of 3pRNA can simultaneously augment both these processes.

Host RIG-I signaling is additionally required for systemic tumor control after anti–CTLA-4 therapy

Because treatment effects of anti–CTLA-4 and local RIG-I activation within the tumor and its microenvironment relied on host IFNaR1 signaling but not tumor-derived IFN-I, we next addressed the relevance of RIG-I signaling in nonmalignant host cells in this context. Hence, we treated cohoused WT or Mavs−/− mice (animals that genetically lack the essential RIG-I adapter molecule MAVS) bearing bilateral flank WT tumors with local 3pRNA injections and/or systemic anti–CTLA-4 (Fig. 5A). Mavs−/− mice failed to inhibit the outgrowth of distant tumors after treatment with 3pRNA and/or anti–CTLA-4, whereas local tumor control remained intact (Fig. 5, B and C). This was associated with a lack of tumor-specific systemic CD8+ T cell expansion and reduced host survival (Fig. 5, D and E). Although the efficacy of anti–CTLA-4 monotherapy was reduced in Mavs−/− mice, this reduction was markedly less pronounced than with 3pRNA or combination of anti–CTLA-4 and 3pRNA, and the reduction of tumor-specific CD8+ T cell frequencies with anti–CTLA-4 monotherapy did not reach significant levels (Fig. 5, D and E).

Fig. 5 Host RIG-I signaling is additionally required for systemic tumor control after anti–CTLA-4 therapy.

(A) Treatment scheme 1: WT and MAVS-deficient (Mavs−/−) mice were bilaterally inoculated with WT B16.OVA cells and were treated with anti–CTLA-4 ± intratumoral 3pRNA as described in Fig. 1D. (B) Tumor growth of 3pRNA-injected and (C) distant WT B16.OVA tumors. (D) Frequency of H-2Kb–SIINFEKL Tetramer+ CD8+ T cells in peripheral blood on day 15 after tumor induction. (E) Overall survival in WT and Mavs−/− recipient mice bearing WT B16.OVA tumors. All figures give data of n = 9 or 10 individual mice per group that were pooled from two independent experiments. (F) Treatment scheme 2: WT and Mavs−/− mice were bilaterally inoculated with RIG-I−/− B16.OVA cells and were treated with anti–CTLA-4 ± intratumoral 3pRNA as described in Fig. 1D. (G) Overall survival in WT and Mavs−/− recipient mice bearing RIG-I−/− B16.OVA tumors. Data give survival of n = 4 to 6 individual mice per group and are representative of two independent experiments.

Overall, our data suggest that systemic immunity and thus regression of distant tumors are dependent on the expansion of tumor-specific CD8+ T cells, for which parallel RIG-I signaling in both tumor and host cells appears to be critical. Either mechanism, tumor, or host-intrinsic RIG-I signaling seems to be able to compensate for the loss of the other to a certain extent, thereby allowing for local tumor control in the absence of functional RIG-I signaling in one or the other cellular subset. To test this hypothesis, we used cohoused WT or Mavs−/− mice harboring bilateral RIG-I−/− flank tumors (Fig. 5F). We observed that Mavs−/− recipients bearing RIG-I−/− tumors completely failed to mount a protective antitumor immune response, as evidenced by complete lack of a therapy-induced survival benefit (Fig. 5G). Together, these data indicate that, in addition to tumor-intrinsic RIG-I signaling, host MAVS signaling is required for systemic antitumor immunity in response to anti–CTLA-4 monotherapy and, even more so, its combination with local RIG-I activation.

Tumor cell–intrinsic STING signaling is dispensable for anti–CTLA-4–mediated antitumor immunity

Given the central role of RIG-I signaling in tumor and host cells for the therapeutic efficacy of checkpoint inhibition, we next investigated the relevance of the dsDNA-sensing cGAS/STING pathway. Activation of STING in host antigen-presenting cells through recognition of tumor-derived DNA was previously shown to be a prerequisite for successful checkpoint inhibitor treatment (16). Because RIG-I–mediated tumor cell death may also lead to the release of tumor-derived DNA, we aimed to test the contribution of STING signaling. To this end, we used cohoused WT and STING Goldenticket mice (harboring a chemically induced loss-of-function mutation in Sting; Stinggt/gt) as recipients for bilateral WT tumors and treated them with local 3pRNA injections and/or systemic anti–CTLA-4 (Fig. 6A). Consistent with previous results, we found that anti–CTLA-4 failed to reduce tumor growth in Stinggt/gt animals bearing WT tumors, whereas 3pRNA monotherapy–mediated tumor regression remained intact (Fig. 6, B and C). As a consequence of the insufficient anti–CTLA-4 response, its synergy with 3pRNA-induced RIG-I activation was abrogated in Stinggt/gt animals, resulting in markedly reduced recipient survival in the combination therapy group (Fig. 6D). These data again underline the importance of host STING signaling to ICB-mediated antitumor responses.

Fig. 6 Tumor cell–intrinsic STING signaling is dispensable for anti–CTLA-4–mediated antitumor immunity.

(A) Treatment scheme 1: WT and STING-deficient (Stinggt/gt) mice were bilaterally inoculated with WT B16.OVA cells and were treated with anti–CTLA-4 ± intratumoral 3pRNA as described in Fig. 1D. (B) Tumor growth of 3pRNA-injected and (C) distant WT B16.OVA tumors. (D) Overall survival in WT and Stinggt/gt recipient mice bearing WT B16.OVA tumors. All figures give data of n = 11 to 13 individual mice per group that were pooled from two independent experiments. (E) Treatment scheme 2: WT mice were inoculated with WT or STING-deficient (STING−/−) B16.OVA cells and were treated with anti–CTLA-4 as described in Fig. 1A. (F) Frequency of H-2Kb–SIINFEKL Tetramer+ CD8+ T cells in peripheral blood was determined on day 15 after tumor induction. All data give values of individual mice + group mean as bar that were pooled from two independent experiments. (G) Overall survival in mice bearing WT or STING−/− B16.OVA tumors that were either treated with anti–CTLA-4 or isotype control antibodies. All data are pooled from at least two independent experiments.

Recent studies have suggested that the cGAS/STING pathway is functional in different tumor cell lines (26, 27). To investigate the contribution of tumor cell–intrinsic STING signaling during checkpoint inhibition, we generated STING-deficient (STING−/−) B16.OVA melanoma cells. STING signaling was active in WT B16.OVA melanoma cells, as we observed STING-dependent induction of IFN-I production by these cells (fig. S5A) after stimulation with interferon stimulatory DNA (ISD; a 45–base pair non-CpG oligomer from the Listeria monocytogenes genome) (28). However, activation of tumor cell–intrinsic cGAS/STING by ISD did not trigger programmed B16 melanoma cell death in vitro (fig. S5, B and C). Next, we used either WT or STING−/− tumors in a unilateral subcutaneous melanoma model in the presence or absence of anti–CTLA-4 (Fig. 6E). Both antitumor efficacy of CTLA-4 blockade and expansion of tumor-specific CD8+ T cells were intact in mice bearing STING−/− tumors (Fig. 6, F and G). Nonetheless, therapeutic targeting of the cGAS/STING pathway by intratumoral application of ISD in the immunogenic B16.OVA model showed synergistic effects with anti–CTLA-4, which resulted in significant tumor growth delay and prolonged host survival (fig. S5, D to F). However, the beneficial effects of ISD and anti–CTLA-4 combination were independent of tumor cell–intrinsic STING signaling and failed to induce complete tumor regression and associated host long-term survival (fig. S5, D to F). We thus conclude that tumor cell–intrinsic STING signaling does not contribute to anti–CTLA-4–mediated antitumor immunity, at least in this model.

Local RIG-I activation renders poorly immunogenic tumors susceptible to anti–CTLA-4 immunotherapy

To test whether the combination of selective RIG-I activation and immune checkpoint inhibition was also effective in a poorly immunogenic tumor model, we treated mice bearing aggressive B16-F10 melanomas that do not express an artificial antigen. In mice challenged with bilateral B16-F10 flank tumors, we combined intratumoral 3pRNA injection with anti–CTLA-4 therapy. Consistent with the literature (29), anti–CTLA-4 monotherapy failed to inhibit tumor growth in the B16-F10 model (fig. S6, A and B). However, intratumoral 3pRNA administration synergized with anti–CTLA-4 blockade to induce systemic antitumor immunity. Combination treatment resulted in rejection of 3pRNA-injected B16-F10 WT tumors and significant growth delay of distant B16-F10 WT tumors, associated with prolonged host survival (fig. S6, A to C). Combining anti–CTLA-4 treatment with local RIG-I activation in the poorly immunogenic B16-F10 model resulted in systemic expansion of endogenous, tumor-reactive CD8+ T cell specific for the melanoma-associated antigen tyrosinase-related protein 2 (TRP2) (fig. S6D). Consistently, depletion of CD8+ but not CD4+ T cells compromised local tumor control and abolished systemic antitumor immunity after combined treatment with 3pRNA and anti–CTLA-4 (fig. S6, E and F).

We next investigated whether local RIG-I activation and associated immunogenic cell death would also augment anti–CTLA-4 treatment efficacy in tumor entities other than melanoma. We found that local application of 3pRNA into pancreatic carcinoma (Panc02) flank tumors synergized with anti–CTLA-4 blockade to induce complete regression of local and distant tumors, associated with long-term survival of all treated animals (fig. S7, A to C). In contrast, anti–CTLA-4 immunotherapy failed to control growth of 4T1 mammary carcinoma cells (fig. S7, D and E). In this model, 3pRNA treatment induced some degree of tumor growth delay but did not synergize with checkpoint blockade. Although all tested tumor cell lines released IFN-I in response to RIG-I activation in vitro (fig. S7, F and G), 4T1 cells were resistant to RIG-I–mediated programmed cell death (fig. S8H). These data demonstrate that targeting of RIG-I enhances checkpoint modulation in poorly immunogenic tumors and distinct tumor entities.

High expression of RIG-I DDX58 in human melanoma correlates with prolonged survival and durable responses to anti–CTLA-4 immunotherapy

To assess the clinical relevance of our findings, we analyzed genome-wide transcriptional programs in 456 primary melanoma patient samples from The Cancer Genome Atlas (TCGA) determined by RNA-seq. Using an unbiased approach, median expression of DDX58 (encoding RIG-I) or TMEM173 (encoding STING) was used as cutoff to classify patients into low- and high-expression subgroups. In line with our preclinical data, we found that high expression of DDX58 but not TMEM173 was associated with prolonged overall survival in patients with malignant melanoma (Fig. 7A). Multivariable Cox regression identified low DDX58 expression in melanoma samples as an independent risk factor for death (hazard ratio, 1.85; P < 0.001) when controlling for stadium of disease (TNM tumor status), age, and gender (table S2). Analysis of differentially expressed genes (DEGs) between patients with melanoma with high or low DDX58 expression revealed 7472 DEGs, of which 994 had expression fold change ≥2 (Fig. 7B). GSEA using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database showed that many of the up-regulated genes clustered in immune-related pathways such as T cell receptor and cytokine receptor signaling, antigen processing and presentation, chemokine signaling, and apoptosis (Fig. 7C), similar to those implicated in our preclinical model.

Fig. 7 High expression of DDX58 in human melanoma correlates with prolonged survival and durable response to anti–CTLA-4 immunotherapy.

(A) Overall survival in 456 patients with melanoma from TCGA by expression of DDX58 (RIG-I) or TMEM173 (STING) in tumor samples. (B) Heat map of DEGs in tumors with high versus low expression of DDX58. (C) KEGG pathway enrichment analysis of DEGs. Fraction of DEGs in pathway is indicated by circle size, significance of enrichment by color. NOD, nucleotide-binding oligomerization domain. (D) DDX58 and TMEM173 expression in tumor samples from 18 patients with durable clinical response to anti–CTLA-4 treatment versus nonresponders. Data give values from individual patients + geometric mean. (E) Overall survival in 18 patients with melanoma undergoing anti–CTLA-4 immunotherapy by expression of DDX58 or TMEM173 in tumor samples.

We next analyzed RNA-seq data from a previously described patient cohort (n = 18) with malignant melanoma undergoing anti–CTLA-4 immunotherapy (13, 30). We found that patients with durable clinical benefit from anti–CTLA-4 therapy (characterized as stable disease or better for at least 6 months as determined by radiography) had higher DDX58 expression (but not TMEM173 expression) than patients who did not respond to treatment (Fig. 7D). We found some evidence that such superior disease control tends to associate with prolonged overall survival in patients with high DDX58 expression (Fig. 7E). However, this did not meet levels of statistical significance, presumably due to the small cohort size. Together, these patient data corroborate our preclinical findings that high transcriptional activity of RIG-I–encoding DDX58 in melanoma tissue associates with prolonged overall survival and durable response to anti–CTLA-4 immunotherapy.

Response to anti–PD-1 does not require tumor cell–intrinsic RIG-I pathway activity

In addition to CTLA-4, inhibitory activity of the PD-1 pathway can also compromise the antitumor efficacy of activated T cells. To address the impact of tumor cell–intrinsic RIG-I pathway activity on therapeutic PD-1 blockade, we treated animals bearing either WT or RIG-I−/− B16.OVA melanoma with anti–PD-1. Blocking the anti–PD-1 pathway resulted in delayed tumor growth and prolonged host survival (Fig. 8, A and B). However, in comparison with anti–CTLA-4 treatment, anti–PD-1 therapy in the B16.OVA model was not as potent in terms of systemic antitumor effects (median overall survival in animals bearing WT tumors 24 versus 35 days; Figs. 1C and 8B). This was in line with low numbers of circulating OVA-specific CD8+ T cells upon anti–PD-1 treatment (Fig. 8C). Despite this, we observed an impaired efficacy of PD-1 blockade in RIG-I–deficient B16.OVA tumors as evidenced by reduced survival of RIG-I−/− tumor–bearing animals (Fig. 8B), but this was not reflected in significantly altered anti–PD-1–mediated tumor growth delay or expansion of tumor antigen–specific T cells in animals bearing RIG-I−/− tumors (Fig. 8, A and C).

Fig. 8 Response to anti–PD-1 does not require tumor cell–intrinsic RIG-I pathway activity.

(A to C) WT mice were implanted with WT or RIG-I–deficient (RIG-I−/−) B16.OVA cells in the right flank. Recipient mice were injected intraperitoneally with anti–PD-1 or isotype control antibodies. (A) Tumor growth and (B) overall survival of mice bearing WT or RIG-I−/− tumors. Data show survival of n = 24 to 30 mice per group that were pooled from four independent experiments. (C) Frequency of H-2Kb–SIINFEKL Tetramer+ CD8+ T cells in peripheral blood. (D) Retrospectively analyzed overall survival in 51 patients with melanoma undergoing anti–PD-1 immunotherapy by gene expression of RIG-I (DDX58) in pretherapy tumor samples. (E) Overall survival of patient cohort from (D) additionally stratified to prior exposure to ipilimumab (Ipi-naïve versus Ipi-progressive). (F) RIG-I (DDX58) expression levels in pretherapy tumor samples in patients with clinical response to anti–PD-1 treatment in comparison with nonresponders. Data give values from individual patients + geometric mean and are shown for the complete patient cohort and Ipi-naïve and Ipi-progressive patients. (G) WT mice were implanted with WT or RIG-I−/− B16.OVA cells and were treated intraperitoneally with anti–PD-1 and anti–CTLA-4 antibodies. Data show overall survival of n = 5 to 20 tumor-bearing mice per group that were pooled from two independent experiments.

To correlate these findings with PD-1 blockade in metastatic melanoma in humans, we retrospectively analyzed tumor RNA-seq data in a cohort of 51 patients that had received anti–PD-1 therapy (7). Some of these patients had shown melanoma progression to prior anti–CTLA-4 therapy. High expression of RIG-I–encoding DDX58 in pretreatment melanoma samples did not associate with beneficial overall survival in response to anti–PD-1 treatment (Fig. 8D), independent of prior exposure to the anti–CTLA-4 antibody ipilimumab (Fig. 8E). In contrast to our data regarding CTLA-4 blockade, we did not find an association between DDX58 transcript copy numbers in tumor tissue and the clinical response to anti–PD-1 treatment (Fig. 8F).

Combination of CTLA-4 and PD-1 blockade can result in particularly strong clinical responses in patients with malignant melanoma (5). We thus performed additional experiments using a combination therapy in mice bearing either WT or RIG-I−/− B16.OVA tumors. We hereby found strong antitumor effects of combinatorial therapy with anti–CTLA-4 and anti–PD-1 with markedly prolonged host survival (Fig. 8G). Moreover, these effects depended on tumor cell–intrinsic RIG-I signaling. Our data therefore suggest that the efficacy of PD-1 blockade can be influenced by RIG-I activity. However, the role of tumor cell–intrinsic RIG-I signaling seems more important for the efficacy of anti–CTLA-4 and, as a consequence hereof, the combinatory treatment with anti–PD-1.

DISCUSSION

Our study identifies a hitherto unrecognized role of tumor-intrinsic RIG-I signaling for checkpoint blockade–mediated antitumor responses. Following up on a genetic screen in patients with melanoma, which suggested an association between increased antiviral gene activity in tumor cells and enhanced clinical responses to anti–CTLA-4 immunotherapy (13), we now provide experimental proof that RIG-I but not STING signaling in melanoma cells promotes the efficacy of ICB with anti–CTLA-4. Mechanistically, we demonstrate that tumor cell–intrinsic, endogenous RIG-I signaling resulted in (i) induction of caspase-3–mediated apoptotic cell death; (ii) spatially confined, enhanced cross-presentation of tumor-associated antigen by CD103+ DCs in local TdLNs; (iii) systemic expansion of tumor-specific CD8+ T cells; and (iv) infiltration of tumors with CD8+ leukocytes (fig. S8). Therapeutic targeting of RIG-I in the tumor microenvironment potently augmented these processes and boosted the efficacy of checkpoint blockade. A lack of tumor cell–intrinsic RIG-I receptor activity resulted in markedly enhanced cancer resistance to anti–CTLA-4 and its combination with anti–PD-1 but demonstrated only discrete effects on anti–PD-1 single-agent treatment, emphasizing the distinct and complementary mechanisms of the two pathways. Our mechanistic data were recapitulated in patients with melanoma: High RIG-I–encoding DDX58 gene expression in tumor samples correlated with transcriptional pathway activity of T cell receptor and cytokine receptor signaling, antigen processing and presentation, chemokine signaling, and apoptosis.

Previous studies have shown an important role of the cGAS/STING pathway within nonmalignant host cells for the generation of antitumor T cell immunity and the efficacy of PD-1/PD-1 ligand blockade (16, 31). It was demonstrated that spontaneous induction of tumor antigen–specific CD8+ T cells depended on IFN-I produced by host DCs, activated by tumor DNA via the STING pathway (16). In line with these results, we found that anti–CTLA-4–induced antitumor immune responses depended on STING- and IFN-I signaling in host cells. In addition, our data indicate a relevant role for MAVS signaling in nonmalignant host cells, which contributed to anti–CTLA-4 efficacy. These effects were less pronounced compared with STING but were accentuated in the context of therapeutic RIG-I targeting. This seems in contrast to the abovementioned study, which claimed that host MAVS was dispensable for the spontaneous development of antitumor T cell responses in the B16.SIY melanoma model (16). However, these findings may be explained by the different immunogenicity of B16.SIY lesions, associated with potent spontaneous antitumor T cell responses and tumor rejection independent of further enhancement by host MAVS. From the present data, we conclude that anti–CTLA-4 immunotherapy depends on tumor cell–intrinsic RIG-I signaling as well as subsequent STING and MAVS pathway activation in the host. We speculate that nucleic acids leaking from disintegrating tumor cells during RIG-I–induced cell death may be engulfed by host myeloid cells to activate cGAS/STING and RIG-I/MAVS. How those nucleic acids would be transported and protected from degradation and whether packaging into extracellular vesicles may play a relevant role remain to be tested.

Synthetic agonists activating the cGAS/STING pathway in the tumor microenvironment such as ISD or cyclic dinucleotides have been successfully used to promote antitumor immunity (32). Furthermore, irradiation of tumors can result in host cGAS/STING-dependent maturation of intratumoral DCs and IFN-I–mediated subsequent T cell cross-priming (15), which can improve responses to anti–CTLA-4 treatment (33). However, focusing on signaling pathways in nonmalignant host cells, none of these studies addressed the impact of tumor-intrinsic nucleic acid receptor signaling on checkpoint inhibitor–mediated antitumor immunity and how this may converge with host cell–mediated antitumor defense. Here, we demonstrate that single-agent anti–CTLA-4 efficacy relies on tumor cell–intrinsic RIG-I signaling but is independent of tumor-intrinsic STING activity in both our murine model and human melanoma transcriptome analysis. It is important to mention that the latter was derived from bulk tumor samples, containing mostly tumor cells and only few infiltrating immune cells. However, tumor cell–intrinsic STING signaling does seem to have the potential to augment anti–CTLA-4 immunotherapy under certain circumstances, which seem to be limited to conditions of simultaneous irradiation or chemotherapy (34). DNA damage in response to such genotoxic cancer therapies, in contrast to RIG-I activation as in our approach, can result in the formation of “micronuclei” that activate cGAS/STING signaling in tumor cells and thereby promote the efficacy of checkpoint inhibitor–based immunotherapy. In agreement with our data, that study also showed that the effects of anti–CTLA-4 therapy alone principally do not depend on tumor-intrinsic STING signaling. Instead, irradiation-induced formation of micronuclei and subsequent tumor-intrinsic cGAS/STING activation augmented treatment efficacy only in the specific scenario of a combined modality approach with genotoxic agents and ICB. Our data do not exclude a role for tumor cell–intrinsic cGAS/STING signaling in ICB-mediated immunotherapy of cancers with high intratumoral expression of the pathway’s components. Proapoptotic functions of intrinsic STING are determined by signaling strength and have been exploited in a murine model of T cell malignancy (35). However, therapeutic targeting of proapoptotic STING signaling is limited by the fact that many cancer cell lines have lost STING expression (20).

Tumor-intrinsic RIG-I signaling triggers the induction of both IFN-I and programmed immunogenic cell death (17, 18, 21). We here demonstrate that tumor control by anti–CTLA-4–mediated checkpoint blockade critically depends on RIG-I–induced, caspase-3–mediated apoptotic rather than necroptotic tumor cell death. The immunogenicity of RIG-I–like helicase-induced apoptosis has previously been attributed to receptor-interacting protein kinase-1 signaling (RIPK1) and nuclear factor (NF) κB–induced transcriptional programs within dying cells (25). According to our data, simultaneous induction of proinflammatory signaling such as IFN-I in host cells and cell death pathways in tumor cells by RIG-I seems to be the basis for promoting efficient antigen cross-presentation by CD103+ DCs. This is also supported by our transcriptome data from murine bulk tumor samples revealing a strong association of active RIG-I signaling induced by 3pRNA with cytokine production (IFN-α and IFN-γ), apoptosis, and inflammatory signaling. Testing tumor cells from human biopsies for in vitro susceptibility to RIG-I–mediated cell death induction might serve as a functional biomarker to predict treatment outcomes of anti–CTLA-4 checkpoint inhibitor treatment and may be helpful before targeting RIG-I in patients in general. Nonetheless, as evidenced by our murine RNA-seq data, alternate pathways including IL-2/STAT5 signaling, inflammation, MYC targets, KRAS signaling, epithelial-mesenchymal transition, and transforming growth factor–β could contribute to modulation of RIG-I signaling and thus influence ICB-mediated tumor immunosurveillance, but this requires further experimental validation.

Our findings that single-agent anti–CTLA-4 treatment is less efficient in mice bearing RIG-I−/− tumors suggest a role for ligand-mediated or otherwise sustained RIG-I signaling within tumor cells. An endogenous source of RIG-I ligands could be responsible for basal, persistent pathway activation within tumor cells in the absence of artificial 3pRNA and thus contribute to the efficacy of anti–CTLA-4–based therapeutic approaches. Previous studies have identified possible RIG-I ligands within tumor cells. Treatment of tumor cells with the DNA methyltransferase inhibitor 5-azacytidine can result in increased expression of endogenous retroviral elements, which can subsequently be detected by tumor cell–intrinsic RIG-I–like helicases and enhance ICB treatment (13, 36). During genotoxic chemo- or radiotherapy, translocation of small nuclear RNAs into the cytoplasm can activate RIG-I signaling in cancer cells, resulting in apoptosis and IFN-I production (37). Alternatively, exogenous factors such as the commensal microbiota, known as relevant modulators of checkpoint blockade efficiency (3841), may also contribute to intratumoral RIG-I/MAVS-dependent regulation of checkpoint inhibition and other cancer therapies.

In sum, we identify both tumor- and host-intrinsic RIG-I signaling as fundamental requirements for anti–CTLA-4–mediated antitumor immunity and establish high tumoral RIG-I–encoding DDX58 expression as a potential biomarker for ICB efficacy. Our data predict that clinical RIG-I targeting in patients may increase response rates of checkpoint inhibitor–based immunotherapy and reduce interindividual variability of treatment outcome.

MATERIALS AND METHODS

Study design

The goal of this study was to evaluate the impact of tumor- and host-intrinsic RIG-I signaling on immune checkpoint inhibitor–mediated tumor immune surveillance. To assess this, we used CRISPR-Cas9–mediated genome editing to generate tumor cell lines that lack nucleic acid receptors or downstream signaling molecules (RIG-I, STING, IRF3/7, and caspase-3) together with available genetically deficient mouse models. For mouse studies, sample sizes were chosen according to the power of the statistical test of each experiment. For all studies, animal numbers are depicted in the figures, and the number of independent experiments is listed in the figure legends. To point out the therapeutic potential of selective RIG-I stimulation in preclinical tumor models, we used in vitro transcribed 3pRNA. Last, we analyzed publicly available transcriptome data of human melanoma samples from patients undergoing checkpoint inhibitor treatment to associate expression of innate nucleic receptors with treatment outcome.

Mice

Female C57Bl/6j and BALB/c mice were purchased from Janvier. Mice genetically deficient in Batf3, IFNaR1, MAVS, and STING (Tmem173/STINGgt/gt) have been previously described (4245). Mice were at least 6 weeks of age at the onset of experiments and were maintained under specific pathogen–free conditions. WT and genetically deficient mice included in the same experiments either were littermates from heterozygous breeding or were cohoused for at least 4 weeks before onset of experiments. Animal studies were approved by the local regulatory agency (Regierung von Oberbayern, Munich, Germany).

Tumor cell lines and in vitro analyses

The B16 murine melanoma cell line expressing full-length chicken OVA (here referred to as B16.OVA) was cultured in complete Dulbecco’s modified Eagle’s medium (DMEM). B16-F10, CT26, and Panc02 cells were cultured in complete DMEM or RPMI 1640 medium, respectively. For in vitro transfection, 3pRNA was complexed with Lipofectamine 2000 (Life Technologies, Darmstadt, Germany) in Opti-MEM (Invitrogen), and tumor cells were cultured in the presence of complexed 3pRNA (3 μg/ml) for 48 hours, if not stated otherwise. Induction of cell death was assessed by staining with annexin V (BD Biosciences) and 7-aminoactinomycin D (BioLegend) or the Fixable Viability Dye eFluor 506 (eBiosience). INF-I release was determined by enzyme-linked immunosorbent assay according to the manufacturer’s (PBL Assay Science) protocol. To initiate necroptosis, cells were pretreated for 30 min with the pan-caspase inhibitor Z-VAD-FMK (20 ng/ml; Invivogen) before the induction of cell death with recombinant murine TNF-α (20 ng/ml; PeproTech) and the SMAC mimetic SM-164 (100 nM; APExBIO) for 4 hours.

CRISPR-Cas–mediated genome editing

Mutant cells were engineered using the CRISPR-Cas9 technology. Briefly, single guide RNAs (sgRNAs) targeting DDX58 (RIG-I), TMEM173 (STING), IRF3 or IRF7, CASP3 (Caspase-3), and MLKL were cloned into the pSpCas9(BB)-2A-GFP plasmid (pX458, Addgene plasmid #48138; a gift from F. Zhang) as previously described (46). The specific sgRNA target sequences were designed with the CHOPCHOP (47) or Benchling tool, optimized for on-target activity (48), and are listed in the supplemental experimental procedures. B16.OVA cells were transiently transfected with pX458 containing individual sgRNAs using Lipofectamine 2000, and green fluorescent protein–expressing single-cell clones were isolated by fluorescence-activated cell sorting 24 hours after transfection. Gene-deficient clones were identified by immunoblotting and functional assays. The target sequence of all used sgRNAs can be found in Supplementary Materials and Methods. RIG-I−/− B16.OVA clone 3.6 (generated with sgRNA 3) was used for all melanoma tumor experiments if not stated otherwise.

Tumor challenge and treatment

For the unilateral tumor model, mice were injected subcutaneously with B16.OVA (2.4 × 105), CT26 (1 × 106), or Panc02 (2 × 106) cells. For the bilateral tumor model, mice were injected subcutaneously with B16.OVA (right flank, 2.4 × 105; left flank, 0.8 × 105) or B16-F10 (1 × 105 each flank) cells on day 0. When tumors were readily visible (volume of about 100 mm3), 25 μg of 3pRNA complexed in 3.5 μl of in vivo-jetPEI (Polyplus) was repeatedly injected into the right-sided tumors, typically on days 6, 9, and 12. Anti–CTLA-4 (clone 9H10), anti–PD-1 (clone RMP1-14), or appropriate isotype controls (200 μg; all from BioXCell, West Lebanon, NH) were administered intraperitoneally (ip) at the same time points. Treatment with anti-CD8α (clone 2.43) or anti-CD4 (clone GK1.5, both from BioXCell) depleting antibodies was initiated 2 days before tumor induction (100 μg ip) and was repeated twice weekly (50 μg ip). For tumor rechallenge experiments, mice were injected intravenously with 105 B16.OVA melanoma cells, and 2 weeks later, superficial pulmonary pseudometastases were counted. Mice were euthanized when the maximum tumor diameter exceeded 15 mm according to the standard legal procedure (responsible state office Regierung von Oberbayern). Mean tumor growth analysis was discontinued when the first animal per group succumbed to tumor progression.

Flow cytometry

Cell suspensions were stained in phosphate-buffered saline with 1% fetal bovine serum. Fluorochrome-coupled antibodies were purchased from eBioscience or BioLegend. Anti-mouse OVA257–264 (SIINFEKL) peptide bound to H-2Kb antibody (clone 25-D1.16) was purchased from eBioscience. iTAg MHC-I murine tetramers detecting SIINFEKL-specific CD8+ T cells were purchased from MBL (Woburn, MA). Murine MHC-I tetramers detecting TRP2-specific CD8+ T cells were provided by D. Busch (Munich, Germany). Terminal deoxynucleotidyl transferase–mediated deoxyuridine triphosphate nick end labeling (TUNEL) was performed with the APO-BrdU Kit (BD Pharmingen). Cleavage of caspase-3 was analyzed using the CaspGlow Flourescein Active caspase-3 staining kit (Invitrogen) according to the manufacturer’s protocol. For intracellular cytokine staining, the Foxp3 Transcription Factor Fixation/Permeabilization Kit (eBioscience) was used. CD103+ DCs were characterized as MHC-IIhighCD11chighCD64Ly6CCD11bCD103+ as described previously (24). Data were acquired on a FACSCanto II (BD Biosciences) and analyzed using FlowJo software (TreeStar).

RNA-seq analysis of primary human melanoma samples

The melanoma patient cohort undergoing anti–CTLA-4 treatment has been described previously (13, 30). For the analysis of clinical long-term benefit, two patients that had additionally received chemotherapy or BRAF-targeted therapy were excluded. The melanoma patient cohort undergoing anti–PD-1 treatment has been described previously (7). RNA-seq data and clinical data from TCGA were downloaded for 472 cutaneous melanoma samples from cBioPortal (49) (study ID “skcm_tcga”) using the R package cgdsr (50). Excluding patients with multiple tumors resulted in 466 patients. DEGs were identified using the R package siggenes (51). Pathway enrichment analysis was carried out with the R packages KEGG.db and KEGGprofile (52).

Statistical analysis

If not stated otherwise, all data are presented as means ± SEM. Statistical significance of single experimental findings was assessed with the independent two-tailed Student’s t test. For multiple statistical comparison of a dataset, the one-way analysis of variance (ANOVA) test with Bonferroni or Dunnett’s posttest was used. For tumor growth, the intergroup comparison of mean tumor volume was calculated on the day the first individual mouse within the respective groups succumbed to tumor progression. The comparison between the four treatment groups for a given tumor genotype was calculated on the day that the mean tumor volume analysis in the “isotype” control group was terminated and one-way ANOVA was used to adjust for multiple testing. In some experiments with only two treatment modalities, the independent two-tailed Student’s t test was used. The comparison of the treatment response of different tumor genotypes to a given treatment modality was calculated on the day that the mean tumor volume analysis in the given treatment group was terminated and the independent two-tailed Student’s t test was used. Overall survival was analyzed using the log-rank test. Significance level was set at P < 0.05, P < 0.01, and P < 0.001 and was then indicated with asterisks (*, **, and ***, respectively). All statistical calculations were performed using Prism (GraphPad Software) except for Cox regression analyses, which were carried out in R (version 3.4.1).

SUPPLEMENTARY MATERIALS

immunology.sciencemag.org/cgi/content/full/4/39/eaau8943/DC1

Materials and Methods

Fig. S1. Characterization of RIG-I–deficient B16.OVA clones.

Fig. S2. Tumor cell–intrinsic RIG-I signaling promotes localized cross-presentation of tumor-associated antigen by CD103+ DCs in TdLNs.

Fig. S3. Tumor cell–intrinsic RIG-I deficiency is associated with reduced TIL frequencies and decreased expression of proteins involved in T cell lytic function.

Fig. S4. Anti–CTLA-4–mediated antitumor immunity does not rely on tumor cell–derived IFN-I.

Fig. S5. Melanoma cell–intrinsic STING signaling induces IFN-I production but not programmed cell death.

Fig. S6. Local RIG-I activation renders poorly immunogenic tumors susceptible to checkpoint inhibition.

Fig. S7. Antitumor synergy between CTLA-4 blockade and local RIG-I activation is not restricted to melanoma.

Fig. S8. Proposed model: Tumor-intrinsic RIG-I signaling promotes checkpoint inhibitor–mediated anticancer immunity.

Table S1. Single-gene data murine tumor RNA-seq (Excel).

Table S2. Low RIG-I–encoding DDX58 expression in melanoma biopsies is an independent risk factor for death.

Table S3. Raw data file (Excel).

References (5357)

REFERENCES AND NOTES

Acknowledgments: First and foremost, we thank all patients and their families who participated in the studies that form the basis of this work. We thank P. Knolle (Munich, Germany) and Life Science Editors for critical reading of the manuscript and editing support. Funding: This study was supported by the Deutsche Forschungsgemeinschaft (DFG; German Research Foundation) (Projektnummer 360372040 – SFB 1335, Projektnummer 395357507 – SFB 1371, and PO 1575/3-1 to H.P.), the Else-Kröner-Fresenius-Stiftung (2012_A61 and 2015_A06 to H.P. and EKFK to S.H.), a Young Investigator Award by the Melanoma Research Alliance (to S.H.), a Feodor-Lynen Scholarship for Experienced Researchers by the Alexander von Humboldt Foundation (to H.P.), the German Cancer Aid (111620 to H.P. and S.H.), the European Hematology Association (to H.P.), a Mechtild Harf Research Grant from the DKMS Foundation for Giving Life (to H.P.), the Dres. Carl Maximilian and Carl Manfred Bayer-Foundation (to S.H.), and a consolidator grant by the European Research Council (CoG-2015_682473_BCM-UPS to F.B.). H.P. was supported by the EMBO Young Investigator Program. P.-A.K. was supported by a postdoctoral fellowship of the TUM University Foundation. Author contributions: S.H., A.W., T.H., and H.P. designed the research and analyzed and interpreted the results. S.H., A.W., F.S., S.B., S.G., T.N., J.C.F., and P.-A.K. performed experiments. K.S. performed histopathologic analysis of tumor samples in a blinded fashion. T.E., R.Ö., and R.R. acquired and analyzed murine tumor RNA-seq data. C.W. analyzed human melanoma RNA-seq data provided by V.M. and T.A.C. S.H., T.H., and H.P. wrote the manuscript. F.B., M.R.M.v.d.B., and J.R. gave methodological support and conceptual advice. S.H., T.H., and H.P. guided the study. This work is part of the doctoral thesis of A.W., F.S., and S.B. at the Technical University of Munich. Competing interests: M.R.M.v.d.B. has received research support from Seres Therapeutics; has consulted, received honorarium from, or participated in advisory boards for Seres Therapeutics, Flagship Ventures, Novartis, Evelo, Jazz Pharmaceuticals, Magenta Therapeutics, Therakos, Amgen, Merck & Co. Inc., Acute Leukemia Forum (ALF), and DKMS Medical Council (Board); and has IP Licensing with Seres Therapeutics and Juno Therapeutics. The other authors declare no competing financial interests. Data and materials availability: Murine tumor RNA-seq data for this study have been deposited in the European Nucleotide Archive (accession no. PRJEB32241).
View Abstract

Navigate This Article