TNFα and Radioresistant Stromal Cells Are Essential for Therapeutic Efficacy of Cyclic Dinucleotide STING Agonists in Nonimmunogenic Tumors

The immune system evolved to recognize, interpret, and combat infection using a series of diverse receptors that recognize evolutionarily conserved viral and bacterial molecules termed pathogen-associated molecular patterns (PAMP). The receptors, termed pattern recognition receptors (PRR), are expressed on a variety of immune and nonimmune cells and are essential for first sensing, then initiating, innate immune responses against a variety of pathogens (1–3). Downstream of PAMP–PRR ligation are a variety of transcription factors, including IRF-3 and NF-κB, that lead to the transcription of innate effector molecules, including IFNβ, TNFα, IFNγ, and IL1 (4). These effector cytokines execute a variety of antimicrobial effects, ranging from activation to apoptosis in both stromal and immune cells.

One such PAMP is cytosolic double-stranded (ds)DNA, which binds to cyclic GMP-AMP (cGAMP) synthase (cGAS). Upon dsDNA binding, cGAS catalyzes the conversion of ATP and GTP into cGAMP (5–7). In turn, cGAMP binds to stimulator of interferon genes (STING), an endoplasmic reticulum–resident protein initially discovered as an essential gene for expression of IFNβ by host cells infected with herpes simplex virus (HSV-1; refs. 8–10). In response to binding cGAMP or other cyclic dinucleotide (CDN) structures, STING phosphorylates the adaptor protein TBK1, which, in turn, phosphorylates IRF-3 and initiates the transcription of a number of interferon-stimulated genes, including type I interferons (9, 11–13) and other coregulated host defense pathways, including the NF-κB and STAT6 pathways (14, 15).

The cGAS–STING pathway initiates innate immune responses to multiple viral and bacterial pathogens. CDNs are expressed ubiquitously by bacteria to signal between microbes and regulate diverse metabolic processes. For example, the intracellular bacterium Listeria monocytogenes (Lm) produces cyclic-di-adenosine monophosphate (AMP), which, when secreted intracellularly through multidrug efflux pumps, binds and activates STING and results in the expression of IFNβ and other coregulated genes (16, 17). However, in mice, enhanced activation of STING suppresses the magnitude of Lm-induced CD8⁺ T-cell priming specific for encoded antigens and diminishes protection against lethal wild-type (WT) Lm challenge, and conversely, mice lacking functional STING are more resistant to infection with intracellular bacteria, including Lm (18). The cGAS–STING signaling axis is also activated by dsDNA viruses, including pox viruses, adenoviruses, gamma herpes viruses such as HSV, and retroviruses such as HIV and HTLV (5, 15, 19–24). In response to infection with dsDNA viruses, induced cGAMP can be transported to neighboring cells through gap junctions and inhibit virus spread, supporting the central role of this pathway as a major antimicrobial defense mechanism (25, 26). However, the mechanisms underlying the observation that cGAS—STING-deficient mice are more sensitive to virus infection but are more resistant to infection with intracellular bacterium are not yet understood.

Results from several laboratories suggest that the cGAS–STING pathway may have evolved to recognize host cell damage by sensing of “self” DNA. STING-deficient mice have a significantly reduced ability to reject immunogenic tumors, and poorly immunogenic tumors have accelerated growth in these mice as compared with WT mice (27). Tumor DNA can be detected in cross-presenting CD8α⁺ DCs, and depletion of this antigen-presenting cell population results in both the significant decrease of IFNβ expression and reduced antitumor immunity (27). These studies suggest that the cGAS–STING is a critical pathway for sensing host cell damage. Sparked by these discoveries, preclinical and clinical studies are testing CDNs as an antitumor immunotherapeutic agent administered either systemically [intravenous (i.v.)] or locally [intratumoral (i.t.); refs. 27–35]. Clinical activity seen with i.t. injection of a modified HSV-1 vector encoding GM-CSF (talimogene laherparepvec, or T-VEC), provides evidence that the STING pathway may regulate tumor immunity in humans and supports the rationale for targeting STING directly with CDNs.

Although previous investigations showed that local or systemic activation of STING results in significant and durable antitumor immunity, the critical innate and adaptive immunologic mechanisms underlying these responses are not completely understood. In part, it is difficult to draw unifying conclusions due to the different CDN structures, dose levels, and treatment regimens used in previous investigations. For example, administration of low doses of CDN i.v. can prime a systemic tumor-specific immune response, but antitumor immunity can be suppressed at high CDN dose levels (33). In contrast, i.t. administration of doses of 25 to 200 μg of certain CDNs can induce local antitumor efficacy, systemic antitumor immunity, and immunological memory (31, 35).

Several CDN molecular structures have been evaluated in mice. The natural product of cGAS is 2′–5′, 3′–5′ cGMP–AMP (2′3′ cGAMP), and CDNs with this internucleotide bridge structure are collectively referred to as “mixed linkage” or “noncanonical linkage” CDNs and have about a 10-fold increased STING-binding affinity. CDNs with a noncanonical 2′3 ′ structure are more potent activators of human STING than the canonical bis 3′-5′ CDNs produced by bacteria (35). The notion that STING can distinguish between bacterial and host cGAS—produced ligands and is activated more profoundly by host-produced 2′3 ′ cGAMP is suggestive that the cGAS–STING pathway evolved principally to sense host damage and dsDNA-based pathogens rather than bacterial produced CDNs. Nevertheless, this dichotomy between metazoan and bacterial CDN structures provided important insights into the development of synthetic CDN compounds for evaluation in humans. In this investigation, we utilized a synthetic CDN known as ADU-S100 that is being evaluated in patients with advanced cancers that are resistant to standard-of-care therapies. The structure of ADU-S100 has been described previously (31, 35) and is disodium dithio-(R_P, R_P)–[cyclic [A(2′,5′)pA(3′,5′)p]], a cyclic di-ribonucleotide composed of two AMP analogues cyclized via a 2′,5′ (noncanonical) and a 3′,5′ (canonical) phosphodiester bond. ADU-S100 has been shown to bind and activate all five human STING polymorphic proteins. Other formulations of CDN and STING agonists have been used preclinically, administered in a variety of vaccination settings, and have shown a range of efficacies in tumor clearance, induction of cytokines, and activation of an adaptive immune response.

Here, we found that TNFα is definitively required for CDN-mediated acute tumor necrosis and modulation of the immune infiltrate. Through the use of bone marrow chimeras, we provide evidence that stromal cells also respond to STING agonists and, thus, play a critical role in mediating acute clearance of tumors during i.t. CDN therapy. Collectively, our results outline a model in which both stromal and hematopoietic cells are involved in the CDN-mediated antitumor immune responses.

B16F10, CT26, MC38, and 4T1 mouse cell lines were acquired from ATCC and were cultured in complete RPMI consisting of 10% FBS, penicillin (100 U/mL), streptomycin (100 μg/mL), 1 mmol/L sodium pyruvate, and NEAA. BT549, H460, T47D, MALME-3M, MCF7, SK-MEL-28, SK-MEL-5, SK-MEL-2, and THP-1 human cell lines were acquired from ATCC and cultured as recommended by ATCC. B16F10, MC38, 4T1, and CT26 cells were validated as Mycoplasma negative using the STAT-Myco test by IDEXX Incorporated. Cells were thawed from liquid nitrogen stocks and passaged 2 to 3 times before being used for in vivo or in vitro experimentation.

Six- to 8-week-old female C57BL/6 and BALBc mice were purchased from The Jackson Laboratory or Charles River, and 6–8-week-old female BL6-CD45.1 mice were purchased from Charles River. cGAS^−/− animals were a generous gift from Dr. Skip Virgin at the University of Washington, St. Louis. STING^−/− mice were the Golden Ticket strain (C57BL/6J-Tmem173^gt/J) and were a gift from Dr. Young Kim. Rag2^−/− animals were a gift from Dr. Jonathan Powell. IFNαR^−/− (B6.129S2-Ifnar1tm1Agt/Mmjax), TNFα^−/− (B6.129S-Tnftm1Gkl/J), and IL6^−/− (B6.129S2-Il6tm1Kopf/J) breeder pairs were purchased from The Jackson Laboratories and bred in Johns Hopkins Facilities. All mouse procedures were approved by the Johns Hopkins University or Aduro Biotech Institutional Animal Care and Use Committee under protocol number M013M08 and were compliant with the Guide for the Care and Use of Laboratory Animals (8th ed., The National Academic Press, 2011).

Chimeric animals were generated by irradiating 6–12-week-old CD45.1, TNFα^−/−, and Golden Ticket mice with two doses of 6 gy separated by 3 hours. Three hours after the second dose of irradiation, mice were reconstituted with 5 to 10 million cells of nonirradiated donor bone marrow via tail vein injection, and left to rest for at least 6 weeks. To confirm reconstitution, mice were bled 5 to 6 weeks after irradiation, and PBMCs were stained for CD45.1 (WT) and CD45.2 (Golden Ticket). All chimeric animals were put on uniprim feed at least 1 week before irradiation and removed from uniprim feed at least 1 week before tumor challenge.

A total of 5 × 10⁵ B16F10 cells were implanted between the skin and peritoneal cavity on day 0. Tumors were monitored until the group average was about 80 mm³ and then treated with 100 μg i.t. injections of CDN in 40 μL PBS or with PBS alone every other day for a total of three treatments. For chimera studies, mice were implanted with tumor cells, and when tumors were palpable, animals were selected and arranged in groups normalized to about 80 mm³. In TNFα blockade experiments, the clinical reagent Enbrel (Amgen) or human IgG1 isotype control (Cat# BE0297; Bio X Cell) was administered intraperitoneally at 1 mg/mL in 200 μL. Tumor outgrowth volume was measured with calipers, and volume was calculated using the equation V = ½ (width² × length). Animals were sacrificed when tumor volumes exceeded 2,000 mm³.

Tumors and tumor-draining inguinal lymph nodes (TDLNs) were excised from animals sacrificed 24 hours after i.t. CDN injection in media containing 10% FBS, penicillin (100 U/mL), and streptomycin (100 μg/mL). These tissues were manually dissociated with scissors and syringe plungers, filtered with 70-μm mesh, manually dissociated over a 70-μm mesh filter a second time if needed, then resuspended for antibody staining. Flow cytometry antibodies included: from BioLegend—CD11b-AF700 (M1/70), CD11c-FITC (N418), CD86-PE (GL-1), CD19–PerCP-Cy5.5 (6D5), Ly6C-BV605 (HK1.4), CD45.2-APC (104), IA/IE–PerCP-Cy5.5 (M5/114.15.2), Ly6G-BV421 (1A8), CD16/32-BV510 (93), F4/80–PE-Cy7 (BM8), Ly6c–PerCP-Cy5.5 (HK1.4), and CD4-BV605 (GK1.5); from EBioscience—NK1.1-PE (PK136) and CD8–PE-Cy7 (53-6.7); CD4-Pacific Orange (RM4-5) from Life Technologies; and CD45.1-FITC (A20) from BD Pharmingen. Samples were analyzed using an LSR II cytometer with FACSDiva software (BD Biosciences), and data were quantified using FlowJo software (Tree Star, Inc.) or Cytobank (Cytobank, Inc.).

Tumor cytokines were quantified by either luminex or by ELISA. Luminex analysis was performed using the Cytokine and Chemokine 36-Plex Mouse ProcartaPlex Panel 1a (Cat# EPX360-26092-901) array. Tumors were collected in 1 mL PBS and disrupted with manual and GentleMacs (Miltenyi Biotec) dissociation. TDLNs were collected similarly in 1 mL PBS and dissociated with syringe plungers and then spun to remove cellular matter. Lysates for luminex analysis were taken directly from the 1 mL solution and plated according to the manufacturer's suggestions. To reduce luminex sample volume, the DropArray LT210 Washing Station MX (Cat# LT-210MX-01-01; Curiox) was used with microbead plates (Curiox, Cat# 96-CC-BD-05). These plates were then run on using the Bio-Plex 200 (Bio-Rad, Cat#171-000201).

Material for ELISAs was prepared by resecting tumors and dissociating in Cell Lytic M (Sigma; Cat# C2978) with Protease Inhibitor Cocktail (Sigma; Cat# S8820). Tumors were incubated at room temperature for 10 minutes, vortexed, and centrifuged at 10,000 × g to pellet DNA and cellular matter. Lysates were removed and frozen at −80°C until ELISA analysis. The following ELISA assay kits were utilized: mouse IL6 (Cat# M6000B; R&D), mouse TNFα (Cat# MTA00B; R&D), VeriKine mouse IFNβ (Cat# 42400-2; PBL Assay Science), Mouse Inflammatory Cytokines Multi-Analyte ELISArray Kit (Cat# MEM004a; Qiagen).

Tumors were collected at indicated time points after i.t. CDN injection in media containing Golgi inhibitors (monensin and brefeldin A) and maintained in Golgi inhibitors throughout processing. Single-cell suspensions were prepared by dissociation with scissors, then by GentleMACs, followed by incubation in 0.2% w/v collagenase IV (Worthington Biochemical) and benzonase (25 U/mL; Sigma-Aldrich) for 30 minutes at 37°C. Samples were mechanically dissociated over the GentleMACS again as before and then cultured for 4 hours. The samples were then stained for surface markers, fixed in BD Cytofix/Cytoperm (BD Biosciences; Cat# 553722) according to the manufacturer's protocol and stained for intracellular cytokines in Permwash (BD Biosciences; Cat# 554723). Samples were analyzed using an LSR II cytometer with FACSDiva software (BD Biosciences), and data were quantified using Cytobank (Cytobank, Inc.).

The following antibodies and reagents were used: from Bio-Legend: Ly-6G-BV510 (1A8), Ly-6C–BV711 (HK1.4), TNFα-PE (MP6-XT22), CD90.2-AF700 (30-H12); From BD: CD11b-BUV395 (M1/70), CD19-BUV737 (ID3), MHC class II-BV605 (M5), CD45-BB515 (30-F11), NK1.1–PE-CF594 (PK136), CD11c–PE-Cy7 (HL3); From PBL Assay Science: purified IFNβ (RMMB-1) conjugated with AF647 protein labeling kit (Thermo Fisher Scientific, Cat# A20186); from Thermo Fisher Scientific: Near-IR Dead cell stain (Cat# L34976).

Graphical Illustrations were created in Prism7 (Graph Pad). Statistical significance for bar graphs was determined using one- or two-tailed, unpaired Student t tests, also using Prism7. Statistical significance for tumor outgrowth curves was calculated by performing Student t tests on individual time points. *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001. ns, not significant.

We sought to build upon our previously published results by further defining the mechanism of action at both macroscopic and microscopic levels for the acute rejection of tumors following i.t. CDN treatment. To study this, we used the CDN derivative ADU-S100 (referred to here as “CDN”). Animals were treated as in Fig. 1A. Established tumors were injected i.t. with CDN for a total of three doses. Within 48 to 72 hours of the first injection, erythema, induration, and necrosis were apparent at the injection site (Supplementary Fig. S1). By 4 to 5 days after injection, an eschar formed, and tumors were no longer palpable. Eight days after the first i.t. injection, mock-treated animals bore tumors approaching or exceeding 2,000 mm³, necessitating sacrifice. In distinct contrast, CDN-treated mice were left with eschars remaining at the site of the initial tumor (Fig. 1B–D). Several weeks posttreatment, mice that had completely cleared tumors showed evidence of eschar healing and injection site depigmentation, suggestive of a melanocyte-specific T-cell response (Fig. 1F). This model was dependent upon tumor volume at the time of initial treatment. Across multiple experiments, 11/14 animals bearing tumors 150 mm³ or smaller experienced full regression of the primary tumor. By contrast, 1/5 animals with a tumor burden greater than 150 mm³ when therapy was initiated experienced complete regression. In some instances, injection and necrosis of these rapidly growing tumors was incomplete and relapse occurred. In such cases, regrowth generally appeared near the periphery of the initial i.t. injection site (Fig. 1E, arrow notes tumor regrowth). We hypothesized that regrowth may be due to two phenomena: (i) the aggressive growth of B16F10 tumor cells or (ii) the lack of priming of an effective tumor-specific CD8⁺ T-cell response in animals where regrowth of B16F10 was observed.

To better understand the immunological effects of i.t. CDN, we performed analyses of the tumor microenvironment (TME) in treated mice. Because such treatment results in significant inflammation and necrosis within 48 to 72 hours of injection (Supplementary Fig. S1), as well as caspase3 activation within 24 hours of injection (Fig. 2B), we hypothesized the involvement of innate responses, rather than rapid infiltration of tumors with adaptive lymphocyte populations. Utilizing the B16F10 model shown in Fig. 1, we first quantified the cellular components of the TME 24 hours after i.t. CDN treatment. Both tumors (Fig. 2) and TDLNs (Supplementary Fig. S2) were analyzed. Although innate cell subsets such as CD11b⁺F4/80⁺ macrophages and CD11b^hiF4/80⁻ neutrophils increased in the TME, T and B lymphocytes were generally, though not always significantly, decreased after i.t. CDN, and NK cells in the TME were essentially unchanged following i.t. CDN treatment (Fig. 2A). The TDLNs exhibited an analogous decrease in T cells, along with a significant increase in high FSC and SSC cells, around 40% of which were CD11b⁺F4/80⁺ macrophages (Supplementary Fig. S2A).

We next quantified i.t. (Fig. 2C) and intra-TDLN (Supplementary Fig. S2B) cytokines 1 hour and 5 hours after CDN treatment. Several innate cytokines, including GM-CSF, IL1, and TNFα, as well as chemotactic factors MCP-1 and RANTES, were increased significantly 5 hours after CDN injection. This broad-based increase in cytokines and chemokines in these compartments at the 5 hours may suggest a contribution of both direct STING signaling, as well as secondary induction of pathways activated by the initial wave of STING-induced effector molecules.

To identify the immune components critical for i.t.-CDN–mediated antitumor responses, we repeated the experiments shown in Fig. 1A using a panel of knockout mouse strains (Fig. 3A–G; Supplementary Fig. S3). As expected, whereas parental WT C57BL/6 mice exhibited complete acute control of tumors following i.t. CDN injection, STING^−/− animals were unresponsive. The complete lack of CDN-dependent antitumor efficacy observed in STING^−/− mice suggested that any CDN-driven signaling and downstream effects in B16F10 tumor cells themselves did not, in isolation, significantly contribute to antitumor efficacy. To confirm this observation, we incubated B16F10, MC38, CT26, or 4T1 tumor cells with CDN (50 μg/mL; Supplementary Fig. S3). CDN treatment of cells did not affect B16F10 cell morphology (Supplementary Fig. S3A), growth kinetics (Supplementary Fig. S3B), or induce cell death in culture (Supplementary Fig. S3C). B16F10 cells were not distinct in their hypo-responsiveness to CDN. Both B16F10 cells and MC38 murine colon carcinoma cells lacked the ability to produce IFNβ after CDN exposure, in contrast to CT26 colon and 4T1 breast tumor lines (Supplementary Fig. S3D). No lines tested produced TNFα in response to CDN (Supplementary Fig. S3E). Human melanoma and breast tumor lines, shown to harbor no STING mutations and to express varying levels of STING (36, 37), were tested in a similar fashion and did not produce IFNβ after incubation with CDN for 4 and 24 hours (Supplementary Fig. S3F), suggesting that B16F10 melanoma is a representative model for human tumors by this measure.

In contrast to STING-deficient mice, tumor reduction and eschar formation following CDN i.t. therapy was indistinguishable between cGAS^−/− and parental C57BL/6 mice, showing that direct engagement of STING by i.t. administration of CDN obviates the requirement for sensing of dsDNA by cGAS to activate the STING signaling pathway. Both RAG2^−/− and IL6^−/− mice also showed similar tumor necrosis and eschar formation as WT animals. Studies in IFNαR^−/− animals were complicated by the fact that B16F10 tumors grew far more rapidly in these animals (Fig. 3G). We, thus, tested a requirement for IFN signaling by injecting CDN into non–tumor-bearing animals. We observed normal eschar formation (Fig. 3H), suggesting that, although IFN signaling is clearly important in tumor outgrowth, it may not be required for CDN activity. The only two knockout strains in which CDN activity was clearly attenuated or absent were the STING^−/− and TNFα^−/− animals (Fig. 3E). To confirm a requirement for TNFα in CDN-mediated antitumor responses, we treated WT C57BL/6 mice with CDN in the presence of a soluble TNF-RII fusion protein that effectively blocks TNFα signaling (ref. 38; Enbrel; Amgen; Fig. 3I). These results were consistent with the results seen in TNFα knockout animals—treatment with CDN alone resulted in tumor clearance in 5/8 mice, but when combined with TNFα blockade, tumors progressed in all eight animals in that cohort. TNFα blockade in 4T1, CT26, and MC38 tumor models dramatically reduced the size of observable eschars after CDN administration. The reduction in necrosis was not as complete as in TNFα KO mice, suggesting that TNFα blockade by this measure incompletely neutralizes the cytokine. In mice-bearing CT26, MC38, or 4T1 tumors, high-dose i.t. ADU-S100-driven tumor clearance was not significantly impacted by TNFα blockade, although acute necrosis was attenuated (Supplementary Fig. S4A and S4B). These data provide evidence that the requirement for TNFα may differ between tumor types.

To determine the cellular source of TNFα induced by i.t. CDN treatment, we generated a series of bone marrow chimera mice by irradiating WT C57BL/6 or TNFα^−/− host animals and reconstituting mice with either WT or TNFα^−/− bone marrow. The results of these studies showed that animals reconstituted with WT bone marrow, in either WT or TNFα^−/− hosts, responded more robustly to i.t. CDN as compared with mice reconstituted with TNFα^−/− bone marrow. Both WT and TNFα^−/− hosts reconstituted with WT bone marrow cleared tumors efficiently (Fig. 4A). Consistent with this observation, mice harboring WT bone marrow had significantly higher i.t. TNFα compared with their counterparts reconstituted with TNFα^−/− bone marrow. The TME of animals with TNFα^−/− bone marrow did not express detectable TNFα 24 hours after CDN i.t. therapy (Fig. 4B). However, some necrosis and inhibition of tumor outgrowth was observed in WT chimeric animals with TNFα^−/− bone marrow, suggesting that stromal cells also have the ability to influence tumor inflammation and clearance to some degree. Taken together, these results demonstrate the importance of TNFα produced by both stromal and bone marrow compartments in response to activation of STING for optimal acute tumor clearance following i.t. CDN injection.

To understand the cellular source of innate cytokines in TME, we treated B16F10 tumor-bearing animals with 100 μg CDN i.t. and performed intracellular staining (ICS). Because bone marrow–derived cells were required for TNFα production (Fig. 4), analyses were restricted to the CD45⁺ leukocyte compartment. To ensure that we were observing primary and direct effects of CDN ligation of STING within these cells and not secondary effects from cytokine-induced downstream interferon-stimulated genes, TNFα and IFNβ levels within tumor infiltrates were assessed at early time points, i.e., at 1 hour and 5 hours post i.t. CDN injection (Fig. 5A). Within 1 hour after CDN administration, both TNFα and IFNβ were expressed by CD45⁺ cells in the TME (Fig. 5B and C). By 5 hours after injection, both TNFα and IFNβ in the CD45⁺ compartment were reduced to near background levels. Using surface markers to determine the lineage of cytokine-producing cells, we found activated monocytic cells (CD11b⁺Ly6C⁺MHCII⁺) were largely responsible for production of TNFα and IFNβ, although other monocytic cells (CD11b⁺Ly6C⁺MHCII⁻), as well as macrophages (CD11C⁺CD11b⁺MHCII⁺), also contributed to production of these cytokines (Fig. 5C). Consistent with our prior data, the TME contained very few T and B cells at these early time points.

Because the results from our bone marrow chimera studies indicated that neither TNFα competent bone marrow–derived cells or stromal cells alone were sufficient to fully recapitulate the phenotype of the acute antitumor response to CDN treatment into WT animals, we sought to better define the relative contributions of these two compartments. To address this question, we made a series of bone marrow chimeras using WT and STING^−/− mouse strains. WT and STING^−/− hosts were first irradiated and then reconstituted with either WT or STING^−/− bone marrow cells. B16F10 tumor cells were implanted into chimeric and parental mice, and CDN was given by i.t. injection when tumors were well established. After CDN therapy, tumor outgrowth was inhibited in all mice except in STING^−/− parental mice (Fig. 6A and B). Treatment in both chimeric groups (STING^−/− bone marrow into WT and WT bone marrow into STING^−/− mice) resulted in similar reduction of tumor burden. As we observed in studies with the TNFα^−/− chimeric mice, only WT bone marrow into WT animals could completely reject tumors, whereas all chimeric animals exhibited partial reductions and eventually succumbed to tumor burden. However, only animals with STING signaling competent stroma (WT animals with STING^−/− or WT bone marrow) became necrotic after i.t. CDN treatment (Fig. 6C). In distinct contrast, necrosis was not observed in any animal that lacked stromal STING signaling (STING^−/− animals with STING^−/− or WT bone marrow), as determined either by visualization (Fig. 6C) or by blind scoring of photographs for necrosis signatures (Fig. 6D). Consistent with this observation, significant CDN-induced expression of IFNβ, IL6, and TNFα were measured by ELISA 24 hours after i.t. injection only in tumor lysates processed from mice with intact stromal STING signaling (Fig. 6E). These data reveal that although CDNs elicit cellular responses from both bone marrow and stromal compartments, a qualitatively different response occurs when the stroma is activated, leading to injection site necrosis.

In prior studies, we found that i.t. administration of CDNs can enhance tumor-specific adaptive responses, and animals that clear tumors can reject a future tumor rechallenge (35). Confirming a putative downstream adaptive immune response, we found that i.t. treatment with CDNs can result in vitiligo (Fig. 1F). To address the role of STING signaling in DC activation in the hematopoietic and stromal compartments, we again generated WT and STING^−/− chimeric mice (Fig. 6) and analyzed the composition and relative APC activation in those animals. A significant infiltration of FSC^hi SSC^hi cells, corresponding to macrophage and monocyte populations, into the TDLNs was noted in all chimeric mice that were evaluated, except for STING^−/−mice with STING^−/− bone marrow (Fig. 7A). FSC^hi SSC^hi cells were further defined using the lineage markers CD11b and CD11c to identify both macrophage and dendritic cell populations (Fig. 7B). Both CD11b⁺CD11c⁺ and CD11b^LoCD11c⁺ macrophage and DC populations, as well as CD11b⁺CD11c⁻ neutrophils were enriched in the TDLNs 24 hours after CDN administration (Fig. 7C). In all chimeras except for STING^−/−mice with STING^−/− bone marrow, upregulation of the costimulatory molecule CD86 on macrophage and DC populations was observed after CDN administration (Fig. 7D). These results suggest that i.t. CDN can result in activation of APCs either via direct activation of the STING pathway in APCs, or, alternatively, indirectly through the production of inflammatory cytokines by STING competent stroma.

We show here that i.t. administration of CDN is a powerful immunotherapy that results in tumor clearance and long-term survival of animals bearing aggressive B16F10 tumors. In these studies, we focused on the acute effects of high-dose i.t. injection CDN therapy, while also noting the potential for APC activation and long-term adaptive immunity. Our bone marrow chimera studies showed that both bone marrow–derived and stromal cells are important in tumor clearance, and neither compartment alone is sufficient for full tumor clearance. These studies represent important progress toward understanding the effects of CDN on immune and nonimmune cell subsets.

I.t. administration of CDN results in significant changes in the TME. Flow cytometry revealed an initial reduction in the proportion of tumor-infiltrating lymphocytes and simultaneous acute increase of tumor-infiltrating macrophages and neutrophils. The tumor cells themselves undergo cell death, as demonstrated by robust IHC staining for cleaved caspase 3 in the tumor. This change in cellular composition is likely both a product of and a cause of the significant increase in multiple i.t. cytokines, including IL6, TNFα, IFNβ, and GM-CSF. This cytokine increase has been reported previously and is hypothesized to be the result of both NF-kB and IRF-3 activity, as well as STAT6 (14, 15). However, whether NF-kB activation is a direct result of STING ligation or rather the result of secondary cytokine signaling remains unclear and is a difficult hypothesis to address due to the rapid induction and abundance of cytokines observed after CDN administration. A report by Gaidt and colleagues showed that the DNA inflammasome is one of the downstream targets of STING activation, leading to IL1β and IP-10 secretion (39).

The data shown here emphasize the potent, ablative effect of high concentrations of innate cytokines induced by CDNs in TME. Studies have shown that IFNαR signaling is an important mediator of antitumor immunity (40), and although our experiments suggested that the acute necrosis following i.t. administration of CDN is not IFNαR mediated, further studies in our lab, as well as in other groups, have shown that IFNαR^−/− animals have reduced capacity to mount long-term immune responses after CDN therapy (33, 35).

Mice lacking TNFα did not develop injection site necrosis when CDN were administered i.t. These findings were corroborated in our studies showing that tumor necrosis can be significantly reduced by administration of the TNFα inhibitor Enbrel. TNFα blockade reduced both the size of the eschar at the injection site, as well as tumor clearance in the nonimmunogenic B16F10 model. In the MC38, CT26, and 4T1 tumor models, TNFα blockade reduced the size of the eschar but did not attenuate tumor clearance. This suggests that there may be distinct phases of antitumor immunity generated by i.t. CDN therapy. TNFα is likely required for acute necrosis, whereas the chronic nature of a tumor burden may require the induction of adaptive immunity for an antitumor effect. We suspect that the difference in tumor rejection between B16F10 and MC38, CT26, and 4T1 models is due to differences in immunogenicity among these tumor lines. At baseline, B16F10 tumors are generally immunologically bland, and this lack of inherent immunogenicity is reflected in the inability of PD-1 or PD-L1 blockade to attenuate tumor growth (41, 42). By comparison, MC38, 4T1, and CT26 tumors are variably infiltrated with a more immunogenic signature (43). As a result, CDN treatment of B16F10 tumors may be hampered by the lack of a preexisting adaptive immune response and, thus, is more reliant on innate mechanisms. Therefore, whereas the acute induction of TNFα by i.t. CDNs is necessary for the clearance of the nonimmunogenic B16F10 tumor line, more immunogenic models may not require such TNFα-mediated tumor destruction for clearance. Instead, TNFα independent mechanisms are sufficient to clear tumors.

We were further able to dissect the TNFα-producing bone marrow–derived populations using linage markers and multidimensional flow cytometry. Using this approach, we ascertained that although multiple lineages of bone marrow–derived cells were capable of producing TNFα, monocytes and macrophages were both more likely to be producing TNFα, based on the frequency of cells positive for TNFα and TNFα MFI. Although these results indicate that innate immune signaling from high-dose CDN can clear tumors in the absence of the adaptive immune system, it is clear that adaptive immunity is critical for long-term efficacy of CDN therapy on the whole and in response to lower dose administrations of CDN.

With this comprehensive study of the critical cellular components and cytokines induced by acute CDN treatment, it has become abundantly clear that the CDN-dependent immune clearance of tumors depends on cross-talk or additive effects from both immune and stromal cells. At first glance, it may seem that the importance of TNFα production by bone marrow–derived cells is not reconcilable with our data that show STING competence in the stroma is necessary. However, it is conceivable that STING signaling in the stroma leads to production of factors that attract bone marrow cells, potentiate their production of TNFα in the tumor, or that STING signaling primes the TME for destruction by upregulation of TNFαR1. Future studies will be required to elucidate these mechanisms and to understand the bone marrow and stromal STING requirements for priming adaptive immune responses by i.t. CDN administration.

As cancer immunotherapies are being approved for treatment for a variety of indications, it is important to recognize the impact on stromal cells, as well as immune cells within tumors. In mice and in humans, stromal cells, such as fibroblasts and endothelial cells, can express and signal through PRRs, shaping the immune response through production of cytokines. In our models, CDN injection acutely leads to the activation of the stroma, an inflammatory immune environment, and tumor clearance. Activation of APCs and observations of long-term vitiligo at injection sites suggest that tumor-specific adaptive responses exist and are functional. The direct injection of CDN is a powerful means to target TME reorganization and the induction of functional immune responses to cancer through activation of STING signaling in stromal and immune cells within the tumor.

A.L. Desbien is a scientist at Aduro Biotech and reports receiving a commercial research grant from the same. K.E. Sivick has ownership interest in a patent with Aduro Biotech. C.G. Drake reports receiving a commercial research grant from Aduro Biotech. No potential conflicts of interest were disclosed by the other authors.

Conception and design: B.J. Francica, A. Ghasemzadeh, A.E. Marciscano, A.B. Sharabi, T.W. Dubensky Jr, D.M. Pardoll, C.G. Drake

Development of methodology: B.J. Francica, A.L. Desbien, A.E. Marciscano, A.B. Sharabi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.J. Francica, A. Ghasemzadeh, A.L. Desbien, D. Theodros, K.E. Sivick, G.L. Reiner, L.H. Glickman, A.E. Marciscano

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.J. Francica, A.L. Desbien, G.L. Reiner, L.H. Glickman, A.E. Marciscano, D.M. Pardoll, C.G. Drake

Writing, review, and/or revision of the manuscript: B.J. Francica, A. Ghasemzadeh, A.L. Desbien, D. Theodros, K.E. Sivick, L.H. Glickman, A.E. Marciscano, A.B. Sharabi, M.L. Leong, S.M. McWhirter, T.W. Dubensky Jr, C.G. Drake

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.J. Francica, A.L. Desbien, A.E. Marciscano

These studies were supported by NIH/NCI R01CA154555, the Prostate Cancer Foundation (PCF), The Patrick C. Walsh Fund and a sponsored research agreement from Aduro Biotech.

The authors of this article would like to thank Ada Tam and Lee Blosser for their help and maintenance of the flow cytometry core. We would also like to thank Dr. Young Kim for his donation of Golden Ticketmice, and Dr. Skip Virgin for his donation of cGAS^−/− mice.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.