Evaluation of Explant Responses to STING Ligands: Personalized Immunosurgical Therapy for Head and Neck Squamous Cell Carcinoma

Surgery is a primary mode of treatment for many patients with head and neck squamous cell carcinoma (HNSCC), but adverse histopathologic features, such as close or positive margins, extranodal spread of tumor in metastatic lymph nodes, multiple lymph node metastases, and perineural and perivascular spread, are negative prognostic indicators for which adjuvant radiotherapy or chemoradiation therapy is recommended (1). However, despite risk-adapted adjuvant therapy, almost half of patients with locoregionally advanced HPV-negative (HPV⁻) HNSCC who undergo surgical resection recur locally or distantly. Local recurrence can be particularly resistant to subsequent therapy, and many patients experience devastating complications from uncontrolled tumor progression and/or the salvage treatment.

Despite a concerted effort on the part of surgeons to extirpate all viable tumor cells, local recurrence develops in approximately 25% of patients who have histologically negative margins (2). Furthermore, approximately 25% of patients with positive margins do not recur. This observation suggests that residual tumor cells can be present following surgery and that some patients can control residual disease following the operation. Our prior studies have demonstrated that immune mechanisms can control minimal residual disease following surgical resection in preclinical models, and that the addition of systemic immunotherapy can enhance local control (3). Moreover, in melanoma we demonstrated clinical immune control of distant metastasis by local injections of CpG after surgical removal of the primary tumor and 1 week prior to the sentinel lymph node procedure (4). Thus, the ability to enhance tumor-specific adaptive immune responses at the time of the initial procedure may be valuable for control of residual microscopic disease, as well as distant tumor deposits (3, 5, 6).

Recently, we and others have demonstrated that intratumoral cyclic dinucleotide (CDN) ligand activation of the stimulator of interferon genes (STING) pathway strongly induces type I IFN and TNFα responses, resulting in rapid regression of a range of tumors (7–9). We found that STING is strongly expressed in HPV⁺ HNSCC cancer cells, but is poorly expressed in HPV⁻ HNSCC cancer cells (10). We and others have shown that local administration of STING ligands to the tumor results in the nonautonomous activation of STING in noncancer cells within the tumor environment, indicating that this therapy can be effective in both STING⁺ and STING⁻ cancers (7, 8, 10). These data also imply that in HPV⁻ HNSCC any response to STING ligand will be entirely dependent on the response of host cells in the tumor immune environment, and therefore has the potential to vary significantly between individuals with widely varying tumor composition.

Recently, novel approaches have been developed that take advantage of the wound-healing response to target systemic immunotherapy to the site of resection (11). STING ligands require local action, and because surgeons are in the unique position of being in direct contact with the tumor during resection, this provides an opportunity to apply therapies directly to the resection margins to target residual cancer cells. Using this close proximity to our advantage, we developed an approach that applied immunotherapies directly in the resection cavity at the time of surgery. Following closure of the resection site, the immunotherapy will remain in place to activate local immune mechanisms to control residual disease. To apply this therapy, we developed a hydrogel-like formulation of STING ligands that is a liquid when refrigerated but forms a gel following in vivo application. We call this STING-targeted surgical immunotherapy STINGblade. Herein, we demonstrate using murine models of purposeful partial tumor resection that STINGblade results in dramatic immune-mediated control of residual disease. We further develop a postresection explant model to identify the variability of responses to STINGblade in murine and patient tumors, and demonstrate that based on these responses, STINGblade can be personalized to improve tumor control.

Six- to 8-week-old female C3H, C57BL/6, or BALB/c mice were obtained from Charles River Laboratories for use in these experiments. STING^−/− (Goldenticket Stock# 017537) and IFNAR1^−/− (Stock# 028288) mice were obtained from The Jackson Laboratory. Survival experiments were performed with 6 to 8 mice per experimental group, and mechanistic experiments with 4 to 6 mice per group. Animal protocols were approved by the Earle A. Chiles Research Institute (EACRI) Institutional Animal Care and Use Committee (Animal Welfare Assurance No. A3913-01).

The SCCVII squamous cell carcinoma cell line was kindly provided in 2014 by Dr. Lee (Duke University Medical Center, Durham, NC). The TC1 squamous cell carcinoma cell line is a Ras and HPV-transformed mouse cell line, commonly used as a C57BL/6 mouse model of HNSCC (12), and was kindly provided by Dr. Hong Ming Hu (EACRI, Portland, OR). The 4T1 mammary carcinoma cell line (BALB/c; ref. 13) was obtained in 2009 from the ATCC. MOC1 and MOC2 were kindly provided in 2017 by Dr. Ravindra Uppaluri (Dana Farber Cancer Institute, Boston MA; ref. 14). Authentication was performed using murine haplotype-specific MHC antibodies, and they were tested for contamination within the past 6 months using a Mycoplasma Detection Kit (SouthernBiotech).

The STING ligands cyclic-di-AMP and rr-cyclic-di-AMP were obtained from Invivogen. Blocking anti-IL10R antibody (BE0050 – BioXCell) was given at a 100 μg dose in the biomaterial. Depleting anti-CD8 (YTS 169.4) antibodies were obtained from BioXCell and were given i.p. 50 μg one day before treatment and again 1 week later. Flow cytometry antibodies included CD45-BV786, CD4-PE, TNFα-BV421 (BD Biosciences), CD90.2-AF700, CD11b-BV605, Ly6C-BV711, Ly6G-PE-594, F4/80-PerCPCy5.5 (Biolegend), CD8-AF647, MHCII-FITC (eBioscience), and CD24-APCe780 (Thermo-Fisher).

Tumors were inoculated subcutaneously into immune-competent female mice at a dose of 5 × 10⁵ SCCVII (C3H mice), 5 × 10⁵ TC1, 2 × 10⁵ MOC1, 2 × 10⁵ MOC2 (all C57BL/6 mice), or 2 × 10⁴ 4T1 (BALB/c mice). Tumors were allowed to develop to 7 to 9 mm in diameter, at which point the mice underwent subtotal resection. Briefly, mice were anesthetized by Isoflurane inhalation, and following surgical skin prep, an incision was made over the tumor and the encapsulated tumor mechanically detached from the skin. The tumor capsule attached to the underlying fascia was left in place and the upper portion of the tumor resected, leaving 1 to 3 mm depth of tumor, approximately 20% of the original material, in the resection cavity. Mice were randomized to treatment with local immunotherapy in a biomaterial. To develop a simple biomaterial for delivery of immunotherapy to the resection cavity, we employed the phase transition property of Matrigel (Corning Inc.), a thermoresponsive hydrogel primarily composed of laminin and collagen IV. Matrigel was mixed 1:4 with PBS or PBS containing 25 μg cyclic-di-AMP (CDN) and kept on ice at 4°C. A 0.01mL volume of this liquid agent was manually placed dropwise into the surgical cavity and allowed to undergo phase transition over 1 to 5 minutes. On application to the surgical cavity, the hydrogel formed an amorphous solid that held position through wound closure. Following application, the wound was closed and mice were allowed to recover. Tumor-bearing mice were monitored a minimum of 3 days per week and euthanized when tumors exceeded 12 mm in any dimension, or when body condition score declined one level. Euthanasia was performed with CO₂ inhalation followed by a second method.

To determine whether normal tissue healing was affected by treatment, we used a classic excisional injury model. Shaved, anesthetized mice were given full-thickness dermal excisional wounds using a 3-mm dermal biopsy punch (Acu Punch, Acuderm). Each 3 mm wound was treated with Matrigel containing PBS or CDN at a dose of 25 μg as above. Wound healing was documented, and at day 7 after wounding, mice were euthanized and the wounds with surrounding tissue were excised and paraffin-embedded for histology. To analyze the degree of re-epithelialization, 10 μm sections were hematoxylin and eosin–stained and scanned for digital microscopy using a Leica SCN400 whole slide scanner. The distance between epithelial tongues was measured at the widest opening.

To determine whether the biomaterials affect the strength of incisional repair, we use a standard approach to measure the mechanical strength of healing incisional wounds. Shaved, anesthetized mice were given a 2 cm incisional wound extending through the dermis and the panniculus carnosus in the paraspinal region. The wound site was treated with Matrigel containing PBS or CDN at a dose of 25 μg as above and closed using sterile surgical clips. Surgical clips were removed after 5 days, and the wounds were harvested 2 days later. From each wound, strips perpendicular to the incision were excised and subjected to tensiometry. In this procedure, sequentially increased tension was applied to the excised wound strips using a M7-10 digital force gauge on an ES20 test stand, and MESUR gauge software, all acquired from Mark-10 Corporation. Wound disruption strength was defined as the load value of wound disruption.

Deidentified human tissue sections and freshly excised surgical material were obtained with written-informed consent from the patients and conducted in accordance with recognized ethical guidelines under IRB# 12-075 approved by the Providence Portland Medical Center Institutional Review Board. The tumors were classified as HPV⁺ if they scored as positive for p16 by immunohistochemistry and originated in the oropharynx. Samples that were negative for p16 and/or originated outside the oropharynx were scored as HPV⁻. Representative tissue blocks for analysis were identified by the reviewing pathologist. Patient details are provided in Supplementary Fig. S1.

Excised murine tumors or surgical samples from human subjects undergoing resection for HNSCC were delivered to the lab on ice following excision. The fresh tumor samples were dissected into 2 mm fragments and each fragment placed into 96-well plates preloaded with PBS or 25 μg rr-CDA, which has a broad spectrum activity to both murine and human STING (8). Explants were overlaid with 100 μL media containing human serum for 24 hours then analyzed for cytokine secretion into the media by multiplex bead assay (Life Technologies) as previously described (15) and read on a Luminex 100 array reader. Cytokine concentrations for replicates of each tumor sample were calculated according to a standard curve.

For immunohistology, murine tumor-draining lymph nodes were fixed in Z7 zinc-based fixative overnight (16). Tissue was then processed for paraffin tissue sections. Note that 5 μm sections were cut and mounted for analysis. Tissue sections were boiled in EDTA buffer for antigen retrieval. Immunohistology antibody to cytokeratin-19 (CK19) was purchased from Santa Cruz Biotechnology. Primary antibody binding was visualized with AlexaFluor 488–conjugated secondary antibodies (Molecular Probes) and mounted with 4′,6-diamidino-2-phenylindole (DAPI; Invitrogen) to stain nuclear material.

Routinely processed paraffin-embedded tumor tissues from surgical specimens of patients tested in the explant assay were used to prepare 5 μm sections for staining. Sections were deparaffinized, rehydrated, and then deprotected before sequential staining with primary antibodies to PDL1 (Cell Signaling Technology), CD163 (Roche), CD8 (Spring Bioscience), Cytokeratin (Dako), CD3 (GeneTex), and FoxP3 (Abcam). Antibody binding was visualized using a horseradish peroxidase–conjugated antispecies antibody (Biocare Medical) and an Opal 7-color manual IHC kit (Perkin Elmer). Slides were counterstained with DAPI and mounted for analysis. Serial sections were stained for expression of STING as previously described (10).

Images were acquired using a Zeiss Axio observer Z1 with attached Nuance Multispectral Image camera and software (Perkin Elmer). All images displayed in the article are representative of the entire tumor and their respective experimental cohort.

MOC1 and MOC2 tumors grown in C57BL/6 mice were harvested, and single-cell suspensions were prepared by dissection into 2 mm fragments followed by agitation in digest solution (1 mg/mL collagenase, 100 μg/mL hyaluronidase, and 20 mg/mL DNase in PBS) for 1 hour at room temperature as previously described (17). The digest was filtered through 100 μm nylon mesh to remove macroscopic debris. Cells were left untreated or stimulated with 25 μg/mL CDA in the presence of brefeldin A for 4 hours at 37°C. Cells were surface stained for a major phenotypic markers including viability dye and CD45, CD90, CD19, CD4, CD8, CD11b, MHCII, Ly6C, Ly6G, CD11c, CD24, and F4/80 to distinguish CD11b⁺MHCII⁺Ly6C⁻Ly6G⁻F4/80⁺ macrophages, CD11b⁺MHCII⁻Ly6C⁺Ly6G⁺ neutrophils, CD11b⁺F4/80⁻Ly6G⁻Ly6C⁺ monocytic cells, and CD24⁺MHCII⁺CD11b^hi DC1, CD24⁺MHCII⁺CD11b^low DC2 (21). Cells were additionally permeabilized and intracellularly stained with anti-TNFα to identify the responsive cell subsets. Samples were run on an LSRII flow cytometer (BD Biosciences) and analyzed using FloJo.

Data were analyzed and graphed using Prism (GraphPad Software). Individual data sets were compared using the Student t test, and analysis across multiple groups was performed using ANOVA with individual groups assessed using Tukey comparison. Overall survival of groups was compared using the log-rank test for differences in Kaplan–Meier survival curves.

We developed a model of subtotal tumor resection using subcutaneous tumors in immunocompetent mice followed by the application of the hydrogel Matrigel into the resection cavity (Fig. 1A). Matrigel was selected based on useful properties, including its solubility in water, ability to absorb water-soluble agents, capacity to form a gel-like solid when elevated to body temperature, and rapid degradation in vivo. In mice bearing established SCCVII squamous cell carcinoma that were treated with subtotal tumor resection and local application of Matrigel, tumors rapidly recurred at the resection site (Fig. 1A). Our locally administered immunotherapy consisted of STING ligands, which strongly induce type I IFN and TNFα, resulting in rapid regression of a range of advanced tumors following direct intratumoral injection (7–9). When STING ligands are incorporated into Matrigel, this resulted in cure of the residual tumor (Fig. 1B) without any signs of toxicity. To evaluate long-term tumor control, mice bearing SCCVII tumors received subtotal resection and were randomized to receive Matrigel containing PBS or CDN to the resection site. All mice receiving PBS recurred locally, whereas those mice receiving STING ligands were cured (P < 0.0001; Fig. 1C). These data demonstrate that STING ligands act as a potent local immunotherapy that controls residual tumors following surgical resection in HNSCC preclinical models and suggests a role for implantable immune-modulating interventions to improve locoregional control in human patients.

To confirm these findings in a different model, we used the TC1 tumor model, which is a Ras and HPV-transformed mouse cell line and is commonly used as a C57BL/6 mouse model of HNSCC (12). As with SCCVII, subtotally resected TC1 tumors regrew locally, but local recurrence was prevented by local administration of CDN in Matrigel (P < 0.01; Fig. 2A). We then used syngeneic C57BL/6 STING^−/− and IFNAR1^−/− mice to investigate the mechanism of tumor cure by STING ligands. Subtotal resection plus local administration of CDN in Matrigel was no longer effective in IFNAR1^−/− mice, which lack the ability to respond to type I IFN, and as expected was no longer effective in STING^−/− mice, where the cancer cells are unchanged, but the host cells lack the ability to recognize CDN (wt CDN vs. IFNAR1^−/− CDN P < 0.001; wt CDN vs. STING^−/− CDN P < 0.0001; Fig. 2A). These data agree with observations in C3H mice bearing SCCVII tumors (Fig. 1B); because SCCVII lack expression of STING (10), the effectiveness of therapy with STING ligands in this model will be entirely dependent on host cell expression of STING. Prior therapies using intratumoral injection of STING ligands demonstrated a similar dependence on host STING and IFNAR1 responsiveness, but were also dependent on CD8⁺ T cells for efficacy (8). The rapid posttreatment tumor regression following STINGblade meant we were not able to evaluate T-cell influx into the residual tumor. Instead, to determine whether subtotal resection plus local administration of CDN resulted in adaptive immune control of residual disease, mice were depleted of CD8⁺ T cells starting 1 day prior to resection. Depletion of CD8⁺ T cells reduced the efficacy of STINGblade (P < 0.05; Fig. 2B), indicating that adaptive immune responses are required for control of residual disease. However, there remained some efficacy in the absence of T cells, which would be consistent with inflammatory action, such as the T-cell–independent TNFα-mediated hemorrhagic necrosis reported following administration of STING ligands to tumors (7, 19). These data demonstrate that application of STING ligands to the resection cavity results in local control of tumor recurrence via activation of STING in host cells, type I IFN signaling in host cells, and results in CD8⁺ T-cell–mediated cure of tumors.

Because STINGblade generates CD8-mediated control of residual disease, it is possible that these CD8 T cells can also control distant tumor. The SCCVII model is spontaneously metastatic, resulting in micrometastases in the tumor-draining lymph node (Fig. 3A; ref. 20), and in untreated mice, lung metastases can develop (not shown). Mice cured of the primary tumor by STING ligands had no evidence of metastatic disease and remained tumor-free long term, suggesting control of metastatic disease with the local therapy. To determine whether CD8 T-cell immunity affected gross distant disease, mice were given tumors on both flanks and only one treated with subtotal resection plus local administration of CDN. We found that growth of the distant tumor was modestly suppressed by CDN treatment within the subtotal resection site, but not by surgery alone, indicating some limited impact of systemic immunity on gross disease (Fig. 3B). To evaluate this in a more aggressive model of metastases, we established subcutaneous 4T1 mammary carcinomas in BALB/c mice, which results in early spontaneous development of lung metastases (Fig. 3C, i; ref. 15). Subtotal resection plus local administration of CDN resulted in improved local control compared with control treatment (Fig. 3C, ii). However, this did not result in long-term survival due to unchanged progression of lung metastases (Fig. 3C, iii). Thus, even mice that had their primary tumor cured by STINGblade did not experience improved overall survival, suggesting that our local therapy was unable to effectively treat this distant disease in the 4T1 model. Development of lung metastases was also unaffected following complete resection of the primary tumor (Fig. 3D, i), indicating that cancer cells had already metastasized prior to the operation and removal of the primary tumor at this time point had no impact on metastatic progression. To address this, we performed an early resection of 4T1 tumors within 5 days of tumor implantation (Fig. 3D, ii). In this setting, STINGblade significantly extended survival over surgery alone (P < 0.01) and resulted in cure of some mice, with no development of metastatic disease in these animals (Fig. 3D, iii). These data indicate that although application of STING ligands to the resection cavity is an effective locoregional therapy leading to control of metastases if performed before systemic spread, our therapy has limited ability to control distant disease despite its CD8⁺ T-cell–dependent mechanism of action.

One concern in developing a postresection immunotherapy is whether it interferes with wound healing. To address effects of STINGblade on wound healing, we tested the effect of Matrigel plus CDN dissolved in PBS or Matrigel plus PBS control in both an incisional wound strength model and an excisional wound-healing model. For the incisional wound strength model, shaved, anesthetized mice were given an incisional wound extending through the dermis and the panniculus carnosus. The wound site was then treated with Matrigel containing PBS or CDN and closed using sterile surgical clips. Surgical clips were removed after 5 days, and the skin surrounding the wound was harvested at day 7. For each wound, strips perpendicular to the incision were excised and subjected to tensiometry (Fig. 4A, i). We found no significant difference in wound strength between PBS and CDN-treated wounds (Fig. 4A, ii). For the excisional model, shaved, anesthetized mice were given full-thickness dermal excisional wounds using a dermal biopsy punch and then treated with Matrigel containing PBS or CDN. There was no difference in wound closure between PBS and CDA-treated animals (Fig. 4B, i). Pathologic assessment did not identify any significant differences in immune infiltrate of wounds between PBS and CDA-treated animals, and histologic examination of the closed wounds demonstrated that in both groups the wounds actively closed with a defined epithelial tongue, resulting in no significant difference in re-epithelialization of the excisional wound between groups (Fig. 4B, ii and iii). These data demonstrate that STINGblade cures tumors without affecting wound healing, suggesting that this is an immunotherapy that can be safely applied without adversely affecting surgical outcomes.

Intratumoral injection of STING ligands has demonstrated efficacy in a range of tumor models and mouse backgrounds (7–9); however, not all mouse tumor models respond to this therapy (9). To evaluate this diversity of responses in tumors, we developed an explant model, where the resected portion of the tumor was tested for its ex vivo response to STING ligands (Fig. 5A). To determine whether this explant approach could be applied to evaluate the response of patient tumors to STING ligands, we acquired surgical samples from patients undergoing resection for HNSCC and treated them ex vivo with STING ligands or PBS. Treatment with STING ligands consistently resulted in statistically significant increases in IFNα secretion from the explant (Fig. 5A, ii; Supplementary Fig. S2). However, IFNβ, TNFα, and IL10 secretion tended to be more patient-specific. To evaluate the effect of treatment with STING ligands on immune cell recruitment to the tumor, we also evaluated chemokine secretion from the explants. Although the tumor samples from 4 of 5 patients significantly increased secretion of CCL3 following treatment with STING ligands, none of the patients significantly increased CXCL10 secretion (Fig. 5A, ii; Supplementary Fig. S2). We have previously demonstrated that increased expression of CCL3 in tumors results in improved T-cell recruitment and improved tumor control (21, 22), potentially linking the inflammatory effects of STING ligands to T-cell–dependent tumor control (Figs. 2 and 3). Because the patient samples included both HPV⁺ and HPV⁻ HNSCC, and because our previous data demonstrate that HPV⁻ HNSCC does not express STING (10), we were also able to identify samples where the response to STING ligands is dependent on the tumor stroma. To confirm this, we stained matched tissue blocks from these tumors for expression of STING (Fig. 5B). HPV⁻ HNSCC patient CRI-2799 did not express STING in the cancer cells (Fig. 5B, i and ii). Therefore, in this tumor, the stromal expressing STING must be critical to the response to STING ligands. In contrast, the HPV⁺ HNSCC expresses STING in both cancer cells and in the stroma (Fig. 5B, iv and v). To assess the mix of immune cells infiltrating these tumors, we performed multiplex immunohistology for immune markers. The multiplex images demonstrate a diverse immune environment in these tumors, consistent with previous results (Fig. 5B, iii and vi; ref. 23). The tumors exhibit not only extensive infiltrates of CD8 T cells, but also abundant FoxP3⁺ T regulatory cells, CD163⁺ macrophages, and high levels of PDL1 on cancer cells and stromal cells (Supplementary Fig. S3). The HPV⁻ HNSCC CRI-2799 had a decreased immune infiltrate compared with the HPV⁺ HNSCC CRI-2751, but further studies are necessary to determine the critical stromal elements that define the cytokine response to STING ligands. These data demonstrate that patient tumor explants treated ex vivo with STING ligands generate a rapid cytokine response that is dependent on the patient-specific immune stroma. We propose that this novel approach can be used to understand the variability of patient responses to STING ligands, investigate the critical cells that drive these responses, and identify therapies to improve local control.

To determine whether this explant assay could be used to personalize therapy according to the tumor-specific response to STING ligands, we tested the tumor explant assay on a panel of murine HNSCC. We included the aggressive MOC2 HNSCC tumor model that has been shown to be unresponsive to STING ligands, as well as the immunogenic MOC1 HNSCC tumor model that can be cured with the same treatment (9). MOC2 has a distinct profile of infiltrating immune cells compared with MOC1 (9, 24), and because our therapy is dependent on the action of STING ligands on host cells, we proposed that the outcome of tumor therapy with STING ligands could be predicted by the response of the immune cells within the tumor. Treatment with STING ligands resulted in rapid induction of TNFα and type I IFN in all HNSCC-related tumor explants (Fig. 6A; Supplementary Fig. S4). These results closely match the cytokine responses observed in tumors following direct injection in vivo (7). Notably, MOC1 tumors exhibited responses that matched SCCVII and TC1, whereas MOC2 tumors uniquely resulted in IL10 secretion following stimulation with STING ligands (Fig. 6A). Using the explant model in mice also allows us to evaluate the role of the cancer cells versus the stroma in the response to STING ligands, by testing explants grown in wild-type mice versus STING^−/− mice. Explants grown in wild-type mice made a strong cytokine response CDA, but this was lost in STING^−/− mice, indicating STING expression in the stroma is key (Fig. 6B). ERAdP has been described as an alternative target for CDA, which can generate responses in STING^−/− mice (25). We see no evidence of a response to CDA in STING^−/−, indicating that ERAdP cannot substitute for STING in this setting. To confirm this, we also treated explants with cGAMP, which does not activate ERAdP (25). We found that cGAMP generated a similar response to CDA and was also dependent on STING expression in the tumor stroma (Fig. 6B), suggesting that STING in noncancer cells is the sole sensor responsible for the activity of cyclic dinucleotides in this model. To identify the principle-responsive cell population, we performed an unbiased assessment of the cytokine-producing cells in the tumor. Tumor-infiltrating cells were stimulated with STING ligands in the presence of secretion inhibitors and surface stained with an immune cell panel and intracellularly stained for TNFα. We found that TNFα was being produced by a subpopulation of tumor-associated macrophages (CD45⁺CD11b⁺Ly6C⁻Ly6G⁻F4/80⁺MHCII⁺ cells) as well as tumor-infiltrating monocytic cells (CD45⁺CD11b⁺Ly6C⁺Ly6G⁻; Fig. 6C). T cells and Neutrophils (CD45⁺CD11b⁺Ly6C⁺Ly6G⁺) did not generate significant responses, and the MOC1 and MOC2 models lacked significant populations of dendritic cells (DC1: CD45⁺Ly6C⁻Ly6G⁻F4/80⁻MHCII⁺CD24⁺CD103⁻CD11b⁺ and DC2: CD45⁺Ly6C⁻Ly6G⁻F4/80⁻MHCII⁺CD24⁺CD103⁺CD11b⁻ cells), using the panel described by Broz and colleagues (Supplementary Fig. S5; ref. 18). In view of the divergent cytokine response of MOC2 and its known unresponsiveness to STING ligands, we evaluated whether the response to STINGblade was similarly divergent in MOC2 tumors. As previously reported, with a similar tumor challenge, MOC2 grows much more aggressively than MOC1 (Fig. 6D). Following purposeful partial tumor resection and application of STING ligands in Matrigel, the highly immunogenic MOC1 tumors were cured by STINGblade like SCCVII and TC1, though a significant proportion were cured by partial resection alone (Fig. 6Ei). By contrast, no MOC2-bearing mice were cured by STINGblade (Fig. 6Eii). Using the explant response to inform therapy, we evaluated whether IL10 production was responsible for the poor response of MOC2 tumors to STING ligands. MOC2 tumors in C57BL/6 mice with subtotal resection plus local administration of CDN alone or in combination with anti–IL10R-blocking antibody incorporated into the biomaterial. Addition of anti-IL10R results in a significant extension in survival and cure of tumors in a portion of the mice (Fig. 6Eii). These data demonstrate that analysis of the cytokine response to tumor explants treated ex vivo identifies tumors with variable immune responses to STING ligands, and enables personalization of the immunotherapy-containing biomaterial to induce tumor cure.

In this study, we demonstrated that in preclinical models of HNSCC, the incorporation of STING ligands into a hydrogel that is delivered into a subtotal resection cavity results in tumor cure. We show that this mechanism occurs through activation of host STING-expressing cells, not cancer cells, and requires induction of type I IFN, resulting in CD8 T-cell–mediated control of residual disease. Although the therapy is effective locally, it can also eliminate locoregional recurrence and impact, but not cure distant disease. Furthermore, we developed a tumor explant model to rapidly screen responses to STING ligands in tumors with divergent immune environments in both mice and in patients, and show that the implanted biomaterial can be personalized according to the explant response to improve outcomes following therapy.

This approach has specific relevance to HNSCC because close or positive resection margins are strongly associated with local tumor recurrence (26), and local recurrence occurs in approximately 30% of HNSCC patients receiving resection with curative intent. In addition, tumor encroachment on or proximal to critical structures, such as the carotid artery, limits the feasibility of a complete and curative resection in some patients. This is a clear area in which immunotherapy can be incorporated to build upon surgical procedures to minimize local recurrence, while potentially generating immune-mediated control of distant metastasis. In our model, therapy had a restricted locoregional action and did not affect distant metastases that were present before treatment started. However, because the final control of local tumors occurs via CD8 T cells, it would be reasonable to combine this therapy with T-cell–targeted therapies such as checkpoint inhibitors to improve distant control, as has proven effective in distant control of immunogenic tumors treated with direct injection of STING ligands (9). In addition, we have previously shown that STING ligands synergize with radiotherapy to control local and distant tumors (7), which is commonly prescribed for HNSCC patients following surgery. Therefore, it is likely that the addition of STING ligands to the surgical site will synergize well with current adjuvant therapies to improve control of both local and distant tumor recurrence.

The biomaterial used in this study was Matrigel, which is not optimized for sustained release, but instead undergoes the default rapid diffusion-based “burst” release of incorporated drug, followed by further low-level release on gradual degradation of the polymer scaffold. In this setting, the primary purpose of the Matrigel is to localize the STING ligands to the resection cavity and prevent leakage before wound closure. An array of alternative biomaterials can be considered for use in patients. Recent studies demonstrated that an alginate scaffold incorporating anti-CD3, anti-CD28, and anti-CD137 microparticles along with NKG2D CAR-engineered T cells placed alongside pancreatic tumors resulted in optimal tumor control when loaded with a mesoporous silica microparticle containing STING agonists (27). These data indicate how biomaterials can be further modified to include additional agents, and potentially personalized. There are a wide variety of suitable biomaterial options, and further refinement could be used to develop formulations with controlled release properties. For example, a sustained release formulation of STING ligands significantly improved treatment of mice when injected into the same area where cancer cells had been injected 3 days previously (28). Since completing these studies, a recent publication has reported on a very similar model, using STING ligands incorporated into an extended release hydrogel in a perioperative setting (29). Distinct from our model, local administration of innate adjuvants was able to control the progression of lung metastases of 4T1-luciferase expressing luciferase (29), suggesting that either there are differences in the complete resection model used in this study or the extended release formulation containing a 4 fold higher dose of STING ligands is critical for efficacy. Despite these data, cyclic dinucleotides may not be appropriate for longer-term release strategies, as they are subject to rapid in vivo degradation by ubiquitous enzymes. Incorporation of small-molecule STING agonists (30, 31) or alternative innate adjuvants (32) could further alter the properties of such biomaterials for improved tumor control.

HNSCCs are known to be immunosuppressive. In this study, we identified IL10 production to be a prominent component of the response to STING ligands in the MOC2 tumor. Although we do not believe that IL10 is the sole immunosuppressive factor operational in this tumor, IL10 production can be indicative of a distinct differentiation profile of tumor-infiltrating macrophages (33–35). STING ligands cause IL10 production in MOC2 tumors, and this likely has a direct impact on the immune cells present in that tumor's microenvironment. IL10 can suppress inflammation in a number of cell types, including antigen-presenting cells, which respond to IL10 by downregulating antigen processing and presentation (36, 37), macrophages, which increase alternative macrophage differentiation (38), and T cells, where costimulation is suppressed (reviewed in ref. 16). Because IL10 shortens the duration of inflammatory responses (39) and suppresses antitumor adaptive immune responses during tumor progression (40), it is plausible that IL10 diminishes many of the positive benefits of STINGblade. Our findings are consistent with previous studies demonstrating that blocking IL10 responses results in more effective immune-mediated control of tumors (41–44). Thus, although IL10 may not be the only suppressive factor expressed in MOC2 tumors, it is a strong target to redirect the inflammatory response to STINGblade.

In summary, local failure is a critical problem in HPV⁻ HNSCC and other immunosuppressive solid tumors. The local delivery of STING agonists, with or without systemic immunotherapy, may enhance the efficacy of surgical resection by minimizing local recurrence, and further serving as a platform to generate systemic immunity to treat or control metastatic disease. We believe that the tumor explant assay is a rapid method that can be used ex vivo to predict the response to treatment with STING ligands in vivo based on the unique environment of patient tumors.

M.J. Gough reports receiving commercial research grant from Bristol Myers Squibb. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.R. Baird, T.C. Blair, M.R. Crittenden, M.J. Gough

Development of methodology: J.R. Baird, D. Friedman, Z. Sun, T.D. de Gruijl, R. van de Ven, R.S. Leidner, M.R. Crittenden

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.R. Baird, V. Troesch, D. Friedman, S. Bambina, G. Kramer, T. Medler, Y. Wu, R.S. Leidner

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.R. Baird, V. Troesch, D. Friedman, G. Kramer, Y. Wu, M.R. Crittenden, M.J. Gough

Writing, review, and/or revision of the manuscript: J.R. Baird, R.B. Bell, T.C. Blair, Y. Wu, T.D. de Gruijl, R.S. Leidner, M.R. Crittenden, M.J. Gough

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.R. Baird, Y. Wu, M.J. Gough

Study supervision: Y. Wu, M.R. Crittenden, M.J. Gough

This work was supported by a postdoctoral fellowship from the American Cancer Society (J.R. Baird) and a research grant from the Oral Maxillofacial Surgery Foundation (M.J. Gough, R.B. Bell, and J.R. Baird).

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.