Established T Cell–Inflamed Tumors Rejected after Adaptive Resistance Was Reversed by Combination STING Activation and PD-1 Pathway Blockade

The stimulator of IFN genes (STING) receptor plays a critical role in Toll-like receptor–independent detection of pathogen-associated molecular patterns and type 1 IFN production in the induction of innate immunity (1, 2). Detection of cytosolic DNA through cGAMP synthase (cGAS) production of a noncanonical cyclic dinucleotide (CDN) and subsequent STING activation explains how innate immunity is initiated after tumor DNA is released into the “sterile” tumor microenvironment (3–5). Synthetic CDNs, designed to have improved in vivo stability and binding activity to both human and mouse STING, have been shown to have antitumor activity as adjuvants or monotherapies in a number of preclinical models (5, 6).

Many head and neck squamous cell carcinomas (HNSCC) have high genetic alteration rates (7) and evidence of a T cell–inflamed phenotype (8, 9) based on histologic and RNA-sequencing analysis (10). Blockade of programmed cell death (PD) receptor interactions with its ligands can induce durable antitumor responses, but only about 20% of head and neck cancer patients respond (11). However, upper aerodigestive tract cancers are often easily accessible for local injection, making them ideal candidates for treatment with intratumoral therapeutics. A phase I clinical trial investigating the safety of locally injected, single-agent CDNs in the treatment of patients with recurrent HNSCC is underway. The potential of STING activation in combination with PD-based checkpoint inhibition in suitable syngeneic models of HNSCC has not been established.

Here, we demonstrate that local injection of the synthetic CDN R_P,R_P dithio c-di-GMP (R,R-CDG) potently induces either primary tumor growth control or rejection of established T cell–inflamed MOC1 tumors. Conversely, non–T cell–inflamed MOC2 tumors lack such responses. Within T cell–inflamed MOC1 tumors, R,R-CDG treatment led to induction of innate and antigen-specific adaptive immunity that was initially dependent upon STING and durably dependent upon CD8 T lymphocytes. Adaptive immune resistance (12) mediated by enhanced expression of tumor microenvironment PD-1 and its ligand PD-L1 after R,R-CDG treatment was reversed with PD-L1 mAb, further enhancing the rate of rejection of established MOC1 tumors and inducing potent systemic antitumor immunity that led to abscopal control of distant tumors. Although others have demonstrated that adaptive immune resistance and tumor rejection can be reversed by checkpoint inhibition in tumors treated with STING agonist–formulated cancer vaccines (6), this study demonstrates such effects with synthetic CDN alone. These findings indicate that adaptive immune resistance following CDN alone can be reversed with PD-1 pathway blockade, with resulting enhanced local and systemic antitumor immunity.

Murine oral cancer (MOC) 1 and 2 cells were provided by Dr. R. Uppaluri (Washington University School of Medicine) in 2014 (13), maintained in culture as previously described for no more than 20 passages (14), authenticated by exome sequencing by others (15) and by cytokeratin immunofluorescence by our laboratory, and tested monthly for mycoplasma. For in vitro experiments, cells were harvested with TrypLE Select (ThermoFisher) at 80% confluence, and viability was determined via propidium iodide exclusion. MOC cells were irradiated to 100 Gy with a ¹³⁷Cs source irradiator (Gammacell-1000). Irradiated MOC1 or MOC2 cells were injected subcutaneously (s.c.) into the flanks of 8-week female wild-type C57BL/6 (WT B6; Charles River Laboratories) mice suspended in 30% complete Freund's Adjuvant (%vol). Four weeks after vaccination, naïve and vaccinated mice were challenged with untreated MOC1 (2 × 10⁶) or MOC2 (1 × 10⁵) cells and were evaluated for tumor engraftment. All experiments were approved by the NIDCD Animal Care and Use Committee (ASP #1364-14).

MOC1 and MOC2 cells were plated, allowed to reach 75% confluence, and treated with either R,R-CDG (Aduro Pharmaceuticals, synthesis described by Baird and colleagues, ref. 16), recombinant murine IFNβ (Biolegend), or control (1× PBS).

MOC1 (2 × 10⁶) or MOC2 (1 × 10⁵) cells were injected s.c. into the flanks of WT B6 mice. Bilateral tumors were implanted for abscopal experiments. All tumors were engrafted to at least 0.1 cm³ (6–7 mm diameter) before treatment. Mice were weighed, and tumors were measured three times weekly. Intratumoral injections of R,R-CDG in HBSS (100 μL volume) were directed at the deep aspect of the tumor. Control mice received HBSS alone. PD-L1 mAb (clone 10F.9G2, rat IgG2b; BioXCell) or rat IgG isotype control antibody treatments were performed via intraperitoneal (i.p.) injection (200 μg/injection). In some experiments, C57BL/6J-Tmem173^gt/J STING-deficient mice (The Jackson Laboratory) were used. In other experiments, cellular or cytokine depletions were performed via i.p. injection of CD8 (clone YTS169.4), IFNγ (clone XMG1.2), or TNFα (clone XT3.11) mAbs (200 μg/injection; BioXCell).

Viability of MOC cells was measured via XTT (Trevigen) assay per the manufacturer's protocol.

Tumor homogenates were generated from snap-frozen tissue using the Tissue Lyser II per manufacturer's protocol. RNA was extracted from MOC cells or tissue homogenates using the RNEasy Mini Kit (Qiagen) or the PicoPure RNA Isolation Kit (ThermoFisher) following the manufacturer's protocol. cDNA was synthesized using high-capacity reverse transcription. Gene expression was determined relative to GAPDH at baseline and following treatments using single tube primers (Life Technologies) on a Vii7 qPCR analyzer (Applied Biosystems).

MOC cells were harvested, and nonspecific binding was blocked with anti-CD16/32 antibodies (Biolegend) prior to staining. Tissues were prepared into single-cell suspensions as previously described (14), followed by anti-CD16/32 antibody staining. Cell surface staining was performed using flourophore conjugated anti-mouse CD45.2 clone 104, CD3 clone 145-2C11, CD4 clone GK1.5, CD8 clone 53-6.7, NK1.1 clone PK136, CD31 clone 390, PD-L1 clone 10F.9G2, H2-K^b clone AF6-88.5, CD107a clone 1D4B, CD69 clone H1.2F3, PD-1 clone RMP1-30, CD11b clone M1/70, CD11c clone N418, CD80 clone 16-10A1, and CD86 clone GL-1 antibodies from Biolegend, 41BB clone 17B5 and OX40 clone OX-86 from eBioscience, and CD8 clone KT15 from MBL. H2Kb:KSPWFTTL (p15E_604–611) tetramer was purchased from MBL. Dead cells were excluded via 7AAD negativity. Isotype control antibodies and a “fluorescence minus one” method of antibody combination were used for specific staining validation. Data were acquired on a FACSCanto using FACSDiva software (BD Biosciences) and analyzed on FlowJo software vX10.0.7r2.

Single-cell suspensions were generated from spleens and lymph nodes from WT B6 mice. Cells were stained and sorted on an autoMACS magnetic sorter (Miltenyi Biotec) using CD19-positive selection (B lymphocyte), pan–T cell–negative selection (T lymphocyte), NK cell–negative selection (NK cells), pan–DC-positive selection (DC cells), and CD11b-positive selection (myeloid cell) kits from Miltenyi. Purity of each cell population was >90% as assessed by flow cytometry.

Tumor tissue was embedded in optimum cutting temperature media. Sections (6 μm) were fixed with methanol, washed, and nonspecific binding was blocked with a solution of 3% BSA and 0.1% Tween 20 in 1×PBS. AF594-conjugated PD-L1 (clone MIH6; Abcam) or isotype control antibody was incubated overnight at 4°C in a humidified chamber. Following washes, slides were counterstained with DAPI and mounted for analysis on an LSM 780 confocal microscope (Zeiss). Photomicrographs were analyzed using Zen 2012 SP1 software.

Splenocytes were harvested from naïve B6 mice and labeled with 2.5 or 0.25 μmol/L CFSE. High CFSE-labeled cells were pulsed with 10 μg/mL of p15E_604–611 (KSPWFTTL) or control OVA_257–264 (SIINFEKL) peptide for 1 hour. Cells were washed, mixed, and adoptively transferred into naïve, control, or treated MOC tumor–bearing mice. Four hours later, splenocytes were harvested, and flow cytometry was used to determine the ratio of CSFE-high to CSFE-low cells. Antigen-specific cell killing was determined as: 1 − (r_naïve/r_vaccinated) × 100, where r = (% CFSE-low cells)/(% CFSE-high cells).

Tests of significance between pairs of data are reported as P values, derived using a Fisher exact test or Student t test with a two-tailed distribution and calculated at 95% confidence. Comparison of multiple sets of data was achieved with one- or two-way analysis of variance (ANOVA). Kaplan–Meier curves were compared using the log-rank/Mantel–Cox test. Error bars reflect standard error of measurement (SEM). Analysis was performed using GraphPad Prism v6.

Patients with HNSCCs demonstrate significant heterogeneity in tumor T cell–inflamed status. Previous work has demonstrated the T cell–inflamed status of MOC1 tumors and non–T cell–inflamed status of MOC2 tumors (17, 18). We performed whole tumor cell vaccination experiments to explore the ability of MOC1 and MOC2 cells to induce immunologic memory and to further validate their use in modeling both T cell–inflamed and non–T cell–inflamed HNSCCs. WT B6 mice vaccinated with irradiated MOC1 efficiently resisted engraftment when challenged with MOC1, but not MOC2, cells (Fig. 1A). This suggests that there are tumor rejection antigens present on MOC1 cells capable of inducing immunologic memory. Either no antigens are shared between MOC1 and MOC2 or MOC2 cells possess an intrinsic insensitivity to T cell–mediated cell killing. In support of this, MOC2 tumors readily engrafted following MOC2 cell challenge after vaccination with irradiated MOC2 cells (Fig. 1B). Thus, the ability of MOC1 and MOC2 cells to induce immunologic memory is consistent with their T cell–inflamed status and suggests that these two cancer cell lines can serve as models of human HNSCC with variable T cell–inflamed status.

We treated established (at least 0.1 cm³, or 6–7 mm diameter) MOC1 and MOC2 tumors, growing in WT B6 mice, with three intratumoral injections of R,R-CDG. This treatment resulted in complete tumor rejection in 5 of 10 established MOC1 tumors and significant growth inhibition in the remaining tumors, leading to significantly prolonged survival (P < 0.001) in MOC1 tumor–bearing mice treated with CDN compared with control (Fig. 2A). Mice that rejected established MOC1 tumors were challenged with MOC1 cells, which resulted in no tumor formation, indicating the formation of immunologic memory (Fig. 2B). Similar to the vaccination experiments, R,R-CDG–treated mice that rejected established MOC1 tumors and subsequent MOC1 challenge readily formed tumors when challenged with MOC2 transplantation. Conversely, R,R-CDG treatment of established MOC2 tumors slightly delayed primary tumor growth in a subset of tumors, but did not result in significant prolongation of survival (Fig. 2C). Mice that rejected established MOC1 tumors developed physical findings consistent with an immune-mediated mechanism, including vitiligo and hair loss at the former tumor site (example shown in Fig. 2D). These data indicate that STING agonists are able to induce antitumor immunity in T cell–inflamed but not non–T cell–inflamed tumors.

We next assessed whether R,R-CDG is directly cytotoxic to MOC cells. Results from XTT viability assays suggest that the observed in vivo effect is not due to direct cytotoxicity as exposure of MOC cells to R,R-CDG at doses up to 20 μmol/L induced no reduction in cell viability (Fig. 3A). Given that STING is known to serve as the cognate CDN receptor (5), we determined STING expression levels in MOC and sorted immune cell subsets. All immune cell subsets expressed STING at significantly higher levels than either MOC1 or MOC2, with myeloid cell subsets (total CD11b⁺ myeloid and CD11c⁺ dendritic cells [DC]) demonstrating the highest relative STING expression (Fig. 3B). Functionally, STING expression correlated with type I interferon production upon exposure to R,R-CDG (Fig. 3C), with dose-dependent induction of IFNβ expression in immune subsets but not MOC cells. IFNβ, but not R,R-CDG, was able to enhance MOC cell expression of genes encoding for antigen processing machinery components (Fig. 3D), as well as cell surface expression of calreticulin, MHC class I and II, and PD-L1 (Fig. 3E). Taken together, these data demonstrate that R,R-CDG exposure leads to type I IFN production by immune cells with subsequent enhancement of MOC cell immunogenicity. These data also suggest that the lack of MOC2 response to CDN is not due to an intrinsic insensitivity to interferon.

We analyzed multiple immune correlates in control and R,R-CDG–treated MOC1 tumors and tumor-draining lymph nodes (DLN). Using flow cytometry, we observed enhanced MOC1 tumor and infiltrating immune cell–specific PD-L1 and tumor cell MHC class I expression (H2-K^b) following treatment (Fig. 4A, validated with immunofluorescence in Fig. 4B). CD8⁺ DCs are critical for cross-presentation of tumor antigen to activate adaptive immunity (19, 20). Figure 4C demonstrates enhanced accumulation of total DLN DCs and accumulation and activation (CD80/86) of CD8⁺ DLN DCs following treatment. Quantification of tumor-infiltrating lymphocytes (TIL) revealed substantially enhanced accumulation and activation of both total (PD-1, CD69, 41BB exhaustion/activation markers) and p15E_604–611 antigen–specific (CD107a activation marker) CD8 TIL into MOC1 tumors following treatment (Fig. 4D). Accumulation and activation (PD-1, OX40 exhaustion/activation markers) of CD4 TIL was also increased after treatment (Supplementary Fig. S1A). Tumor NK cell infiltration was not increased, but activation of infiltrating NK cells was modestly enhanced after treatment (Supplementary Fig. S1B). Representative flow cytometry dot plots for select cellular subsets are shown in Supplementary Fig. S2. Concentrations of IFNα and IFNβ and effector cytokines IFNγ and TNFα from whole tumor or DLN tissue homogenates were determined (Fig. 4E and F). CDN-treated MOC1 tumors had substantially increased IFNβ, the DLNs had modestly increased IFNβ, and effector cytokines were significantly elevated in both. Conversely, treatment of unresponsive MOC2 tumors with R,R-CDG appeared to increase tumor IFNβ and the accumulation of DLN DCs, but not TIL or PD-L1 expression within the tumor microenvironment (Supplementary Fig. S3). Cumulatively, these data provide correlative support for R,R-CDG–induced activation of innate immunity with subsequent development of antigen-specific adaptive immunity within MOC1, but not MOC2, tumors.

We treated established MOC1 tumors in STING-deficient B6 mice in parallel with WT B6 mice (Fig. 5A). Not only did MOC1 primary tumors grow more rapidly in STING-deficient mice, but they also were completely unresponsive to intratumoral R,R-CDG treatments. We next performed antibody-based cellular and cytokine depletions in MOC1 tumor–bearing WT B6 mice. Depletion of CD8 and IFNγ, but not TNFα, abrogated R,R-CDG–induced MOC1 tumor rejection and growth control. Mice with MOC1 tumors and with CD8 or IFNγ depletion initially responded to R,R-CDG, but durable responses were lost: tumor growth rebounded approximately 1 week after completion of treatment. Flow-cytometric analysis of MOC1 tumor tissue from STING-deficient mice revealed near-complete loss of CD8 and CD4 TIL accumulation that was not enhanced with R,R-CDG treatment (Fig. 5B). Analysis of STING-deficient tumor tissues revealed abrogation of R,R-CDG–induced MOC1 tumor cell surface PD-L1 and MHC class I expression and loss of accumulation of type I IFN and effector cytokines in the tumor microenvironment after treatment (Fig. 5C). Baseline tumor IFNγ expression was substantially reduced in STING-deficient mice compared with WT B6 mice. Mechanistically, these data validate the role of STING signaling in the baseline induction of MOC1 antitumor immunity and provide strong evidence that R,R-CDG induces production of STING-dependent type I IFN within the MOC1 tumor microenvironment that subsequently leads to the development of CD8⁺ T cell–dependent adaptive antitumor immunity.

Given the enhanced tumor and immune cell expression of PD-L1 and PD-1 expression on CD8⁺ and CD4⁺ TILs in MOC1 tumors in vivo after R,R-CDG treatment, we hypothesized that combination R,R-CDG and PD-L1 mAb treatment would further enhance rejection rates of MOC1 tumors. We treated established MOC1 tumors with R,R-CDG and PD-L1 mAb, alone or in combination, and measured primary tumor growth and time to 750 mm³ as a statistical readout (Fig. 6A). Treatment with PD-L1 mAb alone delayed primary tumor growth, but tumors eventually rebounded and no tumors were rejected, consistent with previous experiments (ref. 17; Fig. 6B). Following R,R-CDG treatment alone, 5 of 10 tumors were rejected and 4 of 5 remaining primary tumors grew at a reduced rate. However, the combination of R,R-CDG and PD-L1 mAb induced rejection of 90% of MOC1 tumors, with the remaining tumor displaying reduced growth. This combination antitumor effect was significantly more robust (P = 0.041, log-rank/Mantel–Cox analysis for time to 750 mm³) than R,R-CDG treatment alone (Fig. 6C). Thus, the addition of PD-L1 mAb to R,R-CDG treatment appears to reverse adaptive immune resistance and enhance rates of MOC1 tumor rejection.

To validate that R,R-CDG and anti–PD-L1 mAb treatments led to the development of systemic antitumor immunity, we utilized an in vivo CTL assay (Fig. 7A). Selective killing of adoptively transferred splenocytes presenting p15E_604–611 peptide as a model antigen was significantly increased in mice treated with the combination therapy compared with control or either therapy alone. Specificity of the in vivo CTL assay was validated with control (SIINFEKL) peptide-pulsed, adoptively, transferred splenocytes in mice vaccinated with p15E peptide (Supplementary Fig. S4). Mice that rejected MOC1 tumor following combination treatment resisted challenge with MOC1 cells (Fig. 7B). To further establish the development of systemic antigen-specific immunity, we performed abscopal experiments (Fig. 7C). Treatment of right flank tumors with R,R-CDG in the presence of systemic PD-L1 mAb, but not with R,R-CDG or PD-L1 mAb alone, induced near-complete tumor control of contralateral untreated tumors. Thus, reversal of adaptive immune resistance by adding PD-L1 mAb to R,R-CDG results in systemic antigen-specific antitumor immunity that can control distant deposits of tumor.

Previous work has established the role of STING pathway activation as an adjuvant therapy for antitumor peptide and GM-secreting whole tumor cell vaccines (6, 21, 22). Our work reported here demonstrates significant tumor control can be had through STING pathway activation as a monotherapy in a carcinogen-induced model of HNSCC, and that adaptive immune resistance can be reversed, with complete tumor rejection and abscopal control of distant tumors, by treatment with the combination of nonformulated R,R-CDG and checkpoint inhibition. Significant primary tumor growth control and consistent tumor rejection rates of about 50% were achieved with the synthetic noncanonical CDN R,R-CDG, shown to be more potent than natural CDN compounds or other synthetic formulations (5, 6). We demonstrated evidence of adaptive immune resistance (12) following R,R-CDG treatment with increased tumor and immune cell PD-L1, and TIL PD-1, expression. Combining STING pathway activation with PD-L1 mAb led to consistent rejection of 90% of established MOC1 tumors, with the development of more robust antigen-specific immunologic memory and abscopal control of distant, untreated tumors. Enhanced antitumor immunity with vaccine-formulated or other CDNs and checkpoint inhibitors has also been demonstrated in models of melanoma and colon cancer (6, 23). Mechanistically, we demonstrated early STING and late CD8-dependent tumor control, with production of type I IFN appearing to come primarily from infiltrating host cells and not tumor cells, consistent with other studies (4, 16, 24, 25).

Our use of both T cell–inflamed (MOC1) and non–T cell-inflamed tumors (MOC2) allows us to model both immunogenic and poorly immunogenic HNSCCs. Rejection of a subset of T cell–inflamed MOC1 tumors with R,R-CDG treatment alone, which are relatively insensitive to PD-based checkpoint inhibition alone, suggests that lack of innate immune activation may limit responses achievable with PD blockade in some tumors. Lack of response to R,R-CDG in non–T cell–inflamed MOC2 tumors suggests that STING-dependent induction of antitumor immunity is dependent upon the baseline T cell–inflamed status of the tumor and likely the presence of tumor rejection antigen(s). Treatment of MOC2 tumors with R,R-CDG–induced type I IFN within the tumor microenvironment and increased DCs within the DLN, but not accumulation of CD8⁺ TILs, suggesting that the lack of tumor control observed with this model was due to an intrinsic insensitivity of MOC2 to immune control and not due to alteration of innate immune cells within the MOC2 tumor microenvironment leading to insensitivity to CDN. Poor antigenicity of MOC2 or an intrinsic insensitivity of MOC2 to T-lymphocyte killing could explain these findings. Lemos and colleagues have shown the promotion of Lewis lung carcinoma tumor growth that was secondary to IDO production following STING activation (26). These results were reversed with the introduction of a tumor rejection antigen into the cancer cells, providing evidence that STING activation within the tumor microenvironment can have dramatically different results based on tumor cell antigenicity. Previous work by others has established that chronic type I IFN signaling through STING can promote carcinogen-induced tumorigenesis (27) and lead to induction of T-lymphocyte tolerance through IDO production (28, 29). How STING signaling, in the presence or absence of tumor rejection antigen, serves as a switch between protective type I IFN-dependent immunity and tumor-promoting inflammation needs clarification and has significant implications for the clinical translation of STING agonists.

Antigenic heterogeneity of tumor rejection antigens may contribute to variability in responsiveness to STING activation. Because we have not identified all tumor rejection antigens present in MOC1, our use of p15E as a model antigen makes our understanding of antigen spread and TIL response to neoantigens incomplete. However, the addition of PD-L1 mAb largely rescued MOC1 tumors from adaptive immune resistance following CDN treatment, which points to the importance of adaptive immunity in this treatment approach. Additionally, our analysis of immune correlative alterations following R-R-CDG treatment was incomplete. Although we demonstrated increased PD-L1 expression on both tumor and infiltrating immune cells within the MOC1 tumor microenvironment, we did not mechanistically explore which immune cells dominantly express PD-L1 for ligation with PD-1 on CD8⁺ TILs. It is possible that certain immune cell subsets, such as myeloid-derived suppressor cells or myeloid DCs, dominantly contribute PD-L1 at baseline or in response to IFN (30, 31). Further, early responses to CDN may be mediated through vascular disruption and tumor necrosis (16, 32), and these parameters were not explored in our studies. Our finding that early growth control in MOC1 tumors following R,R-CDG treatment was abrogated in STING-deficient mice, but minimally altered by CD8⁺ T cells or IFNγ, suggests mechanisms independent of adaptive immunity but still dependent on host cells, which may include myeloid or NK cell activity (16, 33). This early necrotic response following STING activation may be required for efficient induction of antitumor immunity, as proapoptotic caspases suppress type I IFN production following STING detection of cytosolic DNA (34). Further exploration of these early responses will be important to understand how STING activation can be optimally utilized and possibly sequenced with other anticancer treatments to induce antitumor immunity.

In summary, we demonstrated robust induction of innate and adaptive antitumor immunity following STING activation with a synthetic CDN compound, which led to rejection of a subset of established T cell–inflamed MOC1 tumors. This finding could not be reproduced in non–T cell–inflamed MOC2 tumors. Antitumor immunity was dependent upon a functional STING receptor and durable tumor control was dependent upon CD8⁺ T lymphocytes. Reversal of adaptive immune resistance with combination STING activation and PD-L1 mAb led to rejection of 90% of established MOC1 tumors and control of distant, untreated tumors, indicating the development of systemic, antigen-specific antitumor immunity. The clinical safety of locally injected STING agonists is being tested in phase I studies currently under way. This preclinical data may serve as a rationale for combining CDN with checkpoint inhibition to reverse adaptive immune resistance.

Y.J. Kim reports receiving a commercial research grant from Aduro Biotech. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E. Moore, H. Cash, C.V. Waes, Y. Kim, C. Allen

Development of methodology: E. Moore, P.E. Clavijo, R. Davis, H. Cash, Y. Kim, C. Allen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Moore, R. Davis, H. Cash, C. Allen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Moore, P.E. Clavijo, R. Davis, H. Cash, Y. Kim, C. Allen

Writing, review, and/or revision of the manuscript: E. Moore, P.E. Clavijo, R. Davis, H. Cash, C.V. Waes, Y. Kim, C. Allen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Moore, H. Cash, C. Allen

Study supervision: H. Cash, C. Allen

The authors thank James Hodge and Ravindra Uppaluri for their critical review of this manuscript.

This work was supported by the Intramural Research Program of the NIH, NIDCD, project number ZIA-DC000087. R. Davis was supported through the NIH Medical Research Scholars Program, a public–private partnership supported jointly by the NIH and generous contributions to the Foundation for the NIH from Pfizer Inc, The Doris Duke Charitable Foundation, The Newport Foundation, The American Association for Dental Research, The Howard Hughes Medical Institute, and the Colgate-Palmolive Company, as well as other private donors. C. Allen received further support through the American Academy of Otolaryngology/American Head and Neck Society Duane Sewell Young Investigators Award.

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.