Targeting of CD40 and PD-L1 Pathways Inhibits Progression of Oral Premalignant Lesions in a Carcinogen-induced Model of Oral Squamous Cell Carcinoma

An estimate of 354,000 individuals worldwide are diagnosed with cancers of the oral cavity and lip each year (1) Tobacco, alcohol, and betel nut exposure are well-established risk factors for oral squamous cell carcinoma (OSCC; ref. 2). Interruption of carcinogen use is key intervention to reduce oral cancer incidence. Nonetheless, OSCC risk remains elevated even after alcohol and tobacco cessation and may take up to 20 years to reach baseline levels (3). Thus, there is a need for novel approaches to interrupt or reverse the process of oncogenesis after initiation, but before full malignant transformation.

In patients with oral leukoplakia/erythroplakia, multiple, clonally unrelated premalignant cells have been found in the mucosa as a result of carcinogen-induced field cancerization, oftentimes clinically silent, and distant from an index oral premalignant lesion (OPL; refs. 4, 5). Although surgery is often recommended to treat OPLs (especially with high-grade dysplasia), its impact on long-term oral cancer development remains to be determined (6, 7). Systemic pharmacologic prevention, on the other hand, has been suggested as a strategy to reduce oral cancer incidence by addressing multiple premalignant clones distributed throughout the entire mucosa (8). Despite extensive previous clinical investigations with celecoxib (9) and erlotinib (10), among other candidates, no standard systemic treatment has been developed for this condition.

Immunotherapy with anti-PD-1 antibody as a single agent or combined with chemotherapy improves overall survival in patients with recurrent/metastatic head and neck squamous cell carcinomas (11–13). Dual checkpoint inhibition with PD-1/PD-L1 and CTLA-4 antagonistic antibodies prolongs survival in non–small cell lung cancer (14) and other solid tumors, and is under investigation for head and neck cancers. Given the activity of these agents in advanced disease, modulation of immune checkpoint pathways might prove beneficial as a prevention strategy in patients with OPLs. Immunosuppressive pathways commonly upregulated in OSCC have been shown to be present at the OPL stage, and we have demonstrated that PD-L1 overexpression in patients with OPLs is associated with increased cancer risk (15). Immunoprevention has the advantage of targeting lesions regardless of the dominant dysregulated molecular pathways driving carcinogenesis, thus potentially addressing the multiplicity and genetic heterogeneity of OPLs (16). The evaluation of immune-targeting strategies with diverse mechanisms of action on the prevention of OSCCs in animal models could shed light on mechanisms of OPL immune evasion, inform prioritization of potential targets, and help design human clinical trials for disease prevention. We previously demonstrated that PD-1 blockade decreased the incidence of dysplastic lesions and prevented progression to OSCCs in a carcinogen-induced experimental model using heterozygous TP53-knockout mice (17). These data further support a role for PD-1/PD-L1 axis in the process of malignant transformation in a TP53-mutant background. The effects of targeting other components of the immune microenvironment, as well as the impact of PD-1 inhibition on cancer prevention and immune modulation in TP53 wild-type (WT) mice, remain to be determined.

PD-1 is engaged as a relatively late step of the cancer immunity cycle (18), after antigen presentation, T-cell priming and activation, lymphocyte tumor homing, and cancer cell recognition. OPL immune evasion mechanisms likely include these earlier steps, the targeting of which may represent opportunities for cancer prevention. The CD40/CD40L and OX40/OX40L pathways have been shown to be key regulators of early cancer immunity, as they modulate neoantigen cross-presentation (19) and antigen-primed T-cell survival (20, 21). Agonist targeting the OX40 and CD40 pathways are currently in clinical development for treatment of multiple advanced cancer types. In particular, CD40L binding to CD40 triggers a broad immune response, involving activation and maturation of antigen-presenting cells, cytokine production leading to IFNγ secretion by natural killer cells, cytotoxic T lymphocyte generation mediated by Th cells (CD4⁺), and differentiation of activated B cells in memory B cells (22). CD40 activation has also been found to induce polarization of macrophages toward the proinflammatory M1 phenotype, thus enhancing antitumor activity (23). Previously, we observed that in syngeneic mouse models of invasive oral cancers, OX40 or CD40 agonist monotherapy was at least as effective as anti-PD-1 in reducing flank tumor size and increasing animal survival (24). Their effects on oral cancer prevention have yet to be determined.

We performed this study to further characterize the immunoprevention effects of PD-1, CTLA-4, or dual PD-1/CTLA-4 checkpoint pathway blockade in a TP53 WT, carcinogen-induced mouse model, and to test the primary hypothesis that prevention of OSCC can be achieved by activation of either the CD40 or OX40 pathways. We utilized a 4-nitroquinoline 1-oxide (4-NQO) murine model of oral carcinogenesis, which recapitulates, to some extent, molecular and immune features of human oral cancers (25, 26). CD40 agonist (but not OX40 agonist) antibodies reduced the incidence of OSCC. Blockade of the PD-1/PD-L1 pathway, with or without CTLA-4 blockade, also delayed progression of OPLs to OSCC. Furthermore, we observed a distinct pattern of immune surveillance modulation by CD40 agonist [which generates a lasting expansion of experienced/memory cytotoxic CD8⁺ T lymphocytes (CTL) and proinflammatory M1 macrophages] versus PD-1/CTLA-4 blockade [which largely depletes T regulatory lymphocytes (Treg)]. Finally, during the progression of OPL to OSCC, we noted a significant upregulation of CD40 mRNA using paired clinical samples. These results support clinical development of interventions directed at a variety of immune pathways to prevent OSCCs in patients with OPLs.

All animal studies were carried out according to The University of Texas MD Anderson Cancer Center (MDACC, Houston, TX) Institutional Animal Care and Use Committee–approved protocols. Female C57BL/6J (JAX 000664) mice were purchased from The Jackson Laboratory. PD-L1–knockout (PD-L1^ko) mice were provided by D.L. Gibbons and have been described previously (27). As published previously (28), mice were exposed to 4-NQO via drinking water, at a concentration of 100 μg/mL, at 12 weeks of age and continuing for 8 weeks, after which regular drinking water was resumed. For the time-course experiments characterizing malignant progression, mice were euthanized the same day 4-NQO exposure was terminated, and every 4 weeks thereafter. For treatment cohorts, mice were randomized (n = 10 mice/group) to receive vehicle or immunotherapy once every 3 days for three doses, starting 8 weeks after 4-NQO exposure was terminated. The following antibodies (BioXCell) were injected intraperitoneally: anti-PD-1 (RPM1-14; 250 μg/dose; RRID: AB_10949053), anti-CTLA-4 (9H10; 100 μg/dose; RRID: AB_10950184), an OX40 agonist mAb (OX86; 100 μg/dose; RRID: AB_1107592), a CD40 agonist mAb (FGK4.5; 100 μg/dose; RRID: AB_1107647), and isotype control IgG2a (2A3; 100 μg/dose; RRID: AB_1107769). The combination of anti-PD-1 plus anti-CTLA-4 was given at the same time and at the same doses as single-therapy groups. All mice were euthanized 16 weeks after 4-NQO exposure was terminated. Immediately after sacrifice, tongues were excised and draining cervical lymph nodes (DCLN) were collected for flow cytometry analysis, including two superficial cervical lymph nodes lying at the top poles of the submandibular gland from each side, and, when feasible, one deep cervical lymph node from each side. The total number of lymph nodes collected ranged from five to six per animal.

PD-L1^ko mice and C57BL/6J WT mice were used to study the long-term effects of the PD-L1 pathway on immune surveillance and carcinogenesis. We exposed 25 male and 25 female mice of each strain to 4-NQO for 8 weeks. After 4-NQO exposure was terminated, 10 WT mice and 10 PD-L1^ko mice were sacrificed every 4 weeks, except for the final group at week 16 that included 20 WT and PD-L1^ko mice.

Formalin-fixed tongues were sagittally bisected and embedded in paraffin. Fifty 5-μm-thick sections from each specimen were cut, and the 1st, 10th, 20th, 30th, 40th, and 50th slides were stained with hematoxylin and eosin (H&E) for histopathologic analysis. H&E-stained sections were examined by two pathologists who were blinded to treatment groups, and lesions were classified into OPL and OSCC using an established criteria (Supplementary Table S1). OSCC area quantification was calculated using Aperio ImageScope Software (Leica Biosystems). The total tumor area in each tongue was calculated by adding the tumor area from all the tumors found in the 1st, 10th, 20th, 30th, 40th, and 50th slides. The number of tumors was calculated by counting the microscopic tumors that were not spatially contiguous.

DCLNs of the PD-L1^ko mice and one of three treatment cohorts were homogenized with a 70-μm strainer and depleted of red blood cells. Samples were fixed using the Foxp3/Transcription Factor Staining Buffer Set (eBioscience) and stained with up to 12 antibodies (Supplementary Table S2). Flow data were collected on a four-laser, CytoFLEX flow cytometer and analyzed using CyTexpert Software (Beckman Coulter).

Ten patients enrolled in the Erlotinib Prevention of Oral Cancer study (10) at MDACC (Houston, TX) who had OPLs and subsequently developed cancer were identified. The paired index OPLs and the subsequent invasive oral cancer tissue specimens were retrieved from our biobank for biomarker analysis. The interval between the index OPL biopsy and oral cancer surgery ranged from 1 to 3 years. RNA was obtained from formalin-fixed, paraffin-embedded tissues and analyzed using the HTG EdgeSeq Oncology Biomarker Panel (HTG Molecular Diagnostics).

Data analysis was performed with GraphPad prism (RRID:SCR_002798). In all cohorts, incidence of OSCC between control and treatment groups was calculated using Fisher exact test. Mann–Whitney test was used to compare total tumor area. t Test was used to compare the number of tumors per tongue and flow cytometry statistical significance between control and treatment groups in the second cohort (IgG vs. PD-1) and the PD-L1^ko cohort. Proportion of OSCC between WT and PD-L1^ko mice through weeks 4, 8, 12, and 16 was performed by a mixed model analysis. In the remaining treatment cohorts, due to the multiplicity of treatment groups, one-way ANOVA with Dunnett multiple comparisons test was used to analyze the number of tumors per tongue and flow cytometry statistical significance between control and treatment groups. Paired t test was used to analyze mRNA expression. A P value of 0.05 or less was considered statistically significant.

We first performed a time-course study to characterize the development of hyperplasia, dysplasia, and OSCC after 4-NQO treatment in C57BL/6J mice. Prior studies have characterized the development of OSCC in other mouse strains (29), which may differ in their feeding and drinking behavior (30), consequently altering their exposure to the carcinogen. Immediately after cessation of carcinogen exposure and every 4 weeks thereafter (Fig. 1A), mice were sacrificed, and tongues were excised, processed into six H&E slides, and epithelium was classified as normal, hyperplasia, dysplasia, or OSCC (Fig. 1B). Among mice sacrificed at the time of 4-NQO discontinuation (week 0), 20% displayed hyperplasia, whereas 80% of the tongues had no histologic abnormalities (Fig. 1C). At week 4, 40% of mice had hyperplasia and 60% of mice had normal tongue histology. At week 8, 33% of tongues had dysplasia and 67% had hyperplasia; this was selected as the appropriate timepoint for initiation of immunotherapy for subsequent experiments, as all mice had OPLs, but no tumors were present. At the 16-week timepoint, 40% of tongues had OSCC and 60% had hyperplasia. Thus, we selected 16 weeks after the cessation of 4-NQO as the optimal timepoint for cancer development endpoint analysis in subsequent experiments, as a large fraction of mice had developed invasive cancer, yet without morbidity.

PD-1 checkpoint inhibitors, pembrolizumab and nivolumab, are approved for the treatment of recurrent or metastatic OSCC, and pembrolizumab is currently being evaluated for prevention of OSCC in patients with high-risk OPL in the Immune Prevention of Oral Cancer (IPOC; NCT02882282) clinical trial. Therefore, we initially characterized the impact of PD-1 blockade on the development of OSCC. Forty C57BL/6J mice were treated with 4-NQO as described above. At 8 weeks after cessation of 4-NQO, at which point all mice had developed OPLs, but not OSCC, animals were randomized to receive anti-PD-1 or IgG (Fig. 2A). Anti-PD-1 treatment significantly lowered the total OSCC area when compared with the IgG control group (P = 0.046; Fig. 2B). We observed a trend toward fewer OSCC lesions per mouse (0.85 for anti-PD-1 vs. 1.4 for IgG; P = 0.063; Fig. 2C) and fewer mice with OSCC (60% for anti-PD-1 vs. 85% for IgG; P = 0.155; Fig. 2D) in the anti-PD-1–treated animals as compared with the IgG control group. These findings support the hypothesis that anti-PD-1 treatment can at least modestly prevent malignant progression of OPLs.

We next utilized PD-L1^ko mice to characterize the impact of the PD-1/PD-L1 pathway on oral carcinogenesis. This model has the advantage of sustained PD-L1 suppression in contrast to the three-dose therapeutic model described above using anti-PD-1 antibodies, and may allow us to better ascertain the contribution of the PD-1 pathway to the natural history of OPL immune evasion and malignant transformation. PD-L1^ko mice (n = 50) and WT (n = 50) mice received 4-NQO for 8 weeks. Tongues and DCLNs were collected and analyzed for OPL progression and immune response every 4 weeks after 4-NQO exposure was terminated. During weeks 4, 8, 12, and 16, oral lesions gradually progressed from normal epithelium to OSCC, with an increased proportion of OSCC in WT mice compared with PD-L1^ko mice (P = 0.028; Fig. 3A and B). At week 16, PD-L1^ko mice showed a statistically significant reduction in the number of OSCC tumors per tongue compared with WT mice (P = 0.041; Fig. 3C), accompanied by a trend toward reduced total OSCC area (P = 0.051; Fig. 3D). These results further support that activation of the PD-1/PD-L1 pathway is one of the immune evasion mechanisms utilized by OPLs to facilitate progression to OSCC.

We next performed flow cytometry to evaluate immune cell populations in DCLNs harvested concurrently with the tongues. T-cell subpopulations were characterized as follows: CD4⁺Foxp3⁺ (Tregs), CD4⁺Foxp3⁺CD44^hi [activated/memory-like Treg (_amTreg)], CD4⁺CD44^hiPD1⁺, and CD8⁺CD44^hiPD1⁺ (memory CTLs). PD-L1^ko mice exhibited increased memory CTLs compared with WT mice at weeks 4, 8, and 12 (Fig. 3E), illustrating a role for PD-L1–mediated, CD8-centric mechanism of immune evasion in OPLs. In PD-L1^ko mice, we also observed delayed expansion of Tregs at week 4 (Fig. 3E) and _amTregs at week 4 through 8 (Fig. 3E). These results suggest an additional role for PD-L1 in promoting Treg-mediated immune tolerance in premalignancy, ultimately contributing to progression to cancer. Finally, during weeks 4, 8, 12, and 16, we found sustained expansion of a previously described (31, 32) CD4⁺CD44^hiPD1⁺ cell population (gated from CD3⁺CD4⁺Foxp3⁻) in PD-L1^ko mice (Fig. 3E). This population was the only one that remained significantly increased at every timepoint analyzed, suggesting a novel potential mechanism of PD-L1–mediated OPL immune evasion involving activated, experienced/memory CD4 effector cells. Of note, the greatest differences in both CD4/CD8 effector and CD4 regulatory populations occurred during weeks 4 and 8, and although significant differences in immune milieu were still visible in the following weeks, they were less pronounced, suggesting that PD-1/PD-L1 pathway may be more relevant during early phases of tumor emergence.

We next evaluated the efficacy of different immunotherapy regimens in reducing the incidence of OSCC, with a focus on therapeutic targets currently undergoing clinical evaluation. A total of 180 C57BL/6J mice were exposed to the carcinogen, 4-NQO, and randomized into six treatment groups, control IgG, anti-PD-1, anti-CTLA-4, anti-PD1/anti-CTLA-4 combination, CD40 agonist, and OX40 agonist. The CD40 agonist treatment resulted in the greatest decrease in the incidence of OSCC, with a 47% reduction (P = 0.019; Fig. 4A) observed at 16 weeks compared with the control group. None of the other treatments led to a statistically significant difference in OSCC incidence compared with the control group (Fig 4A). CD40 agonist treatment also yielded the largest reduction in the number of tumors per tongue and OSCC tumor area (Fig. 4B and C). The anti-PD-1 monotherapy regimen (P = 0.039; Fig. 4C) and the anti-PD-1 plus anti-CTLA-4 treatment regimen (P = 0.031; Fig. 4C) also significantly reduced OSCC tumor area, albeit to a lesser extent than the CD40 agonist. Anti-CTLA-4 monotherapy and OX40 agonist monotherapy did not significantly reduce OSCC by any of these measures (Fig. 4A). These studies suggest that in this model, in addition to PD-1, CD40 is a potential therapeutic target for immunoprevention of OSCC, while OX40 agonists were ineffective in this setting.

We next assessed the impact of immunotherapy regimens on the immune milieu and analyzed the DCLN resident lymphoid population by flow cytometry. We observed a significant reduction in the percentage of DCLN resident CD4⁺FOXP3⁺ Tregs from the total CD4⁺ lymphocyte population after PD-1/CTLA-4 dual immune checkpoint blockade (Fig. 4D). This enduring effect was also observed when we assessed the ratio of CTLs and Tregs after PD-1/CTLA-4 therapy (Fig. 4D). The ratio of CTLs and CD4⁺ was similar among groups, demonstrating that this sustained shift between CTLs and Tregs ratio was not associated with a reduction in the parent CD4⁺ population. Overall, these results demonstrate that treatment of OPLs with dual immune checkpoint blockade of the PD-1 and CTLA-4 pathways produced an enduring reduction in the lymph node resident CD4⁺FOXP3⁺ T regulatory population with no significant impact on the other lymphocyte or myeloid populations that were assessed (Fig. 4E).

Given the efficacy of the CD40 agonist in reducing OSCC size and number, we further assessed the immunomodulatory effects of CD40 agonist treatment. Treatment with the CD40 agonist produced a sustained high expression of the CD44 experienced/memory marker on CTLs (Fig. 4E), and this was not observed in the other treatment arms. The durable expansion of DCLN resident antigen experienced/memory CTLs suggests that CD40 agonist therapy of OPLs might create a more permissive environment for the priming of OPL neoantigens on naïve CD8⁺ lymphocytes and facilitate their survival as memory CTLs within the lymph node. CD4⁺ T-cell memory was not affected by CD40 targeting (Supplementary Fig. S1). In contrast to PD-1/CTLA-4 blockade findings, there were no effects of CD40 agonist on Tregs (Fig. 4D). On the other hand, CD40 agonist (but not anti-PD-1/CTLA-4) significantly modulated myeloid cells, as reflected by the increase in the proinflammatory CD45⁺CD11b⁺CD38⁺ macrophage population in DCLN (Fig. 4E). CD45⁺CD11b⁺CD38⁺ macrophages have been characterized previously to represent M1 macrophage phenotype (33). M1 macrophages have direct tumoricidal activity and can promote antitumor T_H1 immune responses (34). These results demonstrate a distinct pattern of immune modulation by the CD40 agonist, which induced a sustained increase in experienced/memory CTLs and M1 macrophages, compared with PD-1/CTLA-4 blockade, which resulted in a striking downregulation of the Treg population.

Next, to assess the potential relevance of these immune targets for human OPLs, we examined their expression in biospecimens from humans prospectively followed long-term within the context of a clinical trial (10). Initial biopsies of 10 index OPLs and paired, subsequent OSCCs arising from those initial lesions were analyzed. The time between the index OPL biopsy and cancer development ranged from 1 to 3 years. Because of the small sample size, these data were considered exploratory. We observed a significant increase in CD40 gene expression from the index OPL to OSCC (P = 0.008; Fig. 5A), accompanied by a trend toward decreased CD40LG gene expression (P = 0.116; Fig. 5A). We also observed trends showing increased PD-L1, PD-L2, and CTLA-4 gene expression in OSCC when compared with the index OPL (P = 0.102, P = 0.078, and P = 0.100, respectively; Fig. 5B). There were no changes in PD-1 or OX40 gene expression between the index OPL and the subsequent invasive cancer. Taken together, these results demonstrate that the CD40 pathway, and to a lesser extent the PD-L1, PD-L2, and CTLA-4 pathways, are modulated as lesions progress from OPLs to OSCC, suggesting they may play a role in the escape from immune surveillance.

Using a mouse model of OSCC prevention, we demonstrated that treatment of mice bearing carcinogen-induced OPLs with a CD40 agonist decreased the incidence of invasive cancers more potently than any of the other immunotherapies evaluated. Anti-PD-1 single agent or combined with anti-CTLA-4 also prevented OSCC, albeit to a lesser extent. The immune changes induced by these two therapies were distinct. CD40 agonist treatment was accompanied by durable expansion of experienced/memory CTLs and M1 macrophages in DCLN, while anti-PD-1 with anti-CTLA-4 mainly caused Treg depletion. We also demonstrated, in human specimens, modulation of CD40/CD40LG, PD-1/PD-L1/PD-L2, and CTLA-4 pathways upon oral epithelium malignant transformation, suggesting multiple checkpoint pathways play a role in OSCC immune evasion, and may serve as complementary and/or alternate targets for immunoprevention strategies in patients with OPLs.

A recent meta-analysis indicated that up to 48% of OPLs may express PD-L1 (35). However, reports included in the meta-analysis were heterogenous and most failed to assess the impact of PD-L1 on subsequent oral cancer development. In one of the largest cohorts reported to date, we have demonstrated that in OPL PD-L1 upregulation was associated with inferior oral cancer–free survival, suggesting a PD-L1–mediated mechanism of immune evasion at the preinvasive stage (15). Herein, we took advantage of a PD-L1^ko mouse model to further characterize the impact of this pathway on the natural history of OPLs. Indeed, we demonstrated that PD-L1^ko mice were more resistant to 4-NQO–induced carcinogenesis in the oral cavity. The mouse model allowed us to further examine the impact of the PD-L1 pathway on the immune milieu. As expected, PD-L1^ko mice showed increased CTLs. Somewhat unexpectedly, however, we observed evidence for immune modulation of populations other than CD8⁺ by the PD-L1 pathway, characterized by contraction of CD4⁺ Tregs and expansion of effector CD4⁺CD44^hiPD1⁺ cells. The expansion of CD4⁺Foxp3⁺ Tregs and its correlation with tumor immune tolerance have been well established by previous studies in tumor models (36–38). Activation of the PD-1/PD-L1 pathway modulates cancer Treg–mediated tolerance (39, 40). _amTregs have been associated with immune tolerance during early phases of tumor development (41), but this has not, to our knowledge, been reported at premalignant stages of carcinogenesis. The depletion of _amTregs and expansion of memory CTLs and CD4_eff observed in PD-L1^ko mice may represent a decrease in Treg-mediated effector T-cell suppression, a mechanism described previously in a colon cancer model (42). This decrease in Treg suppression during the OPL stage may permit the priming and expansion of neoantigen-specific CTL and CD4_eff populations, thus decelerating tumor progression. Higher PD-1 expression in CD4⁺ lymphocytes has been associated with increased overall survival in lung cancer (43). Because this was the only immune change we observed that was sustained long-term throughout the experiment, and because differences in OSCC in PD-L1^ko versus WT mice were more evident at later timepoints, these results suggest a prominent role of these cells in oral immune surveillance.

While the PD-L1^ko mouse model may provide insights on immune editing of early (pre)malignant lesions, it may not mimic current or future clinical interventions under investigation, such as the use of pembrolizumab for immunoprevention of oral cancers (IPOC clinical trial, NCT02882282), in which the treatment is started after OPLs are present. To that end, we used the 4-NQO oral cancer model to induce premalignant lesions on 100% of mice 8 weeks after cessation of the carcinogen, as demonstrated by our time-course experiment. These conditions resemble the clinical scenario of patients with OPLs that seek medical attention and experimental prevention interventions before invasive cancer development. PD-1 inhibitor administered at that time reduced progression of OPLs into OSCC. These results are consistent with our previous findings using TP53 heterozygous–knockout mice (17). Nonetheless, the effects on this TP53 WT background model seemed to be less pronounced than what was observed previously (17). Most patients with OPLs and OSCCs do not harbor germline TP53 mutations, and our current preclinical findings may suggest that predicted efficacy of PD-1 inhibitors for oral cancer prevention in patients may be modest. This illustrates the need for development of optimized immunoprevention strategies that do not rely on PD-1 inhibition alone. We hypothesize that the efficacy of an immune surveillance prevention strategy of OPL/OSCC is directly correlated with the formation of a functional immunologic memory. We did not observe sustained shifts in the Treg, CTL, or CD4 effector populations with anti-PD-1 therapy alone. This prompted us to investigate combined anti-PD-1/CTLA-4 therapy, as well as non-PD-1–based strategies using OX40 agonists and CD40 agonists for cancer prevention.

When CTLA-4 blockade was added to anti-PD-1, we observed a sustained depletion of Tregs (Fig. 4D), accompanied by a reduced total OSCC tumor area (Fig. 4C). These data suggest that the central mechanism implicated on PD-1/CTLA-4 blockade for oral cancer prevention may rely on a favorable CD8⁺/CD4⁺Foxp3⁺ ratio (Fig. 4D), opening up the opportunity to investigate strategies that modulate other immune cell populations that could also have a positive effect in impairing carcinogenesis. This may be particularly important, because CTLA-4 inhibition is associated with significant toxicities, reducing feasibility of such dual targeting strategy in clinical practice for prevention. On the other hand, OX40 agonists have been well tolerated in early-phase clinical trials for patients with advanced cancers (44). Likewise, newer generation CD40 agonists are being developed, which allow for greater tissue penetration and less side effects from systemic CD40 activation (45).

CD40 and OX40 targeting appears to act through mechanisms distinct from the depletion of Tregs observed with anti-PD-1/CTLA-4 therapy. While in our model we did not observe any cancer preventive and/or immunomodulatory effect of OX40 agonist as monotherapy, CD40 agonist showed significant activity and resulted in a reduction in the number and total bulk of OSCC tumors to a greater extent than any of the other regimens tested (Fig. 4A–C). CD40 activation during the premalignant phase created a sustained expansion of DCLN resident memory CTLs and M1 macrophages (Fig. 4E), consistent with the concurrent decline in OSCC incidence. In previous reports, activation of the CD40 pathway in tumor and viral models has been shown to contribute to expansion of antigen-presenting leukocytes, such as M1 macrophages, expansion of effector CTLs (46, 47), and to the clonal expansion, differentiation, maintenance, and recall of memory CTLs (48). In the models tested here, one possible mechanism by which CD40 activation may reduce OSCC is through expanding proinflammatory M1 macrophages, enhancing OPL neoantigen presentation to CTLs during the premalignant stage, broadening the neoantigen-specific effector T-cell repertoire, and immunologic memory formation, thus increasing immune surveillance over a carcinogen-mutated oral field. CD40 agonists are currently under clinical development for cancer therapy, and their safety profiles are being established. Nonetheless, understanding the mechanisms of action of this strategy for cancer prevention, which may distinctively involve myeloid cells and CTLs, as shown here, opens up the opportunity to optimize CD40 targeting (and its downstream effects) for cancer therapy and prevention.

To assess whether CD40 and its ligand are relevant in patients with OPLs and OSCC, we evaluated gene expression profiles in paired clinical samples, and found an increase in the expression of CD40 upon progression of OPL to OSCC, accompanied by a trend toward a reduction in the expression of CD40 ligand. On the basis of our preclinical findings, we, therefore, speculate on a working model, according to which downregulation of CD40 ligand during carcinogenesis may contribute to immune evasion and may lead to compensatory upregulation of CD40. We fully acknowledge the limitations of our clinical data and exploratory nature of these findings, because our clinical sample size was small, given the rarity of cohorts of patients with OPL that have been prospectively followed long term, with biopsies of both the premalignant and the malignant lesions. Our cohort also did not contain longitudinal biopsies of OPLs that did not progress to cancer (an important control group), further limiting our ability to definitively assess the role of CD40/CD40 ligand in human carcinogenesis. Nonetheless, the consistency of the preclinical and clinical data supporting modulations of CD40/CD40 ligand, in addition to PD-1/PD-L1/PD-L2 and CTLA-4, across OPLs and cancers in mouse models and humans points to the existence of multiple pathways regulating immune editing in the process of oral carcinogenesis. Previous studies that evaluated the progression of bronchial premalignant lesions to lung squamous cell carcinoma (49) and atypical adenomatous hyperplasia to lung adenocarcinomas (50) also demonstrated multiple immune pathways involved in antigen processing and presentation, IFN signaling, and immune check points that are modulated during the premalignant stage, and are associated with higher risk of progression. Future studies investigating the potentially synergistic combination of these treatment regimens are warranted. Furthermore, success of any given prevention strategy may be enhanced by targeting the population at highest risk selected on the basis of prognostic biomarkers that may also predict benefit from therapy (8). As such, individuals with OPLs harboring high PD-L1 or low CD40 ligand expression could be ideal candidates for personalized prevention with PD-1- or CD40-targeted agents, respectively, should our findings be confirmed in independent patient cohorts.

Limitations of our study include the incomplete (yet close) resemblance of 4-NQO–induced lesions with human OPLs in terms of molecular and/or immune features (25, 26); the timing of PD-1 axis blockade in knockout models, or short-term treatment administration to the animals (as opposed to anti-PD-1 therapy for several months currently under clinical investigation for cancer prevention); and the lack of mouse tongue immune profiling (due to paraffin block exhaustion) and patient lymph node immune profiling (because neck dissections are not routinely performed as part of OPL surgical treatment), thus reducing our ability to contrast murine and patient findings.

In summary, we found that PD-1 inhibition, with or without CTLA-4 blockade, as well as single-agent CD40 agonists, are potential strategies to prevent OPL evolution into invasive OSCC. We observed a distinct pattern of immune surveillance modulation by CD40 agonist, which generates a lasting expansion of experienced/memory CTLs and M1 macrophages, versus PD-1/CTLA-4 blockade, which largely depletes Tregs. These findings suggest that, at least in preclinical models, there may be multiple mechanisms of immune suppression at work as tumors progress from premalignancy to carcinoma, and that immunoprevention strategies targeting these different mechanisms merit further investigation in the clinic.

J.A. Monteiro de Oliveira Novaes reports grants from CCSG P30 CA016672 during the conduct of the study. M. Nilsson reports personal fees from Spectrum Pharmaceuticals outside the submitted work, as well as has a patent for PTC/US2019/022067 pending and PTCUS2017/062326 and U.S. Provisional Patent Application Nos. 62/423,732; 62/427,692; and 62/572,716 issued, licensed, and with royalties paid from The University of Texas System Board of Regents. D.L. Gibbons reports grants from Takeda, personal fees from Sanofi and Alethia Biotherapeutics, grants and personal fees from Ribon Therapeutics and Astellas, and grants from AstraZeneca and Janssen R& D outside the submitted work. W.N. William Jr reports grants from Barbanti Funds for Cancer Research during the conduct of the study, personal fees from Clovis Oncology, Roche/Genentech, and Eli Lilly, grants, personal fees, and nonfinancial support from AstraZeneca and Merck, grants and personal fees from Boehringer Ingelheim and OSI Pharmaceuticals, and personal fees and nonfinancial support from BMS outside the submitted work. J.V. Heymach reports other from Bristol-Myers Squibb, Merck, EMD Serono, Genentech, Novartis, GlaxoSmithKline, and AstraZeneca during the conduct of the study, grants and personal fees from AstraZeneca and GlaxoSmithKline, personal fees from Boehringer-Ingelheim, Bristol-Myers Squibb, Merck, Catalyst, Genentech, Guardant Health, Foundation Medicine, Hengrui Therapeutics, Eli Lilly, Novartis, EMD Serono, Sanofi, Biotree, and Takeda, and grants, personal fees, and other from Spectrum outside the submitted work. No disclosures were reported by the other authors.

J.A. Monteiro de Oliveira Novaes: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. T. Hirz: Methodology, writing–review and editing. I. Guijarro: Writing–review and editing. M. Nilsson: Project administration, writing–review and editing. M.A. Pisegna: Investigation. A. Poteete: Investigation. H.B. Barsoumian: Methodology. J.J. Fradette: Resources. L.N. Chen: Resources. D.L. Gibbons: Resources. X. Tian: Formal analysis. J. Wang: Formal analysis. J.N. Myers: Conceptualization. M.J. McArthur: Validation. D. Bell: Validation. W.N. William: Conceptualization, supervision, funding acquisition, writing–review and editing. J.V. Heymach: Conceptualization, supervision, funding acquisition, writing–review and editing.

This work was funded by CCSG P30 CA016672 (to J.V. Heymach), the Susan Gold Collins endowment (to J.V. Heymach), Barbanti Funds for Cancer Research (to W.N. William Jr), and the Gil and Dody Weaver Foundation (to J.V. Heymach).

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.