Differentiated State of Initiating Tumor Cells Is Key to Distinctive Immune Responses Seen in H-Ras^G12V–Induced Squamous Tumors

Heterogeneity in tumor immune responses is well documented, but the underlying mechanisms generating distinct tumor immune phenotypes remain poorly understood. Many cancers develop an immunosuppressive microenvironment, dampening antitumor immunity and facilitating tumor growth (1–3), whereas in others, chronic inflammation creates microenvironments that increase the risk of cancer development and enhance tumor growth (4, 5). Molecular and immune phenotyping has revealed the heterogeneity of underlying genetic changes and signaling pathways, which is linked to different immune signatures, responses to therapy, and outcomes for individual patients with the same tumor types (6–9). Different immune microenvironments show heterogeneity at the cellular level: Regulatory T cells (Treg; refs. 10, 11), helper T cells (Th cells; ref. 12), B cells (13, 14), and myeloid cells (15, 16) have been described with both tumor-suppressive and tumor-promoting properties. Between different tumors, this may reflect diverse interactions between multiple cell types within the tumor microenvironment, differences in tumor cell genetics and signaling, tumor type, or progression of tumors through immunoediting to escape (17). This presents challenges in developing immunotherapies, as patients may show improved or worsened prognosis depending upon relationships between the tumor and its immune microenvironment. Patients with melanoma (18), lung cancer (19), gastric cancer (20), renal cell carcinoma (21), and head and neck cancer (22, 23) have all benefited from advances in immune checkpoint blockade, but tumors perpetuated by chronic inflammation, such as colorectal carcinoma (24), hepatocellular carcinoma (25), as well as hematologic malignancies, benefit from reducing tumor-promoting inflammation (26).

It is not known to what extent the immune phenotype of a tumor is affected by the initiating transformed cell. Most solid tumors arise from hierarchically organized epithelia comprised of tissue stem cells, transit amplifying cells, and differentiated cells (27, 28). Studies in murine models of cutaneous cancer in which c-H-Ras mutations drive tumor development show that carcinogen treatment or oncogene activation in different progenitor cell populations can give rise to distinct tumor phenotypes (29–31). We previously showed that benign epidermal squamous lesions with a high frequency for progression to squamous cell carcinoma (SCC) exhibit a gene expression pattern associated with reduced inflammation, relative to lesions with a low frequency for progression (32). To test if oncogene activation in different tumor-initiating cell populations affects the tumor immune response, we generated mice with inducible expression of human H-Ras^G12V targeted to either basal/stem cells or committed progenitor/suprabasal cells of the epidermis (28). Here, we show that within a single epithelium the same driver oncogene can generate opposite immune responses and response to immunotherapies, depending on the origin of the tumor-initiating cell. These distinct immune responses of either immunosuppression or tumor-promoting inflammation are driven by B-cell and CD4 T-cell cooperation within the tumor microenvironment. These results offer insight into the intricacies of modulating the tumor immune microenvironment and provide guidance for the application and scope of immunotherapeutics in cancer treatment.

Tetracycline-regulated transactivators (tTA) driven by involucrin (Inv; InvtTA, ref. 33); obtained from Dr. Julie Segre (NHGRI, NIH National Human Genome Research Institute) or reverse tetracycline-regulated transactivators (rTA) driven by Keratin-14 (K14; K14rTA, ref. 34) mice (The Jackson Laboratory) were crossed with homozygous tetOH-Ras^G12V (35) mice (NCI Mouse Repository), producing double transgenic (DT) InvtTA/tetOH-Ras^G12V (InvRas) or K14rTA/tetOH-Ras^G12V (K14Ras) mice. Suppression of transgene expression in InvRas mice was accomplished with 10 μg/mL doxycycline hyclate (Sigma-Aldrich) administered ad libitum in drinking water. Tumors were induced in shaved 7-week-old mice by reducing doxycycline to 250 ng/mL for InvRas mice and administration of 7.5 μg/mL doxycycline in drinking water to K14Ras mice. Tumors were counted and measured every 48–72 hours and harvested 28–32 days following Ras induction, or if animals were moribund. To control for gender differences, when possible equal numbers of male and female mice were used, and all data represent results from both genders. Cell counts were conducted using a Cellometer Auto T4 Cell Viability Counter (Nexcelom Bioscience). All mice were on the FVB/n background.

To deplete leukocytes, mice were given 500 μg of GK1.5 (anti-CD4), 5D2 (anti-CD20), RB6-8C5 (anti–GR-1), isotype control HB94, or anti-ragweed IgG weekly intraperitoneally beginning 7 days prior to tumor induction, or every other day beginning 1 day prior to H-Ras^G12V induction (anti–GR-1). GK1.5, RB6-8C5, and HB94 hybridomas were obtained from ATCC, and 5D2 and anti-IgG antibodies were obtained from Genentech. Anti-CD40L (BioXCell) was injected 7 days prior to tumor induction, or upon tumor presentation, followed previously defined methods (36). CD40 agonist–treated mice were injected with 100 μg FGK4.5 (BioXCell) weekly beginning at indicated days. Anti-CD40L and CD40 agonist antibodies were tested to be free of endotoxin (< 0.002 EU/μg; BioXCell). Rag1^−/− mice were reconstituted by retroorbital administration of 5 million CD4 T cells or B cells from spleens of healthy age-matched syngenic mice (StemCell Easysep) or phosphate-buffered saline (PBS). Studies were performed in compliance with the U.S. Department of Health and Human Services Guide for the Care and Use of Laboratory Animals, and after approval by The Pennsylvania State University Institutional Animal Care and Use Committee.

Skin leukocytes were isolated as previously described (37), incubated with antibodies in 1% BSA/PBS at 1 × 10⁷ cells/mL. Leukocytes were stimulated for intracellular cytokines in complete RPMI 1640 medium (38). Following 4% paraformaldehyde fixation, cells were permeabilized and stained using 0.2% Saponin/1% BSA/1× PBS at 1 × 10⁷ cells/mL. Cells were analyzed using an LSRFortessa Cytometer (BD Biosciences) and FlowJo version 7. Myeloid cells were gated on live CD45⁺, CD11b⁺, followed by Ly6G⁺/Ly6C⁺ for neutrophils, Ly6G⁻/Ly6C⁺ for Myeloid Derived Suppressor Cells (MDSC), and Ly6G⁻/Ly6C⁻/F4/80⁺ for macrophages. CD4 T cells were gated on live CD45⁺, CD3⁺, CD4⁺, followed by IFNγ (Th1), IL4 (Th2), IL17 (Th17), or FoxP3 (Treg). B cells were gated on live CD45⁺, CD19⁺, CD45R⁺, and MHCII⁺ followed by CD1d^hi/CD5⁺ B Regulatory Cells (Bregs) and CD40⁺/CD80⁺/CD86⁺ (APC⁺). Live cells were identified using the LIVE/DEAD Fixable Yellow Dead Cell Stain Kit (Life Technologies).

Protein and RNA were isolated from tumors and analyzed by immunoblotting and RT-qPCR, respectively, as described (39), using β-actin as a loading control.

Formalin-fixed sections were stained with hematoxylin and eosin (H&E) using an ST5010 Autostainer XL (Leica Biosystems). For anti-CD45 and anti-Ki67 immunohistochemistry, formalin-fixed sections were pretreated with Tris-based antigen unmasking solution (Vector Labs) according to the manufacturer's protocol. Antibodies were detected using VECTASTAIN ImmPRESS (Vector Laboratories). Anti-keratin immunohistochemistry was done on ethanol-fixed sections with the ImmPACT DAB (Vector Laboratories) detection system. For coimmunofluorescence, 7-μm frozen sections were incubated with anti-CD4 (1:100; Biolegend) followed by IgG-Alexa Fluor 647 (1:500; Biolegend), biotinylated anti-CD19 (1:100; Biolegend), streptavidin-Alexa Fluor 488 (1:500; Biolegend) using the Polink Double Staining System (GBI Labs). VECTASHIELD Mounting Medium with DAPI (Vector Laboratories) was used to visualize nuclei.

CD4 T cells and B cells were purified from spleen or tumor tissue and plated onto 24-well plates coated with mAbs to CD3 and CD28 (Biolegend). Lymphocytes were suspended in complete RPMI1640 media, and CD4 T cells were added first followed by hamster IgG, anti-CD40L, or Transwell inserts (BD Falcon), and then by B cells. Cells were incubated for 7 days at 37°C and 5% CO₂ with fresh media added every 48 hours and stained for flow cytometry as indicated.

Negatively selected purified B cells from tumor tissue and CD4 T cells from spleens were cocultured 7 days following staining with 2.5 μmol/L carboxyfluorescein diacetate succinimidyl ester (CFSE). Harvested cells were analyzed by flow cytometry for CFSE dilution. For MDSC suppression assays, mice were placed on doxycycline chow for 4 days (Bio-Serv; K14Ras), or 7 days doxycycline removal (InvRas), splenocytes were stained with αCD11b, αLy6G, and αLy6C, and then sorted using a Cytopeia Influx Sorter (BD Biosciences). Sorted cells were incubated with stimulated naïve CD8 T cells for 72 hours and analyzed by flow cytometry for CFSE dilution.

For experiments with only two groups of study, Student t tests were used. If three or more groups were compared, one-way ANOVA was used with Newman–Keuls post-analysis. All figures, graphs, and statistical analyses were made using GraphPad Prism 4.0. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Papillomas at high risk or low risk for progression to SCC have distinct immune signatures (32). Because these tumors may arise from different initiating cells, we examined whether targeting oncogenic Ras to different cellular compartments of the murine epidermis, using inducible models of squamous tumorigenesis, altered tumor-associated immune responses. These models allow for specific targeting of the activated H-Ras^G12V oncogene to either the basal/stem cell (K14Ras) or committed progenitor/suprabasal epidermal compartments (28). When 7-week-old InvRas and K14Ras bitransgenic mice were placed on threshold doses of doxycycline, focal tumors arose in both groups with similar kinetics (first appearance 7–10 days) and numbers (Fig. 1A), and expression of H-Ras and ERK1/2 (phosphorylated and total; Supplementary Fig. S1A), although by 24 days the average volume of K14Ras tumors was greater than that of InvRas tumors (Fig. 1B). Immunohistochemical staining with H&E and anti-keratin showed that basal expression of H-Ras^G12V generated SCC), whereas suprabasal expression induced squamous papillomas (Supplementary Fig. S1B), similar to previous studies (29). Despite similar expression, targeting H-Ras^G12V to different cell populations within the epidermis caused distinct tumor immune responses. Expression of Th1- (TNFα), Th2- [IL33, thymic stromal lymphoprotein (TSLP), and lymphotoxin (LT)], and Th17- (IL6, TGFβ) associated genes was significantly higher in InvRas than in K14Ras tumors. Although not statistically significant, IL12 was also more highly expressed in InvRas. In addition to the overall reduced level of proinflammatory cytokines, K14Ras tumors expressed more immunosuppressive cytokine IL10, although no differences were observed for IL35 (Fig. 1C–F).

Consistent with these cytokine signatures, InvRas tumors had significantly higher percentages of Th2 cells and more proinflammatory antigen-presenting B cells (APC⁺), whereas K14Ras tumors had significantly higher percentages and numbers of Tregs and Bregs (Fig. 1G–J). Tumor-infiltrating B cells from K14Ras, but not InvRas, tumors caused significant suppression of CD4 proliferation in coculture (Fig. 1K), confirming their specific regulatory capacity. No significant differences in cell numbers or percentages were observed in splenic T or B cells (Supplementary Fig. S1C and S1D). CD45 immunohistochemistry showed significantly more leukocytes infiltrating into InvRas tumors compared with K14Ras (Supplementary Fig. S2).

To rule out that differences observed in immune phenotype were indirectly the result of distinct tumor phenotypes, we induced strong expression of Ras in both models and examined the response after 5 days. High Ras expression caused neutrophil microabscesses in InvRas mice but not K14Ras (Supplementary Fig. S3A). Additionally, patterns of cytokine expression (Supplementary Fig. S3B) were distinct and circulating CD11b⁺ Ly6G⁻ cells that could suppress CD8 T-cell proliferation in vitro were significantly expanded in K14Ras but not in InvRas mice, in agreement with increased arginase1 expression in hyperplastic K14Ras skin (Supplementary Fig. S3C–S3E). Together, these results show that responses to oncogenic Ras are distinct at the earliest time points, even under conditions of supraphysiological Ras expression.

To determine the specific effects of lymphocytes on Ras-driven tumor development, both models were crossed onto a Rag1^−/− background and tumors induced as described above. K14RasRag1^−/− mice developed a significantly greater number of tumors that grew faster than Rag1^+/+ counterparts, with few mice surviving beyond 14 days (Fig. 2A–C). In contrast, tumors in InvRasRag1^−/− mice began to regress 15 days after Ras induction; by 18 days, Rag1^+/+ mice had significantly more tumors, despite an initially similar tumor incidence. Additionally, tumors developing in InvRasRag1^−/− mice were smaller on average than tumors from their Rag1^+/+ counterparts, but variability in tumor size prevented statistical significance (Fig. 2D–F). The percentages of tumor-infiltrating neutrophils in K14Ras mice increased in the absence of lymphocytes (Fig. 2G and H), but there was a striking reduction in neutrophils in InvRasRag1^−/− mice, leading to a change in proportion of myeloid cell populations within the tumor (Fig. 2I and J). Additionally, CD45 staining confirmed reduced leukocyte infiltration into tumors of InvRasRag1^−/− mice, but increased in K14RasRag1^−/− mice, and leukocytes infiltrated the tumor parenchyma in all models (Supplementary Fig. S2). When H-Ras^G12V was strongly induced in the epidermis of mice depleted of neutrophils with anti–GR-1, proliferation of basal cells was significantly reduced compared with isotype control (Supplementary Fig. S2), suggesting that tumor-infiltrating neutrophils could enhance tumor growth. Together, these results suggested that immunoregulation in basal/stem cell Ras-driven tumors was associated with a decreased tumor burden, whereas greater inflammation in committed progenitor/suprabasal-Ras driven tumors led to enhanced tumor burden.

The absence of lymphocytes generated opposite responses between the tumor models. We therefore determined which lymphocyte subsets drove the distinct tumor responses, using adoptive transfer of CD4, CD8, and B cells alone or in combination into Rag1^−/− mice. Although adoptive transfer of naïve CD4 T cells or B cells alone had no significant effect on tumor numbers in K14RasRag1^−/− mice, cotransfer of CD4 T cells with B cells reduced tumor numbers to Rag1^+/+ levels, although suppressive effects were lost after 20 days (Fig. 3A). Adoptive transfer of CD4 T cells or B cells into InvRasRag1^−/− mice suppressed tumor development, whereas cotransfer restored a tumor-promoting phenotype similar to Rag1^+/+ mice until day 20 (Fig. 3B). Adoptive transfer of CD8 T cells or cotransfer of CD8 with B cells reduced tumor numbers in K14Ras and InvRasRag1^−/− mice, but this was not statistically significant compared with Rag1^−/−. Although cotransferred K14Ras mice had fewer tumors, epidermal hyperplasia was more pronounced and mice became moribund more quickly, suggesting that this was not a protective immune response. CD8 T cells caused an overall suppression of both Bregs and APC⁺ subsets, inconsistent with phenotypes observed in Rag1^+/+ mice (Supplementary Fig. S4A–D), suggesting that CD8 T cells have little overall impact on determining the Rag1^+/+ immune responses in either model. Supporting the opposing functions of lymphocytes in these two tumor models, depletion of either CD4 T cells or B cells from K14RasRag1^+/+ mice resulted in a similar increase in tumor number, whereas depletion from InvRasRag1^+/+ mice resulted in suppression of tumor formation (Fig. 3C and D). Together, these data suggest that the nascent tumor microenvironment programs CD4 T and B cells toward distinct cooperating phenotypes that either suppress (K14Ras) or enhance (InvRas) tumor development.

To test this, we examined the phenotype of tumor-infiltrating lymphocytes (TIL) in tumors from mice receiving CD4 T and B cells alone or together. TILs from CD4 T-cell transfers into K14RasRag1^−/− mice were skewed toward a proinflammatory Th1 phenotype with reduced Tregs, whereas cotransfer restored Th1 and Tregs to Rag1^+/+ levels (Fig. 3E). In contrast, TILs from CD4 T-cell transfers into InvRasRag1^−/− mice exhibited polarization toward Tregs, with slightly reduced percentages of Th1 cells. Importantly, cotransfer caused a minor increase in Th1 cells and a significant increase in Th17 cells, but reduced Tregs (Fig. 3F). Cotransfer of B and CD4 T cells selectively restored expression of CD4 associated genes to Rag1^+/+ levels of each model. Cotransfer into K14RasRag1^−/− mice reduced expression of IL12, IL33, TSLP, and IL35 in tumors, whereas cotransfer into InvRasRag1^−/− mice completely restored IL33 and IL35 expression. IL12 expression was higher and TSLP expression was lower than Rag1^+/+ in cotransferred tumors but not statistically different from Rag1^+/+. IL1β expression was not significantly affected by cotransfer (Supplementary Fig. S5A–E). Cotransfer also regulated the phenotype of B-cell TILs (TILB) in opposite directions: B cells transferred into K14RasRag1^−/− mice had increased APC⁺ percentages, whereas cotransfer restored APC⁺ B cells to that of Rag1^+/+ tumors (Fig. 3G). In contrast, cotransfer into InvRasRag1^−/− mice decreased the percentage of Bregs compared with B-cell transfer, although APC⁺ B cells were not affected (Fig. 3H). These altered lymphocyte phenotypes had pronounced effects on tumor-infiltrating neutrophils. Only cotransfer of CD4 T and B cells could reduce neutrophil percentages to that of Rag1^+/+ levels in K14RasRag1^−/− and recover neutrophils in InvRasRag1^−/− (Fig. 3I and J), likely the result of the recovery and suppression of Tregs, respectively. The percentage of dendritic cell (DC) subsets within the skin draining lymph nodes or skin differed little between K14Ras and InvRas mice, except for plasmacytoid DC and CD11b⁻ DC, but these were not closely correlated with the overall immunophenotype, as InvRas tumors had fewer proinflammatory plasmacytoid dendritic cells and K14Ras tumors had fewer immunoregulatory CD11b⁻ lymphoid dendritic cells (Supplementary Fig. S6A–J), indicating that the polarizing effect could not be carried out by dendritic cells alone, and that B cell presence was necessary for proper CD4 T-cell polarization. These results indicate that CD4 T cells and B cells cooperated to produce an immunosuppressive microenvironment in K14Ras mice that constrained tumor outgrowth, whereas cooperation in InvRas mice resulted in a proinflammatory microenvironment supporting tumor growth.

Transferred B cells in both models were largely present within tumors, but not in secondary lymphoid nodes (Supplementary Fig. S7), which prompted us to hypothesize that B cells largely affect CD4 T-cell polarization within the tumor microenvironment. To test the effects of tumor-conditioned B cells on CD4 polarization, we cocultured TILB or splenic B cells from tumor-bearing mice or healthy controls with anti-CD3/CD28–stimulated naïve CD4 T cells from nontransgenic mice. K14Ras TILBs polarized naïve CD4 T cells toward Tregs and away from Th2, whereas InvRas TILB facilitated Th2, but not Treg, polarization. Splenic B cells from tumor-bearing mice had a distinct phenotype, suppressing both Treg and Th1 polarization relative to stimulated controls, but this was not different from B cells of nontransgenic mice. These results suggested that the specific polarization of CD4 T cells by B cells required conditioning by the tumor microenvironment (Fig. 4A–D). Coimmunofluorescence showed these cells in direct contact within the tumor parenchyma of both models, indicating a potential mechanism for interaction within the tumor microenvironment (Fig. 4E and F). We did not observe any tertiary lymphoid structures within tumors of either model, as others have previously reported (13, 40, 41), although this may be due to differences in model systems or time course.

To determine if direct contact was needed for polarization, stimulated naïve CD4 T cells were separated from tumor-conditioned B cells of either K14Ras or InvRas mice using Transwell inserts. K14Ras tumor-conditioned B cells significantly increased the overall percentage of Tregs, but did not affect proinflammatory subsets. Additionally, when direct contact was blocked, the percentage of Th1 cells increased and Tregs significantly decreased. These data suggest that direct contact from K14Ras TILBs could polarize naïve CD4 T cells toward a Treg phenotype and away from Th1 (Fig. 5A–D). In contrast, blocking contact from InvRas TILB caused a decrease in Th1 and Th17 cells and a significant increase in Tregs (Fig. 5A–D). Preventing direct contact had no effect on the ability of InvRas TILB to polarize Th2 cells.

The CD40/CD40L pathway is critical for the establishment of humoral responses as perpetuated by Th2/B cell interactions, is implicated in immune responses to several cancers, and is a potential target for immune checkpoint blockade (42). To test if CD40/CD40L interactions were important for TILB effects on CD4 polarization, a CD40L-blocking antibody (MR-1, BioXcell) was included in coculture assays. Blockade of CD40/CD40L between K14Ras TILBs and naïve stimulated CD4 T cells caused a significant increase in percentage of Th1 and Th2 cells and simultaneously decreased the percentage of Tregs, with little effect on Th17 (Fig. 5E–H). In contrast, blockade of CD40/CD40L from InvRas TILBs caused a marked increase in Treg differentiation and a slight, but not statistically significant, reduction in all proinflammatory CD4 subsets (Fig. 5E–H). Thus, the effects on CD4 polarization caused by blockade of direct cell–cell contact with Transwell inserts was likely due to inhibition of CD40/CD40L interactions between TILBs and CD4 T cells. We conclude that CD40/CD40L interactions were the primary mechanism by which CD4 T cells and B cells interacted within the tumor microenvironment to orchestrate the Th1/Treg balance, and that Th2 polarization by InvRas TILB was a direct result of CD40 ligation.

Given the distinct tumor immune responses generated by targeting H-Ras^G12V to different epidermal cell populations, and divergent effects of CD40/CD40L blockade on TILB polarization of CD4 T cells, we examined how in vivo blockade of CD40/CD40L would affect tumor development. In a pretreatment protocol, mice were given anti-CD40L 7 days prior to H-Ras^G12V induction and were given weekly boosters as described (35). In anti-CD40L–treated K14Ras mice, tumor numbers were slightly elevated compared with isotype control injected mice (Fig. 6A). In contrast, anti-CD40L pretreatment significantly reduced the number of tumors that developed in InvRas mice (Fig. 6B). To determine the effect of CD40 blockade in a therapeutic setting, mice were injected with anti-CD40L or isotype control antibody after tumors became palpable on either day 14 or day 11 for K14Ras and InvRas mice, respectively. Under these conditions, tumor numbers were not significantly affected over time in either tumor model (Fig. 6C and D). Anti-CD40L treatment significantly altered tumor growth in opposite directions. In K14Ras mice, anti-CD40L caused a 10-fold greater increase in tumor volume relative to initial tumor volume compared with isotype control injected mice (200-fold increase vs. 20-fold increase; Fig. 6E). In contrast, anti-CD40L injection in tumor-bearing InvRas mice prevented further tumor growth or caused regression after an initial period of growth. Continuous tumor growth (70-fold over starting volume on day 22) was observed only in isotype control–injected mice (Fig. 6F). The decrease in fold change in InvRas isotype control mice from day 22 to day 24 occurred due to sacrifice of moribund mice at day 22 for humane endpoints.

CD40 agonist antibodies are in clinical trials as therapeutic antibodies (43). To determine if opposite tumor responses would also occur when CD40 was activated, K14Ras or InvRas mice were injected with FGK4.5, a CD40 agonist mAb, after tumors had developed. CD40 agonist mAb significantly reduced the number of tumors in K14Ras mice, but had no effect on tumor number in InvRas (Fig. 6G and H). Furthermore, and in contrast to treatment with anti-CD40L, tumors in CD40 agonist–treated K14Ras mice grew much more slowly compared with isotype control with only a 10-fold increase in volume compared with 60-fold for isotype control relative to starting volume (Fig. 6I). In contrast, tumors from InvRas mice treated with CD40 agonist had on average a 190-fold increase in tumor volume relative to starting volume, as compared with 50-fold increases in isotype controls at study end (Fig. 6J). Together, these data show that the specific tumor immune microenvironment could alter the direction of response to therapeutic antibodies.

Here, we show that oncogenic HRas^G12V expression within cell types with distinct fates in the differentiation hierarchy of the epidermis generated highly divergent immune responses that promoted or suppressed tumor development. When expressed in the basal/stem cell or committed progenitor/suprabasal compartments, oncogenic Ras caused either an immunosuppressive Treg-dominated, or proinflammatory Th2-dominated, microenvironment; B-cell–driven Treg or Th1/2 polarization via CD40/CD40L ligation; and anti-CD40L–mediated promotion or suppression of tumor development. Although the models utilize an inducible system and therefore likely have higher expression of Ras than mutations of the endogenous locus, we have titered doxycycline to threshold doses that give focal tumor development and similar expression of oncogenic Ras and activation of downstream signaling. The mechanism by which these distinct immune responses are generated is not clear. Given the distinct location and properties of dendritic cell subsets in the skin, differential activation of one or more subsets in response to Ras expression in keratinocytes is a possibility, although we did not observe differences between the tumor models in DC subsets in the skin draining lymph nodes shortly after induction of Ras. Differential activation or polarization of skin resident γδ T cells, previously shown to play an important role in squamous tumor development (44), could also be important. Keratinocytes themselves are known to produce inflammatory mediators, such as CD40 (45) and TNFα, and multiple interleukins that can mediate both protumor and antitumor inflammation (46). Although the ERK1/2 MAPK pathway was activated similarly in both models, it is possible that other Ras effector pathways are differentially engaged in keratinocytes, leading to divergent expression of immunoregulatory genes and pathways. That distinct cytokine expression and immune responses occurred rapidly in the epidermis following induction of strong Ras expression supports the idea that the distinct tumor immune phenotypes observed reflect distinct properties of the tumor-initiating cell rather than a consequence of tumor stage or histology. Nevertheless, these results, together with our previous study showing that chemically induced tumors with a high risk and low risk for progression to SCC had distinct immune signatures (32), suggest that heterogeneity in human cancer immune phenotypes within single tissues could reflect the location of the cancer-initiating cell or cancer stem cell within the stem cell hierarchy of the tissue.

As no lymphocyte was able to restore a Rag1^+/+ tumor phenotype when transferred to K14Ras or InvRas Rag1^−/− mice, we cotransferred either CD4 or CD8 T cells with B cells into Rag1^−/− mice of each model. In K14Ras mice, CD8 plus B-cell cotransfer caused fewer tumors, due to a generalized epidermal hyperplasia but with rapid mortality, whereas in InvRas mice, cotransfer was unable to restore tumor development. CD4 T- and B-cell cotransfers restored Rag1^+/+ tumor numbers in both models, without a generalized epidermal hyperplasia, suggesting that the interactions between these two subsets were critical for the observed immunosuppressive and tumor-promoting immune responses. Both CD4 T cells and B cells have been extensively studied in a number of human cancers and mouse models, with each having distinct protumor and antitumor functions. Tregs have a controversial role in SCC. They inhibit antitumor responses through CTLA-4, CD25, CD39, CD73, IL10, and TGFβ in a tumor antigen–specific manner (47) and also increase growth and metastasis directly through RANK signaling (48). On the other hand, Tregs are associated with improved prognosis in SCC, as higher Treg counts in early-stage oral SCC are associated with improved survival and decreased lymph node metastasis (47). Bregs display fewer surface CD20 molecules, causing anti-CD20 administration to enrich for Bregs within the total B-cell repertoire (49). Although Tregs and Bregs are most commonly associated with tumor escape (13), we find that the predominant Treg and Breg immunosuppressive phenotype found in the K14Ras tumors was associated with suppression of tumor outgrowth, in agreement with previous reports in both mouse and human SCC (50–52). The likely mechanism for this tumor suppression is through inhibition of neutrophils that enhance proliferation of transformed keratinocytes, as CD4 T- and B-cell cotransfer significantly reduced percentages of tumor-infiltrating neutrophils. Tregs improve clinical outcome in many hematologic cancers, including follicular lymphoma, diffuse large B-cell lymphoma, and Hodgkin lymphoma (53). Additionally, Th2 cells are also associated with tumor escape in pancreatic, mammary, and colitis-associated cancers, specifically attributed to IL13 (54), IL4- and IL10-inducing M2 macrophages (55), and TGFβ and IL5 (56), respectively. We find that Th2 cells predominate in a proinflammatory setting that promotes tumor formation.

Although the cross-polarizing role of CD4 T cells and B cells has been previously reported (13), we have now directly compared these interactions in a stratified squamous epithelium, comparing tumors of stem and differentiated cell origins. We show that direct interaction and cross-polarization between CD4 T and B cells through CD40/CD40L is essential for establishment of both tumor immune phenotypes. Adoptive transfer studies showed that both lymphocytes were needed to reproduce the tumor immune phenotype of Rag1^+/+ mice. These results also suggest that APC function of innate leukocytes such as dendritic cells is not sufficient to polarize CD4 T cells or B cells toward the immune phenotype seen in Rag1^+/+ mice. Specific depletion of these lymphocytes enhanced tumor numbers in K14Ras and suppressed in InvRas mice. However, although CD4 T cells could be fully depleted, anti-CD20 depletion enriched for Bregs, as shown by others (ref. 49; data not shown), suggesting that simple disruption of the CD4 and B-cell dynamic results in perturbation of the Rag1^+/+ tumor response.

Cotransferred B cells were found only in tumor tissue and not in inguinal lymph nodes, indicating a localized immune response (Supplementary Fig. S7). In many models, including mouse and human SCC, tertiary lymphoid structures are found within the tumor microenvironment, acting as a surrogate localized lymph node (13, 40, 41). We did not observe lymphoid structures within tumor tissue of either mouse model, and lymphocytes appeared to be localized within the tumor parenchyma. Supporting the idea that the critical interaction between B and T cells occurred within the tumor, we found that only B cells from tumor tissue, but not splenic B cells of tumor-bearing mice, could specifically polarize CD4 T cells in coculture toward the Rag1^+/+ phenotype. The specific tumor microenvironment was critical for determining the ability and direction of naïve CD4 polarization by TILBs. Although direct contact in coculture was essential for TILB polarization of Tregs and Th1 cells, prevention of contact had no effect on Th2 percentages, compared with coculture controls. Blockade of CD40L showed similar results as Transwell inserts, but also inhibited B cell effects on Th2 polarization. Activated B cells can secrete soluble CD40 (57); thus, although direct cell contact is necessary for polarizing signals of Th1 and Tregs, it is possible that Th2 polarization is influenced by B-cell–derived CD40. B cells can both promote or suppress tumor formation by acting as plasma cells, APCs, or Bregs, secreting TGFβ, IL10, IL4, and other interleukins to modulate pro- and anti-inflammatory CD4 and CD8 T cells (13). Specific downstream mechanisms beyond CD40/CD40L were not determined in these studies; however, alternative local cytokine expression by B cells is the most likely mechanism driving the direction of CD40-induced CD4 polarization. The CD40/CD40L costimulatory pathway both enhances and prolongs contact between CD4 T cells and B cells, inducing upregulation of CD80 and CD86 (58). It is likely that by inhibiting this pathway, tumor-promoting proinflammatory responses were blocked in InvRas pretreatment protocols, leading to the significant decrease in tumor number. Although our results show the importance of this pathway in regulating polarization both in vivo and in vitro, the studies in mice utilized global CD40L blockade or CD40 activation. While the results agree with our hypothesis, we cannot rule out that blockade or activation of dendritic cell and keratinocyte-based CD40 could have occurred and affected the observed effects on tumor formation. However, the lack of differences in dendritic cell activity, and the inability for Rag1^−/− mice to specifically polarize CD4 T cells or B cells in single adoptive transfers, suggests that neither keratinocytes nor dendritic cells could specifically polarize these lymphocytes, as Rag1^+/+ phenotypes were observed only with both CD4 T cells and B cells simultaneously present.

Our results show that the composition and functionality of the tumor immune microenvironment influence the response to therapeutic mAbs. Treatment of preexisting basal/stem cell–derived tumors with anti-CD40L enhances tumor growth, whereas CD40 agonist treatment prevents new formation and growth of existing tumors. In contrast, treatment of committed progenitor/suprabasal derived tumors with anti-CD40L inhibits tumor growth, but a CD40 agonist enhances growth of established tumors. CD40 agonist and antagonist mAbs such as dacetuzumab (Seattle Genetics) and lucatumumab (Novartis; ref. 59) have been tested in human cancer patients with chronic lymphatic leukemia, multiple myeloma, and non-Hodgkin lymphoma, and approximately 20% of patients with melanoma or pancreatic carcinoma treated with CP-870,893,0 (Pfiser/VLST), a CD40 agonist mAb (43), responded. CD40 agonist Chi Lob 7/4 (University of Southampton) was tested in lymphomas and advanced solid tumors, but stable disease was found to be the best response (43). With the success of immunotherapy with anti-PD1 and anti-CTLA4, as well as novel checkpoint targets in development, our results recommend specifically examining the tumor immune microenvironment prior to administration, as it may be possible to understand why some patients fail to respond, to predict the efficacy of specific immunotherapeutic drugs, to determine which single or combination of immunotherapeutics will be most effective in specific patients, and which may exacerbate conditions and allow tumor escape.

No potential conflicts of interest were disclosed.

Conception and design: M.A. Podolsky, A.J. Gunderson, A.B. Glick

Development of methodology: M.A. Podolsky, A.J. Gunderson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.A. Podolsky, J.T. Bailey, A.J. Gunderson, C.J. Oakes, K. Breech

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.A. Podolsky, J.T. Bailey, A.J. Gunderson, C.J. Oakes, K. Breech, A.B. Glick

Writing, review, and/or revision of the manuscript: M.A. Podolsky, A.J. Gunderson, A.B. Glick

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.A. Podolsky

Study supervision: M.A. Podolsky, A.B. Glick

We thank the Huck Institutes of the Life Sciences Microscopy and Cytometry Core Facilities, the Elsa U. Pardee Foundation for financial support, and Genentech for donation of CD20 mAbs. Michael Podolsky was supported through NIH T32 training grant “Animal Models of Inflammation” and The College of Agricultural Sciences, The Pennsylvania State University.

This work is supported by funds from the NIH T32 Training Grant to M.A. Podolsky, The Pennsylvania State University College of Agricultural Sciences to M.A. Podolsky and A.B. Glick, and The Elsa U. Pardee Foundation.

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.