PPARγ Contributes to Immunity Induced by Cancer Cell Vaccines That Secrete GM-CSF

PPARγ is a widely expressed nuclear receptor that is responsive to multiple lipid ligands. Immune functions of PPARγ were revealed through studies of peritoneal macrophages, where synthetic agonists suppressed NF-κB–mediated increases in proinflammatory factors (1, 2). Further work demonstrated that PPARγ stimulation in myeloid cells induced upregulation of arginase-1 and promoted T helper type 2 (Th2) responses (3, 4). Mice with selective PPARγ deficiency in myeloid cells (LysM-Cre:PPARγ^fl/fl) show impaired tolerance to self-antigens that reflects, at least in part, the reduced phagocytosis of apoptotic cells by macrophages (5). Ly6C^hi monocytes upregulate PPARγ, which contributes to the resolution of acute inflammatory reactions triggered by LPS or thioglycollate (6). PPARγ similarly functions in dendritic cells (DCs), to maintain tolerance through the reduced activation of effector T cells (7–9) and the preferential generation of regulatory T cells (Tregs; ref. 10).

Homeostatic expression of PPARγ in myeloid cells is restricted to splenic and alveolar macrophages (6). PPARγ is required for Granulocyte–macrophage colony stimulating factor (GM-CSF)–driven differentiation of alveolar macrophages from fetal monocytes. Such differentiation is critical for surfactant homeostasis (11). Reduced function of GM-CSF, the α or β chain of the GM-CSF receptor, or PPARγ all cause pulmonary alveolar proteinosis, a disease characterized by surfactant accumulation and pulmonary inflammation (12). These data suggest a functional link between GM-CSF and PPARγ.

GM-CSF has complex roles in immune tolerance and protection that are mediated through several myeloid cell populations. GM-CSF–knockout mice display reduced uptake of apoptotic cells by mononuclear phagocytes, diminished tolerance to self-antigens, and are susceptible to autoimmune diseases, including systemic lupus erythematosus. Similar phenotypes are observed in LysM-Cre:PPARγ^fl/fl mice (5, 13). GM-CSF induces myeloid cells to release TGFβ and IL10 (14) and express coinhibitory receptors such as PD-L1 (15) that together lead to the expansion of Tregs (15, 16). In gut-resident DCs, GM-CSF stimulates production of retinoic acid that supports intestinal Treg differentiation (16). GM-CSF–secretion by cancer cells has been linked to myeloid-derived suppressor cells (17). However, high concentrations of GM-CSF are also generated during acute infections, where it contributes to DC licensing (18) and protective immunity against several pathogens (19). GM-CSF expression in CCR2⁺monocytes and monocyte-derived DCs is required for the pathogenesis of experimental autoimmune encephalitis in murine models of multiple sclerosis (20).

Investigators have tried to exploit the proinflammatory activities of GM-CSF for cancer vaccination. In one strategy, GM-CSF–secreting tumor-cell vaccine (GVAX), tumors are genetically engineered to express GM-CSF, irradiated, and administered subcutaneously to elicit specific and long-lasting protective tumor immunity. Studies of GVAX in murine models and phase I trials in patients with hematologic or solid malignancies have demonstrated that this immunization scheme enhances antitumor antibody and T-cell responses (21). Nonetheless, the limited overall clinical benefit has motivated efforts to improve vaccines based on GM-CSF through attenuating the tolerogenic functions of the cytokine. For example, GM-CSF induces the expression of milk fat globule EGF factor 8 (MFG-E8), a macrophage-derived opsonin that promotes the phagocytosis of apoptotic cells and thereby contributes to tolerance. Coexpression of a dominant-negative mutant form of MFG-E8 with GM-CSF improves vaccine efficacy through attenuating Treg generation (22).

Given the induction of PPARγ by GM-CSF and the similar functions of PPARγ and GM-CSF in maintaining tolerance, we predicted that PPARγ blockade would improve GVAX potency. To explore this idea, we characterized GVAX in mice harboring a deficiency of PPARγ in LysM⁺ myeloid cells. We found that the loss of PPARγ resulted in diminished tumor protection and a reduced ratio of CD8⁺ T effectors to Tregs in vaccinated mice. Mechanistic studies revealed impaired DC responses and increased production of Treg-attracting chemokines in the KO mice. Correspondingly, PPARγ agonists diminished Treg responses in murine and human systems and intensified the potency of GVAX in combination with anti-cytotoxic T lymphocyte–associated antigen-4 (CTLA-4) in two transplantable tumor models.

LysM-Cre (The Jackson Laboratory, 4781) and PPARγ^fl (The Jackson Laboratory, 4584) mice on a C57BL/6 background were interbred to generate viable homozygous LysM-Cre:PPARγ^fl/fl animals. PPARγ deletion was confirmed by PCR for DNA analysis and Western blotting for analysis of PPARγ protein expression. The parental strains were maintained in-house and used as controls. For therapeutic studies, wild-type female C57BL/6J mice were obtained from The Jackson Laboratory at 5 weeks of age and used at 9 to 12 weeks of age.

B16 melanoma and Lewis lung carcinoma cell lines were purchased from ATCC and cultured in ATCC-recommended media for up to several weeks per frozen aliquot. A pMFG-GM-CSF retroviral vector was used to infect both cell lines to create derivatives constitutively expressing GM-CSF, as previously described (23). For prophylactic vaccination experiments, 1 × 10⁶ GM-CSF expressing B16 cells were irradiated and injected into the flank of female mice; 5 to 9 days later, 5 × 10⁵ live parental B16 cells were injected in the contralateral flank. For therapeutic vaccination, the dose of tumor challenge was reduced to 1 × 10⁵ cells. Vaccines were administered 1 day after tumor challenge at a dose of 3 × 10⁶ cells. For combinatorial immunotherapy in the B16 model, mice were injected i.p. with monoclonal antibody (mAb) to CTLA-4 (9D9, Bio X Cell) or isotype control as follows: 200 μg on day 1 (after tumor challenge, same day as GVAX), 100 μg on days 4 and 7. Rosiglitazone (20 mg/kg/day) or DMSO were given in drinking water for 12 days starting on the day of tumor challenge. For combinatorial immunotherapy in the LLC model, mice were injected i.p. with anti–CTLA-4 or isotype control every 3 days for a total of 7 doses. Rosiglitazone or DMSO were given in drinking water for 21 days.

Draining lymph nodes (dLN) were harvested 5 days after vaccination; mice were not challenged with live tumor. dLNs from 4 mice were pooled, and RNA was extracted and subjected to HiSeq as described below (Center for Cancer Computational Biology, Dana-Farber Cancer Institute). RNA was isolated from 2 independent cohorts for controls and 3 independent cohorts for KO mice. RNA quantity was determined on the Qubit using the Qubit RNA Assay Kit (Life Tech), and RNA quality was determined on the Bioanalyzer using the RNA Pico Kit (Agilent). Using the NEBNext Ultra RNA Library Prep Kit for Illumina (NEB), which selects for the poly(A) tail of mRNA, we converted 100 ng of total RNA into a cDNA library following the manufacturer's protocol with no modifications. Library quantity was determined using the Qubit High Sensitivity DNA Kit (Life Tech), and library size was determined using the Bioanalyzer High Sensitivity Chip Kit (Agilent). Finally, libraries were put through qPCR using the Universal Library Quantification Kit for Illumina (Kapa Biosystems) and run on the 7900HT Fast qPCR machine (ABI). Libraries passing quality control were diluted to 2 nmol/L using sterile water and then sequenced on the HiSeq 2000 (Illumina) at a final concentration of 12 pmol/L, following all manufacturer protocols. Reads were mapped to the human genome build hg19 using TopHat. The mapped reads were assembled and annotated using Cufflinks software tools. Transcript abundances were quantified in fragments per kilobase of exon per million fragments mapped by Cufflinks by taking into account both the gene length and the mapped reads for each sample and normalized accordingly. Differential gene expression analysis was performed using the limma software package.

GSEA was performed using expert-curated gene sets in the canonical pathways (C2) to examine whether the set of genes involved in certain pathways is statistically different in one condition versus another. GSEA was also performed using all available gene sets in the Immgen database (∼300 at the time of analysis) to identify modules and associated cell types whose signature were differentially represented in control or KO dLNs.

Tumors were harvested, weighed, chopped into 1 to 3 mm pieces, and incubated with 200 units of Collagenase IV and 10 μg/mL DNAse for 45 to 75 minutes at 37°C. After incubation, the tissue was reduced to a cell suspension by repeated pipetting and the suspension strained with a 70-μm sieve. A gradient for centrifugation was generated using Optiprep (Sigma-Aldrich). A solution (25 mL) containing 0.85% NaCl and 10 mmol/L Tricine in distilled water was mixed with another 5 mL of distilled water and 8.71 mL of Optiprep. This gradient was layered under the tumor cell suspension and spun at 400 g for 25 minutes at room temperature with slow deceleration. The interface was collected and analyzed for flow cytometry or used for coculture.

For flow cytometry for T cells, dLNs were mechanically crushed and filtered through a 70-μm strainer. For DC staining or cultures, dLNs were chopped into 1 to 3 mm pieces and incubated in media containing 100 units Collagenase IV and 10 μg/mL DNAse for 25 to 45 minutes at 37°C. After incubation, the tissue was mechanically digested and strained with a 70-μm sieve. For mixed leukocyte reactions, 250,000 CFSE-labeled Balb/c splenocytes from mechanically digested spleens were coincubated for 4 days with 30,000 CD207⁺ cells pooled from dLNs of 3 to 4 vaccinated mice. Two to four wells were assessed for CFSE dilution for control and KO in two independent experiments.

Human PBMCs were obtained by gradient centrifugation of leukapheresis collars from platelet donors. PBMCs (4 × 10⁶) were plated with 10⁵ irradiated K562-WT or K562-GM-CSF–secreting cells. On days 4 to 6 of culture, floating cells were harvested and adherent cells were detached with 2 mmol/L EDTA. Cells were stained for flow cytometry in the presence of 1 mmol/L EDTA. Dead cells were discriminated by using the Live/Dead Fixable dyes from Invitrogen. For monocyte produced cytokines, 10⁶ monocytes from PBMCs (CD14⁺ magnetic bead isolation, Miltenyi Biotec) were plated with 10⁵ irradiated K562-WT or K562-GM-CSF–secreting cells. Human and mouse CCL22 and CCL17 levels were measured using ELISAs (R&D Systems). FoxP3 expression was detected by using anti-FoxP3 (clone 236A/E7). For Treg suppression assays, CD25^hi cells were collected by flow sorting T cells obtained by negative selection (Miltenyi) and incubated with CD8⁺ or CD4⁺ T cells cultured with activating beads (Miltenyi, 130-092-909), according to the manufacturer's recommendations.

Rosiglitazone (Adipogen) was resuspended in DMSO and 10 μmol/L rosiglitazone or equal volume of DMSO was used on days 0 and 2 of culture.

Tumor incidence was tested in two ways: tempo of tumor incidence was compared between groups using the log-rank test, and the proportion of mice developing tumors by the end of the experiment was assessed between groups using the Fisher exact test. Overall survival was assessed using the log-rank test. Measurements of cytokines and chemokines between groups (such as CCL22 and CCL17), as well as Tregs and other T-cell measures, were compared using the Student t test unless homogeneity of variance did not hold; in that case, the Welch t test (for very small sample sizes) or the Wilcoxon rank sum test were used. In experiments where different treatments were applied to cells, with each donor source treated in each condition, paired t test (one-sample t test) was used to compare treatments, as it permits comparisons within donor source. For all tests, a two-sided nominal P value < 0.05 was considered statistically significant.

We first tested the impact of PPARγ deficiency in myeloid cells using the well-characterized B16 melanoma model that is syngeneic to C57BL/6 mice. Consistent with previous reports, LysM-Cre:PPARγ^fl/fl (KO) mice displayed an approximate 90% reduction in PPARγ expression, as assessed with immunoblotting of cultured peritoneal macrophages (Supplementary Fig. S1A). For the tumor experiments, LysM-Cre and PPARγ^fl mice were used as controls and manifested similar results to wild-type C57BL/6 mice. The subcutaneous implantation of B16 cells resulted in comparable tumor growth in PPARγ-deficient and control mice, establishing that PPARγ expression in LysM⁺ myeloid cells does not affect primary tumor growth in this model (Fig. 1A and B). Next, B16 cells transduced with a retroviral expression vector encoding GM-CSF were irradiated and injected subcutaneously as a prophylactic vaccine; 7 days later, live wild-type B16 cells were implanted subcutaneously at a distal site. Vaccination enhanced tumor rejection in control mice, as previously published (23), but tumor incidence was unexpectedly increased in vaccinated PPARγ-deficient mice (Fig. 1A). Differences from controls were evident within the first week after tumor inoculation and persisted for the duration of the experiment (2 months post tumor challenge). The impaired tumor protection in vaccinated PPARγ-deficient mice was observed in five independent experiments and resulted in reduced survival compared with controls (Fig. 1B).

Diminished vaccine efficacy in the PPARγ-deficient mice was not predicted by published studies. To dissect the underlying mechanisms, we first examined myeloid cell populations recruited to the immunization site with flow cytometry. To increase the number of cells available for characterization, we used live B16-GM-CSF cells that elicit a dense myeloid cell infiltrate. Prior studies had demonstrated many similarities between the infiltrates stimulated with live and irradiated, GM-CSF–secreting tumor cells (21), but we recognize that some differences might be associated with the two systems. A comparison of live versus irradiated cell reactions would be useful, but such analysis would be hampered by the inability to extract sufficient numbers of cells from irradiated tumor cell injections to perform detailed flow cytometry. The total number of isolated hematopoietic cells at live, GM-CSF–secreting B16 tumor cell implantation sites and the frequencies of cells expressing CD11b, CD11c, and Gr-1 were comparable between control and PPARγ-deficient mice (Supplementary Fig. S1B). Further, no changes were observed in the expression of various surface molecules (MHC class II, CD80, CD86, PD-L1, CD1d, and CD103), Ly6C expressing monocytic subsets that can be associated with immunosuppression (Supplementary Fig. S1C), and the activity of retinaldehyde dehydrogenase, an enzyme linked to Treg induction.

Because our candidate selection approach did not identify key differences underlying the impaired vaccine response, we next pursued an unbiased search for defects by performing RNASeq on the draining lymph nodes (dLN), where immune priming was likely to occur. For these studies, dLNs were harvested five days after vaccination with irradiated, GM-CSF–secreting B16 cells (mice were not challenged with live tumors). dLNs from four control and four PPARγ-deficient mice were pooled, and extracted RNA was subjected to HiSeq and gene set enrichment analysis (GSEA). RNA was isolated from 2 independent control and 3 independent PPARγ deficient cohorts. Canonical PPARγ regulated genes as defined in the KEGG module “PPAR signaling pathway” (24) were reduced in the knockout mice, despite the minor myeloid cell representation within the whole lymph node (Supplementary Fig. S2A). Moreover, a gene expression module regulated by PPARγ and expressed in alveolar macrophages (6) was similarly downregulated in the dLNs from KO mice, confirming the loss of PPARγ function (Fig. 1C).

Transcriptional changes in Treg-associated gene modules detected in the dLNs of PPARγ-deficient mice compared to controls affected IL2RA (CD25), FoxP3, CTLA-4, and TIGIT (Fig. 1D). To confirm that these transcriptional changes corresponded to differences in Treg cell populations, we conducted flow cytometric analysis of dLNs from vaccinated control and KO mice. In accordance with the RNASeq data, Tregs were significantly increased in the dLNs of immunized PPARγ-deficient mice compared to controls, whereas the frequency of CD8⁺ T effectors was diminished, resulting in a lower CD8/Treg ratio in the knockouts (Fig. 1E). CD8⁺ T cells in dLNs did not show alterations in IL2 and IFNγ production after stimulation in vitro (data not shown).

Because the CD8/Treg ratio is linked to immune-mediated tumor destruction (25), we investigated whether PPARγ deficiency affected these cell populations at tumor sites using a therapeutic vaccination model. Although GVAX administered to mice with established tumors has only a modest effect on tumor growth, this experimental design allows for a better assessment of tumor-infiltrating lymphocytes. As anticipated from the studies of prophylactic vaccinations, tumors tended to be larger in the PPARγ-deficient mice compared with controls, but at early time points these differences were modest (Supplementary Fig. S2B). Although the total numbers of tumor-infiltrating leukocytes were comparable between vaccinated control and KO mice (Supplementary Fig. S2C), the fraction of CD3⁺ T cells within the infiltrate was significantly reduced in the KOs (Supplementary Fig. S2D). Moreover, tumor infiltrates in PPARγ-deficient mice also showed increased Tregs and a trend toward reduced CD8⁺ T effectors, culminating in a significant reduction in the CD8/Treg ratio (Fig. 1F). Taken together, these results indicate that PPARγ functions in myeloid cells to attenuate GVAX-induced Treg responses.

To test whether PPARγ similarly influences the ability of GM-CSF to stimulate Tregs in humans, we developed an in vitro model using human PBMCs (Fig. 2A). In this system, PBMCs were cultured with autologous monocytes (obtained by prior adherence of PBMC aliquots) and either irradiated wild-type or GM-CSF–secreting K562 tumor cells (Fig. 2A). GM-CSF acts on CD14⁺ monocytes to promote their expansion and differentiation, and after 5 days of culture, total CD14⁺ numbers were approximately 5-fold greater in the presence of GM-CSF–secreting K562 cells compared with parental K562 cells (Supplementary Fig. S3). However, monocyte expression of HLA-DR was reduced in the presence of GM-CSF (Supplementary Fig. S3). CD4⁺CD25^hiCD127^loFoxP3^hi Tregs were increased in the cultures containing GM-CSF (Fig. 2B) and suppressed activation of naïve CD8⁺ and CD4⁺ T cells, as measured with CFSE dilution (Fig. 2C).

The addition of rosiglitazone, a low-molecular-weight PPARγ agonist, to the mixed cultures reduced the number of Tregs (Fig. 2D), demonstrating that the role of PPARγ in modulating GM-CSF induced Treg expansion is conserved from mouse to man. Nonetheless, rosiglitazone did not affect the number of CD14⁺ cells nor the expression of HLA-DR, CD1c, and CD40 on monocytes, irrespective of GM-CSF levels (Supplementary Fig. S3). These results suggest that selective effects of PPARγ on myeloid cell function might be linked to Treg induction, consistent with evidence that monocytes from LysM-Cre:PPARγ^fl/fl mice manifest defects that compromise the resolution of acute muscle inflammation without alterations in surface markers (26).

To explore the role of PPARγ in modulating GVAX induced Treg responses in more detail, we searched the dLN gene expression data from the vaccinated mice for evidence of additional myeloid defects. Changes in multiple myeloid populations were observed in the PPARγ-deficient mice, including the PPARγ-associated macrophage module (Fig. 1C) and Immgen module 165 (ref. 27; Fig. 3A). Canonical macrophage-related genes such as CD163 (Fig. 1C), CD169, and MerTK (and related family member Axl, Fig. 3A) were downregulated in the KOs. Changes were also seen in DC profiles, exemplified by the increased expression of Immgen module 26 (Fig. 3B) with upregulation of zbtb46 (28), a transcription factor that supports DC development, DNGR-1 (clec9A), a scavenger receptor involved in the uptake of necrotic cells (29), and IL12β, which is repressed by PPARγ (30). SOCS2 and CCL22, genes expressed in skin migratory DCs and tumor-resident immunosuppressed DCs (31), were also upregulated in the dLN RNASeq.

Migratory DCs maintain Treg numbers in peripheral lymph nodes (32, 33). Skin-resident DCs migrate to the dLN within 12 to 24 hours of antigen challenge and may be followed by a population of monocyte-derived cells that drain to the lymph nodes at later points in a CCR7⁺-dependent fashion (34). Inflammation driven by monocyte-derived DCs depends on GM-CSF (20). Therefore, we hypothesized that alterations in the DCs present in the dLN might contribute to reduced GVAX efficacy. Flow cytometric analysis showed that most CD11c⁺ cells in the dLNs were MHC class II^hiCCR7⁺, suggesting that they were migratory DC. However, we did not observe any changes in total DC numbers or subpopulations in the dLNs of PPARγ-deficient mice (Supplementary Fig. S4). To evaluate potential functional defects, we sorted CD207⁺ (langerin⁺) DCs from the dLNs and assessed their ability to stimulate proliferation of allogeneic T cells. Langerin⁺ DCs isolated from vaccinated PPARγ-deficient dLNs elicited weaker allogeneic responses compared with controls (Fig. 3C and D), suggesting that impaired vaccination responses reflect altered DC activation.

CCL22 expression by migratory DCs has been linked to the accumulation of Tregs in dLNs through engagement of the cognate receptor CCR4, which also responds to CCL17 (33). Monocyte-derived DCs generated with GM-CSF produce CCL22 (22) and CCL17 (20), and RNASeq analysis of dLNs revealed their expression in vaccinated mice. Consistent with these findings, culture supernatants of dLNs from immunized PPARγ-deficient mice showed higher levels of CCL22 and CCL17 compared with controls, as assessed with an ELISA (Fig. 4A). Correspondingly, activation of PPARγ signaling with rosiglitazone in human PBMC-derived monocytes diminished the production of CCL17 and CCL22 in the presence of irradiated, GM-CSF–secreting K562 cells (Fig. 4B). These studies suggest that PPARγ attenuates GM-CSF–driven Treg responses in part through a reduction of Treg-recruiting chemokines.

Although earlier work suggested that PPARγ agonists may act as direct cytotoxic agents for some cancers (35), our results raised the possibility that rosiglitazone might enhance the potency of GVAX-based immunotherapy. In accordance with the results of the loss-of-function studies, the administration of rosiglitazone (20 mg/kg in drinking water) to wild-type mice augmented the immunostimulatory effects of GVAX delivered therapeutically (after tumor challenge), resulting in an increased ratio of CD8⁺ T cells to Tregs in tumor infiltrates (Fig. 5A). Moreover, PPARγ expression in myeloid cells was required for this activity, as the CD8⁺ T-cell to Treg ratio was not altered when rosiglitazone was administered to LysM-Cre:PPARγ^fl/fl mice. Thus, even though PPARγ functions in multiple cell types, including Tregs resident in adipose tissue (36), that require the transcription factor for survival and differentiation, myeloid cells are the targets in the context of GVAX. Nonetheless, despite the improved CD8⁺ T-cell to Treg ratio, rosiglitazone treatment did not increase the rejection of established tumors with GVAX, suggesting that additional immune defects remain to be addressed.

One strategy that intensifies the activity of GVAX is CTLA-4 antibody blockade, which both enhances T effector cell function and promotes the depletion of intratumoral Tregs (37). The addition of rosiglitazone augmented the therapeutic potency of the GVAX/anti-CTLA-4 combination, resulting in reduced growth of established B16 tumors and improved survival (Fig. 5B), as compared with the double combination of GVAX and CTLA-4 antibody blockade. To examine the generality of these findings, we tested this triple combination therapy against established tumors in the Lewis lung carcinoma (LLC) model. The addition of rosiglitazone to GVAX and anti-CTLA-4 significantly reduced the incidence of tumor ulceration (38) and prolonged survival in this system as well (Fig. 5C). Together, these data establish a role for PPARγ agonists in potentiating the response to GM-CSF–based cancer immunotherapy.

These experiments were undertaken to learn more about the role of PPARγ in GM-CSF–stimulated immune responses. Prior studies of pulmonary alveolar proteinosis had linked PPARγ and GM-CSF in alveolar monocyte/macrophage differentiation, surfactant catabolism, and pulmonary homeostasis (12). Lipid ligands trigger PPARγ signaling to promote immune tolerance through macrophages (1–3, 5), DCs (7–9), and Tregs (36). These properties raised the possibility that PPARγ might skew GM-CSF function toward tolerance and away from protective immunity. We found that PPARγ in myeloid cells supports the immunostimulatory effects of irradiated, GM-CSF–secreting tumor cell vaccines. Mice harboring a selective deficiency of PPARγ in LysM⁺ cells showed impaired tumor rejection after GVAX that was associated with a decreased ratio of CD8⁺ T effectors to Tregs in dLNs and tumor infiltrates. Mechanistic analysis revealed alterations in myeloid cell expression profiles, diminished allostimulatory activity of migratory DCs, and enhanced production of Treg-attracting chemokines. Correspondingly, administration of the PPARγ agonist rosiglitazone augmented the ratio of CD8⁺ T effectors to Tregs in a myeloid cell–dependent fashion, decreased CCL17 and CCL22 production, and augmented tumor destruction in combination with GVAX and CTLA-4 antibody blockade in two therapeutic models where treatment was started after tumor challenge.

The ratio of CD8⁺ T effectors to Tregs in tumor infiltrates has emerged as an indicator of immune-mediated tumor destruction in experimental models and clinical trials (25). Our prior studies had demonstrated that GM-CSF stimulates Treg responses in part through MFG-E8–mediated phagocytosis of apoptotic cells by macrophages and DCs, which promotes CCL22 and TGFβ production (22). Because PPARγ similarly contributes to the uptake of apoptotic cells (5) and is required for adipose Treg fitness (36), the finding that Treg responses are increased in vaccinated PPARγ-deficient mice is surprising. One potential mechanism may be a requirement for PPARγ in full DC maturation. In this regard, fatty acid metabolism in conjunction with glycolysis promotes DC survival and function after TLR ligation (39); lipids may contribute to the expansion of the secretory apparatus that is necessary for optimal cytokine and costimulatory molecule expression. Nonetheless, elevated PPARγ expression is also found in tumor-associated macrophages from non–small cell lung cancer patients, which manifest an immunosuppressive profile (40). These divergent results indicate that PPARγ function depends on context in myeloid cells. Additional studies are required to clarify the signals that determine whether PPARγ contributes to immune activation or tolerance.

GM-CSF may act upon multiple LysM⁺ myeloid cells including monocytes/macrophages, DCs, and granulocytes. The RNASeq analysis of dLNs showed alterations in both DCs and macrophages in the vaccinated PPARγ-deficient mice. Increases were observed in several gene products including SOCS2 and CCL22 that have been linked to a profile of migratory DCs that dampen vaccine responses (31). In this context, studies on DCs implicated GM-CSF as a maturation factor for Langerhans cells (41), and our results suggest that PPARγ signaling contributes to the immunostimulatory effects of migratory langerin⁺ cells. Naïve CD207⁺ cells express modest levels of LysM (Immgen.org), and LysM-Cre has been used to target skin resident DCs (42–44). Future studies to determine whether LysM-Cre–mediated deletion of PPARγ causes cell intrinsic defects in langerin⁺ DCs should provide additional mechanistic insights. Under inflammatory conditions, monocytes can also give rise to DCs in skin and dLNs, and LysM-Cre–mediated deletion of PPARγ could affect this “second generation” of DCs (34). The DC defects in the vaccinated mice might also result from cross-talk with other LysM-Cre–expressing cells, such as macrophages that are impacted by the loss of PPARγ.

Several studies have delineated a role for myeloid cell PPARγ expression in inducing Th2-type inflammation. Mice that develop pulmonary alveolar proteinosis due to a deficiency in PPARγ display enhanced Th1-type inflammation (45). The response of LysM-Cre:PPARγ^fl/fl animals to Leishmania infection similarly involves Th1 skewing (3). A Th2 profile is often accompanied by an alternative activation program in macrophages with high levels of fatty acid oxidation (46). We detected a modest reduction in the mRNA expression of fatty acid synthase in the dLNs of vaccinated PPARγ-deficient mice (P < 0.001), but cytokine profiles in lymph node and tumor-infiltrating T cells were not altered compared with controls. In human PBMCs, PPARγ has also been linked to CD1d-mediated lipid antigen presentation to invariant NKT cells (47), but we did not find defects in this pathway in the vaccinated mice. Differences in the regulation of CD1 family members between mice and humans might account for the lack of NKT-cell alterations in the murine tumor model.

Prior investigations have proposed the use of PPARγ agonists as cancer therapy. Rosiglitazone may enhance the potency of platinum based chemotherapy through modulating metallothionein levels (48), and diminish myeloid derived suppressor cell induced immunosuppression (49) in combination with gemcitabine and phospholipase A inhibition (50). Here, we identified changes in vaccine-induced antitumor responses that are dependent on the expression of PPARγ in LysM-expressing cells. Gain- and loss-of-function experiments in murine and human models have helped to validate PPARγ as a target for cancer immunotherapy. The availability of rosiglitazone as an FDA-approved agent for diabetes management should facilitate the clinical evaluation of PPARγ agonists in combination with GM-CSF–based cancer vaccine strategies.

G. Dranoff is the Global Head of Exploratory Immuno-oncology at and has ownership interest in Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G. Goyal, D. Neuberg, N. Anandasabapathy, G. Dranoff

Development of methodology: G. Goyal, K. Wong, C.J. Nirschl, N. Souders, G. Dranoff

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Goyal, K. Wong, C.J. Nirschl, N. Souders,

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Goyal, K. Wong, N. Souders, D. Neuberg, N. Anandasabapathy, G. Dranoff

Writing, review, and/or revision of the manuscript: G. Goyal, D. Neuberg, G. Dranoff

Study supervision: G. Goyal, G. Dranoff

This study was supported by NCI R01CA143083.We acknowledge Dr. Stephanie Dougan for reagents and Dr. James Akin for experimental assistance. Data assembled by the Immgen Consortium were used extensively in this work.

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.