GM-CSF Promotes Antitumor Immunity by Inducing Th9 Cell Responses

Tumor-specific CD4⁺ T cells trigger tumor immune surveillance by providing costimulation and cytokines required for breaking CD8⁺ T-cell tolerance (1, 2). Various studies have attempted to find MHC II–restricted tumor epitopes (3) and have adopted tumor-specific CD4⁺ T cells for the treatment of cancer (4, 5). Among Th cell subsets, Th1 cells have been considered the most efficacious in rejecting tumors directly or indirectly through the mechanism of activating macrophages and CTLs (3, 6). The role of Th2 cells in tumor immunity is controversial, although Th2 cells are known to exert antitumor effects by activating eosinophils, B cells, and natural killer cells in several models (3, 7, 8). In contrast, regulatory T (Treg) cells create an immunosuppressive microenvironment promoting cancer development (9). Th17 cells inhibit tumors by acquiring a Th1-like phenotype in vivo (10). A distinct Th subset producing IL9 (Th9 cells) has been found to play a role in autoimmunity in the lung, central nervous system, and gut (11, 12). Several cytokines and transcription factors are associated with the development of Th9 cells (11, 12), but a master transcription factor for Th9 cells has yet to be identified. Th9 cells are more potent than other Th subsets in melanoma rejection via mast cells (13) or via activation of dendritic cells (DC) to cross-present tumor antigens (14, 15). A hyperproliferative feature mediated by the PU.1–TRAF6–NF-κB pathway confers persistence on Th9 cells with cytotoxicity directed by Eomes and granzymes in the tumor microenvironment (16). Thus, boosting Th9 responses may be a useful strategy for the treatment of cancer; this strategy is under investigation (11, 15, 17, 18).

GM-CSF was identified as a hematopoietic growth factor causing granulocyte and macrophage colony formation. It is used for DC generation from bone marrow cells or monocytes in vitro. At steady state, however, GM-CSF- or GM-CSFR–deficient mice exhibit no defects in the development of myeloid cells, with the exception of alveolar macrophages and specific DC subsets in nonlymphoid tissues (19–21). Instead, GM-CSF plays a role in tissue inflammation and autoimmune diseases, including rheumatoid arthritis, multiple sclerosis, asthma, psoriasis, and type I diabetes. Blockade of GM-CSF is in clinical trials for the treatment of rheumatoid arthritis and multiple sclerosis (21–23). GM-CSF is mainly produced by T cells, B cells, epithelial cells, and fibroblasts upon activation (22, 23). During inflammation, upregulated GM-CSF acts on macrophages and DCs to generate additional proinflammatory cytokines, which, in turn, affect Th responses (21, 24). Indeed, GM-CSF controls the development and/or function of Th1 and Th17 cells in a mouse model of multiple sclerosis (25–27) and also controls Th2 cell function in mouse asthma models (28, 29) in a context-dependent manner.

In tumor immunity, this immune-stimulating activity of GM-CSF to self in autoimmune diseases has been exploited by using GM-CSF as an adjuvant to inhibit tumors. Irradiated B16 melanoma cells transduced with the GM-CSF gene were used as a cancer vaccine and elicited potent antitumor immune responses in mice (30). On the basis of this finding, autologous cancer vaccines engineered to secrete GM-CSF (GVAX) and tumor peptide and DC vaccines with recombinant GM-CSF have been clinically tested (23, 31). Some trials demonstrated increased immunogenicity and phagocyte and lymphocyte infiltration into the vaccination sites, whereas other studies showed minimal or no effects. One potential explanation for the poor outcomes is that GM-CSF induces the recruitment and expansion of myeloid-derived suppressor cells (MDSC; ref. 32). Alternatively, oncolytic GM-CSF–encoding modified herpes simplex virus (T-VEC) and combination with checkpoint blockades have shown improved outcomes and are under clinical investigation (33–35).

GM-CSF adjuvant augments the recruitment and activation of DCs, which likely help eradicate cancer through the presentation of tumor antigens to T cells. The function of GM-CSF as an adjuvant was supported by the fact that the enhanced antitumor activity by GM-CSF disappeared when CD4⁺ or CD8⁺ T cells were depleted (30). However, the types of Th responses that contribute to this effect remain unclear. In addition, because GM-CSFR is broadly expressed by myeloid cells (22, 23), different myeloid cell types could be involved in this process. Ly6C^hiCCR2⁺ monocytes and monocyte-derived dendritic cells (moDCs) mediate GM-CSF–induced pathogenesis under tissue inflammation (27, 36). Thus, the role of this population in tumor immunity induced by GM-CSF needs to be defined.

In this study, we dissected Th responses in tumor immunity when GM-CSF was administered as an adjuvant and found that GM-CSF selectively induced Th9 development, particularly by acting on moDCs, leading to the promotion of tumor-specific CTL responses that inhibited tumor growth in an IL9-dependent manner.

Male and female C57BL/6n, BALB/c mice were purchased from Charles River Laboratories. C57BL/6j, OT-II, DO11.10, and Ccr2^–/– were from The Jackson Laboratory. CD45.1 congenic mice were kindly provided by Jae-Ouk Kim (International Vaccine Institute, Seoul, Republic of Korea). Mice were used at 6–10 weeks of age and gender-matched in all experiments; mice were bred at Seoul National University (Seoul, Republic of Korea) under specific pathogen-free conditions. All animal experiments were approved by the Institutional Animal Care and Use Committee at Seoul National University (Seoul, Republic of Korea).

A total of 10⁶ B16F10-OVA cells (kindly provided in 2006 by Kenneth Rock, University of Massachusetts Medical School, Boston, MA), 2 × 10⁵ TC-1 cells (express E6/E7 oncoproteins from HPV16; ATCC in 2006), or 3 × 10⁵ CT26-hHER2/neu cells (ATCC in 2002 and transduced hHER2/neu in 2004) negative for Mycoplasma contamination (e-Myco Mycoplasma PCR Detection Kit; iNtRON in 2016) were subcutaneously injected into the flank of mice and tumor-bearing mice were randomly allocated to control or treatment groups in all experiments. Tumor cell lines were validated by their morphology, growth kinetics, and antigen expression before injection and tumor cells with three to four passages were used for inoculation. Sample sizes were determined by Resource Equation method. On day 4 after tumor inoculation, some mice were intraperitoneally treated with cognate 50 μg of OVA (Sigma-Aldrich), total 100 μg of E6/E7 peptide library, or hHER2/neu peptide library with or without 500 ng of carrier-free recombinant GM-CSF (R&D Systems) incubated for up to 3 hours in complete Freund's adjuvant (CFA; Sigma-Aldrich) at a 1:1 volume ratio. Carrier-free recombinant IL9 was from R&D Systems, and 100 ng of IL9 was applied on day 4 and 50 ng of IL9 was applied every other day from day 6 to 16. The sequences of the peptide library, excluding the kinase domain of HER2/neu and consisting of 7 or 8 overlapping peptides, were designed using the peptide library design and calculator tool from Sigma-Aldrich and ordered from GenScript. Two injections of 100 ng recombinant GM-CSF were additionally given on days 3 and 5 after tumor implantation. Tumor volume was calculated as 0.5236 × length × width × height by digital caliper. CD4⁺ T cells were sorted from tumor-draining lymph nodes (TdLN) of B16F10-OVA tumor-bearing mice and stimulated with mitomycin-treated T-cell–depleted total splenocytes (TdS) plus OVA_323-339 peptide (ISQAVHAAHAEINEAGR, 10 μg/mL; Anygen) overnight or for 3 days for mRNA or protein expression analysis, respectively. Isolation of tumor-infiltrating lymphocytes (TIL) was described previously (15).

For analyzing the cytokine-inducing capacity of moDCs and conventional DCs (cDCs), 100 μg of OVA plus CFA was intraperitoneally injected into naïve or 4-day established B16F10-OVA–bearing mice, and splenic moDCs and cDCs were isolated 8 days later and cocultured with naïve or total OVA-specific CD4⁺ T cells plus OVA_323-339 peptide (10 μg/ml) for 48 hours.

For DC vaccine experiments, 1.5 × 10⁵ moDCs or cDCs were isolated from spleens as described above, loaded overnight with 10 μg/ml OVA_323-339 peptide plus or minus recombinant GM-CSF, and intravenously injected into 8-day established B16F10-OVA–bearing mice that had received 2 × 10⁶ OVA-specific naïve CD4⁺ T cells one day prior. The equivalent number of moDCs loaded with the E6/E7 peptide library (10 μg/ml) was transferred into 5-day established TC-1–bearing mice. Tumor volume was measured three times per week. Donor CD45.2⁺Vα2⁺ Th cells were recovered 4 days after DC transfer to analyze Th cell differentiation in CD45.1⁺ tumor–bearing hosts.

For analyzing cell proliferation, 5,6-carboxyfluorescein diacetate succinimidyl ester (CFSE) or CellTrace Violet (CTV, 5 μmol/L each; Invitrogen) was used to label cells according to the manufacturer's instructions. All microbeads (anti-mCD4, mCD11b, mCD11c, hCD3, and biotin) for positive or negative selection were purchased from Miltenyi Biotec. Lipopolysaccharide (0.5 μg/ml) was from Sigma-Aldrich and G-CSF, macrophage colony-stimulating factor (M-CSF; 20 ng/mL each) were from PeproTech. Anti-OX40L (RM134L; BioLegend), anti-TL1A (293327; R&D Systems), and GITR-Fc (Alexis Biochemicals) were used at 10 μg/ml. Synthetic PGD₂ (200 nmol/L) and PGE₂ (50 nmol/L) were from Cayman Chemical.

For neutralization of GM-CSF, mice were intraperitoneally treated with 200 μg of anti-GM-CSF (MP1-22E9) every 3 days starting from day 0. For neutralization of IL9, 200 μg of anti-IL9 (MM9C1) was injected intraperitoneally every other day starting from a day before treatment with DC or peptide vaccines. For neutralization of IFNγ, 200 μg of anti-IFNγ (R4-6A2) was intraperitoneally administered every two days from day 4 to 16 after tumor inoculation. Rat IgG for GM-CSF and IFNγ blockades or mouse IgG (Sigma-Aldrich) for anti-IL9 was used as a control.

OVA-specific naïve CD4⁺CD44^loCD62L^hiCD25^– T cells were sorted from enriched CD4⁺ cells by FACSAria III (BD Biosciences). Purified 5 × 10⁴ naïve CD4⁺ T cells were cocultured with 2 × 10⁵ TdS or 1.25 × 10⁴ moDCs or cDCs plus OVA_323-339 peptide (1 μg/ml) for 3 days under specific Th-polarizing conditions. Cells were grown in RPMI1640 culture medium supplemented with 10% FBS, 2.5% HEPES, 1% sodium pyruvate, 1% nonessential amino acids, 1% penicillin/streptomycin, and 0.1% 2-mercaptoenthanol (all from Gibco by Life Technologies) in the 96-well round bottom plate. moDCs and cDCs were isolated from CD11b- and CD11c-enriched TdS. The concentration of GM-CSF was titrated and used at 20 ng/mL. Polarizing conditions for each Th subset were as follows: Th1 (4 ng/mL IL12), Th2 (10 ng/mL IL4), Th9 (5 ng/mL TGFβ plus 10 ng/mL IL4), Th17 (5 ng/mL TGFβ plus 20 ng/mL IL6), and iTreg (5 ng/mL TGFβ). All cytokines were purchased from eBioscience except TGFβ, which was from PeproTech.

FITC-conjugated antibodies to mouse CD3ϵ (145-2C11), Ly6C (HK1.4), B220 (RA3-6B2), Foxp3 (FJK-16s; Invitrogen) and human CD45RA (HI100), Foxp3 (236A/E7; Invitrogen); PE-conjugated antibodies to mouse CCR2 (475301; R&D Systems), IL13 (eBio13A; Invitrogen), IL17 (TC11-18H10.1), Vα2 (B20.1), DO11.10 TCR (KJ1-26; Invitrogen) and human IL9 (MH9A4); PerCP-Cy5.5–conjugated antibodies to mouse CD3ϵ (145-2C11), CD62L (MEL-14), and I-A/I-E (M5/114.15.2); PE/Cy7-conjugated antibodies to mouse CD44 (IM7), Ly6G (1A8), CD19 (6D5), CD4 (RM4-5) and human CD4 (RPA-T4), CD45RO (UCHL1); APC-conjugated antibodies to mouse CD4 (RM4-5), CD11c (N418), CD25 (PC61), CD45.2 (104), IFNγ (XMG1.2), IL9 (RM9A4) and human CD3 (OKT3), CD25 (BC96); and APC/Cy7-conjugated antibody to mouse CD8α (53-6.7), F4/80 (BM8) and human CD3 (OKT3), Pacific blue–conjugated antibody to mouse CD4 (RM4-5), CD11b (M1/70), Granzyme B (GB11) and human CD4 (RPA-T4) were used. All the antibodies were from BioLegend unless otherwise indicated. Cells were washed and surface molecules stained for 15 minutes at 4°C, followed by intracellular staining. For cytokine staining, cells were restimulated with phorbol 12-myristate 13-acetate (50 ng/mL; Sigma-Aldrich) and ionomycin (500 ng/mL; Sigma-Aldrich) plus GolgiPlug (BD Biosciences) for 4 hours, followed by fixation and permeabilization by Cytofix/Cytoperm Kit (BD Biosciences). For Foxp3 staining, cells were permeabilized with Foxp3 Staining Kit (eBioscience) according to the manufacturer's instructions. Samples were acquired with FACSCalibur or FACSAria III (BD Biosciences) and data were analyzed with FlowJo software (Tree Star).

The following cytokines in culture supernatants were measured by ELISA kits according to the manufacturer's instructions: IL9 (Invitrogen) and IFNγ (BD Biosciences).

Total RNA was isolated using TRIzol reagent (Invitrogen) and reverse-transcribed by SuperScript Reverse Transcriptase and oligo(dT) nucleotides (Invitrogen). Resultant cDNA was further quantified with a SYBR Green Real-Time PCR Kit (Takara) and LightCycler Optical System (Roche). The values of gene expression were normalized to the amount of Hprt1 expression. Primers used in analyses were as follows: mouse Il9 forward, 5′-AAC GTG ACC AGC TGC TTG TGT-3′; mouse Il9 reverse, 5′-CTT GAT TTC TGT GTG GCA TTG G-3′; mouse Ifng forward, 5′-ACA GCA CTC GAA TGT GTC AGG TA-3′; mouse Ifng reverse, 5′-ATT CGG GTG TAG TCA CAG TTT TCA-3′; mouse Il17a forward, 5′-CCG CAA TGA AGA CCC TGA TAG-3′; mouse Il17a reverse, 5′- TCA TGT GGT GGT CCA GCT TTC-3′; mouse Foxp3 forward, 5′-GGA TGA GCT GAC TGC AAT TCT G-3′; mouse Foxp3 reverse, 5′-GTA CCT AGC TGC CCT GCA TGA G-3′; mouse Cox-2 forward, 5′-CCC ACA GTC AAA GAC ACT CAG GTA-3′; mouse Cox-2 reverse, 5′-CCA GGC ACC AGA CCA AAG AC-3′; mouse Ptgds forward, 5′-GTT CCG GGA GAA GAA AGC TGT A-3′; mouse Ptgds reverse, 5′-CTT GGT CTC ACA CTG GTT TTT CC-3′; mouse Ptges forward, 5′-AAG CCT TTT TTC CTG CGT TTT A-3′; mouse Ptges reverse, 5′-TCT AAC TCC AGC AAC TT-3′; mouse Ptgis forward, 5′-GGC TCC TTC TTT TCC TCC TCA A-3′; mouse Ptgis reverse, 5′-CTG TGG GAG TGT GGT CAT CTG T-3′; mouse Hprt1 forward, 5′-AAG ACT TGC TCG AGA TGT CAT GAA-3′; mouse Hprt1 reverse, 5′-ATC CAG CAG GTC AGC AAA GAA-3′.

Human peripheral blood was obtained from healthy volunteers who were previously informed and provided consent to experimental procedures using their samples in accordance with the Declaration of Helsinki. Peripheral blood mononuclear cells (PBMC) were prepared by BD Vacutainer CPT tubes (BD Biosciences) according to the manufacturer's instructions. After centrifugation, naïve CD3⁺CD4⁺CD45RA⁺CD45RO^–CD25^– T cells were sorted from enriched CD3⁺ cells by flow cytometry. The mitomycin-treated allogeneic CD3-depleted cells were used as the antigen-presenting cell (APC) counterparts. For human Th9 cell differentiation, cells were cultured with soluble anti-CD3 (5 μg/ml, OKT3; BioLegend) and soluble anti-CD28 (2 μg/ml, CD28.2; BioLegend) plus hTGFβ (5 ng/mL; PeproTech), hIL4 (20 ng/mL; eBioscience), and hIL2 (20 ng/mL; eBioscience). Sorted naïve CD4⁺ T cells were cultured in the presence or absence of hGM-CSF (40 ng/mL; PeproTech) and analyzed for IL9 expression three days later. The collection of human samples and all human experiments were approved by the ethical committee of Seoul National University (Seoul, Republic of Korea; institutional review board no. 1712/001-003).

Cytotoxicity was measured as described previously (15) with some modifications. Splenocytes and TdLN cells from B16F10-OVA tumor-bearing mice were prepared on day 13 and stimulated with an MHC-I–restricted tumor epitope mix of OVA_257-264 (SIINFEKL) and gp100_25-33 (EGSRNQDWL, 1 μg/ml each; Anygen) peptides. After a 5-day stimulation, equivalent numbers of live cells were cocultured for 4 hours with ⁵¹Cr-labeled B16F10-OVA tumor cells that had been loaded for 1 hour with peptide mix for the use as targets. Specific target cell lysis was calculated as [(Sample lysis count per minute (cpm) – spontaneous lysis cpm) / (Triton X-100 lysis cpm – spontaneous lysis cpm)] × 100 (%) using Wallac 1470 Wizard Automatic γ-Counter (PerkinElmer). For intracellular staining of CD8⁺ CTLs, effector cells were additionally stimulated with the peptide mix plus GolgiPlug for 4 hours and analyzed by flow cytometry.

Cox-2 DNA was amplified using a house mouse Ptgs2 cDNA ORF clone (oMu19313, GenScript) and the following primers: mouse Cox-2 forward (KpnI); 5′- GAG CTC GGT ACC GCC ACC ATG CTC TTC-3′; mouse Cox-2 reverse (NotI), 5′-GAG GCT GAT GCG GCC GCT TAT CAC TTA TCG TC-3′. Sequencing primers were as follows: mouse Cox-2 first, 5′-TGC TGT TCC AAT CCA TGT CA-3′; mouse Cox-2 second, 5′-GGT GAA ACT CTG GAC AGA CA-3′; mouse Cox-2 third, 5′-GAG TAC CGC AAA CGC TTC TC-3′; mouse Cox-2 fourth, 5′-TGT CTG TCC AGA GTT TCA CC-3′. The PCR product was cloned into the pShuttle-CMV-EGFP-C vector (13887; Addgene) with restriction enzymes (KpnI, NotI; both from Enzynomics) and T4 ligase (NEB). Then, PmeI (NEB)-digested plasmid DNA was subjected to homologous recombination with the pAdeasy-1 adenovirus vector (Agilent Technologies) by electroporation. Final plasmid products were transfected into human embryonic kidney (HEK) 293 cells to generate AdMock-GFP or AdCox-2-GFP. A total of 10⁶ plaque-forming units of adenovirus were added to a 96-well round-bottom culture plate.

Statistical analyses were performed using Prism software (GraphPad). Unpaired two-tailed Student t test was used and results with a P value of <0.05 were considered statistically significant. Data are represented as mean ± SEM.

To investigate the mechanism of GM-CSF adjuvanticity in vivo, we employed a therapeutic cancer vaccine model using B16F10 melanoma expressing whole ovalbumin (OVA) as a model tumor antigen. Groups of C57BL/6 mice were subcutaneously injected with B16F10-OVA, and OVA emulsified in CFA was administered 4 days after tumor inoculation. Treatment with OVA plus CFA slightly inhibited tumor growth but failed to control tumors. However, when GM-CSF was added to the vaccine, its efficacy was increased and tumor growth was significantly delayed (Fig. 1A). We further analyzed tumor-specific CD4⁺ T cells in the TdLNs and found that IL9 expression was enhanced by GM-CSF treatment (Fig. 1B; Supplementary Fig. S1A). In contrast, addition of GM-CSF significantly reduced IFNγ production and showed little effects on IL17A and Foxp3 expression by CD4⁺ T cells in this setting. As reported previously (13), administration of IL9 also significantly inhibited tumor growth comparable with GM-CSF treatment in B16F10-OVA challenged mice (Supplementary Fig. S1B). This observation led us to hypothesize that IL9 might be involved in the enhanced antitumor immunity in mice given GM-CSF. To test this hypothesis, we additionally treated the mice with a neutralizing antibody to IL9 and found that anti-IL9 reversed the antitumor effect of GM-CSF (Fig. 1C). Furthermore, we verified IL9-mediated adjuvanticity of GM-CSF in therapeutic models of TC-1 cancer and CT26-HER2/neu carcinoma with a cognate peptide library vaccine (Fig. 1D and E). These results demonstrate a role of IL9 in mediating antitumor immunity induced by GM-CSF in vivo.

The observed role of IL9 in GM-CSF–induced tumor inhibition prompted us to investigate whether GM-CSF plays a role in the differentiation of Th9 cells. For this, we sorted naïve CD4⁺ T cells from OVA-specific TCR transgenic mice and stimulated them with T-cell–depleted total splenocytes (TdS) and cognate peptide under Th9 conditions. Addition of GM-CSF significantly increased the frequency of Th9 cells as early as day 3 (Fig. 2A and B). Accordingly, the amount of IL9 in the culture supernatants was significantly increased by GM-CSF treatment (Fig. 2C). The GM-CSF receptor β chain is not expressed on T cells, whereas the GM-CSF receptor α chain is known to be expressed on activated T cells (24). However, addition of GM-CSF did not increase the frequency of IL9-expressing CD4⁺ T cells stimulated by anti-CD3 plus anti-CD28 in the absence of APCs (Fig. 2D), suggesting that GM-CSF enhances Th9 cell differentiation by acting on APCs. To further investigate which cell type is primarily responsible for the increase in GM-CSF–induced Th9 differentiation, we isolated APC populations and measured their ability to induce Th9 cells in response to GM-CSF (Supplementary Fig. S2). We found that among the APC subpopulations, MHC II⁺ moDCs and cDCs were responsive to GM-CSF in enhancing Th9 differentiation (Fig. 2E). Although moDCs induced substantial numbers of Foxp3⁺ Treg cells under Th9 conditions, GM-CSF suppressed Treg cell differentiation. The increase in Th9 differentiation was possibly due to the enhanced maturation and T-cell–proliferating capacity of APCs caused by GM-CSF. However, T-cell proliferation was not affected by GM-CSF treatment and the addition of LPS rather inhibited Th9 cell development (Supplementary Fig. S3A and S3B). Unlike GM-CSF, G-CSF and M-CSF did not promote Th9 differentiation, although they enhanced IL13 expression (Supplementary Fig. S3B and S3C). In addition, we observed that GM-CSF potentiated the differentiation of human Th9 cells in the presence of APCs (Fig. 2F and G).

Next, we determined the effect of GM-CSF on the differentiation of other Th subsets. In contrast to Th9, GM-CSF had little or no effect on Th1 and Th17 differentiation when T cells were activated by TdS and cognate peptide under Th1 or Th17 conditions (Fig. 3A and B). Moreover, GM-CSF did not induce IL9 production from Th1 and Th17 cells (Supplementary Fig. S4). Under Th2 and Th9 conditions, the induction of IL9⁺, IL13⁺, and IL9⁺IL-13⁺ cells were all augmented by GM-CSF. We found that GM-CSF inhibited iTreg cell generation and produced a small but evident population of IL9⁺Foxp3^– cells (Fig. 3A and B). This effect on the differentiation of Th subsets was similar when moDCs or cDCs were used as APC counterparts except that moDCs slightly increased, whereas cDCs significantly decreased Th17 differentiation in response to GM-CSF (Supplementary Fig. S3D). Collectively, these results suggest that GM-CSF favors the differentiation of Th9 cells, but inhibits that of iTreg cells, presumably by modulating the function of moDCs and cDCs.

To determine whether GM-CSF affects moDCs and cDCs to induce Th9 cells in vivo, we immunized C57BL/6 mice with OVA plus CFA and treated them with control IgG or neutralizing anti-GM-CSF. We isolated moDCs or cDCs from the mice and analyzed their IL9-inducing capacity. Neutralization of GM-CSF significantly inhibited the capacity of moDCs to induce IL9 from naïve and total CD4⁺ T cells, whereas the IFNγ-inducing capacity was increased or unaffected (Fig. 4A and B). For cDCs, IL9 induction from total CD4⁺ T cells, but not from naïve CD4⁺ T cells, was slightly inhibited by anti-GM-CSF.

To examine whether GM-CSF directly acts on moDCs and cDCs to promote IL9-inducing capacity, we isolated moDCs and cDCs from mice immunized with OVA plus CFA, loaded the cells with OVA_323-339 peptide in the presence or absence of exogenous GM-CSF, and adoptively transferred them into recipients that had received OVA-specific naïve CD4⁺ T cells one day before. We found that GM-CSF–activated moDCs and cDCs significantly increased IL9 expression in the donor Th cells compared with GM-CSF–untreated controls (Fig. 4C). Accordingly, IL9 protein production by recovered Th cells was potentiated by GM-CSF–activated DCs, but IFNγ production was not significantly affected (Fig. 4D). Taken together, these results demonstrate that GM-CSF stimulates moDCs and cDCs that are more potent for inducing Th9 in vivo.

Next, we determined whether GM-CSF also regulates the cytokine-inducing activity of moDCs and cDCs in the tumor microenvironment. Consistently, the IL9-inducing capacity of moDCs and cDCs from B16F10-OVA tumor-bearing mice was reduced by anti-GM-CSF, whereas IFNγ induction was increased or unaffected from naïve or total CD4⁺ T cells (Fig. 5A and B). We further investigated the cellularity of myeloid populations including DCs and macrophages in the setting of cancer vaccination. Ly6C^hiCCR2⁺ monocytes and moDCs accumulated in the spleen in response to OVA plus CFA from around a week after tumor inoculation, whereas cDCs were not changed (Fig. 5C; Supplementary Fig. S5A and S5B). Vaccination with OVA plus CFA also led to the early accumulation of neutrophils and the late accumulation of macrophages during tumor progression, and GM-CSF treatment did not further accelerate accumulation of these cells (Supplementary Fig. S5A and S5B). Thus, we next determined whether CCR2⁺ monocytes contribute to GM-CSF–induced antitumor immunity. We challenged WT and CCR2–deficient mice with B16F10-OVA melanoma and treated them with OVA plus CFA with or without GM-CSF, and we found that the antitumor effect of GM-CSF disappeared in the CCR2-deficient hosts (Fig. 5D). In addition, IL9 production by tumor-specific Th cells was not augmented and IFNγ production was also not affected by GM-CSF in CCR2-deficient mice (Fig. 5E). Together, these results suggest that CCR2⁺ monocytes and moDCs play a role in GM-CSF–induced IL9 production and tumor inhibition.

To show the direct action of GM-CSF on DCs in potentiating IL9 production and antitumor activity, we adoptively transferred OVA-specific Th cells and OVA_323-339 peptide–loaded moDCs or cDCs, with or without GM-CSF treatment, into B16F10-OVA tumor-bearing mice. GM-CSF–treated moDCs more efficiently induced IL9 expression in coinjected tumor-specific naïve Th cells in the tumor microenvironment compared with GM-CSF–untreated moDCs (Fig. 5F). In contrast, GM-CSF–treated cDCs did not augment IL9 expression; instead, these cells increased IFNγ expression in transferred tumor-specific Th cells compared with GM-CSF–untreated cDCs. GM-CSF treatment also manipulated moDCs to inhibit the induction of Foxp3–expressing Treg cells, whereas GM-CSF–treated cDCs increased Foxp3 expression in tumor-specific Th cells (Fig. 5F). IL17A expression in transferred Th cells was not changed by GM-CSF treatment. We further investigated whether the transfer of GM-CSF–stimulated moDCs leads to tumor growth inhibition in the therapeutic model and whether its activity is IL9-dependent. As depicted in Fig. 5G, GM-CSF–treated moDCs delayed tumor growth compared with GM-CSF–untreated moDCs. Tumor peptide–loaded cDCs also suppressed tumor growth; however, GM-CSF treatment did not further increase its antitumor activity. To examine whether the enhanced antitumor activity of GM-CSF–treated moDCs was due to IL9, we adopted a neutralizing anti-IL9 and found that treatment with anti-IL9 completely reversed the tumor growth inhibition by GM-CSF–treated moDCs (Fig. 5H). We also observed that E6/E7 peptide–loaded, GM-CSF–activated moDCs significantly inhibited TC-1 tumor growth in an IL9-dependent manner (Supplementary Fig. S6). Collectively, these results indicate that GM-CSF renders moDCs to induce IL9 expression rather than Foxp3 in tumor-specific Th cells, which leads to tumor growth inhibition in an IL9-dependent fashion.

During tumor development, heterogeneous myeloid populations accumulate in the spleen and tumor tissue. To determine whether moDCs and cDCs accumulate locally within tumors and respond to GM-CSF to induce IL9, we analyzed TILs from B16F10-OVA tumors. Along with tumor progression, moDCs, but not cDCs, significantly accumulated in the tumor (Supplementary Fig. S7A). Then, we isolated each tumor-infiltrating DC population and compared their IL9-inducing capacity ex vivo. Tumor-infiltrating moDCs also selectively induced IL9 expression in tumor-specific Th cells in response to GM-CSF (Supplementary Fig. S7B), suggesting that this phenomenon may occur in the local tumor microenvironment.

Previous reports have demonstrated that IL9 or Th9 cells promote tumor-specific CTL responses to eradicate tumors (14, 15, 18). Thus, we analyzed tumor-specific CTL responses in our therapeutic B16F10-OVA model treated with tumor-specific Th cells and moDCs or cDCs. In accordance with the tumor growth inhibition, lymphoid cells from the recipients of GM-CSF–activated moDCs showed more potent cytotoxicity than those from the recipients of GM-CSF–untreated moDCs, which was comparable with those from cDC recipients (Fig. 6A). Anti-IL9 treatment reversed the enhancement of tumor-specific CTL responses in the recipients of GM-CSF–stimulated moDCs. Consistently, expression of granzyme B in CD8⁺ T cells was significantly increased in the recipients of GM-CSF–stimulated moDCs, and this was almost completely abolished by anti-IL9 (Fig. 6B). These results together suggest that IL9 induced by GM-CSF–activated moDCs triggers antitumor CTL responses in vivo.

Next, we sought to determine the underlying mechanism by which GM-CSF promotes Th9 cell differentiation. Studies have shown that engagement of TNF receptor superfamily members including OX40, DR3, and GITR enhances Th9 cell differentiation (15, 37, 38). To test whether GM-CSF–mediated Th9 cell differentiation depends on these molecules, we employed blocking antibodies to each interaction, and found that individual antibodies or the combination of antibodies did not reverse Th9 differentiation induced by GM-CSF (Supplementary Fig. S8A and S8B).

GM-CSF regulates glucose and lipid metabolism in immune cells (39) and COX2 and its metabolic products block Th9 development (40). Therefore, we next investigated the role of COX2-mediated lipid metabolism in Th9 differentiation induced by GM-CSF. The addition of GM-CSF decreased the expression of COX2 and several downstream prostaglandin (PG) synthases under moDC or cDC stimulation (Fig. 7A), and adenoviral transduction of the COX2 gene to moDCs or cDCs significantly abolished the increase in Th9 induction by GM-CSF (Fig. 7B and C). Moreover, when the COX2 metabolic product PGE₂, but not PGD₂, was added to cultures, it significantly reduced the frequency of Th9 cells induced by GM-CSF without affecting cell proliferation (Fig. 7D and E). Collectively, these results demonstrate that GM-CSF potentiates Th9 cell differentiation by regulating the COX2–PGE₂ synthesis pathway.

Although the use of GM-CSF for tumor immunotherapy has been extensively studied in mice and humans, the role of Th responses in GM-CSF–induced antitumor immunity is relatively unappreciated. In this study, we demonstrate that GM-CSF triggers antitumor immunity by enhancing the differentiation of Th9 cells and by concomitantly inhibiting the generation of iTreg cells in therapeutic cancer vaccine and DC vaccine models. This effect of GM-CSF was mediated through moDCs and was found to induce potent tumor-specific CTL responses that suppressed tumor growth in an IL9-dependent manner. The underlying mechanism was that GM-CSF inhibited COX2-mediated lipid metabolism and PGE₂ production, which consequently increased Th9 cell differentiation. Hence, we have elucidated the missing link between GM-CSF and its antitumor activity and identified a pathway for Th9 cell development.

Proinflammatory properties of GM-CSF have been implicated in diverse autoimmune diseases (23). Inflammatory responses during tumor immune surveillance can be beneficial to the host; however, tumors exploit the host immune system to advance tumor growth. Myeloid inflammation induced by MDSCs and tumor-associated macrophages (TAM) is the representative of them, and tumor-derived or exogenous GM-CSF has been shown to trigger the recruitment and differentiation of MDSCs in the tumor microenvironment (32, 41–43), which might account for the inconsistent results in previous clinical studies using GM-CSF (31, 41). In line with this result, we also observed that neutralization of endogenous GM-CSF led to decreased tumor burden. However, administration of GM-CSF does not always exhibit the same detrimental effect observed with endogenous GM-CSF. Indeed, we and other groups have shown the tumor-inhibiting effects of GM-CSF, suggesting that (i) physiologic levels of GM-CSF are not sufficient to induce a specific antitumor immune response or (ii) certain conditions are needed to facilitate the antitumor activity of GM-CSF.

MDSCs are defined as CD11b⁺Gr-1⁺ cells with suppressive activity and consist of a heterogeneous set of immature myeloid cells (44). These include CCR2⁺ monocytes and monocyte-derived cells, and indeed, we observed that IFNγ production by Th cells in response to tumor peptide stimulation was enhanced in CCR2-deficient mice. However, administration of GM-CSF licensed CCR2⁺MHC II⁺ moDCs to drive Th9 cell differentiation, which created favorable conditions for tumor inhibition via IL9. Thus, this result suggests that CCR2⁺MHC II⁺ moDCs are a prerequisite for exerting antitumor activity of GM-CSF, which was achieved by vaccination with tumor antigen plus CFA in our experimental setting. Moreover, these data explain why the antitumor effect of GM-CSF–secreting tumor vaccine disappeared in Gr-1–depleted hosts (45). Further studies are needed to determine the relevant adjuvant in humans that selectively induces the recruitment of CCR2⁺MHC II⁺ moDCs.

The frequency of HLA-DR⁺ classical monocytes in PBMCs may serve as a biomarker for responsiveness to anti-PD-1 therapy (46). This population corresponds to moDCs in our mouse study, implying that combination with GM-CSF may exhibit a synergistic antitumor effect in responders to anti-PD-1 therapy; indeed, this is under investigation (33–35). Vaccination with GM-CSF adjuvant did not appear to abrogate the suppressive functions of tumor-associated neutrophils (TAN) because CCR2-deficient mice given this therapy accumulated neutrophils and no antitumor effect was observed. Thus, combinations with checkpoint blockades or agents selectively depleting TANs would be beneficial for eliciting the full therapeutic efficacy of GM-CSF in patients with cancer.

GM-CSF potentiated the differentiation of Th9 cells rather than Th1 cells, thought to be the most potent antitumorigenic Th cells, to inhibit tumor growth. However, the appropriate induction of type I responses at the early phase of tumor progression is also needed because IFNγ is critical for the differentiation of monocytes into moDCs (47, 48), which were mediators of GM-CSF–induced Th9 differentiation in vivo; indeed, neutralization of IFNγ reversed the antitumor effect of GM-CSF (Supplementary Fig. S9). Furthermore, GM-CSF inhibited the generation of iTreg cells, which are abundant in the tumor microenvironment, and elicited IL9–producing Th cells. This result indicates that the effect of GM-CSF on tumor-specific Th responses has dual advantages of blocking suppressor T cells and promoting effector T cells to eliminate tumors. In addition, GM-CSF increased IL13 production by Th cells, a factor driving the expansion and activation of MDSCs (44), and the addition of neutralizing anti-IL13 may therefore reinforce the antitumor activity of GM-CSF.

It remains a possibility that GM-CSF enhanced Th9 cell differentiation by facilitating differentiation of monocytes into moDCs. However, GM-CSF is dispensable for moDC differentiation in vivo (49). In addition, although GM-CSF can affect the persistent expression of MHC-II molecules on moDCs and increase antigen presentation, it did not promote the differentiation of Th1 and Th17 effector cells. From this result, we suggest that GM-CSF may alter moDC functions to selectively increase Th9 cell differentiation. Meanwhile, consistent with a previous study (50), moDCs were defective in inducing the proliferation and expansion of Th cells, whereas cDCs efficiently induced T-cell proliferation upon antigenic stimulation. Thus, the role of cDCs must not be ignored, with moDCs promoting Th9 cell development in response to GM-CSF and cDCs potentially playing a role in Th9 cell proliferation in vivo.

We showed that GM-CSF facilitated Th9 cell differentiation via regulation of the COX2–PGE₂ pathway. In line with this result, a previous study revealed that COX2 and its metabolic products inhibit Th9 development by decreasing IL17RB expression (40). Further studies are needed to determine how the COX2–PGE₂ pathway regulates IL17RB expression, whether it affects other transcription factors or signaling pathways related to Th9 differentiation, and how Th9 development might be used to manipulate Th cells with antitumor activity.

In this study, we have elucidated a mechanism by which GM-CSF elicits antitumor immune responses. As an adjuvant for cancer vaccines, GM-CSF promotes Th9 cell differentiation and inhibits iTreg cell generation, particularly through its effects on moDCs. This leads to the generation of potent antitumor CTL responses that suppress tumor growth in an IL9-dependent fashion. On the basis of these results, we suggest the frequency of moDCs as a predictive biomarker for GM-CSF therapy and IL9 or Th9 as a biomarker for its effectiveness. Furthermore, as the adoptive transfer of Th9 cells has been shown to eradicate advanced tumors (16), GM-CSF therapy can be applied with a combination of other immunotherapies to treat cancer that is susceptible to Th9 cell responses. Collectively, our findings identify GM-CSF as a Th9 inducer and provide the rationale for the use of GM-CSF in cancer vaccines.

No potential conflicts of interest were disclosed.

Conception and design: I.-K. Kim, C.-Y. Kang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I.-K. Kim, C.-H. Koh, K.-S. Shin, T.-S. Kang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I.-K. Kim, B.-S. Kim

Writing, review, and/or revision of the manuscript: I.-K. Kim, Y. Chung, C.-Y. Kang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.-H. Koh, I. Jeon, K.-S. Shin, E.-A. Bae, H. Seo, H.-J. Ko

Study supervision: C.-Y. Kang

This research was supported by a grant from the Basic Science Research Program (NRF-2015R1A2A1A10055844), Bio & Medical Technology Development Program (NRF-2016M3A9B5941426) through the National Research Foundation of Korea funded by the Ministry of Science, ICT & Future Planning, and Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2016R1A6A3A01007625).

We thank J.-O. Kim (International Vaccine Institute) for CD45.1 congenic mice, J.V. Snick (Ludwig Institute for Cancer Research, New York, NY) for the antibody to IL9 (MM9C1), K. Shortman (The Walter and Eliza Hall Institute of Medical Research) for the antibody to GM-CSF (MP1-22E9), and Kang's laboratory members for technical supports.

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.