Hematopoietic Prostaglandin D₂ Synthase Controls Tfh/Th2 Communication and Limits Tfh Antitumor Effects

The immune system plays a critical role in cancer evolution (1), and understanding the antitumor immune response is essential to better define a patient's prognosis and develop new immunotherapies. CD4⁺ T cells are major players in the adaptive immune system and are known to have both positive and negative roles in antitumor immune responses. Although Th1 cells are predominantly reported as antitumor cells, regulatory T cells (Treg), Th2 cells, and Th17 cells are mostly associated with immunosuppression and tumor progression (2). However, conflicting results are observed with T follicular helper (Tfh) cells.

Th2 cells produce IL4, which is associated with type 2 immune responses (3). IL4, along with IL5 and IL13, confer an immune protective role on Th2 cells by recruiting innate immune cells into mucosal tissues to support epithelial barrier function and integrity (4). Many studies have demonstrated that a high number of Th2 cells in the tumor microenvironment correlates with disease progression due to these cells having the capacity to directly promote tumor growth and neoangiogenesis through IL4 secretion (5). Secreted IL4 directly increases proliferation and resistance to apoptosis of colorectal, breast, head and neck, ovarian, and prostate cancer cells (6,–9). In addition, patients with breast, prostate, and kidney cancer harbor high levels of IL4 in the tumor microenvironment (10 –12).

Physiologically, Tfh cells are located around and inside B-cell follicles in lymph nodes and are required in germinal centers to promote B-cell class-switch recombination, a process crucial for affinity maturation and maintenance of humoral memory through IL4 secretion (13). A recent study shows that IL4 derived from Tfh cells is important for supporting Th2 differentiation during intestinal helminth infection (14). In some reports, Tfh cells associate with a good prognosis for patients with cancer (15,–18). Tfh cells produce IL21, which could explain their antitumor functions. In contrast, other studies suggest that Tfh or intratumoral Tfh-like cells could be associated with tumor growth (19). Thus, mechanisms explaining protumour and antitumor effects of Tfh cells are still elusive.

In this study, we found that Tfh cells have high expression of the hematopoietic prostaglandin D₂ synthase (Hpgds), an enzyme of the eicosanoid pathway involved in prostaglandin D₂ (PGD₂) production. The receptor of PGD₂, CRTH2 (chemoattractant receptor-homologous molecule expressed on Th type 2 cells), is expressed only on CD4⁺ Th2 cells (20). Thus, even when Tfh cells do not produce IL4, their antitumor effects can be abolished in both mouse and humans in a PGD₂/Th2-dependent manner due to the chemoattraction of Th2 cells and induction of cytokine secretion.

B16-F10, LLC1, MC-38, CT26, and 4T1 cell lines were purchased from ATCC between 2017 and 2019; KPC was purchased from Ximbio between 2017 and 2019, and B16-OVA (ovalbumin) was kindly provided by Dr. Apetoh in 2017. Cell lines were cultured at 37°C under 5% CO₂ in DMEM (Dominic Dutscher; KPC, B16-F10, B16-OVA, LLC1, and MC-38) or in RPMI1640 Medium (Dominic Dutscher; CT26 and 4T1) with 10% (volume for volume) FCS (Dominic Dutscher) supplemented with penicillin/streptomycin/amphotericin B (PSA; 10,000 U/mL penicillin, 10 mg/mL streptomycin, 25 μg/mL amphotericin B in 0.85% saline; PAN Biotech), and 4 mmol/L of 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES; 1 mol/L; Gibco). Cell lines were tested for Mycoplasma every week by PCR. Cell passages were carried out three times a week without going further than 10 passages before use. Cell lines were not reauthenticated after purchase. Further information about the cell lines and reagents is detailed in the Supplementary Table S1.

All animals were bred and maintained in accordance with Federation of European Laboratory Animal Science Associations (FELASA) guidelines, and the study was performed after approval from the Ethics Committee for Animal experimentation of the University of Burgundy, France. C57BL/6J and BALB/cJ mice were purchased from Charles River Laboratories. TRP-1, OT-II, C57Bl6 Ly5.1a (referred to hereafter as CD45.1⁺ mice), and B6.CgKitW-sh/HNihrJaeBsmJ mice (referred to hereafter as Kit^W-sh mice) were bred at, and obtained from, the CDTA (Cryopreservation, Distribution, Typage et Archivage animal; Orléans, France). One male and 1 female Hpgds^−/− mouse, each age 8 weeks, were provided by Amelia Trimarco (Division of Neuroscience, San Raffaele Scientific Institute, Milan, Italy). These mice were initially generated by Yoshihiro Urade (Interational Institute for Integrative Sleep Medicine, University of Tsukuba, Tsukuba, Japan). In addition, 1 male and 1 female BCl6^f/fCD4^cre mouse, each age 8 weeks, were provide by James A. Harker (National Heart & Lung Institute, Imperial College London, United Kingdom). Both Hpgds^−/− and BCl6^f/fCD4^cre mice were bred at the Centre de Zootechnie de l'Université de Bourgogne, Dijon, France. We performed the genotyping of the mice resulting from breeding BCl6^f/fCD4^cre/+ mice to identify animals possessing the cre-recombinase and littermate controls according to Dr Harker's recommendations (21).

Female animals between 6 and 10 weeks of age were used in experiments.

To induce tumor lung invasion, 2 × 10⁵ B16-F10 melanoma cells, B16-OVA melanoma cells, or LLC1 lung carcinoma cells were injected intravenously into syngeneic C57BL/6J wild-type (WT), Hpgds^−/−, Kit^W-sh–, Bcl6^f/fCD4^cre/+ and their littermate control mice. In other studies, 2 × 10⁵ CT26 colon carcinoma cells were injected intravenously into syngeneic BALB/cJ mice.

For in vivo cell transfer, mice were injected with 2 × 10⁵ B16-OVA or B16-F10 cells 24 hours before transfer of 2 × 10⁶ T cells generated in vitro from naive T cells (from C57BL/6 OT-II mice) for three days (as indicated below in the related section) or 0.5 × 10⁶ T cells generated in vivo from TRP-1 mice. To generate over-production of Tfh cells in TRP-1 mice, we immunized the mice intraperitoneally with 2 × 10⁹ sheep red blood cells (SRBC, CliniSciences). Tfh cells were sorted from spleen and lymph nodes (inguinal, axial, brachial, cervical, and mesenteric) after 7 days (identified as CD4⁺CXCR5⁺PD1⁺). In each experiment, lung was harvested and tumor foci were enumerated 13 days after T-cell transfer. When indicated, mice were treated intraperitoneally with IL4, IL5, or IL13 blocking antibodies (200 μg, BioXCell; Supplementary Table S1) three times a week or treated with 10 μL i.v. anti-Asialo-GM1 (Sobioda) once a week, starting at the same moment of T-cell transfer.

To induce subcutaneous tumor formation, 8 × 10⁵ MC-38 colon adenocarcinoma, KPC pancreatic cancer, of LLC1 lung cancer cells were injected into the left flank of syngeneic C57BL/6J mice. In other experiments, 8 × 10⁵ CT26 colon carcinoma cells were injected into the left flank of syngeneic Balb/cJ mice.

For the spontaneous lung cancer model, mice received 10 weekly urethane (Sigma-Aldrich) injections (1 g/kg) i.p. All mice were sacrificed after 4 months, and lung tumors were enumerated.

Additional information on the reagents is detailed in the Supplementary Table S1.

Mouse and human cells were cultured in RPMI1640 medium with 10% (volume for volume) FCS supplemented with MEM nonessential amino acids (Gibco; 100X), sodium pyruvate (Gibco, 100 mmol/L), PSA, and 4 mmol/L of HEPES.

Naïve CD4⁺ T cells (CD4⁺CD62L^hi) obtained from spleens and lymph nodes (inguinal, axial, brachial, cervical, and mesenteric) of C57BL/6J, C57BL/6 OT-II or Hpgds^−/− mice were purified after a hemolytic step (1X RBC, eBioscience) using the MACS Cell Separation system [Quadro MACS Separator, LS Columns, Pre-Separation Filter (30 μm) and CD4⁺CD62L⁺ T-cell isolation kit (Miltenyi Biotec)]. Naïve CD4⁺ T cells (5 × 10⁵ cells per condition) were stimulated with plate-bound antibodies against CD3 (145-2C11, 2 mg/mL, BioXCell) and CD28 (PV-1, 2 mg/mL, BioXCell) in the presence of either no cytokines and anti-IL4 and anti-IL12 antibodies (generating Th0 cells) or in the presence of 10 ng/mL of IL12 (R&D Systems) and anti-IL4 (11B11, BioXCell; generating Th1 cells); 10 ng/mL of IL4 (R&D Systems) and anti-IFNγ (XMG1.2, BioXCell; generating Th2 cells); 4 ng/mL of TGFβ (Miltenyi Biotec), anti-IL4 (11B11, BioXCell), and anti-IFNγ (XMG1.2, BioXCell; generating Tregs); 2 ng/mL of TGFβ (Miltenyi Biotec), 20 ng/mL of IL6 (R&D Systems), anti-IL4 (11B11, BioXCell), and anti-IFNγ (XMG1.2, BioXCell; generating Th17 cells); or 50 ng/mL of IL6 (R&D System) and 50 ng/mL of IL21 (R&D Systems), anti-IL4 (11B11, BioXCell), anti-IFNγ (XMG1.2, BioXCell), and anti-TGFβ (1D11.16.8, BioXCell; generating Tfh-like cells). In some experiments, cells were treated with 2 μg/mL anti-IL6 (MP5-20F3, BioXcell), anti-IL21 (AF594, R&D systems) or anti-IL21R (4A9, BioXcell). In some experiments, cells were treated with 50 μmol/L epigallocatechin gallate (EGCG, Sigma-Aldrich) to inhibit STAT1 or 20 μmol/L STA21 (Sigma-Aldrich) to inhibit STAT3. Finally, 10 μmol/L TM30089 (Chemie Tek) was used in some experiments as a CRTH2 inhibitor.

Naive CD4⁺ T cells (CD4⁺CD62L^hi) were obtained from human healthy donor buffy coats from Etablissement Français du Sang (EFS). First, CD4⁺ T cells were purified from the buffy coats using the RosetteSep Human CD4⁺ T-Cell Enrichment Cocktail (STEMCELL Technologies) and Lymphocytes Separation Medium (Eurobio Abcys). Then, the CD4⁺ T cells were stained using the CD45RA-APC, CD62L-PE, CD4-BV421, and FVS700 human antibodies panel (Supplementary Table S1), sorted by FACS (FACSAriaIII Cell Sorter, BD Biosciences) to 95% to 99% purity and restimulated with plate-bound antibodies against CD3 (OKT3, 2 mg/mL) and CD28 (CD28.2, 2 mg/mL).

Additional information on the reagents is detailed in Supplementary Table S1.

All events were acquired using a BD LSRII or a BD LSR Fortessa (BD Biosciences) cell analyzer or sorted by a FACSAria Cell Sorter (BD Biosciences) equipped with BD FACSDiva software (BD Biosciences). Data were analyzed using FlowJo software v10 (Tree Star).

To phenotype and sort distinct CD4⁺ T-cell populations by flow cytometry, lungs (bearing B16-F10, B16-OVA or urethane-induced tumors), subcutaneous tumors (B16-F10, LLC1, CT26, or MC38), lymph nodes and spleens were first harvested and dissociated using a tissue dissociation kit (Miltenyi Biotec; tumors and lungs: tumor dissociation Kit, Mouse; Miltenyi Biotec) and device (gentleMACS Octo Dissociator with Heaters, Miltenyi Biotec), or dissociated through MACS SmartStreiners (70 μm) (Miltenyi Biotec; lymph nodes and spleens). Prior to the gentleMACS dissociation, tumors were mechanically and roughly predissociated using scissors. Then, predissociated tumors and lungs were placed into a gentleMACS C Tube (Miltenyi Biotec) with 2.35 mL of RPMI1640, 100 μL of Enzyme D, 50 μL of Enzyme R and 12.5 μL of Enzyme A. For mechanical dissociation and enzymatic degradation, specific heaters were used and a suitable program selected (37C_m_TDK1) based on the characteristics of the tissues (soft/medium). Following dissociation, the remaining lysates were filtrated through Pre-Separation Filters (70 μm) (Miltenyi Biotec). For spleens and mediastinal lymph nodes, a hemolytic step was performed (1X RBC, eBioscience) prior to staining the cells.

Cells were then stained in Flow Cytometry Staining Buffer (eBioscience) and Brilliant Stain Buffer (BD Biosciences). Anti-CD4-FITC (H129.29) and Fixable Viability Stain 700 (FVS700) were purchased from BD Biosciences. Anti-CD62L L (PE), anti-CD45-VioGreen (30-F11), anti-CXCR3 (CD183)-APC (CXCR3-173) and anti-CCR6 (CD196)-PE-Vio770 (REA277) were purchased from Miltenyi Biotec. Anti-CXCR5 (CD185)-BV421 (L138D7) and anti-PD-1 (CD279)-APC-Cy7 (29F.1A12) were purchased from BioLegend. Anti-T1/ST2 (IL33R)-PE (DJ8) was purchased from mdbiosciences, bioproducts division. Anti-FOXP3-PerCP-Cy5.5 (FJK-16s) was purchased from eBioscience (Thermo Fisher Scientific). FOXP3 intracellular staining buffer carried out using the Intracellular Fixation & Permeabilization Buffer Set according to the manufacturer's protocol (eBioscience).

CD4⁺ T cells from healthy donor buffy coats (EFS) were purified as described previously (see Naïve CD4⁺T-cell purification and in vitro differentiation). Then, the enriched CD4⁺ T cells were stained in Flow Cytometry Staining Buffer (eBioscience) and Brilliant Stain Buffer (BD Biosciences). Anti-CD45RA-BV510 (HI100), anti-CD4-FITC (RPA-T4), anti-CXCR3 (CD183)-APC-Cy7 (G025H7), anti-CCR6 (CD196)-BV605 (G034E3), anti–PD-1 (CD279)-PE (EH12.2H7) and anti-IL7-Rα (CD127)-PerCP-Cy5.5 were purchased from BioLegend. Anti-CD62L-PE (DREG-56), anti-CD4-BV421 (L200) anti-CD25-BV421 (2A3), anti-CXCR5 (CD185)-Alexa Fluor 647 (RF8B2), anti-CCR4 (CD194)-PE-Cy7, and FVS700 were purchased from BD Biosciences. Anti-CD45RA-APC (REA562) was purchased from Miltenyi Biotec.

Fluorescence Minus One controls were used to define background, and identify and control positive populations in multicolor experiments.

Cells were stimulated for 4 hours at 37°C in culture medium containing phorbol 12-myristate 13-acetate (50 ng/mL; Sigma), ionomycin (1 mg/mL; Sigma), and monensin (GolgiStop; 1 mL/mL; BD Biosciences). After the staining of surface markers, cells were fixed and permeabilized according to the manufacturer's instructions (BD Biosciences), and then stained for intracellular products.

Tfh cells (CD4⁺CXCR5⁺PD1⁺) from C57BL/6J or TRP-1 mice were sorted from spleen and lymph nodes (inguinal, axial, brachial, cervical, and mesenteric). CD4⁻CRTH2⁺ and CD4⁺CRTH2⁻ were used respectively as positive and negative controls in the transwell experiments. They were isolated from spleen and lymph nodes of C57BL/6J mice. Characterization of these cells indicated CD4⁻CRTH2⁺ cells were mainly represented by eosinophils CD45⁺CD3⁻Ly6c⁻CD11b⁺CD11c⁻ and CD4⁺CRTH2⁻. A hemolytic step was performed (1X RBC, eBioscience) prior to staining the cells. Purity was between 95% and 100%. Sorting was performed with a FACSAria Cell Sorter (BD Biosciences) equipped with BD FACSDiva software (BD Biosciences). Finally, CD45⁺CD3⁻Ly6c⁻CD11b⁺CD11c⁻ (eosinophils) were used as positive controls for hPGDS protein expression analysis and were also sorted in the same conditions.

Additional information on the reagents is detailed in Supplementary Table S1.

Total RNA was extracted from in vitro differentiated Th1 and Tfh cells and from Tfh cells isolated from tumors according to membrane markers using TRIzol (Invitrogen). rRNA was removed using the Ribo-zero rRNA Removal Kit (Illumina). A total of 100 ng of rRNA-depleted RNA was used for library preparation using the TruSeq Stranded Total RNA Library Prep kit (Illumina) following the manufacturer's instructions. RNA sequencing (RNA-seq) was performed on a NextSeq device (Illumina). RNA-seq libraries were sequenced with paired-end 75 bp reads. FASTQ files were mapped by using Burrows-Wheeler Aligner (ref. 22; mm10 version of Mus musculus genome). Analysis was performed using TopHat 2.1.1 (23). Generated files were processed with Cufflinks 2.2.1 (24) to obtain annotated expressed genes in each studied CD4⁺ subtype. Then, differential expression between the samples was analyzed with Cuffdiff (25). Genes with an absolute fold change greater than 1 and a P value below 0.005 were considered as significantly differentially expressed.

Additional information on the reagents is detailed in Supplementary Table S1.

Secreted cytokines were measured by ELISA for mouse IL4, IL5, IL10 (BD Biosciences), and IL13 (eBioscience) after 72 hours of in vitro cell differentiation in the culture medium or in the supernatant of dissociated tumor-bearing lungs. ELISA assays were performed according to manufacturers’ instructions. PGD₂ concentration was measured by an EIA assay (Interchim) according to manufacturers’ instructions on mouse and human cells differentiated for 72 hours in vitro differentiated cells. Additional information on the reagents is detailed in the Supplementary Table S1.

Total RNA from T cells (from mouse and human samples) was extracted with TRIzol (Invitrogen). A total of 300 ng of RNA was reverse transcribed into cDNA by M-MLV reverse transcriptase (Invitrogen), random primers (Invitrogen), and RNaseOUT inhibitor (Invitrogen). cDNAs were quantified by real-time qPCR with a PowerUp SYBR Green Master Mix (Applied Biosystems) on a Fast7500 detection system (Applied Biosystems). Relative mRNA expression was determined with the 2^ΔΔC_t method relative to β-actin. All primers used are listed within the Supplementary Table S1, section “Oligonucleotides.” Additional information on the reagents is detailed in Supplementary Table S1.

Tfh cells were differentiated for 24 hours prior to analysis of hPGDS and Bcl6 protein expression and were analyzed at various timepoints after cell stimulation (3, 6, 10, 12 minutes) for STAT1 and STAT3 phosphorylation. Proteins were prepared by lysing cells for 10 minutes at 4°C in boiling buffer (1% SDS, 1 mmol/L sodium orthovanadate, 10 mmol/L Tris, pH 7.4, complete protease inhibitors from Roche Diagnostics). A total of 50 μg protein lysates in loading buffer (125 mmol/L Tris-HCl, pH 6.8, 10% β-mercaptoethanol, 4.6% SDS, 20% glycerol, and 0.003% bromophenol blue) were heated at 95°C for 5 minutes, separated by SDS-PAGE, and transferred by electroblot onto a nitrocellulose membrane (Amersham Protran; GE Healthcare Life Science). After a 2-hour incubation at room temperature with 5% BSA (South America, Dominic Dutscher) in TBS and 0.1% Tween 20 (Sigma-Aldrich; TBS-Tween), membranes were incubated overnight with primary antibody diluted in 5% BSA in TBS-Tween, washed with TBS-Tween, and then incubated for 1 hour at room temperature with secondary antibody. Membranes were then washed before analysis with luminol reagents (ImmunoCruz: Santa Cruz Biotechnology or SuperSignal West Femto Maximum Sensitivity Substrate: Thermo Fisher Scientific) and signals were detected with ChemiDoc XRS+ (Bio-Rad). All primary and secondary antibodies and additional reagents are listed in Supplementary Table S1.

A total of 1 × 10⁶ cells (mouse and human, in vitro/ex vivo) and eosinophilic cells (mouse, ex vivo) were washed, fixed for 10 minutes at room temperature with 4% paraformaldehyde (Agar Scientific), and permeabilized for 10 minutes on ice with methanol 100% glacial [for hPGDS staining and proximity ligation assay (PLA)] or for 45 minutes with 3% BSA and 0.2% saponin (Sigma-Aldrich) in PBS (for PGD₂ staining). Nonspecific binding was blocked at room temperature for 1 hour in blocking buffer made of 5% FBS in PBS [for immunofluorescence (IF)] or 30 minutes at room temperature in blocking buffer made of 0.5% BSA in PBS (for PLA). Samples were incubated overnight at 4°C with primary antibodies at the manufacturer recommended dilution in their respective blocking buffer. Antibodies are listed in Supplementary Table S1. For IF experiments, cells were washed two times with 0.05% Tween 20 in PBS (PBS-Tween) and incubated for 1 hour at room temperature with secondary antibody diluted at the manufacturer recommended dilution in the blocking buffer. Cells were washed two times with PBS-Tween and two times with ultrapure water. For PLA experiments, after washing, cells were incubated for 1 hour at 37°C with the appropriate probes (Duolink In Situ PLA Probe Anti-Rabbit PLUS and Anti-Mouse MINUS; Sigma-Aldrich) according to the manufacturers’ recommendations. For both IF and PLA, stained cells were deposited (10 to 20 μL of stained cell suspension) on microscopy slides (Superfrost Ultra Plus, Thermo Fisher Scientific) and incubated at room temperature until water evaporated. Slides were mounted with a drop of Mounting Medium containing DAPI (Molecular Probes). Slides were imaged with a charge-coupled device–equipped upright microscope (ZEISS Apotome.2, Carl ZEISS Microscopy), and images were analyzed with ZEN system software (Carle ZEISS microscopy) and ImageJ 64-bit Java 1.8.0_172 software (ImageJ). Additional information on the reagents is detailed in the Supplementary Table S1.

Twenty-four hours after tumor inoculation, C57BL/6J mice bearing lung B16-OVA tumors received OT-II cells polarized to a Th2- and Tfh-like phenotype in vitro. These cells were stained with CellTrace Far Red and CellTrace CFSE (Thermo Fisher Scientific), respectively. The next day, lungs were harvested and washed with PBS, included in OCT matrix (Leica Biosystems), cut in slides, and mounted on microscopy slides (Superfrost Ultra Plus, Thermo Fisher Scientific). Samples were then stained with DAPI (Sigma-Aldrich) and imaged using spectral microscopy technology (Mantra Workstation, MantraSnap from PerkinElmer) at 20X magnification. DAPI-only positive cells, CellTrace CFSE only, or CellTrace FarRed cells were phenotyped using a trainable algorithm from the inForm software (PerkinElmer). Additional information on the reagents is detailed in the Supplementary Table S1.

For both mouse and human experiments, 5 × 10⁵in vitro differentiated Th2 cells or sorted CD4^– CRTH2⁺ cells (used as positive control) and sorted CD4⁺CRTH2⁻ cells (used as negative control) were plated in RPMI (Dominic Dutscher) supplemented with 10% FBS (Dominic Dutscher) in 8.0 μm transwell inserts (Greiner bio-one) and placed into a 24-well plate. PGD₂ (25 nmol/L; Cayman Chemical), naive CD4⁺ T cells, in vitro differentiated WT Th0 or WT or Hpgds^−/− Tfh-like cells were placed on the bottom, underneath the transwell inserts [5 × 10⁵ cells/well, in RPMI supplemented with 10% FBS (Dominic Dutscher)]. After a 24-hour incubation at 37°C under 5% CO₂, the transwell grid was released, loaded on microscopy slides (Superfrost Ultra Plus, Thermo Fisher Scientific), mounted with a drop of Mounting Medium containing DAPI (Molecular Probes), and imaged with a charge-coupled device–equipped upright microscope (Zeiss) and 40× objective with a numerical aperture of 1.4. Images were analyzed with ImageJ 64-bit Java 1.8.0_172 software (ImageJ). Additional information on the reagents is detailed in Supplementary Table S1I.

Chromatin shearing of mouse naïve CD4⁺ T cells and Tfh-like cells 4 hours after in vitro polarization was performed with the truChIP Chromatin Shearing Kit (Covaris) using the Focused-Ultrasonicator M220 device (Covaris). A chromatin immunoprecipitation (ChIP) assay was performed with the ChIP-IT kit (Active Motif Europe) according to manufacturer's instructions. In brief, sheared chromatin was immunoprecipitated with 1 μg of phospho-(p)STAT1 or pSTAT3 antibodies (Cell Signaling Technology) or 1 μg of rabbit negative control immunoglobulin G (IgG) at 4°C overnight. After addition of protein G beads, the protein G/antibody/chromatin mixture was washed and eluted from the protein G with supplied buffers. Then, cross-link was reversed and samples were analyzed by quantitative PCR (see qPCR analysis).

A chromatin reimmunoprecipitation (ChIP-on-ChIP) assay was performed with the Re-ChIP-IT kit (Active Motif Europe) according to manufacturer's instructions, with 1 μg of pSTAT1 and 1 μg pSTAT3 antibodies (Cell Signaling Technology). As for the ChIP assay, samples were analyzed by qPCR.

Additional information on the reagents is detailed in Supplementary Table S1.

Putative STAT3 binding sites were predicted via MatInspector (Genomatix software) using Matrix Library 11.5. A total of 677 bp of Hpgds promotor (>gi|372099104:c65145015-65117293 Mus musculus strain C57BL/6J chromosome 6, GRCm38.p4 C57BL/6J) containing putative STAT3 binding sites were cloned into the pGl3 basic vector (Promega). Mouse 4T1 cells were transiently transfected for 6 hours with reporter plasmids (pHpgds or pGl3 basic vector) and pSV-β-galactosidase control vector (Promega) using Lipofectamine 2000 (Thermo Fisher Scientific). In these cells, STAT1 and STAT3 phosphorylation was activated by IFNγ (10 μmol/L) and IL6 (10 μmol/L; both from R&D Systems). β-galactosidase activity was measured using the β-galactosidase Enzyme Assay System (Promega), and luciferase activity was determined using the Luciferase Assay System (Promega) according to the manufacturer's instructions. Firefly luciferase was measured with an EnVision 2105 Multimode Plate Reader (PerkinElmer). Additional information on the reagents is detailed in Supplementary Table S1.

A total of 5 × 10⁵ naïve CD4⁺ T cells isolated from C57BL/6J mice lymphoid organs were transfected with validated control siRNA (SR30004) or Hpgds-specific siRNA (SR402083; Life Technologies) for 48 hours using Lipofectamine 2000 transfection reagent according to manufacturer's instructions (Mirus Bio). Additional information on the reagents is detailed in Supplementary Table S1.

RNASeqVA data with RSEM normalization (26) and corresponding clinical data were downloaded from The Cancer Genome Atlas (TCGA) data portal (https://portal.gdc.cancer.gov/) using the TCGA2STAT R package (http://www.liuzlab.org/TCGA2STAT/). Genes used to compute metagenes corresponding to the different cell populations (Tfh: B3GAT1, BCL6, CCR7, CD200, CD83, CD84, CDK5R1, FGF2, GPR18, PDCD1 and Th2: ASB2, CALD1, CCR2, CSRP2, DAPK1, DLC1, DNAJC12, DUSP6, GATA3, GNAI1, HTR2B, LAMP3, NRP2, OSBPL1A, PDE4B, PHLDA1, PLA2G4A, RAB27B, RBMS3, RNF125, SIGLEC10, SKAP1, SMAD2, TMPRSS2, UBASH3A) were taken from already published articles without further curation (27). For each tumor sample, the metagene value for one cell population was obtained by computing the average expression of corresponding genes. The prognostic value of the metagene estimation was tested using Cox proportional hazards models for overall survival (OS). Survival probabilities were estimated using the Kaplan–Meier method, and survival curves were evaluated using the log-rank test. We censored the OS at 60 months. The cutoffs to build different patients groups were defined using the cutoff finder algorithm (Supplementary Table S2; ref. 28). Metagene variables were dichotomized using the methodology of Hothorn and colleagues via the maxstat library. For each variable, the first group containing at least 30% of the population was used (29). Version 3.3.3 of R was used for statistical analysis.

Results are shown as mean ± SD or SEM, and datasets were compared using unpaired Student t test (test group vs. control group) or two-way ANOVA when appropriate. Differences in tumor foci numbers were assessed using Kruskall–Wallis test according to group numbers. Statistical calculations were done using GraphPad Prism 6. All P values were two tailed. A P < 0.05 was considered statistically significant for all experiments.

Data from RNA-seq analysis are available under the Gene Expression Omnibus access number GSE134875 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE134875). The data generated in this study are available within the article and its Supplementary Data files or from the authors upon request.

IL4-producing Th2 cells have protumor activity (30). In C57Bl6 mice treated with anti-IL4, we observed that fewer tumor lung foci were induced by intravenous injection of B16-F10 cells or by intraperitoneal injection of urethane (Fig. 1A and B). Antibodies blocking IL5 and IL13, known to be produced by Th2 cells, had little effect on B16-F10 lung tumor foci (Fig. 1A). The analysis of CD4⁺ T-cell subpopulations by flow cytometry in healthy or B16-F10 tumor-bearing lungs highlighted a significant increase in Th1, Th2, and Tfh cells in the lungs of tumor-bearing mice (Supplementary Fig. S1A; Fig. 1C and D). The increase in Th1, Th2, and Tfh cells was confirmed in subcutaneous CT26 and LLC1 tumors (colorectal carcinoma and Lewis lung cancer, respectively; Supplementary Fig. S1B and S1C). Tfh cells isolated from tumors (CD4⁺PD1⁺CXCR5⁺ cells) were similar to Tfh cells isolated from lymph nodes. Intracellular staining analysis showed that these cells expressed Bcl6 but did not express significant levels of RORγt, Tbet, and GATA3 (Supplementary Fig. S2A and S2B). After 24 hours of restimulation, Tfh cells did not produce IL4 or IL17 but did produce IL21 and low IFNγ (Supplementary Fig. S2C). Correlation analysis between the different CD4⁺ T-cell populations showed a correlation between Tfh and Th2 numbers in lungs of B16-F10 tumor-bearing mice (Fig. 1E) and urethane-induced tumor-bearing mice, but this was not seen with other T-cell subsets (Supplementary Table S3).

We investigated the potential link between Tfh and Th2 cells using adoptive cell transfer. OVA-specific Th1-, Th2-, and Tfh-like cells were generated in vitro by stimulation of naïve CD4⁺ T cells from OT-II mice. The global transcriptome of Tfh-like cells generated in vitro was compared with the transcriptome profile of Tfh cells isolated from tumor-bearing mice. These cells exhibited similarity in their transcriptional program (Supplementary Fig. S3A). Enrichment analysis of differentially expressed genes showed that Tfh-like cells differentiated in vitro did not harbor Th1, Th2, Th17, or Treg transcriptomic profiles, transcription factors, or cytokine expression (Supplementary Fig. S3B–S3D). In mice bearing B16-OVA lung tumor foci, we observed that Th2 cells and IL4 levels in lungs were increased after Tfh-like cell adoptive transfer (Fig. 1F and G). Few differences were observed in other cell subpopulations identified by flow cytometry (Fig. 1F). After adoptive transfer of CD45.2⁺ OT-II Tfh-like cells into CD45.1⁺ recipients showed that transferred cells from lung tumors did not switch to Th2-like cells expressing Il4 and Gata3 (Fig. 1H and I). OVA-specific Tfh-like cells induced a 70% decrease of B16-OVA lung tumor foci, whereas Th2-cell transfer had a protumor effect (Fig. 1J). Intraperitoneal treatment with anti-IL4 increased Tfh-like antitumor efficacy (Fig. 1K). We confirmed these results using naïve CD4⁺ T cells isolated from TRP-1 mice, which express a transgenic T-cell receptor (TCR) that recognizes tyrosinase-related protein 1 (TRP-1), a melanocyte differentiation antigen expressed by B16-F10 melanoma cells. We differentiated these cells in vitro into Tfh-like cells and found they harbored antitumor effects against lung tumor foci when B16-F10 cancer cells were injected intravenously into WT mice (Fig. 1L), and Tfh-like cell transfer induced an increase in Th2 cells in the lungs (Supplementary Fig. S4A). Again, CD4⁺ T cells differentiated into Th2 cells exhibited protumor activity, as shown by the increased number of tumor foci detected in the lungs (Fig. 1L). Combination of anti-IL4 and Tfh-like cells improved antitumor efficacy (Fig. 1L), as previously observed in the B16-OVA/OT-II model, where reduced numbers of B16-F10 lung tumor foci were observed. Similar results were obtained using Tfh cells isolated from spleens of TRP-1 mice immunized with SRBCs (Supplementary Fig. S4B). Together, these data highlight that Tfh cell accumulation in tumors correlates with Th2-cell accumulation, and that IL4 blockade improves the antitumor efficiency of Tfh-like adoptive transfer.

Th2 cell numbers increased in B16-OVA lung tumors after adoptive transfer of ex vivo generated Tfh-like cells (Fig. 1F) but sorted Tfh cells from tumors did not produce IL4 (Supplementary Fig. S2C). Thus, we asked if Tfh cells could attract Th2 cells into the tumor bed. Using RNA-seq, we analyzed Tfh and Tfh-like cells for the expression of genes encoding proteins known to attract Th2 cells (Ccl17, Ccl22, Il33, eotaxin, Hpgds, and Pacap; refs. 31 –34). Among these genes, we observed that only Hpgds was expressed in Tfh cells isolated from tumors or generated in vitro but not in Th1 cells (Fig. 2A). Using qPCR, we found that in vitro generated Tfh cells and Tfh cells from tumors expressed higher Hpgds than other CD4⁺ T-cell subsets (Fig. 2B and C). Western blot and IF analyses showed that Tfh cells expressed hPGDS at the protein level (Fig. 2D and E). Human (h)PGDS is the enzyme that catalyzes the conversion of PGH₂ to PGD₂. Comparing hPGDS expression at the protein level using IF and PGD₂ production using EIA in the different CD4⁺ T-cell subsets differentiated in vitro or isolated from B16-F10 tumors indicated hPGDS and PGD₂ were mostly expressed and produced by Tfh cells (Fig. 2F–I). Moreover, we performed IF costaining and observed colocalization of hPGDS and PGD₂ in Tfh-like cells (Fig. 2J), and hPGDS expression correlated with PGD₂ production in Tfh cells (Fig. 2K). Together, these data showed that Tfh cells expressed biologically active hPGDS and produced PGD₂.

CRTH2 and PGD₂ receptor 1 (DP1) are both receptors of PGD₂ (35). Therefore, we analyzed, using IF, the presence of CRTH2 and at the membrane of Th2 cells generated in vitro. We observed that CRTH2 was expressed on Th2 cells but not on Th0 or Tfh cells, and DP1 expression was not found on CD4⁺ T cells (Fig. 3A).

PGD₂ can recruit CRTH2-expressing cells by chemotaxis (36). We next performed adoptive transfer of in vitro differentiated OT-II Tfh-like cells (labeled with CellTrace CFSE) and OT-II Th2 cells (labeled with CellTrace FarRed) in mice bearing B16-OVA lung tumor foci. We observed, using fluorescence microscopy, that WT Tfh-like and Th2 cells accumulated in the same clusters in WT and Hpgds-deficient mice, suggesting a chemoattraction mechanism of Th2 cells by Tfh-like cells (Fig. 3B and C). Using a trainable algorithm from inForm software (PerkinElmer), OT-II Tfh-like CFSE⁺ only or OT-II Th2 FarRed⁺ cells were phenotyped and counted. Tfh-like and Th2-cell proportions correlated in lungs of tumor-bearing mice but not when Hpgds-deficient Tfh-like cells were transferred (Fig. 3D). To validate the chemoattractant role of PGD₂ produced by Tfh cells on Th2 cells, we performed transwell migration assays. We observed that synthetic PGD₂ alone could induce Th2-cell migration. We also observed that Tfh-like cells could attract Th2 cells whereas Th0 cells could not. Other CD4^–CRTH2⁺ cells isolated from mouse spleen (mainly represented by CD45⁺CD3⁻Ly6c⁻CD11b⁺CD11c⁻ eosinophils) were used as controls and were also attracted by Tfh-like cells, whereas CD4⁺CRTH2⁻ cells were not recruited by Tfh-like cells. Th2-cell migration was completely abolished when Tfh-like cells were differentiated from Hpgds^−/− mice or when the pharmacological CRTH2 inhibitor TM30089 was added (Fig. 3E).

To determine whether PGD₂ produced by Tfh cells had additional effects on Th2 cells, we incubated in vitro differentiated Th2 cells with increasing doses of synthetic PGD₂ or with supernatant of in vitro differentiated Tfh-like cells. Th0 supernatant was used as a control. We observed that the synthetic PGD₂ as well as the Tfh-like supernatant increased Th2-specific cytokine expression Il4, Il10, and Il5 at the mRNA level detected after 24 hours of treatment (Fig. 3F; Supplementary Fig. S5A and S5B), at the protein level detected after 3 days by ELISA (Fig. 3G; Supplementary Fig. S5C and S5D), and by intracellular staining for IL4 (Fig. 3H). Supernatants from Tfh-like cells deficient for Hpgds (siHpgds or Tfh-like cells generated from Hpgds-deficient mice) lost ability to increase Th2 cytokine expression and production, whereas the deficiency in Hpgds did not affect Tfh-cell frequency, differentiation, proliferation, or function (Supplementary Fig. S6A–S6D). Similar results were observed when we targeted CRTH2 using TM30089 (Fig. 3I and J). Tfh cells produce IL21, which is known to positively affect Th2 cells (37). The ability of Tfh-like cells to increase Th2 cytokine expression and production was not disrupted when Tfh-like cells were treated with a blocking anti-IL21 (Fig. 3I and J). Together these data showed that PGD₂ produced by Tfh cells could increase cytokine production of Th2 cells and induce Th2-cell recruitment to the tumor site in a CRTH2-dependent manner.

Because Hpgds was expressed in Tfh cells and not in other CD4⁺ T-cell subsets (Fig. 2B), we postulated that IL6 and IL21 required for Tfh differentiation might trigger Hpgds expression in Tfh cells. After 24 hours, IL6 and IL21 alone were responsible for a 2-fold induction of Hpgds expression, but when the cells were treated with both IL6 and IL21, Hpgds expression was increased nearly 10-fold at mRNA and protein levels. IL6 is described to induce IL21 production (38). Using a blocking anti-IL21R when cells were cultured in presence of IL6, we observed nonsignificant decrease of Hpgds induction (Fig. 4A; Supplementary Fig. S7A). The amount of PGD₂ detected by EIA was consistent with these results (Supplementary Fig. S7B). Conversely, the use of blocking anti-IL6 and anti-IL21 blunted Hpgds expression in cells treated with both antibodies (Fig. 4B). Thus, both signals were required to induce Hpgds in Tfh-like cells.

IL6 and IL21 can induce STAT1 and STAT3 phosphorylation and activation during Tfh differentiation (39, 40). We observed that both STAT1 and STAT3 were phosphorylated on Y701 and Y705, respectively, as early as 3 minutes after Tfh-like polarization (Fig. 4C). We also confirmed that STAT1 and STAT3 were both activated in ex vivo Tfh cells isolated from tumors (Fig. 4D). Using pharmacologic inhibitors (EGCG and STA21 for STAT1 and STAT3, respectively), we blocked Hpgds induction at the mRNA and protein levels, as well as PGD₂ production (Fig. 4E–G). We postulated that pSTAT1 and pSTAT3 could modulate Hpgds expression together. This cooperation was confirmed by PLA, which showed a proximity between pSTAT1 and pSTAT3 in Tfh-like cells, but not in naïve cells (negative controls; Fig. 4H).

We next bioinformatically analyzed the Hpgds promoter with Genomatix and observed a putative STAT3 binding site (−839 to −820) and a putative pan-STAT binding site (−394 to −375). ChIP experiments on in vitro differentiated Tfh-like cells indicated that pSTAT3 and pSTAT1 could both bind to the identified binding sites, suggesting that they could cooperate at the same DNA sites (Fig. 4I and J). To demonstrate the possibility and necessity of cooperation between pSTAT1 and pSTAT3 for the induction of Hpgds, we performed a ChIP-on-ChIP experiment. The results indicated that the complex involving pSTAT1 and pSTAT3 was only able to bind to the STAT3 putative binding site (Fig. 4K). Finally, we used a luciferase activity reporter assay to validate that the presence of both STAT1 and STAT3 was required for an optimal activation of Hpgds promoter. To this end, we cloned the Hpgds promoter upstream of the luciferase gene and transfected it into 4T1 cells. In these cells, STAT1 and STAT3 phosphorylation was activated by IFNγ and IL6. We observed that STAT activation allowed optimal activity of the Hpgds promoter (Fig. 4L). Together, these data highlight that a protein complex composed, at least, of pSTAT1/pSTAT3 can modulate Hpgds expression in Tfh cells.

Our data suggested that PGD₂ produced by Tfh cells promoted Th2-cell recruitment and function. To test the pathologic relevance of such an observation, B16-F10 melanoma cells were injected intravenously into WT and Hpgds^−/− C57Bl6 mice to induce tumor lung foci. We observed significantly fewer tumor foci in Hpgds^−/− mice (Fig. 5A). hPGDS is expressed at the protein level in the skin and hematopoietic tissues (https://www.proteinatlas.org/ENSG00000163106-HPGDS/tissue). Using EIA, we observed that in vitro differentiated Tfh-like cells produced significant amount of PGD₂ compared with B16-F10 cells (Fig. 5B). Hpgds^−/− mouse lungs had a reduced Th2 infiltrate compared with WT mice but similar Tfh-cell accumulation (Fig. 5C). We then evaluated the effect of Hpgds deficiency on other cells harboring CRTH2 expression by flow cytometry. Th2 cells were the most frequent IL4-producing CRTH2⁺ population in the lungs of tumor-bearing mice, and the recruitment of other cell populations (Tc2, ILC2, eosinophils, basophils, and macrophages) was not statistically affected by the loss of PGD₂ production in Hpgds^−/− mice (Fig. 5D). In the urethane-induced tumor model, we observed fewer lung tumor foci in Hpgds^−/− mice than in WT mice (Fig. 5E), with a decreased Th2 infiltrate but similar Tfh accumulation (Fig. 5F). These results demonstrate that hPGDS has a negative impact on tumor growth and promotes Th2 accumulation.

We previously showed that IL4 is responsible for the protumor effect of Th2 cells in some mouse cancer models, such as the B16-F10 model (30). As expected, we observed here that anti-IL4 had an antitumor effect on lung tumor foci induced by intravenous injection of B16-F10 cells. In contrast, an IL4-blocking antibody resulted in loss of antitumor effects in Hpgds^−/− mice (Fig. 5G), supporting the hypothesis that the hPGDS/PDG₂ pathway is required for IL4 protumor effects. Similarly, when the TM30089 was injected 5 days after B16-F10 tumor cells, it reduced tumor growth only in WT mice but not in Hpgds^−/− mice (Fig. 5H). CRTH2 inhibitor and anti-IL4 combination therapy had no synergistic effect, suggesting that these two molecules act on the same pathway (Fig. 5I). We validated our findings using additional tumor models. We observed that IL4 depletion using anti-IL4 had antitumor activity in subcutaneously injected CT26 colorectal cancer and LLC1 lung cancer models (Balb/cJ and C57Bl6 mice, respectively; Fig. 5J). LLC1 tumor growth was also delayed in Hpgds^−/− mice, and CT26 was inhibited by TM30089 treatment (Fig. 5J). In contrast, in KPC pancreatic cancer and MC38 colorectal cancer models, whose growth was not impacted by the depletion of IL4, we did not observe effects on tumor growth in Hpgds-deficient mice (Supplementary Fig. S8A and S8B), again suggesting that IL4 and Hpgds acted on the same pathway.

Mast cells could be a potential source of PGD₂ (41). In Kit^W-sh mice, which carry the spontaneous Kit “sash” mutation associated with mast-cell deficiency, the CRTH2 inhibitor and/or IL4-blocking antibody had the same antitumor effects on B16-F10 lung tumor foci as in WT mice, suggesting that the role of PGD₂ in our model was not attributable to mast cells (Fig. 5K). Natural Killer (NK) cells have been shown to express DP1, and PGD₂ inhibits cytotoxicity and IFNγ production of NK cells (42). In our B16-F10 lung tumor model, depletion of NK cells by an anti-asiolo-GM1 in WT or Hpgds^−/− mice bearing lung tumors had the same effect in WT mice as in Hpgds^−/− mice, suggesting that NK cells were not inhibited by PGD₂ in this model (Fig. 5L).

To validate the link between Tfh and Th2 cells in vivo, we used Bcl6^f/fCD4^Cre/+ mice and their littermate controls (Bcl6^f/f CD4^+/+). These mice have little or no Tfh cells (43). Using the B16-F10 lung tumor model, we observed an accumulation of Tfh cells in the lungs of tumor-bearing Bcl6^f/fCD4^+/+ mice. Such an increase was not observed in Bcl6^f/fCD4^Cre/+ mice, where these cells were almost absent (Fig. 5M). Th2-cell accumulation was observed in Bcl6^f/f CD4^+/+ mice but not in Bcl6^f/fCD4^Cre/+ mice (Fig. 5N). In addition, tumor growth was impaired in Bcl6^f/fCD4^Cre/+ compared with control mice, and the CRTH2 inhibitor led to loss of antitumor effects in Bcl6^f/fCD4^Cre/+ mice (lack Tfh cells; Fig. 5O). Finally, to confirm Hpgds expression in Tfh cells associated with a lower antitumor efficacy, we transferred OT-II Tfh-like cells in Hpgds^−/− mice bearing B16-OVA lung tumors. The transfer increased the number of lung tumor foci. Conversely, the adoptive transfer of OT-II Tfh-like cells transfected with siHpgds (Tfh siHpgds) exhibited antitumor effects and reduced lung tumor foci (Fig. 5P and Q). These data were confirmed using Tfh-like cells differentiated from TRP-1 mice (Supplementary Fig. S8C and S8D). Together, these results demonstrate that Hpgds expression in Tfh cells is responsible for their protumor effect, whereas in the absence of Hpgds, Tfh cells exert antitumor functions.

We sorted Th1, Th2, Tfh, Th17, and Treg memory cells from the blood of heathy donors according to chemokine receptor labeling (Supplementary Fig. S9A). Using PCR, we observed that only memory Tfh cells expressed a significant level of HPGDS mRNA compared with Th1, Th2, Th17, and Tregs (Fig. 6A), and all Tfh cell subsets expressed the same level of HPGDS (Supplementary Fig. S9B). Using immunofluorescence and EIA, we also validated that only human memory Tfh cells, and no other Th cells, could produce PGD₂ (Fig. 6B and C). The supernatant of human memory Tfh cells increased IL4 production by human memory Th2 cells, as assessed upon PCR after 24 hours (Fig. 6D). Using transwell experiments, we also validated that human Tfh cells could attract Th2 cells, whereas CD4⁺CRTH2⁻ cells used as controls were not attracted (Fig. 6E). To address the in vivo relevance of such an observation on cancer prognosis, we have performed a pan-cancer analysis. We downloaded RNA-seq data from TCGA. Using the previously reported metagene strategy (27), we evaluated the correlation between B-cell, CD8⁺ T-cell, Th2-cell, and Tfh-cell accumulation in different cancer types. The expression of HPGDS and Tfh and Th2 metagenes correlated in six cancer types (bladder, kidney, lung, pancreas, colorectal, and stomach; Fig. 6F; Supplementary Fig. S9C). We focused on colorectal cancer (COAD READ) because the pathology for Tfh/Th2/Hpgds were the most significantly correlated. We observed that patients with high intratumoral Tfh cells and that patients with high Th2 cells in their tumors had a poor prognosis (Fig. 6G and H). Together, these data emphasize that, as in mice, Tfh cells express HPGDS and induce Th2-cell activation. In colorectal cancer, high Tfh and Th2 cells associated with unfavorable patient outcome.

In this study, we showed that Tfh cells produced PGD₂ and that this led to chemoattraction and activation of Th2 cells, which limited Tfh-cell antitumor effects. We found that Hpgds expression was regulated by a pSTAT1/pSTAT3 complex, detected in Tfh cells following IL6 and IL21 pathway activation, both necessary for Tfh differentiation. Targeting the PGD₂–CRTH2 axis in mouse tumor models reduced tumor growth by decreasing Th2-cell recruitment and their IL4 production. In humans, similar properties of Tfh cells were observed, and analysis of TCGA RNA-seq data supported the cross-talk between Tfh and Th2 cells in bladder, kidney, lung, pancreas, colorectal, and stomach cancers.

Studies have previously demonstrated the potential antitumor role of Tfh cells in some cancers, such as breast cancer (2, 15), colorectal cancer (16), hepatocellular carcinoma (17), and non–small cell lung cancer (18). In these cancers, Tfh-cell accumulation associates with a good prognosis. In our study, we have observed Tfh-cell accumulation not only in implanted tumors, but also in chemically induced tumors in mice, suggesting that tumor Tfh-cell recruitment is a general phenomenon observed in humans and in mice.

In the context of adoptive transfer of antigen-specific Tfh-like cells, we observed antitumor activity. The main cytokine produced by Tfh is IL21 (44). The effects of IL21 are pleiotropic due to the wide cell distribution of IL21 receptor. This cytokine plays an essential role in Tfh-dependent human B-cell differentiation and in the generation of humoral immune responses (17). A recent report also suggests that Tfh cells may promote CD8⁺ T-cell activation in the context of human colorectal cancer in an IL21-dependent manner (45) and could improve anti–PD-1 immunotherapy (46). In contrast, another report detected Tfh-cell IL4 production in the draining lymph nodes of some transplantable mouse cancer models. IL4 from these Tfh cells induces myeloid cells to differentiate into protumor M2 macrophages (47). Similarly, in mice and human melanomas, Tfh-like cells have been observed in tumors to suppress CD8⁺ T-cell activation (19). The mechanism behind their immunosuppressive effect remains elusive. We found here that Tfh cells could recruit Th2 cells through an IL4-independent mechanism. Comparing the therapeutic effect of adoptive transfer of WT or Hpgds^−/− Tfh-like cells, we observed a stronger antitumor effect of Hpgds^−/− Tfh-like cells. We therefore believe that the protumor effect of Tfh cells depends on their capacity to produce high levels of PGD₂ and recruit Th2 cells. IL4 produced by Th2 cells could then restrain the antitumor effect of Tfh cells. In our adoptive transfer experiments, Tfh-like cells had an antitumor effect that could be explained by the number of cells transferred. Indeed, we injected between 0.5 to 2 × 10⁶ Tfh-like cells, but in tumors (transplantable or urethane induced), we estimated the number of Tfh cells at a few tens of thousands. Therefore, the large number of Tfh-like cells during adoptive transfer may explain their antitumor effect, which would be masked by the deleterious effects of Th2 cells when the number of Tfh cells is low.

We performed in vitro differentiation of Tfh-like cells by activation of TCR in the presence of IL21 and IL6. Whether these cells are close enough to bona fide Tfh cells may require further evaluation. However, by comparing the transcriptome of in vitro differentiated cells with Tfh cells isolated from tumor-bearing mice, we observed that 13.2% of the genes showed different expression. Therefore, our Tfh-like cells could not be considered as true Tfh cells. Interactions between Tfh cells and B cells, as well as with other antigen-presenting cells, occur in germinal centers and are involved in the development of Tfh cells in vivo. The absence of these signals is probably responsible for the 13.2% of differentially expressed genes. Nevertheless, our in vitro differentiated Tfh-like cells produced similar amounts of IL21, and they expressed specific Tfh transcription factors and membrane markers. In addition, our results seem to indicate that adoptive transfer of in vivo differentiated Tfh cells sorted from mouse spleen (CD4⁺CXCR5⁺PD1⁺ cells) had antitumor activity. Finally, the overall transcriptomic analysis indicated that the cells differentiated in vitro did not resemble any other CD4⁺ T-cell subset. Tfh-like cells did not produce IL4, contrary to Tfh cells isolated from spleen. It is probable that this cytokine production was induced by signals present in vivo, which were not mimicked in vitro (i.e., cell interactions). However, our data and other studies in the literature indicate that Tfh cells present in tumors do not produce IL4. Tfh-like cells showed equivalent levels of HPGDS expression to bona fide Tfh cells.

The regulation of Hpgds was dependent on a STAT1/STAT3 dimer in both cell types. Therefore, the in vitro differentiated cells were not identical to the Tfh cells present in lymphoid organs, and it is probable that the results in the context of Tfh-like cells are incomplete. Nevertheless, the similarity of Tfh-like cells with bona fide Tfh cells allowed us to explore the Tfh–PGD₂–Th2 axis. Mouse and human Tfh cells expressed hPGDS, the enzyme responsible for PGD₂ production. Tfh-cell polarization in vivo requires intercellular interactions between CD4⁺ T cells, dendritic cells, and B cells in the presence of IL6 and IL21 (48). In vitro, using IL6 and IL21 under CD3 and CD28 stimulation, naïve CD4⁺ T cells can be polarized into Tfh-like cells (49). We confirmed, using transcriptomic analysis, that such Tfh-like cells were transcriptionally similar to intratumor Tfh cells. Moreover, even if our Tfh-like cells were not identical to the Tfh cells generated in vivo, the mechanisms responsible for the induction of Hpgds was similar. Indeed, we showed in mouse models that the signaling pathways induced by IL6 and IL21 allowed for Hpgds expression. In agreement with the literature, we observed early activation of STAT1 and STAT3 in Tfh-like cells (40). These transcription factors interact and bind to Hpgds promoter, inducing its activation. The pSTAT1/pSTAT3 complex has already been described after IL6 stimulation (50). However, its role remains unclear. Our main hypothesis is that STAT heterodimers compete with STAT homodimers, thereby reducing the efficacy or strength of STAT homodimer signals (51). The pSTAT1/pSTAT3 complex also exists in Th17 cells, which do not express Hpgds. Tfh cells are differentiated with IL6 and IL21, whereas Th17 cells are differentiated with IL6 and TGFβ. Because TGFβ is described to have no effect on PGD₂ production (52) in mast cells, it is conceivable that IL21 could explain the differences in Hpgds expression between Tfh and Th17 cells. At the same time, our results indicated additive effect of IL6-dependent and IL21-dependent pathways in Tfh cells. IL21 also induces STAT5 phosphorylation, and it would be interesting to determine whether STAT5 has a role in regulating Hpgds expression.

PGD₂ is a major inflammatory mediator (53) mainly produced by mast cells (54). In our report, we provide evidence that mouse and human Tfh cells expressed the hPGDS enzyme. PGD₂ is involved in allergic asthma or atopic dermatitis. Very few studies have examined the role of PGD₂ in cancer. Nevertheless, it is described that certain cancer cells, such as Lewis lung carcinoma in mice (55) or certain gastric cancers in humans, have CRTH2 expression (56). PGD₂ has been described as having antiangiogenic and antitumor properties in genetically modified mice (57). In addition, group 2 innate lymphoid cells (ILC2) also express CRTH2. A recent study assessed the immunosuppressive role of the PGD₂-mediated ILC2/M-MDSC (monocytic myeloid-derived suppressor cell) axis in acute promyelocytic leukemia. In this context, PGD₂ drives ILC2 activation, expansion, and engagement in M-MDSCs (58). Thus, although our findings point to a prominent role of Th2/Tfh cross-talk, the potential role of ILCs will need to be investigated.

We observed in this study that targeting the PGD₂–CRTH2 axis decreased tumor growth in a Th2-dependent manner. In humans, we confirmed that Hpgds expression correlated with Tfh and Th2 metagene expression in pancreatic cancer. The Tfh/Th2 balance associated with patient prognosis, suggesting a similar negative role for HPDGS in human pancreatic cancer.

The role of IL4 and Th2 cells in cancer remains a matter of debate, with some reports suggesting a protumor effect for IL4 (5–9) and others describing an antitumor effect of IL4 (59). Large amounts of Th2 cytokines are observed in the tumor microenvironment and peripheral blood of patients with prostate, bladder, and breast cancers (10–12). IL4 can directly induce the proliferation of colon, breast, head and neck, ovarian, and prostate cancer cells in vitro (6–8). In other contexts, IL4 may blunt tumor proliferation (60) or promote antitumor immunity (61). Th2 cells suppress antitumor immune responses through cytokine secretion. Indeed, all cytokines produced by Th2 cells participate in immune escape (5). In this study, as in our previous report (30), our results indicated that Th2 cells had protumor activity and that IL4 had deleterious activity in mice bearing implanted or chemically induced tumors.

Inhibition of the PGD₂–CRTH2 axis blocks Th2-cell chemoattraction and activation not only in vitro, but also in vivo. From a therapeutic point of view, our in vivo mouse experiments indicated that PGD₂ had protumor effects and that blocking the PGD₂–CRTH2 axis limited tumor growth in B16-F10 melanoma, LLC1 lung cancer, CT26 colorectal cancer, and urethane-induced tumors. In contrast, the antitumor effect was not seen in KPC pancreatic and MC38 colorectal models. We observed, using blocking antibody, that the KPC and MC38 growth was not impacted by IL4, unlike in B16-F10 melanoma, LLC1 lung cancer, CT26 colorectal cancer, and urethane-induced tumors. Thus, the effect of Tfh–PGD₂–Th2 axis blockade on the antitumor response appears to be determined by the tumor dependence on IL4. This parameter may explain why the two models of colorectal cancer did not respond in the same way. Th2 cells associate with poor outcome in adrenal carcinoma, pancreatic cancer, glioma, lung cancer, and sarcoma, suggesting that IL4 can only promote tumor growth in a subset of cancers (62), as we observed in mice. In our study, we found a correlation between Th2, Tfh, and HPGDS in human bladder, kidney, lung, pancreas, stomach, and colorectal cancers. Our data are thus transposable for the lung and the colon, but the difference observed for melanoma and pancreatic cancer models is most likely a reflection of cancer heterogeneity.

In conclusion, by uncovering the link between Tfh and Th2 cells, our results highlight targets that could not only reduce the deleterious effect of Th2 cells but also increase the antitumor effect of Tfh cells. CRTH2 antagonist agents such as OC000459 (timapiprant) (63), BI 671800 (64), AZD198 (65), fevipiprant (66), and a humanized anti-huCRTH2 (67) have already demonstrated significant results in asthma preclinical models (67) and in randomized trials (63–66). Therefore, it could be very interesting to determine whether these CRTH2 inhibitors could have potent effects on tumor growth and OS of patients with cancer, such as pancreatic adenocarcinoma.

J.A. Harker reports grants from Asthma UK, Wellcome Trust, Imperial College London BRC, CW+, and National Heart and Lung Institute Foundation outside the submitted work. L. Apetoh reports grants from Ligue contre le Cancer CCIR-GE during the conduct of the study; personal fees from Orega Biotech and grants from Sanofi outside the submitted work. No disclosures were reported by the other authors.

R. Mary: Formal analysis, investigation. F. Chalmin: Formal analysis, investigation. T. Accogli: Formal analysis, investigation. M. Bruchard: Formal analysis, investigation. C. Hibos: Formal analysis, investigation. J. Melin: Investigation. C. Truntzer: Formal analysis, visualization. E. Limagne: Methodology. V. Derangdol: Investigation. M. Thibaudin: Investigation. E. Humblin: Investigation. R. Boidot: Data curation, supervision, validation, methodology. S. Chevrier: Investigation. L. Arnould: Validation, investigation. C. Richard: Formal analysis. Q. Klopfenstein: Formal analysis. A. Bernard: Investigation. Y. Urade: Resources. J.A. Harker: Resources. L. Apetoh: Resources. F. Ghiringhelli: Conceptualization, validation, writing–original draft, writing–review and editing. F. Végran: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft.

Authors would like to thank the “Plateforme de Cytométrie” (http://www.cytometrie-dijon.fr) for providing reagents and access to cytometers and the “Centre de Zootechnie” (http://recherche.u-bourgogne.fr/potentiel-de-recherche/plateformes-technologiques/centre-de-zootechnie.html) for providing animal bred facilities and devices to perform ex vivo and in vivo experiments. In addition, authors thank Amelia Trimarco and Yoshihiro Urade for respectively providing and designing Hpgds^−/− mice and James A. Harker for providing Bcl6^f/fCD4^cre mice. Finally, authors thank Isabel Grégoire for article editing.

This work was supported by Fondation ARC pour la recherche sur le cancer, Ligue contre le Cancer CCIR-GE, French Government grant managed by the French National Research Agency under the program “Investissements d'Avenir” with reference ANR-11-LABX-0021, Ministère de l'Enseignement supérieur et de la Recherche and by the Fondation de France. FG team is “Equipe labélisée Ligue Nationale Contre le Cancer.”

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.