The COX-2–PGE₂ Pathway Promotes Tumor Evasion in Colorectal Adenomas

Chronic inflammation is one of the known risk factors for colorectal cancer. Clinical and epidemiologic evidence suggests that long-term use of NSAIDs reduces the risk for developing colorectal adenomas and colorectal cancer and suppresses colorectal cancer development (1). Nonselective NSAIDs are known to exert certain anti-inflammatory effects by targeting COX enzymes that include COX-1 and COX-2. COX-1 is constitutively expressed in most tissues. In contrast, COX-2 is an immediate-early response gene usually absent in healthy tissues and organs. However, it is highly inducible at sites of inflammation and in the tumor microenvironment of certain cancers, including colorectal cancer. For example, COX-2 expression is elevated in approximately 50% of colorectal adenomas and 85% of adenocarcinomas (2, 3) and associated with poorer overall survival in patients with colorectal cancer (4). Both COX-1 and COX-2 convert arachidonic acid to prostanoids, including prostaglandins (PG) such as PGE₂, PGD₂, PGF₂α, PGI₂, and thromboxane A₂ (TxA₂) by a two-step process. Among prostanoids, PGE₂ is the most abundant found in various types of human malignancies, including colorectal cancer, and its presence is often associated with a poor prognosis (5–8). More importantly, only PGE₂ and PGI₂ levels are elevated in colorectal cancer specimens compared with matched normal tissues (9).

Moreover, a urinary PGE₂ metabolite (PGE-M) has been evaluated as a biomarker for colorectal cancer (10, 11). More importantly, a recent epidemiologic study revealed that regular use of NSAIDs, including aspirin, significantly reduced the risk of developing colorectal adenomas in women with high PGE-M levels. This effect was not observed in those with low PGE-M levels (12). PGE₂ exerts its cellular effects by binding to specific G protein–coupled receptors, EP1, EP2, EP3, and EP4. Among prostanoids, only PGE₂ has been demonstrated to promote intestinal adenoma formation and growth by induction of tumor epithelial cell proliferation and survival, as well as promoting angiogenesis in mouse models of intestinal adenomas (13). However, all the mechanisms that PGE₂ accelerates colorectal adenoma development are not fully understood. For example, relatively little is known about the impact of PGE₂ on host defenses against tumor cells within the tissue microenvironment.

Tumor immune evasion can occur through several mechanisms, including tumor cell resistance to immune attack, activation of immune checkpoint pathways in effector T cells, suppression of macrophage phagocytosis of tumor cells, and induction of massive infiltration of immunosuppressive cells. For example, tumor cells can evade immunosurveillance by directly impairing CD8⁺ T-cell cytotoxic activity and proliferation through immune checkpoint receptors such as PD-1. Interaction of PD-1 with its ligands, PD-L1 and PD-L2, suppresses CD8⁺ T-cell cytotoxicity and proliferation. It has been documented that a strong and persistent expression of PD-1 is observed in tumor-infiltrating effector T cells (14–16). Tumor-associated macrophages (TAM) are a significant subpopulation of tumor-infiltrating immune cells (17). Increased infiltration of TAMs is recognized as a poor prognostic sign in patients with colorectal cancer (18). A recent report revealed that PD-1 expression in macrophages inhibited their phagocytosis of tumor cells (19). However, the mechanisms by which PD-1 is regulated in CD8⁺ T cells and macrophages in the tumor microenvironment are still largely unknown.

Immunotherapies using checkpoint inhibitors offer great promise for treating some malignancies, but their effectiveness in many solid tumors has been disappointing. For example, the therapeutic effect of a PD-1 mAb was mainly observed in patients with microsatellite instability (MSI) colorectal cancer but not in patients with microsatellite stability (MSS) colorectal cancer (20). However, MSI is only found in about 15% of sporadic colorectal cancer, whereas approximately 85% of sporadic colorectal cancer is classified as MSS. Interestingly, the elevation of COX-2 expression was observed in 79% of MSS colorectal cancer, but only 48% of MSI colorectal cancer (21). Other studies also revealed that MSI colorectal cancer has a low or absent COX-2 expression (22–24). One potential explanation for the poor efficacy of checkpoint inhibitors in MSS colorectal cancer is that the presence of COX-2–derived PGE₂ might attenuate their effect by reducing tumor-infiltrating CD8⁺ T-cell abundance and cytotoxicity and/or by impairing TAM phagocytosis of tumor cells. In this study, we investigate whether activation of the COX-2-PGE₂ pathway induces tumor immune evasion by reducing tumor-infiltrating CD8⁺ T-cell cytotoxicity and abundance via PD-1 and/or by impairing macrophage phagocytosis of tumor cells via PD-1.

All animal experiments were completed according to our animal protocols approved by the Institutional Animal Care and Use Committee at the Medical University of South Carolina (MUSC, Charleston, SC). Apc^Min/+ mice were purchased from Jackson Laboratory (catalog no. 002020). For experiments evaluating celecoxib or Ono-AE3-208 treatment, Apc^Min/+ male mice at the age of 6 weeks old were randomly divided into two groups fed with a control diet, a diet containing 500 ppm of celecoxib, or provided with water containing Ono-AE3-208 (10 mg/kg) for 10 weeks. No significant sex variable exists in the intestinal tumor burden of Apc^Min/+ mice.

All fat and Peyer's patches were removed from excised intestines under a dissecting microscope for intestinal immune cell preparation. Mouse normal intestinal tissues and adenomas were minced and digested with RPMI1640 medium containing 5% FBS, 1 mmol/L MgCl₂, 1 mmol/L CaCl₂, 2.5 mmol/L HEPES, and 200 units/mL collagenase I (Gibco). The immune cells from intestinal tissues were enriched by using a discontinuous (44% and 67%) percoll (GE Healthcare) separation method. Isolated immune cells were subjected to Flow Cytometry or in vitro culture. Excised spleens were smashed using a 40 μm cell strainer. The red blood cells (RBC) in the spleen were lysed with RBC lysis buffer (eBioscience). Mouse CD8⁺ T cells were isolated from the spleen using a mouse CD8⁺ T-Cell Isolation Kit (Miltenyi Biotec Inc.) according to the manufacturer's instructions; isolated CD8⁺ T cells with a purity of 98% were subjected to experiments listed below.

Intestinal tissues were homogenized in PBS with 10% 2,6-di-tert-butyl-p-cresol as described previously (25, 26). The levels of PGs, including PGE₂, were measured using mass spectrometry and normalized with protein concentration as described previously (25, 26).

Isolated mouse splenic CD8⁺ T cells were cultured in RPMI1640 medium with 10% FBS (Gibco, catalog no. 10082147), 25 mmol/L HEPES, 100 IU/mL IL2 (Peprotech, catalog no. 200-02), 1:100 100× non-essential amino acids (Sigma-Aldrich) and 1 mmol/L sodium pyruvate, and 50 μmol/L β-mercaptoethanol. Mouse CD8⁺ T cells were activated by adding anti-CD3/CD28 mAb-coated beads (Dynabeads Mouse T-Activator CD3/CD28, Gibco) according to the manufacturer's instructions. Then cells were starved for 24 hours. Human monocytic THP-1 (catalog no. TIB-202) and HCT-116 (catalog no. CCL-247) cells were obtained from ATCC and maintained in RPMI1640 with 10% FBS. THP-1 monocytes were treated with 50 mg/mL phorbol 12-myristate 13-acetate (PMA) 24 hours to stimulate their differentiation into macrophages, followed by incubation for 24 hours in RPMI medium. Macrophages were treated with 20 ng/mL IL4 (R&D Systems) and 20 ng/mL IL13 (R&D Systems) for 24 hours. THP-1–derived macrophages were starved for 48 hours followed by treatment with the indicated dose of PGE₂ (Cayman Chemical), 0.1 μmol/L Ly294002 (Calbiochem-Novabiochem Corp.), 1 μmol/L MK2206 (Selleck Chemicals), and/or 1 μmol/L BAY110785 (ENZO) for the times indicated. For bone marrow–derived macrophages (BMDM), bone marrow cells were flushed aseptically from the femurs of Apc^Min/+ mice (Jackson Laboratory, catalog no. 000664) and cultured in Falcon Petri dishes (BD Biosciences) with DMEM supplemented with 10% FBS and 10 ng/mL of MCSF for 4 days. After reseeding cells, cells were cultured in DMEM with 10% FBS, 10 ng/mL MCSF (Life Technologies), and 20 ng/mL IL4 (R&D Systems) for 2 days. BMDMs were cultured in serum-free DMEM for 2 days and treated with the indicated dose of PGE₂ and/or inhibitors for indicated time for 24 hours. All cell lines were tested using a MycoProbe Mycoplasma Detection Kit (R&D Systems) and authenticated before each experiment using the ATCC short tandem repeat database.

Total RNA was isolated from indicated cells using an RNeasy Mini Kit (Qiagene), and 1 μg of RNA was reversely transcribed to cDNA using High-Capacity cDNA Reverse Transcription Kits (Bio-Rad). qPCR was performed with a SYB Gene Expression Assay Mix (Bio-Rad) using a QuantStudio 7 Flex Real-Time PCR System (Thermo Fisher Scientific). The sequences of the specific PCR primers were as follows (5′ to 3′): mPdcd1 (PD-1), forward: CCTCTGACACTGTGAGCCAG, reverse: GCAGGTACCCTGGTCATTCA; hPDCD1, forward: CTCCGATGTGTTGGAGAAGC, reverse: CGGCCAGGATGGTTCTTAG; β-actin, forward: AGAAAATCTGGCACCACACC; reverse: AGAGGCGTACAGGGATAGCA. The relative expression of each target gene represents an average of triplicates that are normalized against the transcription levels of β-actin.

Whole-cell or nuclear extracts were prepared from the indicated cells in each experiment. Transfer membranes were blocked with 5% dry milk in TBS-T buffer for 1 hour and then incubated with anti-pAkt (1:1,000, Cell Signaling Technology, catalog no. 4058), anti-Akt (1:1,000, Cell Signaling Technology, catalog no. 4691), anti-NFκB (1:1,000, Santa Cruz Biotechnology, catalog no. SC372), anti-GAPDH (1:5,000, Cell Signaling Technology, catalog no. 8884S), or anti-LaminA (1:1,000, Santa Cruz Biotechnology, catalog no. SC56137) antibody for overnight at 4°C.

For single color flow cytometry analyses, each indicated single-cell suspension was incubated with anti-mPD-1-PE (1:100, BioLegend, catalog no. 135206), anti-hPD-1-PE (1:100), or the corresponding isotype controls in staining buffer (BD Biosciences). For multicolor flow cytometry analyses, mouse intestinal immunocyte cells were incubated with the appropriate combination of the following antibodies in staining buffer at the following dilution for 30 minutes on ice: PD-1–PE (1:100), CD45-PE-Cy7 (1:250, BioLegend, catalog no. 103114), CD8-FITC (1:50, BioLegend, catalog no. 100706), CD4-AF700 (1:100, BioLegend, catalog no. 100536), CD3–Percep5.5 (1:100, Invitrogen, catalog no. 45-0031-82), and V450 (1:1,500, Invitrogen, catalog no. 65-0863-14) for lymphocytes; PD-1–PE (1:100), CD45-APC (1:100, BioLegend, catalog no. 103112), CD11b-FITC (1:10, Miltenyi Biotech, catalog no. 130-081-201), Ly6G-Alexa700 (1:50, BD Biosciences, catalog no. 561236), CD11c-PE-Cy7 (1:80, BioLegend, catalog no. 117318), F4/80-Percep5.5 (1:100, Invitrogen, catalog no. 45-4801-32), V450 (1:1,500) for granulocytes/monocytes. Dead cells were excluded using V450 staining. To analyze INFγ expression on CD8⁺ T cells, immunocyte cells were stained with cell surface markers for CD8⁺ T cells as described above. Then, the cells were fixed and permeabilized using a Cytofix/Cytoperm kit (BD Biosciences, catalog no. 554714) followed by intracellular cellular staining with anti-mouse INFγ-FITC antibody (1:50, BD Biosciences, catalog no. 554411) in permeabilization buffer for 30 minutes on ice. After incubation of antibodies, the cells were analyzed on a Fortessa X-20 cytometer (BD Biosciences). The flow cytometric profiles were analyzed by counting 30,000 events using FlowJo X software (Tree Star).

The Magna ChIP A/G chromatin immunoprecipitation (ChIP) Kit (EMD Millipore, catalog no. MAGNA0017) was used for ChIP assays according to the manufacturer's instructions. Briefly, cells were treated with 1% formaldehyde for 10 minutes at room temperature to cross-link proteins to DNA, then quenched by adding glycine to 0.125 mol/L for 5 minutes at room temperature. Chromatin was sonicated to an average length of 400–600 bp. A total of 5 μg of chromatin was immunoprecipitated by 0.5 μg IgG or anti-NFκB p65 antibody (Cell Signaling Technology, catalog no. 8242). After being reverse cross-linked, the DNA was purified and eluted into 50 μL of elution buffer. Immunoprecipitated DNA was detected by PCR. The primers used for PCR were as follows (5′ to 3′): mPDCD1 (PD-1), forward: 5′-GTGAGACCCACACATCTCATTGC-3′ and reverse: 5′-CCTCACCTCCTGCTTGTCTCTC-3′ and hPDCD1 (PD-1), forward: 5′-AGA GAC ACA GAG GAG GAA GG-3′ and reverse: 5′-AGG GAC TGA GAG TGA AAG GT-3′. All ChIP assays were performed in three independent experiments.

The mouse PD-1 luciferase reporter plasmids with wild-type (WT) and mutant NFκB binding sites (mNFκB1 or mNFκB2) were obtained from Dr. Jeremy Boss (27). Mutant NFκB binding sites were generated by replacing NFκB binding sites with the scrambled sequences. These scrambled sequences were predicted to have no NFκB and other transcription factor binding potential (27). The human PD-1 promoter (−884 to +140) was cloned into a PGL3 vector (Promega, catalog no. E1751). The NFκB1 (GGGGATGGGCC) and NFκB2 (GGGGAGACCCC) elements within the PD-1 promoter were mutated to mutant NFκB1 (GTTTATGGGAA) and mutant NFκB2 (GTTTAGACAAA) sites using the QuikChange II XL Site-Directed Mutagenesis Kit (Agilent Technologies). For dual-luciferase reporter assays, mouse CD8⁺ T cells or THP-1 cells were transfected with 5 μg of a mouse or human PD-1-Luc plasmid and 0.2 μg pRL-TK (Promega, catalog no. E2231) DNA as a control using a Cell line Nucleofector Kit V (Lonza, catalog no. VCA-1003) according to the manufacturer's instructions. After transfection, CD8⁺ T cells were starved for 24 hours and treated with 0.1 μmol/L PGE₂ for 24 hours. Transfected TPH-1 cells were treated with 50 ng/mL PMA for 24 hours. TPH-1–derived macrophages were treated with 0.1 μmol/L PGE₂ for 24 hours after a 48-hour starvation period. Following treatment, cells were lysed using cell lysis buffer provided in the kit (Promega, catalog no. E1960). Luciferase activity was measured using a Dual-luciferase reporter assay kit (Promega) with a Monolight 3010 luminometer (BD Biosciences/Pharmingen). The relative luciferase activity was determined and normalized to Renilla luciferase activity.

Mouse splenic activated CD8⁺ T cells were labeled with 5 μmol/L carboxyfluorescein succinimidyl ester (CFSE; Life Technologies, catalog no. C34554) according to the manufacturer's instructions. CFSE-labeled CD8⁺ T cells were starved for 24 hours and treated with 0.1 μmol/L PGE₂, 20 μg/mL IgG, 20 μg/mL anti-PD-1 (Bio X Cell, catalog no. BE0273), 0.1 μmol/L Ly294002, 1 μmol/L MK2206, and/or 1 μmol/L BAY110785 for 48 hours. For intestinal CD8⁺ T cells, immune cells isolated from intestinal normal tissues or tumors were labeled with 0.5 μmol/L CFSE. CFSE-labeled immune cells were cultured in RPMI1640 medium with 10%FBS for 24 hours. Then the cells were stained with the anti-CD8, anti-CD3 antibodies, and V450. CD8⁺ T cells proliferation was measured by a Fortessa X-20 cytometer.

THP-1–derived macrophages were starved for 4 hours and then treated with 0.1 μmol/L PGE₂, 20 μg/mL IgG, 20 μg/mL anti-PD-1 (nivolumab, BioVision, catalog no. A137-100), 0.1 μmol/L Ly294002, 1 μmol/L MK2206, and/or 1 μmol/L BAY110785 for 24 hours. After treatment, 0.5 × 10⁶ macrophages were cocultured with 1 × 10⁶ GFP-labeled HCT-116 cells for 3 hours. For mouse intestinal macrophages, 5 × 10⁵ mouse intestinal immune cells were cocultured with 2 × 10⁵ GFP-labeled MC26 cells for 3 hours. Human cells were incubated with the antibodies indicated, including hCD45-PE-Cy7 and hCD11b-APC, as well as V450 in flow cytometry buffer. Mouse cells were incubated with antibodies, including mCD45-PE-Cy7, mCD11b-APC, and mF4/80-PerCP/Cy5.5 as well as V450 in flow cytometry buffer. Human macrophages that phagocytose GFP-labeled tumor cells were referred to as CD45⁺CD11b⁺GFP⁺ cells, whereas mouse macrophages that phagocytose GFP-labeled tumor cells were indicated as CD45⁺CD11b⁺F4/80⁺GFP⁺ cells by flow cytometry analysis. The phagocytic index was calculated as the percentage of CD45⁺CD11b⁺GFP⁺ cells in total CD45⁺CD11b⁺ cells for human cells and the percentage of CD45⁺CD11b⁺F4/80⁺GFP⁺ cells in total CD45⁺CD11b⁺F4/80⁺ cells for mouse cells.

Mouse splenic activated CD8⁺ T cells were treated with 30 U/mL IL2, 0.1 μmol/L PGE₂, 20 μg/mL IgG, 20 μg/mL anti-PD-1 (Bio X Cell), 0.1 μmol/L Ly294002, 1 μmol/L BAY110785, and/or 1 μmol/L MK2206 for 48 hours. After treatment, CD8⁺ T cells were cocultured with 2.5 × 10⁵ MC26 in 6-well plate at ratios (E:T = 2:1) for 16 hours. Then, the cells were incubated with anti-mEpCAM-PE-Cy7 (1:1,000, BioLegend, catalog no. 118216) and anti-mCD3-APC (1:100, BioLegend, catalog no. 103112) antibodies in staining buffer (BioLegend) for 30 minutes on ice. The cells were then washed twice with 1 mL of staining buffer and stained with propidium iodide (PI) using a TACS Annexin V-FITC Apoptosis Detection Kit according to the manufacturer's instructions (R&D Systems). The EpCAM⁺CD3⁻PI⁺ cells were dead epithelial cells and the percentage of dead epithelial cells in total epithelial cells (EpCAM⁺CD3⁻) was presented.

All in vitro experiments were performed at least three times. All animal experiments were repeated at least twice. Results were expressed as mean + SEM. In general, we used the Bonferroni test by factorial analysis of variance to compare outcomes among multiple groups of mice. Student t test or Mann–Whitney U test were used to compare two groups. For all tests, P < 0.05 was considered statistically significant.

The data generated in this study are available within the article and its Supplementary Data file. Raw data generated in this study are available from the corresponding author upon request. The data analyzed in this study were obtained from The Cancer Genome Atlas (TCGA) Colorectal Adenocarcinoma Provisional dataset and in Gene Expression Omnibus (GEO) at GSE17537 and GSE17538.

To determine whether the COX-2–PGE₂ pathway regulates immune checkpoint receptors, we first examined the effect of a COX-2 selective inhibitor, celecoxib, on the expression of immune checkpoint receptors in mouse intestinal immune cells. Treatment of Apc^Min/+ mice with celecoxib reduced PD-1 expression in CD8⁺ T cells, which are resident in both small and large intestinal adenomas and matched normal tissues (Fig. 1A). Interestingly, CD8⁺ T-cell abundance in adenomas is much less than that found in normal tissues; however, celecoxib treatment restored CD8⁺ T-cell abundance in intestinal adenomas as well as matched normal tissues (Fig. 1B). Celecoxib treatment also significantly reduced intestinal PG levels, including PGE₂ (Supplementary Fig. S1A), providing evidence that PGE₂ levels correlated with CD8⁺ T-cell abundance by induction of PD-1 in intestinal tumors and matched normal tissues. Similarly, celecoxib treatment also reduced PD-1 expression in macrophages, resident in both small and large intestinal adenomas, and matched normal tissues (Fig. 1C). In contrast, celecoxib treatment does not appear to affect PD-1 expression in other intestinal immune cells (Supplementary Fig. S1B). CTLA4 expression in intestinal CD8⁺ T and CD4⁺ T cells was not significantly affected following celecoxib treatment as well (Supplementary Fig. S1C).

Expression of EP2 and EP4 receptors is higher in human colorectal cancer specimens than in normal colon tissues (28–30). In contrast, the expression of EP3 receptors is reduced in human colorectal cancer specimens compared with adjacent normal colon tissues (30). Multiple lines of in vivo evidence further demonstrated that PGE₂ promotes intestinal adenoma formation and growth via EP2 and EP4 receptors, but not EP3 (31). The involvement of EP1 in intestinal tumorigenesis remains unclear because different groups have reported conflicting results (31). Importantly, immune cells mainly express EP2 and EP4 receptors. For example, macrophages express EP2 and EP4, but not EP1 or EP3 (32). Moreover, the EP4 receptor has a higher affinity for PGE₂ than EP1 or EP2 (33). Indeed, EP2 and EP4 mRNA levels were higher than EP1 and EP3 in splenic CD8⁺ T cells and BMDMs isolated from Apc^Min/+ mice as well as in THP-1–derived macrophages (Supplementary Fig. S2A). Notably, only EP4, but not EP2, is expressed on the cell surface of splenic CD8⁺ T cells and BMDMs isolated from Apc^Min/+ mice (Supplementary Fig. S2B). In human THP-1–derived macrophages, EP2 is marginally detectable, whereas EP4 is highly expressed (Supplementary Fig. S2B). Therefore, we examined whether inhibition of PGE₂ signaling pathways by targeting its EP4 receptor has a similar effect as that seen following treatment with COX-2 inhibitors. Like celecoxib, treatment of Apc^Min/+ mice with an EP4 antagonist (Ono-AE3-208) led to a reduction of PD-1 expression in CD8⁺ T cells and macrophages, which are resident in both small and large intestinal adenomas and matched normal tissues accompanied by increased intestinal CD8⁺ T-cell abundance (Fig. 1D–F). As expected, Ono-AE3-208 treatment did not affect PD-1 expression in other intestinal immune cells or CTLA-4 expression in intestinal CD8⁺ T and CD4⁺ T cells (Supplementary Fig. S2C and S2D).

To further evaluate whether PGE₂ directly induces PD-1 expression in CD8⁺ T cells and macrophages, in vitro experiments were performed. PGE₂ induces PD-1 mRNA levels in mouse splenic activated CD8⁺ T cells and THP-1–derived macrophages (Fig. 2A and B) as well as mouse BMDMs (Supplementary Fig. S3A and S3B). Significantly, PGE₂ induces PD-1 expression levels on the cell surface (Fig. 2C; Supplementary Fig. S3C and S3D). Moreover, time- and dose-dependent effects of PGE₂ on PD-1 expression were observed in these cells (Fig. 2A–C; Supplementary Fig. S3A–S3D).

The PD-1 promoter contains two potential NFκB binding sites located at NFκB1 (−560 to −551) and NFκB2 (−240 to −231) for human and NFκB1 (−1,265 to −1,256) and NFκB2 (−974 to −965) for mouse. LPS has been shown to induce PD-1 expression via binding of NFκB to the NFκB2 binding site within the PD-1 promoter in mouse macrophages (27). In addition, our previous results demonstrated that PGE₂ could activate NFκB via an EP4-dependent PI3K pathway in colorectal carcinoma cells (34, 35). These results prompted us to postulate that PGE₂ induces PD-1 expression by activating NFκB via an EP4–PI3K pathway. To test this hypothesis, we first examined whether PGE₂ induces AKT and NFκB activation. As shown in Fig. 2D, treatment of mouse splenic activated CD8⁺ T cells and THP-1–derived macrophages with PGE₂ led to increased phospho-AKT levels, and nuclear translocation of NFκB. PGE₂ induction of NFκB nuclear translocation was also observed in mouse BMDMs (Supplementary Fig. S3E). Moreover, inhibition of PI3K, AKT, or NFκB by each respective inhibitor attenuated the effect of PGE₂ on induction of PD-1 expression in mouse splenic activated CD8⁺ T cells, THP-1–derived macrophages, and mouse BMDMs (Fig. 2E; Supplementary Fig. S3F). Interestingly, Ly294002 treatment at 0.1 μmol/L inhibited PGE₂ induction of PD-1 mRNA expression in these cells (Supplementary Fig. S3G).

To further examine whether PGE₂ induces PD-1 transcription by increasing binding of NFκB to cis-elements in the PD-1 promoter, ChIP and PD-1 luciferase reporter assays were performed. As shown in Fig. 3A and B, PGE₂ induced binding of NFκB to the PD-1 promoter in both mouse splenic activated CD8⁺ T cells and THP-1–derived macrophages. Moreover, PGE₂ induces PD-1 transcription in both mouse splenic activated CD8⁺ T cells and THP-1–derived macrophages (Supplementary Fig. S3H). As expected, site-directed mutations of the NFκB2 binding element within the PD-1 promoter abolished PGE₂ induction of transcription in these cells, whereas mutations of the NFκB1 binding site within the PD-1 promoter did not affect PGE₂ induction of transcription (Fig. 3C and D). These results demonstrate that only the NFκB2 binding site is required for PGE₂ induction of PD-1 transcription. Because EP4, but not EP2, is expressed on the surface of these cells, our results reveal that PD-1 induction is regulated mainly via the EP4–PI3K–AKT–NFκB pathway.

We first evaluated the impact of PGE₂ on CD8⁺ T-cell proliferation and cytotoxicity against tumor cells and found that PGE₂ inhibited proliferation of mouse splenic activated CD8⁺ T cells in a time-dependent manner (Fig. 4A). In addition, PGE₂ suppressed mouse splenic activated CD8⁺ T-cell cytotoxicity against MC26 cells in a dose-dependent manner (Supplementary Fig. S4A). Blockage of PD-1 signaling by using a neutralizing antibody attenuated the effect of PGE₂ on suppression of CD8⁺ T-cell proliferation and cytotoxicity against MC26 cells (Fig. 4B and C). Moreover, inhibition of PI3K, Akt, or NFκB by each respective inhibitor also blocked the effect of PGE₂ on suppression of CD8⁺ T-cell proliferation and cytotoxicity against MC26 cells (Fig. 4B and C). Like CD8⁺ T cells, PGE₂ reduced THP-1–derived macrophage phagocytosis of HCT-116 cells (Fig. 4D) and mouse BMDM phagocytosis of MC26 cells (Supplementary Fig. S4B). Treatment with a PD-1 mAb, a PI3K inhibitor, an Akt inhibitor, or an NFκB inhibitor attenuated the effect of PGE₂ on inhibition of THP-1–derived macrophage phagocytosis of HCT-116 cells (Fig. 4D) and BMDM phagocytosis of MC26 cells (Supplementary Fig. S4B). These results demonstrate that PGE₂ suppresses CD8⁺ T-cell proliferation and cytotoxicity and impairs macrophage phagocytosis of tumor cells via an EP4–PI3K–AKT–NFκB–PD-1 pathway in vitro.

Because inhibition of the COX-2–PGE₂–EP4 pathway reduces PD-1 expression in intestinal CD8⁺ T cells and macrophages, it was conceivable that inhibition of this pathway could induce intestinal CD8⁺ T-cell activation and macrophage phagocytosis of tumor cells. As shown in Fig. 5A–D, treatment of Apc^Min/+ mice with celecoxib or Ono-AE3-208 resulted in increased proliferation and IFNγ production in CD8⁺ T cells in intestinal adenomas and matched normal tissues, demonstrating that inhibition of the COX-2–PGE₂–EP4 pathway increases CD8⁺ T-cell activation in intestinal adenomas and matched normal tissues. Treatment of Apc^Min/+ mice with celecoxib or Ono-AE3-208 also enhanced the ability of macrophages isolated from intestinal adenomas and matched normal tissues to phagocytize mouse MC26 cells (Fig. 5E to F). As expected, treatment of Apc^Min/+ mice with celecoxib or Ono-AE3-208 reduced both small and large adenoma burden (Fig. 5G and H). These results suggest that PGE₂ accelerates colorectal adenoma formation by suppressing CD8⁺ T-cell cytotoxicity and macrophage phagocytosis against transformed epithelial cells via PD-1 in normal tissues and promotes colorectal adenoma growth by suppressing tumor-infiltrating CD8⁺ T-cell cytotoxicity and TAM phagocytosis of tumor cells via PD-1.

To further assess whether our preclinical results may have clinical relevance in humans, we first examined the correlation between mRNA levels of PDCD1 (PD-1) and PTGS2 (COX-2) in a human colorectal cancer database. We found a positive correlation between COX-2 and PD-1 mRNA levels in two TCGA and GEO databases (Fig. 6A and B). An evaluation of the GEO database reveals that patients with colorectal cancer with high levels of both COX-2 and PD-1 in tumor tissues experienced a much poorer overall survival than those with low levels of both COX-2 and PD-1 (Fig. 6C). In TCGA database, patients with colorectal cancer with high levels of both COX-2 and PD-1 in tumor tissues tended to have poorer survival compared with patients with low levels of both COX-2 and PD-1. However, the difference does not reach statistical significance (P = 0.068; Fig. 6D).

Some of the roles of PGE₂ in regulating immunity and host defense against viral, fungal, and bacterial pathogens have been reported previously (36). For example, one study showed that PGE₂ suppressed CD8⁺ T-cell survival and function during a chronic lymphocytic choriomeningitis virus infection (37). It was not clear whether PGE₂ induces tumor immune invasion by inhibiting CD8⁺ T-cell proliferation and cytotoxicity in vivo. This report provides the first in vivo evidence demonstrating that inhibition of the COX-2–PGE₂–EP4 pathway increases intestinal CD8⁺ T-cell abundance and activation accompanied by a reduction of PD-1 and tumor burden. Moreover, two in vivo studies suggest that PGE₂ might promote intestinal adenoma formation and growth via its effects on macrophages. One study revealed that treatment with celecoxib resulted in a reduction of polyp burden accompanied by conversion of TAMs from the M2 to M1 in the Apc^Min/+ mice (38). Another study reported that deletion of EP4 in myeloid cells resulted in reducing adenoma burden in the Apc^Min/+ mice (39). However, the mechanisms by which PGE₂ promotes colorectal adenoma formation and growth via macrophages were not fully understood or examined. Here our results demonstrate for the first time that inhibition of the COX-2–PGE₂–EP4 pathway increases intestinal macrophage phagocytosis of tumor cells accompanied by reduced levels of PD-1 in intestinal macrophages as well as decreased tumor burden. The most striking of our findings is that inhibition of the COX-2–PGE₂–EP4 pathway promotes intestinal CD8⁺ T-cell cytotoxicity and macrophage phagocytosis of epithelial cells via PD-1, which occurs well before tumor formation. This finding provides a potential explanation for the chemopreventive effect of NSAIDs on colorectal cancer that occurs very early during precancer development.

Following immune activation or infection, PD-1 expression is elevated in effector T cells, macrophages, dendritic cells, B cells, and natural killer T cells. PD-1 expression is induced in CD8⁺ T cells following T-cell receptor (TCR) stimulation via NF-AT (40). However, TCR-dependent PD-1 induction is weak and transient. It has been well documented that a strong and persistent expression of PD-1 is observed during chronic viral infection (41). Several cytokines such as INFα, IL6, and IL12 have enhanced TCR-induced PD-1 expression in CD8⁺ T cells (42, 43). In addition to CD8⁺ T cells, PD-1 expression can be regulated by cytokines and Toll-like receptor ligands such as LPS in macrophages. For example, LPS, TNFα, INFα, and IL1β induced PD-1 expression in macrophages in vitro (27, 44, 45). However, how PD-1 is regulated in CD8⁺ T cells and macrophages in the tumor microenvironment is still largely unknown. Previous studies have shown that PD-1 expression in CD8⁺ T cells can be upregulated by different transcription factors such as FoxO1, IRF9, Notch, and c-Fos (42, 46–48), and we found that activation of NFκB by PGE₂ via an EP4–PI3K–AKT pathway initiates PD-1 transcription via binding to the PD-1 promoter in CD8⁺ T cells. Our finding extends the scope of our understanding of the underlying mechanisms for PD-1 regulation in CD8⁺ T cells and macrophages. Moreover, our results also reveal that PGE₂ leads to AKT activation in both CD8⁺ T cells and macrophages. Although AKT has been shown to activate NFκB by phosphorylation of IKKα (49), further research is needed to determine the molecular mechanisms by which PGE₂-induced AKT activation induces NFκB activation in CD8⁺ T cells and macrophages. In addition, our results also reveal that PGE₂ only induces PD-1, but not CTLA4, in CD8⁺ T cells. This finding indicates that combined inhibition of PGE₂ signaling and CTLA4 may have a greater antitumor efficacy than combined inhibition of PGE₂ signaling and PD-1.

Previous in vitro studies showed that PGE₂ regulates CD8⁺ T-cell proliferation and function. For example, treatment of CD8⁺ T cells with PGE₂ resulted in suppression of cell proliferation by inducing replicative senescence (50). PGE₂ treatment also inhibited CD8⁺ T-cell cytotoxicity by inducing CD94 and the NKG2A C-type lectin receptor complex (51) or by inhibiting IFNγ release induced by TCR (52). Moreover, PGE₂ produced by carcinoma cells also blocked initial priming of naïve CD8⁺ T cells to cytotoxic T cells by tumor cells (53). Here we reveal a novel mechanism by which PGE₂ inhibits CD8⁺ T-cell proliferation and cytotoxicity against tumor cells by inducing PD-1 via an EP4–PI3K–AKT–NFκB pathway. In addition, in vitro studies have shown that PGE₂ promotes M2 macrophage polarization via a CREB/CRTC pathway in BMDMs (54) and the COX-2–PGE₂ pathway mediated the effect of bladder tumor cells on induction of PD-L1 in BMDMs and MDSCs in a coculture system (55). These PD-L1–positive cells were immunosuppressive (55). Moreover, PGE₂ also inhibited alveolar macrophage phagocytosis against bacteria (56). However, the mechanisms underlying PGE₂ regulation of macrophage phagocytosis have not been determined. This study found that PGE₂ suppresses macrophage phagocytosis of tumor cells by induction of PD-1 via an EP4–PI3K–Akt–NFκB pathway.

PD-1⁺ tumor-infiltrating lymphocytes are associated with poor prognosis in human breast cancer (57) and soft-tissue sarcomas (58) and correlate with poor survival in patients with high-grade upper tract urothelial carcinomas (59) and gastric cancer (60). PD-1⁺ tumor-infiltrating immune cells are also associated with advanced tumor–node–metastasis stage and poor survival in patients with renal cell carcinoma (61) and operable breast cancer (62). Consistent with these reports, our results showed that COX-2 levels are positively associated with PD-1 in human colorectal cancer specimens. A recent study revealed that PGE₂ levels are also associated with PD-1 levels in tumor-infiltrating CD8⁺ T cells in human lung cancer specimens (63), consistent with our results. Moreover, our results reveal that high levels of both COX-2 and PD-1 are associated with poorer overall survival in patients with colorectal cancer. Recent studies showed that combined inhibition of PGE₂ signaling and PD-1 increased CTL proliferation in vitro (64), and EP4 antagonists enhanced antitumor efficacy of PD-1 in a syngeneic mouse model of colorectal cancer (65) and in a mouse model of colitis-associated tumorigenesis (66).

In summary, we report that PGE₂ promotes tumor immune evasion by suppressing CD8⁺ T-cell cytotoxicity and macrophage phagocytosis against tumor cells by induction of PD-1 via an EP4–PI3K–AKT–NFκB pathway. In contrast, inhibition of the COX-2–PGE₂–EP4 pathway suppresses tumor immune evasion by increasing CD8⁺ T-cell abundance as well as cytotoxicity and enhances macrophage phagocytosis of tumor cells. These findings uncover a novel role of PGE₂ in tumor immune evasion and may provide a rationale for the development of new therapeutic approaches to subvert tumor immune evasion by targeting immune checkpoint pathways using EP4 antagonists. Given that EP4 antagonists have similar efficacy as celecoxib in the prevention and treatment of colorectal adenomas and colorectal cancer, the use of these agents will likely avoid the serious cardiovascular side effects associated with other NSAIDs and COXIB use.

J.D. Lang reports grants from NIH outside the submitted work. R.N. DuBois reports grants from Department of Defense during the conduct of the study. No disclosures were reported by the other authors.

J. Wei: Investigation. J. Zhang: Investigation. D. Wang: Conceptualization, supervision, writing–original draft, writing–review and editing. B. Cen: Investigation. J.D. Lang: Data curation. R.N. DuBois: Supervision, funding acquisition, writing–review and editing.

This work was supported in part by NIH R01 DK047297 and Flow Cytometry & Cell Sorting Unit, Hollings Cancer Center, Medical University of South Carolina (P30 CA138313).

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.