Constitutive IDO1 Expression in Human Tumors Is Driven by Cyclooxygenase-2 and Mediates Intrinsic Immune Resistance

Clinical results from cancer therapy based on the stimulation of antitumor immune responses has introduced a paradigm shift in oncology. By releasing brakes that repress T lymphocytes, antibodies able to block inhibitory receptors CTLA-4 or PD1 expressed at the surface of T cells can unleash antitumor immunity, inducing durable tumor responses and improving survival of advanced cancer patients (1). However, this favorable outcome is observed in only 20% to 40% of patients. In other patients, tumors appear to establish an immunosuppressive microenvironment that prevents the action of antitumor T lymphocytes. Further progress therefore depends on our ability to understand and target these immunosuppressive mechanisms, which are likely manifold and involve a variety of cellular and molecular pathways.

Tryptophan catabolism has emerged as a factor orchestrating immunosuppression in the tumor microenvironment. Aberrant tumoral expression of cytosolic tryptophan-degrading enzyme indoleamine 2,3-dioxygenase 1 (IDO1) in tumors results in rapid conversion of tryptophan into kynurenine. The resulting microenvironment, depleted in tryptophan and enriched in kynurenine, is immunosuppressive (2, 3). Indeed, tryptophan shortage induces an arrest of T-lymphocyte proliferation, apparently linked to the induction of a GCN2-mediated integrated stress response, whereas kynurenine and other tryptophan catabolites can induce T-cell apoptosis and/or favor the induction of regulatory T cells (4–6). In humans, IDO1 is normally expressed only in a restricted set of cells, including placental endothelial cells and mature dendritic cells. Studies in mice demonstrated the key role of placental IDO1 to prevent immune rejection of the fetus (7, 8). IDO1 expression in mature dendritic cells is believed to contribute to the retro-control of immune responses (9). In line with this function is the fact that IDO1 expression can be induced in most tissues and cell types upon exposure to IFNγ (10, 11). Because IFNγ is produced by T lymphocytes at the peak of their response, the ensuing induction of IDO1 ensures a negative feedback of the T-cell response. Indeed, IDO1-deficient mice develop exacerbated autoimmune pancreatitis triggered by the injection of Freund's adjuvant (12).

The availability of small-molecule inhibitors able to block IDO1 enzymatic activity allowed confirmation of the immunosuppressive role of tumor IDO1: in a number of preclinical models, the inhibition of IDO1 activity resulted in antitumor effects that were synergistic with various immunotherapy approaches, including vaccination and checkpoint blockade (13, 14). In a wide range of human tumors, IDO1 expression was also correlated with a worse prognosis and a reduced T-cell infiltration (15–17). These results have prompted the clinical development of IDO1 inhibitors, which are now in phase III trials.

Although mouse studies highlighted increased IDO1 expression in tumor-draining lymph nodes, this was not observed in humans (9). In addition, mechanistic studies indicated that the relevant IDO1 for tumor-induced immune suppression was present at the tumor site (18). There are distinct patterns of IDO1 expression in human tumors (9). A number of tumors show IDO1 expression associated with a T-cell infiltrate and signs of inflammation. This expression likely results from induction by IFNγ produced by infiltrating T cells. Such a pattern of expression, indicative of adaptive immune resistance, has been observed for other immune counter-regulatory molecules such as PD-L1 (19). In contrast, many human tumors express IDO1 in the absence of any inflammation and T-cell infiltrate, representing a state of intrinsic immune resistance (3, 9). This may account for the clinically relevant group of “cold” or “noninflamed” tumors, which fail to respond to checkpoint inhibitors due to the lack of T-cell infiltrate. This pattern of IDO1 expression is recapitulated in a number of human tumor cell lines that express IDO1 in a constitutive manner, i.e., in the absence of IFNγ (13). Little is known about the signaling mechanisms accounting for constitutive IDO1 expression in human tumors.

Here, we report the characterization of the signaling pathways accounting for constitutive IDO1 expression in a series of human tumor cell lines. Our results show that autocrine prostaglandin E₂ triggers PKC and PI3K signaling, which in turn leads to IDO1 transcription. PGE₂ production results from constitutive expression of COX-2 triggered by MAPK signaling. These tumors harbor oncogenic mutations in one or several of these signaling pathways, which likely contribute to constitutive IDO1 expression making up an immunosuppressive tumor phenotype. Blocking COX-2 activity in vivo prevents IDO1 expression and restores immune rejection of human tumors in a mouse model reconstituted with human lymphocytes. Our results provide a mechanistic explanation for the reported role of COX-2 in tumor evasion of immunity (20–22). Our results also highlight a link between oncogenic signaling and immunosuppression in human tumors and suggest the use of COX-2 inhibitors as immune modulators to increase the clinical efficacy of cancer immunotherapy.

KUL98-MELA and KUL98-MELB are melanoma nodules that were obtained from UZ Leuven Gasthuisberg in 2003 and 2004, respectively. Authentication by DNA finger printing was last done in September 2014. Mycoplasma tests were done in July 2004 for KUL98-MELA and April 2005 for KUL98-MELB. LB919-SCCHN was derived at the Ludwig Institute Brussels Branch from a patient sample donated by Cliniques universitaires Saint-Luc in February 1994 and tested for mycoplasma in April 1994. No authentication was carried out. LB1610-MEL was derived at the Ludwig Institute Brussels Branch from a patient sample donated by Cliniques universitaires Saint-Luc in July 1997 and last tested for mycoplasma in November 2012. Authentication was last carried out in May 2011 by DNA finger printing. LB188-SAR was derived at the Ludwig Institute Brussels Branch from a sample donated by Hôpital Jolimont in May 1991 and last tested for mycoplasma in January 1992. No authentication was done. LS411N was received from collaborators in Lausanne in 1990. Neither mycoplasma test nor authentication has been done. NCI-H596 was purchased from the ATCC in 1995 and last tested for mycoplasma in November 2012. No authentication has been carried out. U87 was purchased from CLS Cell Lines Services GmbH in 2011. Neither mycoplasma test nor authentication has been done. A172 was purchased from Antisense Pharma GmBH in 2004 and last tested for mycoplasma in November 2012. No authentication has been carried out. MZ-CHA-3 was received from collaborators in Mainz in 1991 and last authenticated in May 2011 by DNA finger printing. No mycoplasma test has been done. SKOV3 was purchased from CLS Cell Lines Services GmbH in 2014 and last authenticated by DNA finger printing in August 2016. No mycoplasma test has been done. Cells were cultured under standard conditions. For PGE₂ metabolite quantification, cells were cultured in X-VIVO 15 medium (Lonza). All mycoplasma tests were done by qPCR.

LY294002 (4402-4) (used at 20 μmol/L for 24 hours) and R0-31-8425 (557514) (20 μmol/L for 24 hours) were obtained from Calbiochem. GDC0941 (957054-33-0) (5 μmol/L for 72 hours) was bought from Axon Medchem. A66 (S2636) (20 μmol/L for 72 hours), TGX221 (S1169) (30 μmol/L for 72 hours), MK2206 (S1078) (10 μmol/L for 72 hours), and ruxolitinib (S1378) were bought from Selleckchem. CAY10598 (13281) (10 μmol/L for 24 hours) was obtained from Cayman Chemicals. H-89 (B1427) and rapamycin (R3781) (200 nmol/L for 8 hours) were bought from Sigma. PLX4032 (S1267)(10 μmol/L for 72 hours) was bought from Selleckchem. U0126 (BML-EI282-0001) (20 μmol/L for 72 hours), and SB203580 (EI-286-0005) (10 μmol/L for 72 hours) were obtained from ENZO. Celecoxib (used at 5 μg/mL) was bought by prescription from Pfizer. MF63 (10 μmol/L) was bought from AbMole BioScience. AH6809 (14050), GW627368X (10009162) both used at 10 μg/mL were obtained from Cayman Chemicals. Celecoxib, MF63, AH6809, and GW627368X were added twice (days 0 and 2) on a 3-day period. Epacadostat was obtained from Chemietek. Human recombinant IL1β (200-01B) (1 μg/mL for 12 hours) and human IFNγ (300-02) were bought from Peprotech. Anakinra (Kineret 10 mg/0.67 mL), used at 50 μg/mL for 72 hours, was bought by prescription from Sobi.

Cells were lysed in homemade lysis buffer [0.1% SDS, 1% deoxycholic acid, 0.5% NP40 complemented with protease and phosphatase inhibitors (Roche)]. Proteins were separated on precast NuPage 4% to 12% Bis-Tris gels (Invitrogen) in MOPS running buffer. Dry transfer was done with the iBlot from Thermo Scientific. Membranes were blocked in PBS or TBS containing 5% dry milk and 0.1% Tween 20 and then probed with the primary antibodies at 4°C overnight. Images were acquired by Fusion Fx camera from Vilber Lourmat. Quantification was obtained with the Bio-1D (version 15.06b) software from Vilber Lourmat.

p-AKT (Ser 473) rabbit mAb (#4060 from Cell Signaling Technology), AKT rabbit mAb (#4691 from Cell Signaling Technology), p-GSK3β (Ser9) rabbit mAb (#5558 from Cell Signaling Technology), GSK3β mouse mAb (#9832 from Cell Signaling Technology), p-p70S6K (T389) rabbit mAb (#9234 from Cell Signaling Technology), p70S6K rabbit mAb (#2780 from Cell Signaling Technology), p-CREB (Ser133) rabbit mAb (#9191 from Cell Signaling Technology), CREB rabbit mAb (#9197 from Cell Signaling Technology), p-p44/42 MAPK (ERK1/2) (Thr202/Tyr204) rabbit mAb (#4370 from Cell Signaling Technology), p44/42 MAPK rabbit polyclonal Ab (#9102 from Cell Signaling Technology), β-catenin rabbit mAb (#8480 from Cell Signaling Technology), p-STAT3 (Y705) rabbit mAb (#9145 from Cell Signaling Technology), STAT3 rabbit mAb (#4904 from Cell Signaling Technology), p-STAT1 (Tyr 701) rabbit polyclonal Ab (#9171 from Cell Signaling Technology), and STAT1 rabbit polyclonal Ab (#9172) were diluted at 1/1,000. IDO1 mouse antibody (4.16H1 clone 7) was homemade and used at 2 μg/mL (9). Actin mouse antibody (A5441) bought from Sigma was used at 1/10,000. Vinculin mouse mAb (hVIN-1) antibody (V9131) bought from Sigma was used at 1/2,000.

Secondary antibodies were incubated 1 hour at room temperature before revelation with the cheluminescent substrate West Pico SuperSignal sold by Thermo Scientific. Anti-rabbit HRP-linked IgG (#7074 from Cell Signaling Technology) was used at 1/3,000, and anti-mouse HRP-conjugated IgG (HAF007 from R&D Systems) was used at 1/5,000.

RNA was extracted by Tripure (Roche), and the retro-transcription was done with the revertAid kit from Thermo Scientific (K1691). RT-qPCR was performed with the Takyon Rox probe core kit dTTP from Eurogentec (UF-RPCT-C0201). RT-qPCR for IDO1 (Forward: GGTCATGGAGATGTCCGTAA; Reverse: ACCAATAGAGAGACCAGGAAGAA; Probe: CTGTTCCTTACTGTCAACTCTCCAAGAAACTG), mPGES1 (F: GCTGGTCATCAAGATGTACG; R: CCAGGTAGGCCACGGTGTCT), EP2 (F: CCACCTCATTCTCCTGGCTA; R: CGACAACAGAGGACTGAACG), EP4 (F: TGGTATGTGGGCTGGCTG; R: GAGGACGGTGGCGAGAAT), and GAPDH (F: TCAACGACCACTTTGTCAAGC; R: CCAGGGGTCTTACTCCTTGG; Probe: CCTGGTATGACAACGAATTTGGCTACAGC) was performed with an elongation condition of 62°C for 1 minute. RT-qPCR for COX-2 (F: CAGGCTAATACTGATAGGAGAGA; R: TTGACAACTATCAACATAAAGACCAGT), EP1 (F: GGTATCATGGTGGTGTCGTG; R: GGCCTCTGGTTGTGCTTAGA), EP3 (F: CTTCGCATAACTGGGGCAAC; R: TCTCCGTGTGTGTCTTGCAG), IL1β(F: CTGCCCACAGACCTTCCA; R: GGACCAGACATCACCAAGC; Probe: AATGACCTGAGCACCTTCTTTCCTT), IL8 (F: CTGAGACTGATTGAGAGTGGA; R: GTGTTGAAGTAGATTTGCTTGAAG; Probe: AGCTCTGTCTGGACCCCAAGGAAAAC) CD3ϵ (F: GGCAAGATGGTAATGAAGAAATGG; R: AGGGCATGTCAATATTACTGTGGTT), CD8β (F: CAGCTGAGTGTGTGGTTGATTTCC; R: ACCGGCACACTCTCTTCTTGA; Probe: CCCACCACTGCCCAGCCCAC), and Actin (F: GGCATCGTGATGGACTCCG; R: GCTGGAAGGTGGACAGCGA; Probe: TCAAGATCATTGCTCCTCCTGAGCGC) was performed with an elongation condition of 60°C for 1 minute.

NSG mice (from The Jackson Laboratory) were bred at the animal facility of the Ludwig Institute for Cancer Research, Brussels, Belgium. Handling of mice and experimental procedures were conducted in accordance with national and institutional guidelines for animal care. NSG mice were injected subcutaneously with 4 × 10⁶ SKOV3 or LS411N cells. After 7 days (d-3), the tumors were measured and randomized groups of mice (9 for SKOV3 and 8 for LS411N) were treated or not with celecoxib (10 mg/kg once a day) and/or IDO1 inhibitor epacadostat (100 mg/kg twice a day) by oral gavage (d-3). Control mice received methocel (100 μL). Three days later (d0), the mice were injected or not with 3 × 10⁶ human allogeneic peripheral blood mononuclear cells (PBMC; d0) in the tail vain. Under those conditions, mice did not develop graft-versus-host disease (GVHD) before day 40. Tumors were measured at regular intervals with an electronic caliper. Mice were sacrificed after 40 days and tumors were resected and frozen for immunochemistry, RNA, and protein extraction. RNA and proteins were extracted after crushing the tumors in liquid nitrogen.

PBMCs were purified from the blood of a hemochromatosis donor (LB2419), courtesy of Saint-Luc University Hospital, by centrifugation on a Lymphoprep gradient and then frozen at −80°C.

Recombinant lentiviruses were produced by transient cotransfection of 293T cells in 100-mm tissue culture dishes with 4.2, 2.1, and 2.5 μg of pGAG-POL, pREV, and PVSV-G, respectively, for viral encapsidation. At the same time, the shSTAT3 pGIPZ vector (V3LHS_641817 from Dharmacon) or the pGIPZ empty vector (RHS4351 from Dharmacon) was transfected too. The transfection was performed using the calcium phosphate precipitation method. The medium was replaced 4 hours later and virus-containing supernatants were harvested after 24, 48, and 72 hours after transfection. Cancer cells were then plated in a 6-well plate and 500 μL of viral supernatant and 500 μL of fresh media were added after 4 hours. Seventy-two hours later, supernatants were discarded and fresh media were added with 3 μg/mL of puromycin for selection.

Cells were treated with 5 μg/mL of celecoxib or 10 μg/mL of GW627368X every 2 days, 20 μmol/L LY294002 or 10 μmol/L MK2206, 20 μmol/L Ro-31-8425, or 10 μmol/L SB203580 every 3 days. Cell culture supernatants were harvested up to 7 days after plating and analyzed by HPLC as previously described (23). Time point 0 corresponds to cell-free medium.

Cells were cultured in X-VIVO 15 medium (Lonza) for 5 days before supernatants were collected and PGE₂ metabolites (13,14-dihydro-15-keto PGA₂ and 13,14-dihydro-15-keto PGE₂) quantified by a prostaglandin E metabolite EIA kit from Cayman Chemical. Control corresponds to cell-free X-VIVO 15 medium.

KUL98-MELA and KUL98-MELB cells were cultured in X-VIVO 15 medium for 3 days. The supernatant was harvested, and IL1β was quantified with the human IL1β ELISA Ready-SET-Go from Affymetrix-eBioscience (88-7261-88).

siRNA (100 nmol/L) directed against CTNNB1 and the negative control siRNA diluted in Opti-MEM medium (ThermoFisher Scientific) were first incubated with lipofectamin 2000 (Invitrogen) before adding the lipofectamin-siRNA mixture on the cancer cells for 6 hours. The medium was changed after 6 hours, and cells harvested after 48 hours were analyzed. CTNNB1 Silencer validated siRNA (AM51331) and Silencer Negative Control siRNA (AM4611) were purchased at Ambion ThermoFisher Scientific.

RNA samples were prepared from cell lines using the guanidinium isothiocyanate/cesium chloride procedure (24). Mutations in KUL98-MELA, KUL98-MELB, MZ-CHA-3, and LB1610-MEL cell lines were visually detected in selected genes from RNAseq data performed on an in-house SOLID platform. Mutations in LB188-SAR and LB919-SCCHN cell lines were visually detected in selected genes from RNAseq data performed on an Illumina TrueSeq platform. Coverage was of 470M reads (2 × 125 bp paired end) at Beckman Coulter Inc.

Expression data and mutation data were extracted from the Cancer Cell Line Database from the Broad Institute in collaboration with Novartis. The database was validated by quantifying IDO1 expression by RT-qPCR in a series of cancer cell lines available at the Ludwig Institute. Significance of the comparisons between the IDO1-positive and IDO1-negative populations was assessed by unpaired t-test with Welch correction (unequal variance, F test was significant). Correlation between IDO1 and COX-2 or mPGES1 was evaluated by a Spearman correlation test. The differences in proportions between IDO1-positive mutated cell lines and IDO1-positive nonmutated cancer cell lines were tested by a unilateral Fisher test. The mutations taken in account for this study were mined from the CCLE database and located in oncogenic hotspots: PI3K, AKT, PTEN, HRAS, BRAF, NRAS, and KRAS.

On day 0, SKOV3 cells were plated in a 96-well flat bottom plate and treated with different concentrations of epacadostat. Human allogeneic PBMCs (LB2419) were stimulated with CD3⁺/CD28⁺ beads and IL2 for 10 minutes before coculturing them with SKOV3 cells at a 1:2 SKOV3/PBMC ratio. On day 1, supernatants were harvested for tryptophan/kynurenine quantification assay by HPLC and ³H-thymidine was added in each well. On day 2, PBMC proliferation was measured by thymidine incorporation (TopCount counter Perkin Elmer). In a separate experiment, human allogeneic PBMCs (LB2419) prestimulated with CD3⁺/CD28⁺ beads and IL2 for 10 minutes were cultivated either in fresh medium supplemented or not with celecoxib 5 μg/mL, or in conditioned medium from SKOV3 cells cultured 72 hours in the presence or absence of celecoxib 5 μg/mL. On day 1, ³H-thymidine was added and PBMC proliferation was measured as above. On day 2, the supernatants were also used to quantify the Th1 and Th2 cytokines with the Bio-Plex Pro Human cytokine Th1/Th2 assay from Bio-Rad.

Tumors were extracted at the end point of each experiment and frozen. Frozen sections (5 μm) were incubated 5 minutes in 4% formaldehyde for fixation. Endogenous peroxidases were blocked with DAKO peroxidase blocking solution for 15 minutes, and nonspecific staining was prevented by incubation in a 5% BSA, 2% milk, and 1% human IgG solution in TBS-T for 1 hour. Sections were then incubated with a 1/200 dilution in REAL antibody diluent (#S2022 from Agilent) of the rabbit anti-human CD3 antibody (Clone SP7, # ab16669 from Abcam, concentration 1–3 mg/mL) or 1/200 dilution of the mouse anti-human CD8 antibody (Clone C8/144B, #M710301 from Agilent, concentration 157 mg/L). The anti-mouse goat polymer HRP (#K4005 from Agilent) or the anti-rabbit goat polymer-HRP (#K4009 from DAKO) was then added and incubated for 1 hour. Finally, sections were stained with the AEC+ chromogenic substrate (#K4005 from Agilent) for 20 minutes and counterstained with a solution of hematoxylin for 5 minutes. Stained sections were digitalized with a Pannoramic P250 Flash III slide scanner (3DHISTECH) and were analyzed with CaseViewer (3DHISTECH). Sections were quantified with the ImageJ software.

For the analysis of the effect of the different inhibitors, an unpaired t-test was done; the Welch correction was applied in the cases where the variance was not equal. The P values are annotated as follows: *, P < 0.05; **, P < 0.01; ***, P < 0.005; and ****, P < 0.0001. For the mouse in vivo experiments, a nonparametric “Anova Wilcoxon each pairs” test was done on the last time point.

We selected a series of human cancer cell lines that constitutively express IDO1 and, accordingly, degrade tryptophan and produce kynurenine in equimolar amounts (13) (Supplementary Table S1). We focused our initial experiments on a BRAF V600E-mutated melanoma cell line, named KUL98-MELA. Because IFNγ induces IDO1 expression, we first considered the possibility that KUL98-MELA might display a constitutive activation of the IFNγ signaling pathway, which entails JAK1/JAK2-mediated phosphorylation of STAT1. We observed that ruxolitinib, a JAK1/JAK2 inhibitor, did not reduce the constitutive levels of IDO1 transcripts in KUL98-MELA cells (Supplementary Fig. S1A), although it prevented both IDO1 induction and STAT1 phosphorylation triggered by addition of IFNγ (Supplementary Fig. S1B). These results indicated that constitutive IDO1 expression was not triggered by the IFNγ signaling pathway in these melanoma cells.

COX-2 shapes the immunosuppressive tumor microenvironment of mouse melanoma, although the downstream mechanisms remain unclear (20). In addition, PGE₂ contributes to induction of IDO1 expression in dendritic cells during the maturation process (25). These observations prompted us to consider that the PGE₂ signaling pathway might be involved in constitutive IDO1 expression by human tumor cells. We observed that celecoxib, a COX-2 specific inhibitor, reduced the levels of IDO1 transcripts by 3-fold in KUL98-MELA cells (Fig. 1A). Moreover, the IDO1 protein was not detectable in treated cells, which no longer degraded tryptophan nor produced kynurenine (Fig. 1A and B). As another means of preventing PGE₂ formation, we then used MF63, an inhibitor of membrane-associated PGE₂ synthase (mPGES1), and observed again a reduction of IDO1 transcripts and protein levels (Fig. 1A). We then considered that PGE₂ might trigger autocrine signaling and investigated the expression of the four isoforms of PGE₂ receptors, EP1-4, on our melanoma cells. We observed expression of EP4 transcripts (Fig. 1C). We therefore inhibited EP4 with a competitive antagonist, GW627368X, and observed again a decrease in IDO1 transcripts, IDO1 protein, and enzymatic activity in the treated melanoma cells (Fig. 1A and B). Altogether, these results suggested that constitutive IDO1 expression in KUL98-MELA cells was triggered by an autocrine loop of PGE₂ secretion. In line with this, we observed the production of up to 400 pg/mL of PGE₂ metabolites in the culture supernatant of those cells (Fig. 1D). Moreover, addition of this PGE₂-containing supernatant to celecoxib-treated KUL98-MELA cells restored IDO1 expression by these cells (Fig. 1E).

To characterize the link between PGE₂ signaling and IDO1 expression, we investigated the signaling pathways downstream of the EP receptors, which are G protein–coupled receptors able to activate the PKA, PI3K, and PKC pathways (26–28).

H-89, a competitive inhibitor of PKA, did not reduce IDO1 transcript levels in KUL98-MELA cells, although it reduced phosphorylation of PKA downstream effector CREB (Supplementary Fig. S1C).

In contrast, LY294002, a pan-PI3K inhibitor, reduced IDO1 transcript levels and abolished both IDO1 protein expression and enzymatic activity (Fig. 1F and G). Similar results were obtained using GDC-0941, a pan-PI3K inhibitor used in clinical trials (ref. 29; Fig. 1F). PI3K-subunit specific inhibitors A66 (p110α inhibitor) and TGX221 (p110β inhibitor) also reduced IDO1 transcript levels (Fig. 1F). All these inhibitors effectively reduced phosphorylation of AKT, a downstream effector of the PI3K pathway (Supplementary Fig. S2A–S2D). MK2206, an AKT inhibitor, also reduced IDO1 expression and enzymatic activity (Fig. 1F and G). Because AKT can phosphorylate both GSK3β and p70S6K, we assessed the phosphorylation of these proteins in treated cells. Although p70S6K phosphorylation was abrogated by MK2206 treatment, GSK3β phosphorylation was not affected (Supplementary Fig. S2E). We observed similar results with the different PI3K inhibitors (Supplementary Fig. S2A–S2D). These results suggested that, although EP receptors signal through the PI3K pathway, this pathway is not involved in GSK3β phosphorylation, but rather activates mTOR, consistent with the fact that p70S6K phosphorylation is abrogated by all the PI3K and AKT inhibitors. This was confirmed by treatment of KUL98-MELA cells with rapamycin, which led to reduction of IDO1 expression (Fig. 1F) and abrogation of phospho-p70S6K (Supplementary Fig. S2F). We suggest a possible signaling pathway for constitutive IDO1 expression in human tumors in Fig. 2.

Finally, we tested the PKC pathway, which can inactivate GSK3β by phosphorylating its serine 9 in the same manner as PI3K (30). Upon treatment with pan-PKC inhibitor Ro-31-8425, KUL98-MELA cells reduced their IDO1 expression and enzymatic activity and dephosphorylated GSK3β (Fig. 3A–C). GSK3β regulates a number of transcription factors, including β-catenin, which binds the IDO1 promoter (30, 31). Silencing β-catenin with siRNA reduced levels of IDO1 transcript and protein (Fig. 3D). We confirmed this PKC pathway to be independent from the PI3K/AKT pathway described above, by testing a combination of the AKT inhibitor MK2206 and the pan-PKC inhibitor Ro-31-8425: we observed an additive inhibitory effect on IDO1 expression, indicating that both pathways act independently downstream of the EP4 receptor (Fig. 3E). Altogether, we conclude that PGE₂-triggered signaling involves both the PKC pathway, which activates IDO1 via β-catenin, and the PI3K pathway, which activates IDO1 via mTOR. The mTOR activation may depend on the transcription factor ETV4/PEA3, which functions downstream of mTOR and binds the IDO1 promoter (32) (Fig. 2). Of note, melanoma-intrinsic β-catenin signaling was found to be linked to immune evasion and resistance to immunotherapies (33).

Because KUL98-MELA cells harbor a BRAF V600E mutation, we also investigated the role of the MAPK pathway in constitutive IDO1 expression. We used a BRAF V600E inhibitor (PLX4032), a MEK1/2 inhibitor (U0126) and a p38MAPK inhibitor (SB203580). All reduced IDO1 expression in KUL98-MELA cells (Fig. 4A and B). Because the expression of COX-2 is also controlled by the ERK1/2 pathway, we also measured COX-2 expression and found that it was reduced in cells treated with these inhibitors (34) (Fig. 4C). In contrast, the PGE₂ and PI3K pathway inhibitors we used above did not reduce COX-2 expression (Supplementary Fig. S3). Altogether, these results suggest an indirect involvement of the MAPK pathway in constitutive IDO1 expression, ensuring constitutive expression of COX-2 to sustain autocrine production of PGE₂ (Fig. 2).

KUL98-MELB is another melanoma line derived from a different metastasis from the same patient. KUL98-MELB cells do not express IDO1. As compared to KUL98-MELA cells, they express COX-2 at a similar level, but mPGES1 at a much lower level (Fig. 5A). Accordingly, they produce no PGE₂ (Fig. 5B), likely explaining why they do not express IDO1. To understand why KUL98-MELB cells express less mPGES1 than KUL98-MELA, we considered IL1β, which can control expression of both COX-2 and mPGES1, and we compared the expression of IL1β and its target gene IL8 (35–37). We observed that KUL98-MELB cells expressed less of IL1β and IL8 transcripts (Fig. 5C and D) and produced less IL1β in their supernatant (Fig. 5E). To determine whether this lack of IL1β production was involved in the reduced mPGES1 and IDO1 expression by KUL98-MELB cells, we treated these cells with recombinant IL1β and observed increases in IDO1, COX-2, and mPGES1 expression (Fig. 5F). The observed IDO1 induction was blocked with the p38MAPK inhibitor SB203580, but also with the EP4 inhibitor GW627368X (Fig. 5G). Hence, IL1β contributes to COX-2 and mPGES1 expression, which in turn allows PGE₂ production and IDO1 expression. These results suggest that in KUL98-MELA cells, besides the autocrine PGE₂ loop, an IL1β autocrine loop contributes to constitutive COX-2 and mPGES1 expression and therefore indirectly supports IDO1 expression (Fig. 2). Accordingly, we observed that blocking IL1β in KUL98-MELA cells with recombinant IL1ra, anakinra, reduced expression of IL8, a classical IL1β target, but also reduced IDO1 expression as well as COX-2 and mPGES1 expression (Fig. 5H and I).

So far we have demonstrated the importance of PGE₂ and downstream signaling pathways in the KUL98-MELA melanoma line. Next, we tested whether this mechanism also accounted for constitutive IDO1 expression in other human tumor lines, including a cholangiocarcinoma line, a head and neck carcinoma line, a sarcoma line, another melanoma line, an ovarian carcinoma line and a non–small cell lung carcinoma line (Supplementary Table S1). We observed a decrease in IDO1 expression upon treatment of these lines with celecoxib (Fig. 6A). We also profiled the expression of the four EP receptor isoforms in the tumor lines (Supplementary Fig. S4A), and treated them with the inhibitor of the most expressed isoform. In all cases, the EP inhibitor decreased IDO1 expression (Fig. 6B). We also observed that CAY10598, an agonist of the EP4 receptor, restored IDO1 expression in celecoxib-treated SKOV3 cells (Fig. 6C). In addition, this agonist was able to induce low levels of IDO1 in IDO1-negative glioblastoma cell lines A172 and U87 (Fig. 6D). Finally, we confirmed the production of PGE₂ by detecting the presence of PGE₂ metabolites in the supernatant of the 4 tumor lines that we tested (Supplementary Fig. S4B). Therefore, we conclude that the COX-2/PGE₂ pathway is responsible for constitutive IDO1 expression in a wide range of different human cancer types.

We then mined the transcriptomics data of 1,041 different human cancer cell lines from the Broad Institute, looking for correlations between IDO1 expression and activation of the COX-2/PGE₂ axis. We first took 21 of these cell lines, measured IDO1 expression by RT-qPCR, and matched the result with the value in the RNAseq database (Supplementary Fig. S5A). This validation allowed us to use percentile 80 as a cutoff for IDO1 positivity, with 20% of tumor lines being positive. This was in line with our previous work showing IDO1 expression in tumor cells in 3% to 56% of tumor samples depending of the type (9). We then clustered the database according to percentile 20/80 to separate IDO1-negative (below percentile 20) and IDO1-positive lines (above percentile 80), and interrogated COX-2 and mPGES1 expression. We observed more expression of COX-2 and mPGES1 in the IDO1-positive group, but no correlation with housekeeping gene VINCULIN (Fig. 6E). Even though the number of samples became limiting when we started analyzing the database per histological tumor type, we also observed correlations between IDO1 and COX-2 (PTGS2) or mPGES1 (PTGES) expression in six different cancer histotypes (Supplementary Fig. S5B).

Most of the IDO1-expressing tumor cell lines we have characterized in this work had mutations in genes of the PI3K or MAPK pathways or of receptors leading to the activation of these pathways, such as ERBB2 (Supplementary Table S1). This suggested that activation of the PI3K pathway downstream of the EP receptors or of the MAPK pathway upstream of COX-2 expression was required to jump-start constitutive IDO1 expression. We therefore interrogated the CCLE database asking whether IDO1-expressing cell lines were enriched in PI3K or MAPK mutations. Indeed, we observed enrichment in MAPK or PI3K mutations in the IDO1-expressing cell lines. This was true in the whole set of cell lines and was confirmed in the subset of lung cancer, for which the number of cell lines was large enough (Fig. 6F).

One of the human tumor lines that we tested above, ovarian carcinoma SKOV3, was previously analyzed by Litzenburger and colleagues, who proposed that constitutive IDO1 expression was driven in this cell line by STAT3 activation, triggered by an autocrine loop of interleukin-6 (38). In fact, STAT3 has been described as a downstream effector of tyrosine kinase receptors, such as the EGF receptor or the IL6 receptor, but it can also be activated by the EP receptors (39). We therefore knocked down STAT3 in both SKOV3 and KUL98-MELA cells using a lentiviral shRNA construct. Although STAT3 was silenced in both cell lines, we observed no effect on IDO1 expression, neither at the mRNA level nor at the protein level (Supplementary Fig. S6A and S6C). Accordingly, IDO1 enzymatic activity was not reduced by STAT3 knockdown, although it was reduced in both cell lines by treatment with celecoxib (Supplementary Fig. S6B and S6D). Litzenburger and colleagues observed reduced IDO1 expression in SKOV3 cells after treatment with AG490, which was used as a JAK inhibitor. However, AG490 also inhibits EGFR (40) and thereby may inhibit the MAPK pathway and repress COX-2 expression. We observed that ruxolitinib, a more specific JAK1/2 inhibitor, abrogated the active phosphorylated form of STAT3 but did not reduce IDO1 transcripts in both KUL98-MELA and SKOV3 cells (Supplementary Figs. S1A, S1B, and S6E). Altogether, these results do not support a direct role for STAT3 in IDO1 expression in SKOV3 and KUL98-MELA, but rather support the notion that constitutive IDO1 expression is driven by autocrine PGE₂. We do not exclude the possibility that STAT3 can induce IDO1 expression in specific contexts, but we propose that this effect might be indirect through the induction of COX-2 by STAT3 (41).

We set out to assess the relevance of our findings for cancer therapy. Because we could not identify a murine tumor line that expresses IDO1 constitutively, we set up an in vivo model using the human ovarian carcinoma line SKOV3 characterized above, which expresses IDO1 in a constitutive manner that is sensitive to celecoxib (Supplementary Fig. S7A and S7B). We injected SKOV3 cells into immunodeficient NOD/Scid/Il-2rg^−/−(NSG) mice (42). After 10 days, tumors were visible and mice received an injection of allogeneic human lymphocytes (PBMCs). Mice were also treated with celecoxib, and tumor growth was monitored (Fig. 7A). In the absence of celecoxib, tumor growth continued unabated at a rate similar to tumors of mice that did not receive human lymphocytes, indicating that the antitumor effect of allogeneic lymphocytes was blocked, possibly because of IDO1 expression by SKOV3 cells (Fig. 7B). In contrast, tumor growth was controlled in mice that received lymphocytes and celecoxib. Because the COX-2/PGE₂ axis may also favor tumor development by non-immune pathways such as angiogenesis, we also administered celecoxib to SKOV3 tumor–bearing mice that did not receive human lymphocytes. In this setting, celecoxib did not alter the course of tumor growth (Fig. 7B). In line with the notion that the tumor rejection promoted by celecoxib resulted from a blockade of IDO1 expression, we observed an identical antitumor effect in mice treated with allogeneic lymphocytes and IDO1 inhibitor epacadostat, which we confirmed can effectively block IDO1 enzymatic activity and restore proliferation of T lymphocytes cocultured with SKOV3 cells (Supplementary Fig. S7C and S7D). Conversely, when we repeated this experiment with IDO1-negative tumor colorectal carcinoma line LS411N, we observed some immune control with allogeneic lymphocytes in untreated mice, in line with the absence of IDO1 expression, but there was no additional effect of celecoxib or epacadostat (Fig. 7B, right). These results suggested that celecoxib treatment of mice bearing SKOV3 tumors repressed IDO1 expression in these tumors and thereby allowed lymphocyte infiltration and proliferation in the tumor. To confirm this interpretation, we first excluded a direct effect of celecoxib on human lymphocytes by measuring their proliferation and cytokine production in vitro in the presence or absence of celecoxib. No difference was observed in terms of proliferation and Th1 cytokine production, although Th2 cytokine production was reduced in the presence of celecoxib (Supplementary Fig. S7E and S7F). In contrast, proliferation and Th1 cytokine production were impaired when lymphocytes were cultured in SKOV3-conditioned medium, but not when they were cultured in medium conditioned with celecoxib-treated SKOV3 cells (Supplementary Fig. S7E and S7F). This is in line with the effect of celecoxib on IDO1 expression and tryptophan catabolism in SKOV3 cells (Supplementary Fig. S7A and S7B). Second, we tested IDO1 expression by Western blot in SKOV3 tumors collected from mice at day 7 of the in vivo experiment (Fig. 7C). In mice treated with celecoxib, IDO1 expression was reduced, particularly in mice that did not receive lymphocytes. Mice that did receive lymphocytes expressed residual amounts of IDO1 despite celecoxib. This residual expression, which probably resulted from IDO1 induction by IFNγ produced by lymphocytes, did not affect lymphocyte activity in this model, because we observed equal tumor control in mice that received lymphocytes with celecoxib, IDO1 inhibitor or both (Fig. 7B). Lastly, we analyzed T-cell infiltration in SKOV3 tumors collected at day 40 of the in vivo experiment. We stained tumor sections for CD3 and CD8 and observed a higher infiltration of both CD3-positive cells and CD8-positive cells in tumors from mice treated with celecoxib or epacadostat as compared to untreated mice (Fig. 7D and E). This was confirmed by RT-qPCR analysis of additional tumors (Fig. 7F). Altogether, these results confirmed the mechanism whereby celecoxib induced the rejection of SKOV3 tumors in this model. Celecoxib repressed IDO1 expression in SKOV3 tumor cells, thereby releasing the T-cell suppression imposed by tryptophan catabolism in the tumor microenvironment, and unleashing T cells that can now survive and proliferate at the tumor site to achieve tumor rejection.

Certain human tumors of various histotypes express IDO1 in a constitutive manner, thereby setting up a mechanism of innate or intrinsic immune resistance that can prevent T-cell infiltration resulting in uninflamed or “cold” tumors. Such a mechanism differs from the adaptive immune resistance, initially described in human melanoma, in which IDO1 expression is associated with T-cell infiltration and triggered by the interferon-gamma produced by infiltrating T cells. Because immune checkpoint inhibitors act by unleashing existing antitumor T lymphocytes, they are more efficient in patients bearing T-cell inflamed tumors than in those with uninflamed or cold tumors. Defining new approaches to trigger T-cell infiltration in cold tumors will be key to increasing the efficacy of cancer immunotherapy. This requires a detailed understanding of the molecular mechanisms responsible for this lack of T-cell infiltration.

Here, we characterized the signaling mechanisms responsible for constitutive IDO1 expression in a series of human tumors of various types. We found that constitutive IDO1 expression was dependent on an autocrine loop of PGE₂ production, leading to activation of the PI3K and PKC pathways and the subsequent activation of IDO1 transcription by factors such as β-catenin and, most likely, ETV4. Autocrine PGE₂ production depends on constitutive expression of mPGES1 and COX-2. The latter is controlled by the MAPK pathway, which can be activated in tumor cells by growth factors such as IL1 or EGF, but can also be activated by oncogenic mutations such as BRAF V600E. Indeed, inhibition of mutant BRAF in melanoma line KUL98-MELA suppressed both COX-2 and IDO1 expression. Our results therefore provide a conceptual framework to explain the role of COX-2 as a driver of tumor-induced immunosuppression (20, 21). In Zelenay and colleagues, inhibition or ablation of COX-2 in a BRAF V600E mouse melanoma model rendered tumors sensitive to immune control (20). In Hou and colleagues, PGE₂ inactivation sensitized tumors to immunotherapy (21). Our results suggest that at least part of the effect reported in these studies may have been mediated by the repression of IDO1 expression in those tumors.

We observed that most of the tumor lines that express IDO1 constitutively also bear oncogenic mutations in one of the signaling pathways involved in the autocrine loop we describe, including PI3K and MAPK. An analysis of the CCLE database confirmed that IDO1-expressing tumor lines harbored mutations in the PI3K or MAPK pathways more frequently than IDO1-negative lines, suggesting that oncogenic signaling jump-starts IDO1 expression.

A similar case has been described for PD-L1, another immunosuppressive factor whose expression can either be induced by interferon-gamma produced by infiltrating T cells, resulting in adaptive immune resistance, or be constitutive, leading to intrinsic (or innate) immune resistance (19). The mechanisms responsible for constitutive PD-L1 expression in human tumors have only begun to be characterized, and appear to involve the PI3K/AKT axis in glioblastomas (43) and triple-negative breast carcinomas (44), whereas Hodgkin lymphomas express PD-L1 as a result of the amplification of the encoding gene (45).

Our results suggest potential therapeutic avenues that may be useful to treat cold tumors. Besides inhibitors of IDO1 enzymatic activity, which are in phase III of clinical evaluation, COX-2 inhibitors can prevent IDO1 expression in those tumors. We validated this prediction using human SKOV3 tumors in NSG mice reconstituted with human allogeneic lymphocytes. These tumors were not rejected unless mice were treated with a COX-2 inhibitor or with an IDO1 inhibitor. Although PGE₂ may trigger other immunosuppressive mechanisms in different contexts (46), the lack of additive antitumor effect of targeting both COX-2 and IDO1 supports the notion that the effect of celecoxib in our model is essentially mediated by repression of IDO1 expression. Although we did not test the effect of PI3K, mTOR, MAPK, or BRAF V600E inhibitors in this in vivo model, it is possible that IDO1 repression contributes to the antitumor effects observed with such molecules in other contexts. PI3K contributes to the induction of long-term immune tolerance by mouse plasmacytoid dendritic cells (pDC) that have been exposed to TGFβ. PI3K promotes phosphorylation of IDO1 by the Fyn tyrosine kinase, thereby inducing a signaling pathway that results in sustained expression of IDO1 and TGFβ (47). PI3K inhibition would therefore not only reduce IDO1 constitutive expression in cancer cells but also abrogate tolerance induction by pDCs in the tumor microenvironment. This double effect would favor tumor rejection. In sum, our results further highlight the potential of COX-2 and, potentially, PI3K inhibitors as useful adjuvants to combine with cancer immunotherapy.

Imatinib, which is used in the treatment of gastrointestinal stromal tumors (GIST) for its ability to inhibit oncogenic signaling due to mutant KIT or PDGFR, also represses IDO1 expression in those tumors and thereby favors their control by antitumor T lymphocytes (32). In line with our findings in other tumor types, oncogenic KIT signaling in GIST involves the PI3K–AKT–mTOR pathway and culminates in the induction of ETV4, a transcription factor whose expression is repressed in imatinib-treated GIST cells. ETV4 binding sites were identified in the IDO1 promoter and confirmed to bind ETV4 by chromatin immunoprecipitation (32). Hence, ETV4 may link the PI3K–AKT–mTOR pathway to the activation of the IDO1 promoter (Fig. 2). However, other transcription factors downstream of mTOR may also play a role.

Our results also indicate that β-catenin likely links the PKC-GSK3β pathway to the activation of the IDO1 promoter (Fig. 2). In melanoma, intrinsic β-catenin signaling is a key factor mediating immune resistance and lack of T-cell infiltration (33, 48). This study identified ATF3-mediated repression of chemokine CCL4 as responsible for the lack of recruitment of CD103⁺ dendritic cells in the tumor and the resulting lack of T-cell infiltration. Based on our results, we speculate that β-catenin signaling also prevents T-cell infiltration by inducing constitutive expression of IDO1.

By deciphering the oncogenic signaling pathways responsible for constitutive IDO1 expression in human tumors, our results shed light on the mechanisms responsible for the problem of cold tumors, which are not infiltrated by T lymphocytes and therefore fail to respond to immunotherapy. Our results also suggest clinically applicable therapeutic options to counteract such immunosuppressive mechanisms.

B.J. Van den Eynde is a member of the board of directors for, has ownership in, and is a consultant/advisory board member of Iteos Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Hennequart, L. Pilotte, E. De Plaen, B.J. Van den Eynde

Development of methodology: M. Hennequart, L. Pilotte, S. Cane, E. De Plaen, B.J. Van den Eynde

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Hennequart, L. Pilotte, D. Hoffmann, V. Stroobant, E. De Plaen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Hennequart, L. Pilotte, V. Stroobant, E. De Plaen, B.J. Van den Eynde

Writing, review, and/or revision of the manuscript: M. Hennequart, L. Pilotte, E. De Plaen, B.J. Van den Eynde

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Hennequart, L. Pilotte, E. De Plaen, B.J. Van den Eynde

Study supervision: B.J. Van den Eynde

We thank Guy Warnier, Gilles Gaudray, Julien Gossiaux, and Laurent Hermans for production of NOD/Scid/Il-2rg^−/− (NSG) mice; Sophie Lucas, Catherine Uyttenhove, and Stéphanie Liénart for advice on the NSG experiment; Céline Bugli for statistical advice; Pierre van der Bruggen and Pierre Coulie for critical comments; and Auriane Sibille for editorial assistance.

This work was supported by Ludwig Cancer Research, Walloon Excellence in Life Sciences and Biotechnology (WELBIO, Belgium), FNRS-Télévie (Belgium), Foundation Against Cancer (Belgium), de Duve Institute and Université catholique de Louvain (Belgium). M. Hennequart was supported by FNRS-Télévie (Grant number: 7.4590.15), L. Pilotte by de Duve Institute, S. Cane by FNRS-Télévie (Grant number: 7.4538.14), D. Hoffmann by FRIA (Grant number: 1.E082.14), and V. Stroobant, E. De Plaen, and B.J. Van den Eynde by the Ludwig Institute for Cancer Research.

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.