The COX2 effector microsomal PGE2 synthase-1 is a regulator of immunosuppression in cutaneous melanoma

Abstract

PURPOSE:

Microsomal prostaglandin E2 synthase 1 (mPGES1) was evaluated as an important downstream effector of the cyclooxygenase 2 (COX2) pathway responsible for tumor-mediated immunosuppression in melanoma.

EXPERIMENTAL DESIGN:

The analysis of a Stage III melanoma tissue microarray (n=91) was performed to assess the association between mPGES1, COX2, CD8, and patient survival. Pharmacological inhibitors and syngeneic mouse models using mPGES1 knockout mouse melanoma cell lines were used to evaluate the mPGES1-mediated immunosuppressive function.

RESULTS:

We observed correlations in expression and co-localization of COX2 and mPGES1, which are associated with increased expression of immunosuppressive markers in human melanoma. In a syngeneic melanoma mouse model, mPGES1 knockout increased melanoma expression of PD-L1, increased infiltration of CD8a+ T cells and CD8a+ dendritic cells into tumors and suppressed tumor growth. Durable tumor regression was observed in mice bearing mPGES1 knockout tumors that were given anti-PD-1 therapy. Analysis of a stage III melanoma tissue microarray revealed significant associations between high mPGES1 expression and low CD8+ infiltration, which correlated with a shorter patient survival.

CONCLUSIONS:

Our results are the first to illustrate a potential role for mPGES1-inhibition in melanoma immune evasion and selective targeting in supporting the durability of response to PD-1 checkpoint immunotherapy. More research effort in this drug development space is needed to validate the use of mPGES1 inhibitors as safe treatment options.

Introduction

The understanding of mechanisms and resultant improvements for immunotherapies of melanoma continue to emerge, resulting in progress toward reducing the overall lethality of melanoma (1–4). Although many immunotherapies produce an initial response, most advanced melanomas (and other late-stage cancers) develop resistance; since this often leads to relapse, there is an urgent need to identify and counteract the mechanisms responsible. Considerable research effort is engaged in revealing the molecular and biologic mechanisms that drive resistance and identifying the most appropriate adjuvant therapies that promote durable responses (5).

Melanoma initiation and progression are strongly associated with local inflammatory cells and marker signatures (6,7). In melanoma and other cancers, the pro-inflammatory enzyme cyclooxygenase 2 (COX2) and its downstream metabolic pathway product, prostaglandin E2 (PGE2), are reported to enhance carcinogenesis and tumor progression and to support immunosuppression (8,9). PGE2 produced by tumor cells and/or their surrounding stromal cells induces immunosuppression through several mechanisms, including downregulating antitumor T helper 1 (TH1) cytokines and upregulating immunosuppressive TH2 cytokines; inhibiting CD8+ T cell proliferation and activity, which suppresses the antitumor activity of natural killer cells and stimulates the expansion of regulatory T cells (Treg cells) and myeloid-derived suppressor cells (MDSCs); and inhibiting CD8+ T cell antitumor functions by impairing the ability of tumor cells to directly present tumor antigens, which inhibits dendritic cell differentiation and switches the function of dendritic cells from induction of immunity to T cell tolerance (10–17). In addition, tumor-derived COX activity in melanoma is a crucial suppressor of type I interferon (IFN) - and T cell–mediated tumor removal and the inducer of an inflammatory signature associated with cancer progression (18,19).

Importantly, COX inhibitors (aspirin and celecoxib), in combination with an anti–PD-1 monoclonal antibody, promoted much more rapid and durable regression of Braf^V600E melanoma cells than anti–PD-1 alone (18). Thus, there is considerable motivation to identify or develop drugs that selectively target and inhibit the COX2/PGE2 pathway axis for use as adjuvants to immune-based therapies for melanoma patients. As most COX inhibitors exhibit cardiovascular and other toxic effects, the downstream PGE2 synthase can be considered a more specific and potentially less toxic target. Although three human PGE2 synthase isoforms exist, one of them, microsomal prostaglandin E2 synthase 1 (mPGES1), is conditionally coupled with COX2 gene expression that is markedly induced by pro-inflammatory stimuli (20) and upregulated in several cancers (21,22). Our recent studies showed that mPGES1 expression is increased in progressive human melanoma and that high expression levels of the protein are associated with shorter patient survival. In addition, mPGES1 inhibition suppressed melanoma cell survival in vitro and in vivo (9). However, little is known about the immunosuppressive or immune evasion functions of mPGES1 in melanoma tumors. Therefore, we carried out this study to identify any potential role for mPGES1 in regulation of immune evasion in melanoma.

MATERIALS AND METHODS

TCGA and CCLE data analysis.

Skin Cutaneous Melanoma data from TCGA (https://www.ncbi.nlm.nih.gov/pubmed/26091043) was obtained from public TCGA repositories. Melanoma cell line data was downloaded from the CCLE (https://www.ncbi.nlm.nih.gov/pubmed/22460905]). Correlation analyses with immune markers were performed using the R language and visualized using Tableau Desktop as described elsewhere (https://www.ncbi.nlm.nih.gov/pubmed/28670312).

Cell culture.

Human melanoma cell lines LOXIMVI and WM793 were purchased from ATCC (Manassas, VA, USA). Mouse melanoma cells established from the transplantable tumor cell line established from a Braf+/LSL-V600E;Tyr::CreERT2+/o;p16INK4a/ mouse were obtained from Dr. Zelenay (The University of Manchester) (18). Cells were maintained in Dulbecco modified Eagle medium containing 10% fetal bovine serum in a 5% carbon dioxide atmosphere.

PGE2 measurement.

PGE2 levels in culture supernatants were determined by using a commercially available enzyme-linked immunosorbent assay kit (R&D Systems, Minneapolis, MN, USA) according to the manufacturer’s recommendations.

RNA interference.

Two siRNAs targeting mPGES1 were purchased from Sigma-Aldrich (St Louis, MO, USA). Cells were transfected with 20 nmol/L of mPGES1 siRNA or non-targeting siRNA using Lipofectamine RNAiMAX (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. The efficacy of knockdown was confirmed by Western blot analysis.

Protein cytokine array.

Secreted proteins in each culture supernatant were measured using a Human XL Cytokine Array kit (R&D Systems). Protein expression dots were scanned using a computer scanner, and dot pixel density was quantified by using Image Studio Lite Ver 4.0 (LI-COR Biosciences, Lincoln, NE, USA).

Western blotting.

Total proteins extracted from cell lysates were resolved on a 10% sodium dodecyl sulfate polyacrylamide gel and were transferred to a nitrocellulose membrane. Membranes were blocked with 5% non-fat dry milk and were incubated with primary antibodies. Secondary antibody conjugated to horseradish peroxidase (HRP; 1:3,000; Santa Cruz Biotechnology, Santa Cruz, CA, USA) was used to detect primary antibodies, and enzymatic signals were visualized by chemiluminescence.

Establishment of mPGES1 knockout cell lines.

A CRISPR/CAS9 knockout kit for mouse mPGES1 (KN314172) was purchased from Origene Technologies Inc. (Rockville, MD, USA) and used according to the manufacturer’s specifications. Finally, the success/efficiency of mPGES1 knockout were confirmed by Western blot analysis, and PGE2 level in the cells was measured.

Co-immunofluorescent staining and immunohistochemistry.

Paraffin-embedded tumor specimens were treated with xylene to remove the paraffin and dehydrated with ethanol. The slides were immersed in Borg Decloaker solution (Biocare Medical, Inc., Pacheco, CA, USA) and boiled in a pressure cooker at 125°C for 5 min for antigen retrieval. Endogenous peroxidase activity was blocked by incubating the slides for 10 min in phosphate-buffered saline solution (PBS) containing 3% hydrogen peroxide. The slides were blocked with 5% normal goat serum and were incubated with primary antibodies (COX2 monoclonal antibody, BD Biosciences, Franklin Lakes, NJ, USA; mPGES1 polyclonal antibody, Novus Biologicals, Littleton, CO, USA) overnight at 4°C. HRP-conjugated secondary antibodies were then applied to the slides. Alexa Fluor 488–labeled tyramide for inducible nitric oxide synthase and Alexa Fluor 594 for mPGES1 were used to detect the specific signals. The nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI). Signal localization was then calculated using the Manders colocalization coefficients found within the Manders Calculator plugin of Image J. To quantify CD3 or CD8a expression, the slides were blocked with 5% normal goat serum and incubated with anti-CD3 or CD8 (both, Cell Signaling Technology, Danvers, MA, USA) at 4°C overnight. After washing with PBS, the slides were incubated with HRP-conjugated goat anti-rabbit antibody (Vector Laboratories, Burlingame, CA, USA). After washing, the slides were developed with 3,3′ diaminobenzidine reagent (Vector Laboratories), followed by counterstaining with hematoxylin.

Syngeneic mouse tumor model.

All mice were housed and treated in accordance with protocols approved by the Institutional Animal Care and Use Committee at MD Anderson. Six- to 8-week-old female C57bl/6 mice (n=8 per group) were injected subcutaneously in the flanks with control (scramble) or PTGES-KO melanoma cells (1 × 10⁶ per flank). After 8 days, the mice began intraperitoneally receiving anti-PD-1 monoclonal antibody (200 μg per mouse, BE0416, Bio X Cell, NH) or IgG (control, BE0086, Bio X Cell, NH) every 3 days for 2 weeks. Tumor size was measured with an external caliper; the length (L) and width (W) was measure by caliper and the volume (V) of each tumor was calculated according to the equation (V = (L×W²) × 0.5). The mice were euthanized by carbon dioxide asphyxiation. Tumors were excised, weighed, and subjected to immunohistochemical (IHC) or immune cell profile analysis.

Isolation of immunocytes from tumors.

After careful excision of mouse tumor tissues, portions were minced into 2-to 4-mm pieces. These tissues were further dissociated into single-cell suspensions by combining mechanical dissociation with enzymatic degradation of the extracellular matrix using a Mouse Tumor Dissociation Kit (Miltenyi Biotec Inc, Auburn, CA, USA). The discontinuous Percoll (GE Life Sciences, Pittsburgh, PA, USA) gradient (44% and 67%) separation method was used to enrich immunocytes.

Flow cytometry analysis.

For multicolor flow cytometry immunophenotypic analysis, cells were stained with indicated antibodies and analyzed on a Gallios flow cytometer (Beckman Coulter, Brea, CA, USA). The flow cytometric profiles were analyzed by counting 20,000 events using the Kaluza software (Beckman Coulter). Information on antibodies is presented in Supplemental Methods and gating strategies for immunophenotyping showed in supplementary Figure 1. The number of specific immune cells per gram of tumor was calculated by total cell numbers from 0.1g tissues × 10 × frequencies of CD45 positive cells × frequencies of specific immune cells acquired from flow cytometry for DCs and MDSs. The number of specific immune cells per gram of tumor was calculated by total cell numbers from 0.1g tissues × 10 × frequencies of CD3+CD8+ subsets for T cells.

PCR array.

Total RNA was isolated with an RNeasy Mini kit (QIAGEN) from mouse tumor tissues. We synthesized cDNA from total RNA (2 μg) by using High-Capacity cDNA Reverse Transcription kits (Applied Biosystems), mixed the cDNA with SYBR Green PCR Master Mix (QIAGEN) and then applied to the mouse cancer inflammation and immunity PCR array (QIAGEN). PCR was performed using MasterCycler RealPlex (Eppendorf) and the mRNA expression of target genes relative to the mRNA of actin was calculated.

Melanoma tissue microarray.

A tissue microarray (TMA) consisting of lymphadenopathy standard of care tumor samples comprising lymph node metastases from 118 patients with stage III melanoma was employed. The TMA was analyzed for mPGES1 and CD8 expression in previous studies (9,23). From this cohort, 27 patients were excluded from the outcome analysis because their mPGES1 or CD8 expression measurements were non-evaluable or required clinical data were missing. The immunostaining of the evaluable 91patients’ samples for PGES1 was scored separately for two variables, first according to the percentage of melanoma cells with positive staining (<5% = 0, 5%–25% = 1, 26%–75% = 2, and >75% = 3) and then according to the overall intensity of the immunoreactivity of cells that stained positive (no staining = 0, light staining = 1, moderate staining = 2, and marked staining = 3) (9). These slides were manually scored independently by two researchers without prior knowledge of tumor stage or other data. Patients were categorized into two CD8+ infiltration groups: the low group had a CD8+ infiltration value less than or equal to the median, and the high group had a CD8+ infiltration value greater than the median. For mPGES1, patients were categorized in two ways: 1) low, intensity 0 or 1, and high, intensity 2 or 3; and 2) low, intensity 0, 1, or 2, and high intensity 3. Finally, the patients were categorized into four groups based on mPGES1 and CD8 infiltration values: 1) both low; 2) mPGES1 low/CD8 high; 3) mPGES1 high/CD8 low; and 4) both high.

Statistical analysis.

Each experiment was done at least 3 times, and data are presented as the mean ± SD. Statistical significance was determined using Student t test and one way or two way ANOVA, where applicable. In Figures, * indicates P < 0.05; ** indicates P < 0.01; and *** indicates P < 0.001. Overall survival (OS) was computed from stage III diagnosis to last known vital status. Patients alive at the last follow-up date were censored. Similarly, RFS was computed from the date of stage III diagnosis to the first date of a subsequent disease event (local/regional, distant, or both) or death (if the patient died without disease progression) or to the last date of follow-up in which the patient remained with no evidence of disease. Patients with no evidence of disease at the last follow-up date were censored. The Kaplan-Meier method was used to estimate OS and RFS, and the log-rank test was used to assess differences between groups. In addition, associations between OS/RFS and expression groups were evaluated by univariate and multivariable Cox proportional hazards regression models. Survival analyses were performed using SAS 9.3 for Windows (SAS Institute Inc., Cary, NC, USA).

RESULTS

mPGES1 is linked to COX2 to produce PGE2 in human melanoma

We analyzed the expression of PTGS2 (the gene that encodes COX2) and PTGES (the gene that encodes mPGES1) in cutaneous melanomas from The Cancer Genome Atlas (TCGA) (Figure 1A). The expression of PTGS2 and PTGES was weakly but significantly correlated in primary (r=0.262, p=0.007), metastatic (r=0.304, p<0.001), and combined (r=0.296, p<0.001) cutaneous melanoma samples. We next performed double immunostaining of mPGES1 and COX2 with immunofluorescence-conjugated antibodies in tumor specimens from four human melanoma patients, and calculated the Manders colocalization coefficients. The coefficients were 0.887, 0.859, 0.908, and 0.913, indicating that COX2 and mPGES1 dual expression was co-localized in these specimens (Figure 1B). This mPGES1 expression was clearly evident in melanoma cells (S100 positive) in melanoma tumor tissues (Supplementary Figure 2). Analysis of mPGES1 and COX2 RNA expression in data on 60 human melanoma cell lines extracted from the Cancer Cell Line Encyclopedia (CCLE) database revealed that their expression is weakly correlated (r=0.3214, p=0.020; Figure 1C). To test whether the link between COX2 and mPGES1 is important for PGE2 production, mPGES1 was knocked down or inhibited in WM793 human melanoma cells, which express high levels of COX2 and mPGES1 and produce high levels of PGE2 (9). The selective COX2 inhibitor celecoxib and selective mPGES1 inhibitor PF9184 (24) significantly suppressed PGE2 production (Figure 1D, left), as did knockout of mPGES1 (Figure 1D, right). These data suggest that mPGES1 and COX2 together form a key enzyme axis that regulates PGE2 production in melanoma.

Figure 1. Association between COX2 and mPGES1 in human melanoma.

A) Correlation between expression of PTGS2 (COX2) and PTGES (mPGES1) mRNA in primary or metastatic melanomas in the TCGA datasets. TPM, transcripts per million. B) Representative images of staining for mPGES1 (red), COX2 (green), and DAPI (blue) in human melanoma specimens. C) Correlation between PTGS2 (COX2) and PTGES (mPGES1) mRNA levels in 60 human melanoma cell lines from the CCLE. D) LOXIMVI human melanoma cells were treated with specific COX2 inhibitor celecoxib (1 μM) or specific mPGES1 inhibitor PF9184 (5 μM) for 24 h (left) or transfected with one of two mPGES1 siRNAs (right). PGE2 levels were determined in the cell culture media. Each experiment was done at least 3 times, and data are presented as the mean ± SD..**P <0.01; ***P <0.005 (One way ANOVA test).

COX2/mPGES1 pathway is associated with immunosuppressive signatures in melanoma

To determine whether there is an inflammatory immune signature related to COX2 and mPGES1, we analyzed Skin Cutaneous Melanoma (SKCM) data obtained from TCGA data repositories. In Figure 2A, we included selected immune markers based on the literature, including from our previous data (25). Levels of several markers for immune checkpoint-related genes, including CTLA4, IL4I1, and c10orf54, or immune suppressive cells, including CD1A and THBD for dendritic cells, CD14 and ITGAM for MDSCs, and FOXP3 and TGFB1 for Tregs, were positively correlated with mPGES1 levels in these samples (Figure 2A). In addition, levels of mPGES1 and COX2 were significantly associated with numerous pro-inflammatory chemokines such as CXCL1 and CXCL2 and cytokines IL8 and IL6 (Figure 2B and Supplementary Figure 3). Therefore, mPGES1 expression is related to an immunosuppressive signature in human melanoma.

Figure 2. Immunosuppressive signatures related to COX2 or mPGES1 in melanoma.

A) Correlation between markers for dendritic cells, myeloid-derived suppressor cells (MDSC), and Tregs and mPGES1 or COX2 in melanoma samples from the TCGA dataset. NS, no significant correlation. B) Correlation between expression of various chemokines/cytokines and COX2 or mPGES1 mRNA in melanoma samples from the TCGA dataset. Correlation was analyzed by calculating the Spearman rank correlation coefficients and the correlations of the genes listed were significant (p value <0.05).

mPGES1 promotes the expression of pro-inflammatory factors but not PD-L1 in human melanoma cells

As shown in Figure 3A, treatment of melanoma cells with the inhibitor, PF9184, reduced or eliminated expression of several pro-inflammatory cytokines and chemokines, including CCL, CXCL5, ANG1, and ANG2, suggesting that mPGES1 regulates and promotes features of inflammation. Because of the heightened awareness of immune checkpoints and their inhibition in the treatment of melanoma (26,27), we further analyzed the expression of PD-L1 RNA in relation to COX2 and mPGES1 expression in the CCLE data on 60 human melanoma cell lines. These analyses revealed significant associations between PD-L1 and COX2 (r=0.312, p = 0.014) or mPGES1 (r=0.316, p = 0.012). Despite the significant associations in the large dataset, treatment with celecoxib or PF9184 did not affect IFNγ–induced PD-L1 expression in A375 and SB2 melanoma cells or the constitutive PD-L1 levels expressed in LOXIMVI cells (Figure 3C and supplementary Figure 4), suggesting that the COX2/mPGES1 pathway does not directly regulate PD-L1 expression.

Figure 3. Regulation of cytokines and chemokines by mPGES1 inhibition.

A) LOXIMVI human melanoma cells were treated with specific mPGES1 inhibitor PF9184 (5 μM) for 48 h, after which the conditioned media were collected for analysis on a chemokines/cytokines protein array. The densities of dots were quantified by using Image Studio. B) Correlation between the expression of CD274 (PDL1) and PTGS2 (COX2) or PTGES (mPGES1) in 60 human melanoma cell lines from CCLE. C) A375 and SB2 melanoma cells were pretreated with selective COX2 inhibitor celecoxib (1 μM) or PF9184 (5 μM) for 1 h and treated with IFNγ (100 IU) for 24 h to induce PD-L1 expression. LOXIMVI human melanoma cells were treated with PF9184 (5 μM) for 24 h, and cell lysates were subjected to Western blotting for PD-L1 and actin.

mPGES1 deletion suppresses tumor growth and enhances T cell and dendritic cell infiltration in a syngeneic mouse model

Because PGE2 produced by tumor cells and/or their surrounding stromal cells is known to induce immunosuppression through several mechanisms, we examined the impact of mPGES1 on immunosuppression by using a previously described transplantable melanoma cell line (18) that was established from a Braf+/LSL-V600E;Tyr:CreERT2+/o;p16INK4a/ mouse, expresses high levels of COX2 and mPGES1, and produces high levels of PGE2. Since the available mouse-specific mPGES1 inhibitors did not block PGE2 production in these cell lines (data not shown), we established mPGES1-knockout mouse melanoma cell lines by using CRISPR-CAS9 systems. In the established knockout clones (PTGES-KO), mPGES1 was undetectable and PGE2 production was dramatically suppressed compared to parent or scramble control cells (Figure 4A). As our previous study showed that mPGES1 inhibition suppressed human melanoma growth in a mouse xenograft model, we injected mice with PTGES-KO cells to create a model of mPGES1-deficient melanoma. Tumor growth was dramatically reduced in the KO model compared to tumors in mice injected with scramble control cells (Figure 4B) although their growth rates were not different in vitro (Supplementary Figure 5). Next, to determine whether mPGES1 modulates the tumor immune environment, we analyzed the immune cell profile regulated by mPGES1 in this mouse model of melanoma. CD3 and CD8a staining were dramatically more prominent in the PTGES-KO tumors than in the scramble controls (Figure 4C). Flow cytometry analysis also showed that CD3+CD8a+ T cell infiltration into the tumor was significantly higher in the PTGES-KO model than in the scramble controls. We performed further analysis of the infiltrating CD8a+T cells for effector markers and/or molecules or exhaustion markers. About 30% of CD8a+ T cells expressed granzyme b and IFNγ (effector makers), but around 30% of CD8a+ T cells showed the exhaustion phenotype expressing Tim3 (Figure 4D). Additionally, we found that the frequencies of intratumoral conventional dendritic cells (DCs) (CD45+CD11b+CD11c+Ly6clo/-Ly6G-F4/80-) and lymphoid DCs (CD45+CD11b-CD11c+Ly6G-F4/80-) were significantly higher in PTGES-KO tumors than in control tumors. Since the basic leucine zipper ATF-like transcription factor 3 (Batf3)–dependent subfamily of dendritic cells characterized by CD8a expression are important for anticancer responses, we assessed the effect of mPGES1 deletion on these cells and found that they were expressed at significantly greater levels in the PTGES-KO tumors (Figure 4E, a-c). The frequencies of monocytic MDSCs (mMDSCs) (CD45+CD11b+Ly6c+Ly6Glo/-) were significantly higher in PTGES-KO tumors than in control tumors, whereas the frequencies of granulocytic MDSC (gMDSCs) (CD45+CD11b+Ly6clo/-Ly6G+) did not differ significantly in control and PTGES-KO tumors (Figure 4E, d-e). To explain how mPGES1 deletion led to the tumor infiltration with CD8a+ DC and T cells, we performed the mouse cancer inflammation and immunity PCR array which includes 84 key genes involved in mediating communication between tumor cells and the cellular mediators of inflammation and immunity. mPGES1 knockout induced the expression of STAT1, CCL2, CCL4, and CCL5 which are products of type I IFN pathway activation and the expression of CXCL9 and CXCL10 which are T cell attracting chemokines secreted by CD8a+ DCs. In addition, granzyme a and b (markers for effector T cells) was increased in PTGES-KO tumors (Figure 4F, Left). In contrast, PTGES deletion decreased the expression of several tumor promoting factors including CXCL1, CXCL5 and SPP1 (Figure 4F, Right). Collectively, these results suggest that the loss of mPGES1 leads to the CD8a+ DC infiltration in tumor via the activation of type I IFN pathway and CD8a+ DCs attract effector CD8+T cells by secreting T cell-attracting chemokines.

Figure 4. Effect of mPGES1 deletion on tumor growth and the tumor immune microenvironment.

A) Mouse melanoma cell lines were co-transfected with gRNA for mPGES1 or donor (parent) or scramble control; mPGES1 knockout clones (PTGES-KO) were isolated using GFP and puromycin selection. PTGES knockout was validated by Western blotting, and PGE2 levels were determined in these cells. B) Scramble control or PTGES-KO mouse melanoma cells were injected subcutaneously into the flanks of C57bl/6 mice and tumor size was measured with an external caliper. After 4 weeks, the animals were sacrificed, and tumors were excised and weighed. 7 mice per group were used and three independent experiments were performed. C) The mouse tumor tissues from two different mice each group (n=5/group) were immunostained for CD3 and CD8a. Scale, 400X (Left). Percentiles of CD3 or CD8a positive cells were measured using the Vectra 3.0 spectral imaging system (Right). D) Immune cells isolated from tumor tissues (n=8/group) by using the discontinuous Percoll separation method were subjected to flow cytometry analysis of cell surface markers to define T cells (CD3+ CD8a+) (Left and middle), activation cells (granzyme b and IFNγ) and exhaustion phenotype (Tim3) of T cells (Right). E) Flow cytometry analysis of cell surface markers to define DCs and MDSCs; a) numbers of conventional DCs, b) numbers of lymphoid DCs, (c) numbers of CD8a+ DCs d) numbers of mMDSCs, e) numbers of gMDSCs. (F) The mouse cancer inflammation and immunity PCR array was achieved and the ratio of relative mRNA expression between Scramble and PTGES KO mouse tumor tissues was calculated. Data are presented as the mean ± SD. * P<0.05; **P <0.01; ***P <0.005 (Student t test).

mPGES1 deletion improves anti-PD-1 efficacy in a syngeneic mouse model

The infiltration of CD8+ T cells into invasive tumors is an indicator of potential response to anti-PD-1 treatment in melanoma patients (28). In addition, the PD-1 ligand on tumors, PD-L1, has been proposed to serve as an important biomarker for predicting patient response to anti-PD-1 therapy (29). We found that expression of PD-L1 was higher in PTGES-KO tumor tissues than in controls (Figure 5A), showing the same pattern as CD8+ T cell infiltration. Since, in a published study, combination of a COX inhibitor with an anti–PD-1 antibody promoted regression of melanoma cells to a greater extent than anti–PD-1 alone (18), we tested whether deletion of mPGES1 also enhances the anti-PD-1 effect on tumor regression. Mice bearing a scramble control or PTGES-KO tumor were treated with anti-PD-1 antibody. Although anti-PD-1 treatment suppressed about 50% of tumor growth in the control model, tumors continued to grow. Notably, however, the anti-PD-1 treatment induced durable tumor regression in the mice bearing a PTGES-KO–derived tumor which led to an increased survival (Figure 5B and C).

Figure 5. Enhancement of PD-1 blockade efficacy by mPGES1 knockout.

A) Representative images of staining for PD-L1 in two different mouse tumors derived from scramble control or mPGES1 knockout (PTGES-KO) cells (Scale, 400X) (Left). Percentiles of PD-L1 positive cells were measured using the Vectra 3.0 spectral imaging system (Right). B and C) Mouse models were developed by injecting scramble control or PTGES-KO mouse melanoma cells subcutaneously into the flanks of C57bl/6 mice. Beginning on day 8 after injection, mice received an anti-PD-1 monoclonal antibody (200 μg per mouse) or IgG every 3 days for around 2 weeks (5 mice/group, duplicate experiments); B) Tumor size was measured with an external caliper (Left), and at the end of the experiment, the animals were killed and the tumors were excised and weighed (Right). Mean ± SD. * P<0.05; **P <0.01; ***P <0.005 (ANOVA followed by Bonferroni’s multiple comparisons test). C) Survival was estimated by the Kaplan-Meier method and significance was determined by Log-rank (Mantel-Cox) test.

Elevated mPGES1 expression is associated with low CD8+ T cell infiltration into melanomas and poor patient survival

To test for association between mPGES1 expression and lymphocyte infiltration in human melanoma tissues, we first performed IHC staining for CD8 as an indicator of lymphocyte infiltration on samples from a TMA of lymph node metastases from patients with stage IIIB or IIIC melanoma. Our use of lymph node specimens is based on the fact that nodes are generally the first sites of metastasis, and these are from the patients with lower tumor burden and considered to include those most amenable to treatments. Curiously, the biology of these is still being studied, especially with respect to immune therapy. Using mPGES1 expression data from our previous study (9), a Spearman correlation analysis showed that mPGES1 staining intensity expression was inversely associated with CD8 levels in these tissues (r= −0.272, p=0.013) (Figure 6A). In Figure 6B, representative images show this inverse relationship between the mPGES1 expression and tumor-infiltrating lymphocyte infiltration. This finding is interesting as the lymph nodes are considered rich in lymphocytes overall, however differences in melanoma sites based on mPGES1 expression was observed. We also found in these patients, that survival duration was similarly associated with this relationship between mPGES1 expression and immune cell infiltration. The shortest median overall survival (OS) was seen in the group whose tumor showed a high level of mPGES1 expression and low CD8+ infiltration (n=8, 11.0 months), while the longest median survival was seen in the patients whose tumor showed low mPGES1 expression and high CD8+ infiltration (n=41, 4.5 years; p=0.003 [Figure 6C]). Patient tumors that exhibited low mPGES1 expression and high CD8+ infiltration experienced a significant decreased risk of death compared with patient tumors exhibiting a high level of mPGES1 expression and low CD8+ infiltration (hazard ratio[HR]=0.10; p<0.001). This significant decrease risk of death remained the same when adjusting for gender, age, stage of diagnosis, ulceration status, and Breslow thickness (Supplementary Table 1). As only one patient (of 91) was in the high mPGES1 /high CD8+ group, that category was not included in the analysis. Differences in recurrence-free survival (RFS) among the groups paralleled those in OS, with the shortest median survival in the group whose tumors displayed high mPGES1 expression and low CD8+ infiltration (3.6 months) and the longest median survival in the patients whose tumors showed low mPGES1 intensity and high CD8+ infiltration (10.4 months; p=0.012 [Figure 6D]), as well as a decreased risk of recurrence or death (HR=0.19; p=0.003). We conclude that low mPGES1 expression and high CD8+ cell infiltration in tumors predicts a better outcome.

Figure 6. Reverse association of mPGES1 with CD8+ T cell infiltration in stage III melanoma.

A) Correlation between mPGES1 and CD8 staining in a tissue microarray of stage III melanoma samples determined by the Spearman rank correlation. B) The inverse relationship between mPGES1 expression and tumor-infiltrating lymphocyte (CD8+) infiltration in these melanoma samples is shown in these representative images: a) high mPGES1 expression and low CD8+ infiltration (0.7%); b) high mPGES1 expression and low CD8+ infiltration (0.1%); c) low (no) mPGES1 expression and high CD8+ infiltration (13.2%); and d) low (no) mPGES1 expression and high CD8+ infiltration (25.1%). C-D) Overall survival and recurrence-free survival were estimated by the Kaplan-Meier method. The indicated p-values shows the association between groups and OS/RFS were evaluated using univariate Cox proportional hazards regression models. Groups were defined as follows: for CD8+ infiltration, low was ≤ median CD8 percentage and high was > median CD8 percentage; for mPGES1, low was intensity 0, 1, or 2 and high was intensity 3. Since only one patient was in the high/high group, that category was not included in the analysis. Red lines, high mPGES1 expression, low CD8+ infiltration; black lines, low mPGES1 expression, low CD8+ infiltration; green lines, low mPGES1 expression, high CD8+ infiltration.

DISCUSSION

Despite a reasonable understanding of the influence of PGE2 on inflammation and risk status in many cancers (8), its role in melanoma progression and immunosuppression has not been tested. The present study builds upon our previous observations on the clinical impact of mPGES1 in melanoma, which reported the significant increase in mPGES1 expression during melanoma progression and its association with poor survival of patients with stage III melanoma (9). The current study reveals that mPGES1 mediates most, if not all, the previously observed COX2/PGE2 axis–dependent evasion of immunity.

The influence of PGE2 on immunosuppression appears to be as complicated as its pro-inflammatory effects, and the mechanisms are even less well understood (30). Our study provides that elevated mPGES1 expression is associated with low CD8+ T cell infiltration into melanoma-rich areas of tumor positive lymph nodes, especially in those patients having poor survival in the Stage III melanoma samples tested. In our previous publication (9) mPGES1 protein expression also was suggested as a poor prognostic marker in stage III melanoma from non-lymphoid tissues. It has been reported that T-cell activation markers, such as interleukin 2 receptor α (CD25) and OX40 (CD134), were significantly lower in melanoma metastasis in distant visceral metastases, compared with non-metastatic or lymph node metastatic tumors (31), suggesting that the lymph node tumors may have preferential T cell availability. At this time, we propose that the role of mPGES1 in tumor microenvironment of the lymph node could differ as CD8+ T cells may be more migratory, in and out of the node in comparison to other sites of metastasis. While it is well known that such nodes are usually the first site of metastatic spread, our finding may be useful in patient prognostication, if validated. A more recent analysis of Stage IV melanoma, whose patients had uniformly poor prognosis, did not show any such association as T cells constituted only a minority of the overall tumor population in these visceral metastases environment as it could be seen in tumor HE and mPGES1 IHC pairs (Supplementary Figure 6). The presence of CD8+ T cells in melanoma is correlated with a positive patient prognosis (32,33). Our current data provides the first evidence that mPGES1 is an important factor in regulating CD8+ T cell infiltration into these tumors. This effect may be either direct or indirect. In terms of a direct effect, PGE2 signal transduction is mediated by four G-protein coupled receptors, EP1–4 (34). PGE2 acting through EP2 and EP4 prostanoid receptors is a key inducer of T cell cAMP levels (34). The activation of the EP2 and EP4 pathways by PGE2 may also positively regulate the level of PD-1 in infiltrating CD8+ T cells, which results in immune tolerance (35). Indirect PGE2 effects, in contrast, can involve a number of immune cell factors, including MDSC, dendritic cells, natural killer cells, and macrophages. Levels of markers for MDSCs were positively correlated with mPGES1 levels in the human melanoma TCGA data set (Figure 2A). In contrast, mouse mMDSCs, although a minor presence compared to the lymphoid DCs, significantly increased in numbers in the PTGES KO melanoma compared to scramble control in a syngeneic mouse experiment (Figure 4E) representing a dichotomy in these systems. The differences in mouse and human MDSC markers and function remains and exciting and evolving field, but to date our data suggests that in the mouse, the few MDSC present may not be functional. It is known that MDSC express PGE2 receptors and their immunosuppressive functions are regulated by PGE2 (36,37). Therefore, PGE2 depletion by PTGES KO may reduce MDSCs’ immunosuppressive activity on activated T cells. DCs play a significant role in the control of cancer by adaptive immunity, which is influenced by PGE2 (16,38,39) and our finding that deletion of mPGES1 enhanced the accumulation of tumor-infiltrating conventional, lymphoid and CD8+ DCs supports dendritic cell–mediated immune suppression. The Batf3 transcription factor expression selectively supports the acquisition of CD8+ DC phenotype and function (40) and the intratumoral presence of Batf3+ DCs promotes tumor-reactive effector T cell recruitment (41). Type I IFN pathway-mediated chemokines, CCL2, CCL4, and CCL5 are important for Batf3+ DCs infiltration (42). PGE2 inhibits type I IFN production and suppresses PGE2 synthesis while mPGES1 inhibitor treatment increased type I IFN (43) Our genetic deletion of mPGES1 increased these chemokines that appears to enable CD8a+ DC infiltration in tumor via the activation of type I IFN pathway. In addition, CD8a+ DCs attract effector CD8+T cells by secreting T cell-attracting chemokines such as CXCL9 and 10 (41), which were upregulated in the tumor tissues we analyzed that lacked PTGES. Collectively, these findings suggest that mPGES1 can support a shift from antitumor to immunosuppressive responses within the tumor microenvironment.

Cancer immunotherapies aim to shift the balance back to dominant antitumor immunity through various approaches, including antibody blockade of immunosuppressive signaling pathways, vaccination, or adoptive transfer of activated or engineered T cells (44). Tumor-associated PD-L1 expression is predictive of clinical response to PD-1-directed immunotherapy. In addition to our earlier findings regarding mPGES1 expression in melanoma (9), human COX2 expression was found by others to correlate significantly with PD-L1 expression in both primary and unmatched metastatic melanomas (45). Similarly, human melanoma cancer cell lines express elevated levels of both COX2 and PD-L1 (45). In our current study, inhibition of COX2 and mPGES1 did not directly affect PD-L1 expression in vitro, although expression of COX2 and mPGES1 was correlated with that of PD-L1 in the CCLE human melanoma cell dataset. However, PD-L1 staining was dramatically greater in PTGES-KO–derived tumor tissues than in control tumors. Since IFNγ is a strong inducer of PD-L1 and is secreted mainly by T cells, macrophages, and natural killer cells, we expect that the increased PD-L1 expression in PTGES-KO tumors is mediated by IFNγ from the increased T cell infiltration into tumor sites. The blockade of the PD-1 pathway reinvigorates exhausted T cells (46) and our results demonstrated that about 30% of CD8+ T cells showed the exhaustion phenotype. These findings provide a rationale for the combination with anti-PD-1 antibodies to enhance T cell –mediated elimination of tumor cells. It has been reported that both aspirin and celecoxib synergize with anti-PD-1 blockade in mouse models of melanoma and colorectal cancer (18). Comparisons between human, rat and mouse mPGES1 show structural differences in their active site that suggest more selective mouse inhibitors are needed to achieve the same degree efficacy when combined with PD-1 in a syngeneic mouse setting (47,48). This prompted us to perform KO of mPGES1 in syngeneic melanoma cells to achieve the same effects as aspirin or celecoxib. Further studies are ongoing to identify and evaluate mPGES1 inhibitors that are selective for mouse mPGES1. Knockout of mPGES1, as shown in Figure 5 of the present study, more dramatically enhanced the antitumor effect of anti-PD1 therapy than either of these agents (18). Also, we preliminarily tested whether the anti-tumor effect of mPGES1 inhibition with anti-PD-1 antibody is immune-mediated by re-inoculating the tumor-free mice with wild type melanoma live cells. Our preliminary study showed that tumor did not growth in half of the mice (2 of 4) which were remaining tumor free by mPGES1 inhibition with anti-PD-1 antibody and tumor growth rate in other half mice were much slower than in control mice (data not shown). Therefore, these exciting results have helped focus our future efforts to test that the combination of PGE2 inhibitory strategies and anti-PD-1 antibodies is able to provide long-term protection and immunological memory against melanoma. We also report that there is a tendency (p= 0.087) for higher levels of mPGES1 in progressive disease or partial response tumors than in complete response tissues in the public RNA-sequencing dataset (49), including 38 melanoma patient specimens treated with anti-PD-1 therapy, either pembrolizumab or nivolumab (Supplementary Figure 7). Therefore, much more data are needed to evaluate the association between mPGES1 expression and the efficacy of PD-1 therapy in melanoma patients and provide a justification for developing new therapeutic approaches, which include checkpoint blockade to reactivate tumor-inhibited effector T cells in combination with an mPGES1 inhibitor to impair tumor-induced immunosuppression.

Our data suggest that the selective mPGES1 inhibitors will supplant COXIBs and NSAIDs as safer and more effective adjuvants for immunotherapy. There is increasing interest in the development and use of mPGES1 inhibitors for treating inflammatory diseases, and pharmaceutical companies have initiated extensive mPGES1 drug development programs (50,51). The first clinical trial with the mPGES1 inhibitor LY3023703 showed more potent inhibition of inducible PGE2 synthesis than celecoxib, without PGI2 or thromboxane A2 suppression (52). Another mPGES1 inhibitor, compound 35, is currently in phase III clinical trials for osteoarthritis (50,51). However, only recently these inhibitors are finding use in clinical trials, and none has yet reached the general market. More preclinical and clinical studies on mPGES1 inhibitors are needed if they are to serve as novel treatments for inflammation and cancer and as more effective adjuvants for immunotherapy (50,51).

In conclusion, our results support a potential role for mPGES-1-derived PGE2 production in regulating immune evasion and provide a rationale for therapeutic strategies targeting selected PGE2-driven inflammatory mediators as adjuvants with immune-based therapy. Clearly, more research and drug development efforts are necessary to implement and validate the use of mPGES-1 inhibitors as safe treatment options, but for now this approach appears to be very promising.

Translational relevance.

While many melanoma patients are initially responsive to systemic treatments, including various forms of immunotherapy, most advanced patients demonstrate either resistance to or relapse from these treatments. Much research is underway addressing the molecular and biologic mechanisms that drive resistance, and our data presented here provides strong evidence of a novel approach to overcome immune evasion by regulation of tumor intrinsic mPGES1, with detailed examples provided from both human and mouse models. Therefore, these results support development of a rational therapeutic strategy targeting specific PGE2 inflammatory mediators, and identify the mPGES1 inhibition as a useful approach for improving clinical responses to check point inhibition, and possibly later to other types of immune based therapy.