Inhibition of Microsomal Prostaglandin E2 Synthase Reduces Collagen Deposition in Melanoma Tumors and May Improve Immunotherapy Efficacy by Reducing T-cell Exhaustion Open Access

Melanoma, which originates epidermal melanocytes, is the deadliest type of skin cancer. In fact, patients with melanoma who develop distant metastases have a dismal prognosis, with a 5-year survival rate of less than 25% (1, 2). Recent advances in drug discovery have led to new molecular targeted therapies, therapeutic vaccines, adoptive T-cell therapy, and immune checkpoint inhibitors (ICI), all of which have revolutionized treatment for melanoma (3, 4). In particular, the advent of ICIs has reaffirmed the potential role of tumor immunity, as a subset of patients has gained durable responses and long-term survival benefits from ICIs (5–7). However, a certain number of patients will experience primary resistance, and the majority of patients will develop acquired resistance regardless of their initial response to ICIs (8). It was recently shown that these resistance mechanisms are largely dependent on the negative switching of tumor-immune phenotypes induced by ICIs, such as T-cell exhaustion, the activation and stimulation of regulatory T cells, myeloid-derived suppressor cells, and the repolarization of macrophages into the M2 phenotype (9, 10). Therefore, many efforts are underway to develop effective strategies that can enhance immunotherapy by favorably altering the tumor immune microenvironment (TIME).

Various molecular components of the inflammatory process unquestionably contribute to the establishment of an immune-suppressive network in melanoma (11). Arachidonic acid metabolite pathways trigger inflammatory responses and consequently facilitate immunosuppression in the TIME (12–14). In particular, cyclooxygenase 2 (COX-2) is a well-known proinflammatory enzyme responsible for the production of multiple prostaglandins, including prostaglandin E2 (PGE2), which plays a central role in regulating the mechanisms of immune tolerance in cancer (15–20). Recently, seminal studies by Zelenay and colleagues demonstrated that genetic ablation of COX-2 reinforced the antitumor type I immune response and the accumulation of Batf2-dependent CD103⁺ dendritic cells (DC) in the TIME in a mouse model of Braf^V600E melanoma (15, 21). In addition, they showed that pharmacologic inhibition of COX-2 synergized with anti-PD-1 therapy and represented a promising adjuvant for immunotherapies (15). However, the specific COX-2 inhibitors are associated with an increased risk for major adverse cardiovascular events (22, 23). Therefore, our group has targeted microsomal prostaglandin E2 synthase 1 (mPGES-1; refs. 24, 25), a downstream PGE2 synthase, for which an inhibitor is likely to avoid cardiotoxicity and is expected to use in a clinical setting. Our earlier report showed that elevated mPGES-1 expression in tumor samples was linked with low CD8⁺ T-cell infiltration and unfavorable survival outcomes in patients with stage III melanoma (25). In addition, genetic depletion of mPGES-1 significantly slowed the growth of Braf^V600E murine melanoma tumors via the increased infiltration of CD8a⁺ T cells and CD8a⁺ DCs and thus improved anti-PD-1 efficacy in a syngeneic mouse model (25). In the current study, our goals were to distinguish the roles of mPGES-1 from those of COX-2 in tumor immunity and to directly test the potential of mPGES-1 inhibitors as an adjuvant to immune-based therapies for melanoma in a preclinical mouse model.

Human melanoma cell line WM793 (Coriell, catalog no. WC00062, RRID: CVCL_8787) was purchased from the ATCC. Murine Braf^V600E melanoma cells established from a C57BL/6 Braf^+/LSL-V600E;Tyr::CreERT2^+/o;p16^INK4a-/− mouse was generously provided by Dr. S. Zelenay (The University of Manchester, Manchester, England; refs. 15, 21). Cell lines were cultivated in DMEM supplemented with 10% heat-inactivated FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin. Cell cultures were maintained in 5% CO₂ at 37°C. All the cell lines had been authenticated using short tandem repeat DNA fingerprinting using the AmpFLSTR Identifiler kit, within the last 3 years. All experiments were performed with Mycoplasma-free cell lines.

Knockout (KO) of ptgs1, ptgs2, and ptges genes coding for COX-1, COX-2, and mPGES-1 proteins, respectively, was performed using CRISPR/CAS9 knockout kits (OriGene Technologies, catalog no. KN314182 for ptgs1, KN314183 for ptgs2, and KN314172 for ptges) according to the manufacturer's specifications. Stable KO cells were selected with puromycin, and subclones were obtained with the limiting dilution method. KO of ptgs1, ptgs2, and ptges genes were confirmed by measuring the mRNA and protein levels using qRT-PCR and Western blot analysis, respectively.

Cells were seeded in 6-well culture plates at 2 × 10⁵ cells/well. After incubating for 24 hours, cells were treated with two different concentrations of celecoxib or mPGES-1 inhibitor CAY10678 (Cayman Chemical) or 0.1% DMSO and incubated for another 24 hours. Cell culture supernatants were collected and centrifuged at 3,000 × g for 10 minutes. Tumor lysates (100 mg) were washed three times with PBS and prepared in 500 μL RIPA buffer supplemented with 1% protease inhibitor cocktail and 1% phosphatase inhibitor cocktail (Thermo Fisher Scientific). Then, the tissue samples were sonicated with a sonicator thoroughly on ice, followed by centrifugation at 10,000 × g for 20 minutes. Protein concentration was determined, and the tumor tissue lysates were analyzed for ELISA. Levels of released prostaglandins and thromboxane in supernatants and tumor tissues were quantified using Prostaglandin E2 ELISA kit, Prostaglandin D2 ELISA kit, Prostaglandin F2α ELISA kit, 6-keto Prostaglandin F1α ELISA kit, and Thromboxane B2 ELISA kit (Cayman Chemical), according to the manufacturer's recommendations.

Total RNA was isolated with the RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. cDNA was synthesized with iScript Reverse Transcription Supermix for RT-PCR (Bio-Rad) according to the manufacturer's recommendations. Then, qRT-PCR was performed using specific oligonucleotide primers for murine gapdh, ptgs1, ptgs2, ptges, col3a1, col4a1, col4a2, col8a1, col16a1, and col18a1 genes (all purchased from Qiagen). Murine gapdh served as a reference gene. Amplification products were detected with Applied Biosystems Fast SYBR Green Master Mix (Thermo Fisher Scientific). PCR was performed with the following protocol using Mastercycler Realplex (Eppendorf): initial denature for 60 seconds at 95°C, amplification with 40 cycles of 10 seconds at 95°C for DNA denature and 30 seconds at 60°C for primer annealing and extension. The relative fold gene expression of samples was calculated using the 2^−ΔΔC_t method.

Total proteins were extracted with RIPA buffer supplemented with 1% protease inhibitor cocktail and 1% phosphatase inhibitor cocktail. Protein aliquots were electrophoresed on either Novex WedgeWell 10% for COX-1, COX-2, and GAPDH or 16% gel for mPGES-1 (Thermo Fisher Scientific), followed by transfer to nitrocellulose membranes. After blocking with 5% nonfat milk for 1 hour at room temperature, membranes were incubated overnight at 4°C with primary Abs (1:1,000 for COX-1 Ab (clone D2G6, Cell Signaling Technology, catalog no. 9896, RRID:AB_10860249), 1:200 for COX-2 Ab (clone 33, BD Biosciences, catalog no. 610203, RRID:AB_397602), 1:250 for mPGES-1 Ab (Novus Biologicals), and 1:3,000 for GAPDH Ab (clone 6C5, Santa Cruz Biotechnology). Subsequently, membranes were incubated with IgG horseradish peroxidase (HRP)-linked secondaryAbs (1:1,000, R&D Systems) for 1 hour at room temperature. Enzymatic signals were visualized with SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific). The membranes were stripped with 2% SDS for 30 minutes at 60°C and reincubated with specific Abs as needed.

Melanoma cells were seeded in 96-well culture plates at 4 × 10³ cells/well and incubated for 24 hours. Then, cells were exposed to different concentrations of celecoxib and CAY10678. After 24, 48, and 72 hours of treatment, cells were incubated with PrestoBlue Cell Viability Reagent (Thermo Fisher Scientific) for 2 hours according to the manufacturer's recommendations. The absorbance was monitored with a spectrophotometer at 570 nm using 600 nm as a reference wavelength.

The animal experimental protocol was approved by the Institutional Animal Care and Use Committee at MD Anderson. All animal experiments were performed according to the NIH guidelines. Six-week-old female C57BL/6J mice, purchased from the Jackson Laboratory, were anesthetized with isoflurane and were subcutaneously inoculated 100 μL suspensions of murine Braf^V600E melanoma cells (wild type, scramble control, ptgs1 KO, ptgs2 KO, and ptges KO) into the right flank (1 × 10⁶ per mouse). In treatment-control experiments, 8 days after cell injection, mice were randomly divided into two groups for the first study: vehicle or CAT10678, and six groups for the second study: IgG control, anti-PD-1 Ab (αPD-1), celecoxib+IgG, CAY10678+IgG, celecoxib+αPD-1, or CAY10678+αPD-1. αPD-1 (200 μg per mouse, BE0416, Bio X Cell) or IgG control (BE0086, Bio X Cell) was intraperitoneally administered every 3 days for a total of four times, and 50 mg/kg celecoxib or 100 mg/kg CAY10678, suspended with 1% Tween 80 and 0.5% carboxymethylcellulose in normal saline, was intraperitoneally administered daily for 13 days. Tumor length (L) and width (W) were measured every 3 or 4 days with an external caliper, and the volume (V) of each tumor was calculated according to the equation [V = (L × W²) × 0.5]. Body weight was also measured at the same time. The mice were euthanized by carbon dioxide asphyxiation. Tumors were then excised, weighed, and subjected to Masson's trichrome staining or multiplex fluorescent IHC (mfIHC) staining.

All sections were scanned using Vectra Polaris Imaging System (Akoya Biosciences). At least two and up to five of regions of interest (ROI; each ROI was 1.1×1.5 mm) were randomly selected from each mouse tumor, depending on the tumor size, using InForm software (PerkinElmer) for the quantitative analysis of collagen deposition and tumor-infiltrating lymphocyte (TIL) distribution (Supplementary Fig. S1A). ROIs for the analysis of TIL distribution were selected at the same location as those for the collagen deposition analysis (Supplementary Fig. S1B).

To evaluate the collagen deposition in mouse tumor tissues, sections of formalin-fixed paraffin-embedded (FFPE) tissues with a thickness of 5 μm were stained by Masson's trichrome stain kit (Polysciences Inc.) according to the manufacturer's recommendations. Collagen positivity (%) was calculated using automated digital pathology and image analysis software (Visiopharm). On each ROI, background glass and vessel area were excluded from analysis and collagen positivity (%) was automatically calculated as the ratio of the Masson's trichrome–positive area to the entire area (Supplementary Fig. S2).

mfIHC staining was performed using Opal 7-Color Manual IHC Kit (Akoya Biosciences), according to the manufacturer's recommendations. Concretely, 5 μm FFPE tissues were baked for 60 minutes at 60°C. After deparaffinization, a heat-mediated stripping procedure using AR6 buffer (Akoya Biosciences) at 95°C for 15 minutes was performed with the EZ-Retriever systems V.3 (BioGenex) between each Ab staining cycle. After blocking with Protein Block (Agilent Technologies Inc) for 20 minutes, slides were incubated for 30 minutes at room temperature with primary Abs. Sequentially, Opal Polymer HRP Ms + Rb was introduced for 10 minutes and preselected opal fluorophores were applied for 10 minutes at room temperature. The staining procedure was carried out in a step-by-step manner. After all staining was completed, slides were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) solution (Supplementary Fig. S3A–S3D). Supplementary Table S1 contains the Abs and reagents for the four panels utilized.

The number of signal-positive c CD45⁺CD8a⁺IFNγ⁺ells was counted using Visiopharm software. First, each cell was stained with DAPI and segmented using the software's nuclei detection application, and the total number of cells in each ROI was automatically counted. Next, the signal positivity threshold for each marker was manually optimized on a one-by-one basis, and the number of signal-positive or -negative cells was automatically counted in a step-by-step manner (Supplementary Fig. S4). By this method, a total of 10 cell lineages were assessed from four mfIHC panels (Supplementary Table S2).

The extracted RNA from ptgs1-KO, ptgs2-KO, ptges-KO, and scramble cells using the RNeasy Mini Kit were checked for overall quality. Only RNA samples with high quality (RNA Integrity Number [RIN] > 8.0) and high purity (Optical Density [OD] 260/280 = 1.8–2.0) were used. mRNA libraries were generated with a Bioo Scientific NEXTflex Rapid Directional RNA sequencing (RNA-seq) Library Prep Kit. The mRNA libraries were then sequenced on an Illumina HiSeq 2500 in rapid mode using 101 bp paired-end reads.

RNA-seq normalized data were utilized to compare ptgs1, ptgs2, or ptges mRNA levels across all the control (scramble) and KO conditions. Of the 28 types of collagens encoded by 43 genes in the genome, only collagen genes with average normalized counts greater than 10 (col3a1, col4a1, col4a2, col4a5, col5a1, col8a1, col16a1, and col18a1) were included and analyzed by principal component analysis (PCA) and heat maps using ClustVis software (ClustVis, RRID:SCR_017133).

RNA-seq expression data for The Cancer Genome Atlas (TCGA)-Skin Cutaneous Melanoma (SKCM) were obtained from https://xenabrowser.net/ in May 2022. All the data were normalized to log₂ normalized counts+1. Only patients with metastatic melanoma from TCGA-SKCM were included in the analysis. Patients with metastatic melanoma were divided into four quantiles based on PTGES mRNA levels. Patients in the lower quartile (Low-PTGES, n = 92) and the upper quartile (High-PTGES, n = 92) were included and compared for PTGES, COL1A1, COL1A2, COL3A1, COL4A1, COL4A2, COL8A1, COL16A1, and COL18A1 mRNA levels.

Statistical analyses were conducted using GraphPad Prism 8 (GraphPad Software Inc., RRID:SCR_002798) or R version 4.1.2 (https://www.R-project.org/) in a two-tailed way. Continuous variables were expressed as mean ± 95% confidence interval. Statistical differences between two groups were compared using Student t test. Multiple groups were analyzed by one-way ANOVA and Tukey multiple comparisons test. Chronologic changes in tumor volume between two groups were compared by repeated measures one-way ANOVA, followed by Tukey post hoc test. Survival curves were estimated via the Kaplan–Meier method and compared via the log-rank test. P values < 0.05 were considered statistically significant.

The RNA-seq data generated and discussed in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) and are accessible through the GEO Series accession number GSE236204.

Several enzymes convert the arachidonic acid present in the phospholipids of cell membranes into active metabolites (Fig. 1A). To evaluate the specific roles of mPGES-1 in melanoma, we established murine Braf^V600E melanoma cells, which knocked out ptgs1, ptgs2, and ptges genes (coding for COX-1, COX-2, and mPGES-1 proteins, respectively), using CRISPR/CAS9. As reported previously, murine Braf^V600E melanoma cells have elevated levels of COX-2 and mPGES-1 and produce a large amount of prostaglandins in normal conditions (15, 24). KO efficacy for ptgs1, ptgs2, and ptges genes and corresponding proteins were respectively assessed by qRT-PCR and Western blotting (Fig. 1B and C). Then, we compared the prostaglandin and thromboxane levels released in supernatants. As thromboxane A2 (TxA2) and prostaglandin I2 (PGI2) are highly labile and rapidly degraded, we instead measured thromboxane B2 (TxB2) and 6-keto prostaglandin F1α (PGF1α), respectively (Fig. 1A). Suppression of PGE2 release was most effective with ptgs2 KO and moderately effective with ptges and ptgs1 KO (Fig. 1D). Deletion of ptgs2 gene slightly but significantly decreased the release of TxB2 (Fig. 1E). The release of prostaglandin D2 (PGD2), prostaglandin F2α (PGF2α), and 6-keto PGF1α were dramatically reduced by ptgs2 KO and slightly decreased or unchanged by ptgs1 KO. In contrast, ptges KO dramatically increased PGD2, PGF2α, and 6-keto PGF1α formation (Fig. 1F–H).

To elucidate the impact of COX-1, COX-2, and mPGES-1 expression on tumor burden, we implanted ptgs1-KO, ptgs2-KO, ptges-KO, or scramble cells into the flank of syngeneic immunocompetent C57BL/6J mice and compared the tumor growth. Before cell implantation, four subclones of ptgs1-KO, ptgs2-KO, ptges-KO, and scramble cells with similar proliferation rates in vitro were selected (Fig. 2A). The tumor growth originating from ptgs2- or ptges-KO cells was significantly slower than that of tumors originating from scramble or ptgs1-KO cells (Fig. 2B), whereas the tumor growth was not significantly different between ptgs1 KO- and scramble-derived tumors and between ptgs2 KO- and ptges KO-derived tumors. Moreover, tumor-free survival (tumor volume < 1,000 mm³) was longer in ptgs2 KO- and ptges KO-derived tumors compared with ptgs1 KO- and scramble cell-derived tumors (Fig. 2C).

The similar tumorigenicity between ptgs2- and ptges-KO cells (Fig. 2B) despite their difference in PGE2 suppression potencies (Fig. 1D) suggests the presence of other factors which may specifically be involved in the regulation of tumor progression and immune cell responses by mPGES-1. To address this, we performed RNA-seq analysis using three independent samples from each scramble, ptgs1-KO, ptgs2-KO, and ptges-KO murine Braf^V600E melanoma cells cultured in vitro. The mRNA levels of ptgs1 showed a decreased in ptgs1-KO, but not ptgs2-KO or ptges-KO cells compared with scramble control cells (Supplementary Fig. S5A). Similarly, mRNA levels of ptgs2 were reduced in ptgs2-KO cells, but not in ptgs1-KO or ptges-KO cells, compared with scramble cells (Supplementary Fig. S5B). Also, the mRNA levels of ptges were reduced in ptgs1-KO and ptges-KO cells, but not ptgs2-KO cells, compared with scramble cells (Supplementary Fig. S5C). In summary, all these results validated the results observed in qRT-PCR and Western blot analysis (Fig. 1B and C). We focus on analyzing genes that code for collagen molecules, a total of 43 genes in the human/mouse genome. In the RNA-seq dataset, only collagen genes with average normalized counts greater than 10 (col3a1, col4a1, col4a2, col4a5, col5a1, col8a1, col16a1, and col18a1) were included and analyzed. In PCA, we observed that the eight collagen genes clustered each of the cell line conditions analyzed (Supplementary Fig. S5D). Also, heat map (Fig. 2D) and boxplots (Supplementary Fig. S6A–S6H) showed that all the eight collagen genes were significantly downregulated in ptges-KO cells compared with scramble control cells. RNA-seq data were validated by qRT-PCR. Although some qRT-PCR results were inconsistent with RNA-seq data, mRNA levels of some of these genes, including col3a1, col4a1, col4a2, col8a1, col16a1, and col18a1 were reduced in ptges-KO cells compared with scramble cells (Supplementary Fig. S6I and S6J). To evaluate the clinical relevance of this finding, we assessed a cohort of patients with metastatic melanoma from TCGA-SKCM dataset. Patients with high and low mRNA levels of PTGES (n = 92 each) were compared (Supplementary Fig. S7A). In accordance with the results of mouse cells, mRNA levels of collagen-related genes COL3A1, COL4A1, COL4A2, COL8A1, and COL18A1 were significantly lower in patients with low PTGES than in those with high PTGES (Supplementary Fig. S7B–S7F). In contrast, COL16A1 levels were similar between the two groups (Supplementary Fig. S7G). Intriguingly, mRNA levels of COL1A1 and COL1A2, coding for collagen type 1, which is most abundant of the collagens, were also linked with PTGES in patients with melanoma (Supplementary Fig. S7H and S7I). Data integration from RNA-seq, qRT-PCR, and TCGA-SCKM RNA-seq showed that four collagen genes (col3a1, col4a1, col4a2, and col8a1) were consistently downregulated when the ptges mRNA levels were decreased (Supplementary Fig. S6J). Furthermore, ptges-KO murine Braf^V600E melanoma cells, cultivated in vitro, were less aggregated than scramble, ptgs1-KO, and ptgs2-KO cells (Supplementary Fig. S8A). Although different cell aggregation may be influenced by different expression of cell-cell adhesion molecules, ptges-KO cells cultured in collagen IV-coated well (Thomas Scientific) were more aggregated than those cultured in normal well (Supplementary Fig. S8B), suggesting that ptges-KO cells released the lowest amount of collagen which was partially associated with cell aggregation. Of note, cutaneous melanoma data obtained from the TCGA-SKCM dataset demonstrated that mRNA levels of some collagen-related genes were closely correlated with immunosuppressive signatures (Fig. 2E). These results suggest that mPGES-1 regulates the production and release of collagen, which may be associated with immune suppression in melanoma.

To identify the distinctive role of intrinsic mPGES-1 from that of COX-2 at the tumor site, we compared the collagen deposition and TIL distribution among tissues from similarly sized tumors derived from scramble, ptgs2-KO, and ptges-KO murine Braf^V600E melanoma cells. First, we performed Masson's trichrome staining to compare the intratumoral collagen deposition across the three groups. Collagen positivity was significantly diminished by ptges KO, but it was not altered by ptgs2 KO (Fig. 3A). Next, the distribution of TILs was assessed by mfIHC staining (Supplementary Fig. S9A–S9F and Supplementary Fig. S10A–S10C). Tumor-infiltrating cytotoxic CD8a⁺ T cells were drastically increased by ptgs2 and ptges KO; however, no significant difference was observed between tumors derived from ptgs2-KO and ptges-KO cells (Fig. 3B; Supplementary Fig. S9A).

Subsequently, we analyzed effector and exhaustion markers of cytotoxic CD8a⁺ T cells. The total number of tumor-infiltrating granzyme B⁺ (GrB⁺) effector CD8a⁺ T cells was greatly improved by ptgs2 and ptges KO, although the percentage of GrB⁺CD8a⁺ T cells in total CD8a⁺ T cells was similar between three groups and no significant difference was observed between ptgs2-KO and ptges-KO cells (Fig. 3C; Supplementary Fig. S9B). The number of infiltrating IFNγ⁺CD8a⁺ T cells was significantly higher in tumors derived from ptges-KO versus scramble cells (Supplementary Fig. S10A). Notably, the number and frequencies of tumor-infiltrating Tim3⁺ exhausted CD8a⁺ T cells were significantly fewer in tumors derived from ptges-KO versus scramble and ptgs2-KO cells (Fig. 3D; Supplementary Fig. S9C). Conversely, the number of infiltrating IFNγ⁺Tim3⁻CD8a⁺ T-cell population was significantly higher in tumors derived from ptges-KO cells compared with scramble and ptgs2-KO cells (Supplementary Fig. S10B). The number of Tim3⁺PD-1⁺LAG3⁺ exhausted CD8a⁺ T cells was lowest in ptges-depleted cell-derived tumors, followed by ptgs2-depleted and scramble cell-derived tumors although the difference did not reach statistically significant (Supplementary Fig. S10C). Furthermore, we compared tumor-infiltrating classic CD8a⁺ DCs population, synonymous with Batf3⁺ DCs, because this subset is a key orchestrator of immune responses in cancer and our previous work demonstrated that ptges depletion could enhance this population in melanoma (25). In agreement with our earlier work, tumor-infiltrating CD8a⁺ DCs were more abundant in ptges-depleted cell-derived tumors compared with scramble cell-derived tumors, but this was not specific to ptges as ptgs2 depletion equivalently upregulated the frequencies (Fig. 3E; Supplementary Fig. S9D). Finally, the frequencies of tumor-infiltrating natural killer (NK) and B cells were examined because tumor-derived PGE2 inhibited the accumulation of these two immune cell types in previous studies (21, 26, 27). However, in the current study, neither ptgs2 nor ptges depletion affected the trafficking of NK cells and B cells into tumors (Fig. 3F and G; Supplementary Fig. S9E and S9F).

Slight differences in the amino acid compositions of human and mouse mPGES-1 demands the use of a mouse/human dual mPGES-1 inhibitor for preclinical in vivo study. CAY10678 was selected as a potential candidate to selectively inhibit recombinant human (IC₅₀ = 90 nmol/L) and rat mPGES-1 (IC₅₀ = 900 nmol/L; ref. 28). CAY10678 attenuated the release of PGE2 from both human WM793 and murine Braf^V600E melanoma cells by up to 50%, albeit to a lesser degree than celecoxib (Fig. 4A). The effect of these inhibitors on the release of other prostaglandins and thromboxane in supernatants was similar to that of genetic depletion of ptgs2 and ptges in murine Braf^V600E melanoma cells (Fig. 1E–H); only celecoxib decreased TxB2 production (Supplementary Fig. S11A), whereas PGD2, PGF2α, and 6-keto PGF1α production were suppressed by celecoxib and increased by CAY10678 (Supplementary Fig. S11B–S11D).

Exposure to different concentrations (5 and 10 μmol/L) of celecoxib and CAY10678 did not influence proliferation of murine Braf^V600E melanoma cells in vitro (Fig. 4B). Only CAY10678 effectively reduced the mRNA levels of multiple collagen genes (Fig. 4C). In C57BL/6J mice injected with murine Braf^V600E melanoma cells, CAY10678 treatment significantly delayed tumor growth compared with vehicle treatment (Fig. 4D). CAY10678-treated mice had significantly lower tumor weight than untreated mice at the time of euthanasia (Fig. 4E). In addition, PGE2 levels were lower in CAY10678-treated tumor tissue versus untreated tumor tissue (Fig. 4F). Mice treated with CAY10678 displayed transient body weight loss, most relevant at 3 days of treatment, yet this change was not significant compared with untreated mice through the observation period (Fig. 4G).

Our recent study demonstrated that genetic deletion of mPGES-1 enhanced PD-1 blockade in a syngeneic mouse model of Braf^V600E melanoma (25), indicating the potential of mPGES-1 inhibitors for improving immunotherapy. In addition, COX-2 inhibitors significantly synergized with immunotherapy in the same mouse model (15). We therefore set up a head-to-head study to directly compare the impact of mPGES-1 and COX-2 inhibitors on anti-PD-1 therapy. Eight days after inoculation with parent murine Braf^V600E melanoma cells, mice were randomized into six groups: IgG control, αPD-1, celecoxib+IgG, CAY10678+IgG, celecoxib+αPD-1, CAY10678+αPD-1 (Fig. 5A). Single-drug therapy with αPD-1, celecoxib, and CAY10678 moderately and similarly slowed tumor growth; suppression of tumor growth was more powerful using combinations of celecoxib or CAY10678 with αPD-1 (Fig. 5B). Tumor weight of mice treated with combination therapies was significantly lower than that of mice treated with single-drug therapies at the time of euthanasia (Fig. 5C). Tumor-free survival (tumor volume < 500 mm³) was most extended in mice treated with combination therapies and somewhat extended with single-drug therapies (Fig. 5D). Tumor growth speed and weight as well as tumor-free survival were all similar between mice treated with celecoxib+αPD-1 and CAY10678+αPD-1 (Fig. 5B–D). Consistent with the results of CAY10678 single-drug therapy (Fig. 4G), mice treated with CAY10678 and CAY10678+αPD-1 experienced slight body weight loss of less than 10% (Fig. 5E)

Finally, we compared the collagen deposition and TIL distribution using tumors of drug-treated mice at the time of euthanasia. αPD-1 monotherapy significantly induced the collagen in tumors (Fig. 6A), which may be associated with earlier findings that long-term exposure to αPD-L1 accelerated intratumoral collagen deposition in a murine lung cancer model (29). In contrast, CAY10678 effectively reduced collagen deposition, even when combined with αPD-1. Consequently, the difference of TIL distribution was evaluated with mfIHC staining and quantified by Visiopharm software. Total number and frequencies of tumor-infiltrating cytotoxic CD8a⁺ T cells were comparably augmented by any of our treatments (Fig. 6B; Supplementary Fig. S12A). The number of tumor-infiltrating GrB⁺ effector CD8a⁺ T cells was highest in mice treated with combinations of αPD-1 plus celecoxib or CAY10678 and somewhat increased in mice treated with αPD-1, celecoxib, or CAY10678 monotherapy (Fig. 6C). The percentage of GrB⁺CD8a⁺ T cells in total CD8a⁺ T cells was highest in mice treated with CA10678 monotherapy and combinations of αPD-1 plus celecoxib or CAY10678 (Supplementary Fig. S12B). In contrast, the number of tumor-infiltrating Tim3⁺ exhausted CD8a⁺ T cells was significantly lower in mice treated with single-agent CAY10678 or CAY10678 plus αPD-1 compared with any other treatments (Fig. 6D), and the percentage of Tim3⁺ exhausted CD8a⁺ T cells in total CD8a⁺ T cells was significantly suppressed by CAY10678-based treatments compared with celecoxib-based treatments although αPD-1 monotherapy also strongly suppressed the frequency (Supplementary Fig. S12C). The accumulation of CD8a⁺ DCs was induced by both celecoxib- and CAY10678-based treatments (Fig. 6E; Supplementary Fig. S12D). Neither celecoxib nor CAY10678 increased the number of tumor-infiltrating NK cells, as we observed in our KO models (Fig. 3F); however, intriguingly, tumor-infiltrating NK cells were significantly upregulated in mice treated with αPD-1–based regimens (Fig. 6F; Supplementary Fig. S12E). Celecoxib plus αPD-1 treatment was the only combination that enhanced the infiltration of B cells (Fig. 6G; Supplementary Fig. S12F). Finally, we assessed the presence of apoptotic tumor cells by evaluating the levels of cleaved caspase-3–positive cells (Fig. 6H), whose positivity was almost inversely related to the tumor weight (Fig. 5C).

The current study demonstrated that despite the common upstream/downstream enzymes responsible for the production of arachidonic acid metabolites, intrinsic mPGES-1 displays a specific function, distinct from that of COX-2, in modulating TIME in melanoma. Besides the recent intensive researches showing the roles of COX/PGE2 axis in evading tumor immunity (15–21), we, first time, uncovered the intriguing roles of mPGES-1 apart from mPGES-1/PGE2 axis. Genetic or pharmacologic inhibition of mPGES-1 was unique in that it could shift the TIME to favor cytotoxic CD8⁺ T-cell activation by excluding collagen production and Tim3⁺ exhausted phenotype. Accordingly, despite the inadequate inhibitory function of PGE2 synthesis, CAY10678 displayed noninferior therapeutic effects compared with celecoxib and was capable of boosting anti-PD-1 treatment without an increase of serious adverse events in a syngeneic mouse model.

Different arachidonic acid metabolite profiles induced by COX-2 and mPGES-1 inhibition have been proposed. Bergqvist and colleagues showed that compound III, a mPGES-1 inhibitor (same chemical structure with CAY10678), reduced PGE2 and increased PGF2α and TxB2 production while NS-398, a COX-2 inhibitor, blocked the production of them all (30). In addition, mPGES-1 inhibitor LY3023703, but not celecoxib, increased systemic PGI2 synthesis in humans (31). Similarly, genetic and pharmacologic inhibition of COX-2 and mPGES-1 conferred different metabolite profiles in our study. Inhibiting mPGES-1 may repartition PGH2 to the remaining prostaglandin synthases, resulting in increased PGD2, PGF2α, and PGI2 generation in a murine Braf^V600E melanoma model. Considering that each metabolite has a specific role in moderating tumor cells and neighboring immune cells (15–17, 32–36), it is likely that different profiles of prostaglandins released by tumor cells during genetic or pharmacologic depletion of COX-2 and mPGES-1 may lead to different phenotypes or functional properties of antitumor immune cells. In particular, PGD2 has been recognized as a tumor suppressor via the PGD2 receptor in colitis-associated colon cancer (33), and PGI2 is capable of recruiting CD4⁺ T cells in MHC class II–expressing murine lung cancer cells (34). Increased release of PGD2 and PGI2 by mPGES-1 inhibition may partially support the shift to favorable immune periphery in melanoma. In addition, previous studies have shown that each prostaglandin has positive or negative impact on collagen synthesis (37, 38). Therefore, it is of interest to determine whether the different balance of these prostaglandins is associated with the different collagen-related genes and collagen deposition.

Collagen is a predominant component of the extracellular matrix that may directly and indirectly influence the TIME to be advantageous for cancer development and progression. Collagen confers a physical barrier that makes it difficult for immune effectors to infiltrate the tumor microenvironment (39). In addition, recent studies have shown that collagen derives T-cell suppression through leukocyte-associated immunoglobulin-like receptor-1 (LAIR-1), an inhibitory collagen receptor expressed on immune cells (29, 40). David and colleagues discovered that collagen binds to LAIR-1 on T cells and activates the src homology region 2 domain–containing phosphatase 1 pathway, thereby inducing a T-cell exhaustion alternative to the PD-1 pathway in lung cancer (29). In our current study, we found that intrinsic mPGES-1 was associated with the expression of collagen-related genes, especially col3a1, col4a1, col4a2, and col8a1, and that genetic and pharmacologic depletion of mPGES-1 effectively decreased collagen deposition in the tumor microenvironment in a mouse model. TCGA-SKCM dataset corroborated the close correlation between mPGES-1 and these collagen mRNA levels. Moreover, frequencies of Tim3⁺ exhausted cytotoxic CD8a⁺ T cells were decreased by genetic depletion of mPGES-1 as well as by the treatment with an mPGES-1 inhibitor, suggesting collagen reduction by mPGES-1 inhibition may facilitate reinvigoration of exhausted T cells. The selection of ROIs at the same locations using serial sections for Masson's trichrome and mfIHC staining allowed us to assess how changes in collagen deposition affect T-cell exhaustion in this study (Supplementary Fig. S1B). In addition, the significant association between collagen and Tim3 gene expression (HAVCR2) in the dataset (Fig. 2E) also supports our suggestion. However, further investigation is warranted to clarify the mechanism by which mPGES-1 inhibition improves T-cell exhaustion and enhances anti-PD-1 therapy.

Building on previous research by our group and others (15, 25–27), we also accessed whether associations exist between DC, NK-cell, and B-cell infiltration into the tumor site and mPGES-1 and COX-2 expression in tumor cells. Compatible with findings from earlier studies (15, 25), genetic or pharmacologic depletion of COX-2 and mPGES-1 similarly enabled the recruitment of the classic CD8a⁺ DC subset, which is essential in initiating CD8⁺ T-cell priming and directing, into the tumor site by suppressing PGE2 production and provoking type I IFN-mediated chemokines; however, this subset was also trafficked by anti-PD-1 monotherapy, probably because of IFNγ produced by drug-activated T cells (41), and no synergistic effect of mPGES-1 and COX-2 inhibitors was observed. Although it has been noted that PGE2 was involved in suppressing NK-cell and B-cell activation and proliferation (21, 26, 27), neither the COX-2 nor mPGES-1 inhibitor was sufficient to recruit these cells to the tumor site in this study, indicating the insufficient stimulation of appropriate chemokines by these drugs. We did not examine the impact of the mPGES-1 inhibitor on the infiltration of immune-suppressive cell populations—including regulatory T cells, myeloid-derived suppressor cells, and tumor-associated macrophages—because our earlier study has already shown that genetic depletion of mPGES-1 did not suppress the number of tumor-infiltrating immune-suppressive cells (25).

The current study included several limitations. First, although we showed that genetic depletion of ptges and pharmacologic inhibition of mPGES-1 reduced collagen deposition and T-cell exhaustion, we did not prove the direct association between the decrease of collagen deposition and improvement of T-cell exhaustion. Investigation this association is of great importance to better understanding the specific roles of mPGES-1. Second, we evaluated only tumor cell–producing collagen deposition in TIME. Considering that collagen is also produced by peritumoral stroma, it is conceivable that genetic depletion of ptges or pharmacologic inhibition of mPGES-1 indirectly decrease collagen deposition by regulating the function of peritumoral stroma. Third, we suggested in this study that intrinsic mPGES-1 expression in melanoma was linked with several collagen genes by RNA-seq data analysis using a mouse melanoma cell line and the analysis of TCGA-SKCM dataset. However, the data of TCGA-SKCM dataset should be interpreted cautiously as human melanoma cells included not only tumor cells but also peritumor cells including stromal and immune cells, and these cells may contribute to the expression of collagen genes.

In conclusion, intrinsic mPGES-1 uniquely functions as an inducer of collagen deposition and a modulator of cytotoxic T-cell exhaustion; and specific pharmacologic inhibition of mPGES-1 may hold therapeutic promise as a safer alternative to COX-2 inhibitors for improving immune checkpoint–based therapies in melanoma. However, to date, no current mPGES-1 inhibitor has been approved for human use, with early candidates exhibiting unique noncardiac toxicities (31, 42). Therefore, further development of mPGES-1 inhibitors for human use is encouraged and is expected to broaden adjuvant treatment options for immunotherapy in melanoma.

Y. Fukuda reports grants from The University of Texas, MD Anderson Cancer SPORE in Melanoma (P50-CA093459), MD Anderson Cancer Center Support Grant (P30-CA016672), Foundation for the National Institutes of Health (FNIH)-PACT, Dr. Miriam and Sheldon G. Adelson Medical Research Foundation, Jim Mulva Foundation, and the AIM at Melanoma Foundations during the conduct of the study. M.A. Bustos reports grants from Dr. Miriam and Sheldon G. Adelson Medical Research Foundation during the conduct of the study. No disclosures were reported by the other authors.

Y. Fukuda: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. S.-H. Kim: Conceptualization, data curation, formal analysis, validation, investigation, methodology, project administration, writing-review and editing. M.A. Bustos: Formal analysis, validation, investigation, methodology, writing-review and editing. S.-N. Cho: Formal analysis, visualization, methodology, writing-review and editing. J. Roszik: Data curation, formal analysis, validation. J.K. Burks: Resources, formal analysis. H. Kim: Formal analysis, methodology. D.S.B. Hoon: Resources, supervision, funding acquisition, writing-review and editing. E.A. Grimm: Conceptualization, resources, funding acquisition, investigation, writing-review and editing. S. Ekmekcioglu: Conceptualization, resources, supervision, funding acquisition, investigation, project administration, writing-review and editing.

This work was supported by The University of Texas, MD Anderson Cancer SPORE in Melanoma, P50-CA093459 (S. Ekmekcioglu and E.A. Grimm), MD Anderson Cancer Center Support Grant, P30-CA016672 (S. Ekmekcioglu and J.K. Burks), Foundation for the National Institutes of Health (FNIH)-PACT (S. Ekmekcioglu), Dr. Miriam and Sheldon G. Adelson Medical Research Foundation (E.A. Grimm and D.S.B. Hoon), Jim Mulva Foundation, and the AIM at Melanoma Foundations. Scientific and financial support for the PACT project are made possible through funding support provided to the FNIH by AbbVie Inc., Amgen Inc., Boehringer-Ingelheim Pharma GmbH & Co. KG, Bristol-Myers Squibb, Celgene Corporation, Genentech Inc., Gilead, GlaxoSmithKline plc, Janssen Pharmaceutical Companies of Johnson & Johnson, Novartis Institutes for Biomedical Research, Pfizer Inc., and Sanofi.

We would like to thank Dr. Yared Hailemichael in the Melanoma Medical Oncology Department at The University of Texas MD Anderson Cancer Center for his expert opinion on immune phenotypes suggestions and for Mr. Bryan Tutt, MA, ELS, of the Research Medical Library at The University of Texas MD Anderson Cancer Center for the scientific edition.