Cyclooxygenase-2 Inhibition Potentiates the Efficacy of Vascular Endothelial Growth Factor Blockade and Promotes an Immune Stimulatory Microenvironment in Preclinical Models of Pancreatic Cancer

Primary tumors and metastases require nutrients and oxygen delivered by blood vessels (1). Although angiogenesis is complex, it is widely recognized that vascular endothelial growth factor-A (VEGF-A) is the predominant angiogenic factor that promotes tumor neovascularization (2, 3). Inhibitors of angiogenesis have become a central part of systemic therapy for a variety of malignancies (4, 5). However, angiogenesis inhibition has, in general, resulted in only modest gains in clinical outcomes in cancer patients, as many patients treated with antiangiogenic/anti-VEGF therapy either fail to respond or relapse on therapy (6, 7). Additionally, antiangiogenic therapy has been implicated in promoting tumor progression and accelerating metastasis in preclinical models (8, 9).

Pancreatic cancer, the third leading cause of cancer-related death (10), is highly metastatic and poorly responsive to standard therapy (11, 12). It is also an immunologically “cold” tumor that has remained largely refractory to immune-checkpoint blockade (12, 13). Anti-VEGF therapy has been studied in pancreatic cancer patients (4); however, it has not provided significant clinical benefit in combination with gemcitabine, the standard chemotherapy for pancreatic ductal adenocarcinoma (PDA; refs. 14–16). Previously, we investigated the efficacy and biology of anti-VEGF therapy in preclinical models of PDA using the antibody mcr84 (9, 17). We found that mcr84 alone or in combination with gemcitabine slowed the growth of PDA but induced hypoxia-induced epithelial plasticity that resulted in a less differentiated tumor cell phenotype and continued metastatic burden (9). These observations reinforce the need to develop companion therapies that combat therapy-induced epithelial plasticity.

Inflammation is a pathologic phenotype that facilitates the “hallmarks” of cancer (18). Further, the incidence of several cancers is associated with inflammation, which contributes to tumor initiation and cancer cell survival by producing reactive oxygen species, cytokines, and proinflammatory mediators (19). Among mediators of inflammation that are associated with tumor progression is cyclooxygenase-2 (COX-2), an inducible enzyme that catalyzes the rate-limiting step in the synthesis of the prostaglandin E₂ (PGE₂). COX-2 is induced at sites of inflammation and during the process of tumor progression (20). Multiple studies have demonstrated that elevated COX-2 expression is prevalent in human malignancies, including PDA (21, 22). In addition, elevated expression of COX-2 in tumors correlates with advanced stage and worse outcome by promoting chemoresistance, metastasis, and angiogenesis (23, 24). COX-2 has also been identified as a potential mediator of VEGF-independent tumor angiogenesis (25). Thus, targeting COX-2 has been explored as a potential anticancer therapy (26). Additionally, COX-2 blockade can enhance the efficacy of antiangiogenic treatments in breast cancer, which supports the investigation of COX-2 inhibitors with VEGF blockade in other tumors (27).

Apricoxib is a selective COX-2 inhibitor that has shown significant antitumor activity in various xenograft models (28) and has been under clinical investigation. Previously, we demonstrated that apricoxib improved the efficacy of standard therapy in preclinical models of PDA (29). Further, we found that inhibition of COX-2 reversed epithelial–mesenchymal transition (EMT), leading to a shift toward a more epithelial phenotype in xenograft models of PDA (29). In the present study, we investigated the combination of anti-VEGF therapy and COX-2 inhibition as a therapeutic strategy in robust preclinical models of PDA with the hypothesis that apricoxib would prevent or reduce therapy-induced epithelial plasticity. We also investigated the effect of anti-VEGF and COX-2 inhibition on the immune landscape of PDA given that prior reports have demonstrated VEGF and PGE₂ can limit T-cell infiltration into tumor cell nests (30) and reports that EMT can be a significant driver of immune suppression in tumors (31–34).

Human pancreatic cancer cell lines AsPC-1 and HPAF-II were obtained from ATCC. Colo357 was obtained from the MD Anderson Cancer Center. AsPC-1 was grown in DMEM, and Colo357 and HPAF-II were grown in MEM. All cell lines were grown in a humidified atmosphere with 5% CO₂, at 37°C, and were DNA fingerprinted for provenance using the Power-Plex 1.2 kit (Promega) and confirmed to be the same as the DNA fingerprint library maintained by ATCC. Cell lines were confirmed to be free of mycoplasma (e-Myco kit, Boca Scientific) before use. In vitro PGE₂ and VEGF response to apricoxib treatment was evaluated by enzyme-linked immunosorbent assay (ELISA; R&D Systems) of conditioned media over different time points. To induce EMT, cells were grown on collagen I–coated plates and treated with 50 ng/mL transforming growth factor-β (TGFβ) for 24 hours (9). EMT changes were confirmed by probing cell lysates for E-cadherin, N-cadherin, and Vimentin (Cell Signaling Technology; see Supplementary Table S1).

All animals were housed in a pathogen-free facility with 24-hour access to food and water. Experiments were approved by and conducted in accordance with the Institutional Animal Care and Use Committee at UT Southwestern (Dallas, TX). Kras^LSL-G12D; Cdkn2a^fl/fl; Ptf1a^Cre/+ (KIC) mice were generated as previously described (35). At 3 weeks of age, mice were randomized to receive saline, mcr84 500 μg/dose i.p. once weekly, apricoxib 10 mg/kg by oral gavage daily or mcr84 plus apricoxib. All mice were sacrificed when they were 7 weeks old. Four-to-6-week-old female NOD/SCID mice were obtained from a campus supplier. A total of 1 × 10⁶ Colo357 cells were injected orthotopically and tumor growth was monitored by ultrasound. Mice with established tumors were randomized to receive therapy. Treatment groups were the same as described above. Mice bearing Colo357 tumors received 4 weeks of therapy prior to sacrifice. Tissues from all animal experiments were fixed in 10% formalin or snap-frozen in liquid nitrogen for further studies.

Formalin-fixed tissues were embedded in paraffin and cut into 5-μm sections. Sections were evaluated by Masson's trichrome staining and Immunohistochemistry (IHC) analysis using antibodies specific for VEGF, COX-2 (Abcam), E-cadherin, N-cadherin, Slug, Snail (Cell Signaling Technology), CD3 (Bio-Rad), CD8 (Bioss), FoxP3 (eBioscience), CD31 (dianova), FasL (Santa Cruz Biotechnology), CD11b (Abcam), iNOS, Arginase 1 (Thermo Fisher), endomucin (Santa Cruz Biotechnology), NG2 (Millipore), and F4/80 (Novus Biologicals). Fluorescent images were captured with Zeiss Aixoscan Z1 using ZenLite software. Color images were obtained with Hamamatsu Nanozoomer 2.0HT using NDPview2 software. Pictures were analyzed using NIS Elements (Nikon) and Fiji software.

Data were analyzed using GraphPad software. Results are expressed as mean ± SEM. Data were analyzed by ANOVA with the Dunn's test for multiple comparisons, and results are considered significant at P < 0.05.

To investigate the efficacy of COX-2 inhibition with apricoxib and VEGF blockade with mouse chimeric r84 (mcr84) (17) in preclinical models of PDA, we used a genetically engineered mouse model of PDA and SCID mice bearing established orthotopic pancreatic xenografts. Therapy was initiated in 3-week-old KIC mice. Mice were randomized to receive saline, mcr84, apricoxib, or mcr84 + apricoxib and were sacrificed after 4 weeks (7 weeks old). Therapy with mcr84 or apricoxib reduced primary tumor weight by ∼30%, whereas mcr84 + apricoxib reduced primary tumor weight by 62% compared with the control group (P < 0.0001; Fig. 1A). At the time of sacrifice, the extent of liver metastasis was determined based on gross metastasis. Seven of 10 evaluable mice in the control group had at least 1 macroscopic metastasis; this number was reduced to 1/5, 2/7, and 1/6 for the mcr84, apricoxib, and combination therapy groups, respectively (Fig. 1A). To further define the effect of COX-2 inhibition and anti-VEGF therapy on tumor burden and liver metastases, we established human PDA xenografts in mice by orthotopically injecting Colo357, a human pancreatic cancer cell line, into the pancreas of SCID mice. Similar to in vitro data published previously (29), Colo357 cells showed high COX-2 expression and were responsive to apricoxib (data not shown). Mice with established tumors, which was confirmed by ultrasound, were randomized to receive treatment as described above. After 4 weeks of therapy, we found that single-agent therapy had a minimal effect on primary tumor growth (Fig. 1B) and metastatic incidence, although the mean metastatic events per treatment group were reduced by mcr84 or apricoxib (Fig. 1B). In contrast, combination therapy significantly reduced primary tumor weight (P < 0.05 vs. control) and substantially limited metastases (P < 0.01 vs. control; P < 0.05 vs. single-agent therapy; Fig. 1B). H&E analysis of livers confirmed metastatic lesions in Colo357 tumor–bearing mice (Supplementary Fig. S1A). The effect of mcr84 + apricoxib on primary tumor growth compared favorably to the effect of gemcitabine + erlotinib in the same model reported in our prior study (ref. 29; Supplementary Fig. S1B).

COX-2 activity has been implicated in promoting angiogenesis (25, 36, 37). Previously, prostaglandins, products of COX-2 activity, were shown to elevate VEGF expression, and inhibition of COX-2 was shown to contribute to antiangiogenic effects (38, 39). Furthermore, fibroblasts from Cox-2-deficient mice were reported to produce significantly less VEGF than fibroblasts from wild-type or Cox-1-deficient animals (40). Additionally, treatment of wild-type fibroblasts with a selective COX-2 inhibitor resulted in a 90% reduction in VEGF production (40). However, recently Xu and colleagues (25) determined that PGE₂ can contribute to angiogenesis in a VEGF-independent manner in colon cancer models. Given these data, we sought to investigate the relationship between COX-2 activity and VEGF production in PDA cell lines; we selected a COX-2 negative cell line, AsPC-1, and 2 COX-2 positive cell lines, one with a high expression of COX-2, Colo357, the other with moderate COX-2 expression, HPAF-II (29). Cells were treated with 500 nmol/L apricoxib and the level of VEGF produced was determined by ELISA. Only in the high COX-2 cell line Colo357, COX-2 inhibition reduces VEGF production transiently. In HPAF-II and AsPC-1 cells, VEGF production was unaffected by apricoxib, with VEGF production in HPAF-II cells elevated over time (Fig. 1C). To determine if EMT induction altered VEGF production and/or the effect of apricoxib, we plated Colo357 and AsPC-1 cells under conditions that stimulate EMT. Under normal culture conditions, we observed similar trends as shown before. However, under EMT-inducing conditions VEGF production was elevated significantly and was largely independent of COX-2 inhibition in Colo357 cells. In AsPC-1 cells, VEGF production increased faster over time under induced EMT conditions compared with normal conditions (Fig. 1C). We also investigated the effect of apricoxib on PGE₂ production by Colo357 and AsPC-1 cells under normal and EMT-inducing culture conditions (Supplementary Fig. S2A and S2B). The induction of EMT was confirmed by evaluating the expression level of E-cadherin, N-cadherin, and Vimentin (Supplementary Fig. S2C). The induction of EMT reduced the effect of apricoxib on PGE₂ production in Colo357 cells. In contrast, AsPC-1 cells produced minimal PGE₂ under either culture condition (Supplementary Fig. S2A and S2B). We corroborated these findings by examining the level of VEGF expression in Colo357 pancreatic xenografts by immunofluorescence staining and found that VEGF expression was not affected by apricoxib (Fig. 1D). Importantly, apricoxib did reduce COX-2 expression in Colo357 tumors, supporting the pharmacodynamic activity of the drug. The induction of hypoxia by mcr84 is consistent with prior studies (9) and the reduction of microvessel density by mcr84 in Colo357 tumors (Supplementary Fig. S3A and S3B). We found that apricoxib alone did not reduce microvessel density in Colo357 tumors (Supplementary Fig. S3A and S3B), which further supports that apricoxib antitumor activity is not mediated by inhibition of angiogenesis. However, we did observe that apricoxib alone or in combination with mcr84 increased the percentage of pericyte-associated blood vessels in Colo357 tumors (Supplementary Fig. S3C). These data suggest that COX-2 functions in a VEGF-independent manner in PDA to promote tumor progression.

Although anti-VEGF therapy with mcr84 restricts tumor growth and improves the survival of KIC mice (9), therapy-induced hypoxia results in a less differentiated tumor cell phenotype (9). We previously found that COX-2 inhibition with apricoxib reverses EMT in HT29 xenografts (28) and Colo357 tumor–bearing mice (29). To determine whether apricoxib can prevent or reduce hypoxia-induced epithelial plasticity as a result of mcr84 treatment, we analyzed tumor tissue from KIC mice in each treatment group. Treatment with mcr84 alone increased the expression of N-cadherin, a common marker of mesenchymal cells and Slug, an EMT-inducing transcription factor (EMT-TF). Apricoxib alone or in combination with mcr84 significantly reduced N-cadherin expression and downregulated Slug expression to the same level of control group. Although the expression of Snail, another EMT-TF (41) was not affected by mcr84, treatment with apricoxib or apricoxib combined with mcr84 decreased Snail expression significantly (Fig. 2A). We also observed that collagen deposition was increased in KIC and Colo357 tumors after treatment with mcr84, a feature we identified previously that is associated with epithelial plasticity (9). This effect was attenuated by apricoxib alone or in combination with mcr84 (Fig. 2B).

Eicosanoids, including PGE₂, contribute to the immune microenvironment of solid tumors (20). For example, PGE₂ can induce a shift in cytokine expression in myeloid-derived suppressor cells (MDSC) and macrophages toward an immune-suppressive profile (e.g., IL4, IL10, and IL6) and PGE₂ can directly reduce T effector cell activity (20). Furthermore, EMT is also associated with an immunosuppressive tumor microenvironment (31, 32, 34). Thus, given our observations that COX-2 inhibition with apricoxib reduces PGE₂ production and decreases therapy-induced EMT, we investigated the immune landscape in KIC tumors from the different treatment groups shown in Fig. 1. Tumors harvested from mice that received mcr84 or apricoxib alone had an increase in the number of tumor-associated CD3⁺ and CD8⁺ T cells. Combination therapy further elevated CD3⁺ and CD8⁺ T-cell levels (Fig. 3A). Additionally, apricoxib alone and in combination with mcr84 significantly decreased FoxP3⁺ regulatory T cells (Treg; Fig. 3A). Motz and colleagues (30) previously reported that selective expression of the death mediator Fas ligand (FasL) on endothelial cells in human and mouse solid tumors was associated with scarce T-cell infiltration. They also identified that FasL was induced on endothelium by VEGF, IL10, and PGE₂. Thus, we evaluated FasL expression in the vasculature of KIC tumors by dual staining of the endothelium for CD31 and FasL. We found that FasL was indeed present on CD31⁺ endothelial cells in control-treated KIC tumors and that treatment with mcr84, apricoxib, or the combination significantly reduced endothelial FasL expression (Fig. 3B). To determine the effect of VEGF blockade and COX-2 inhibition on macrophages in the tumor microenvironment, we stained for CD11b, iNOS, and Arginase 1. We found that mcr84 alone and mcr84 in combination with apricoxib reduced CD11b⁺iNOS⁺ macrophages but apricoxib alone did not. In contrast, mcr84 or apricoxib alone decreased CD11b⁺Arg1⁺ macrophages, while the effect was more significant with combination therapy (Fig. 3C). Although the number of total myeloid cells that were marked by CD11b was elevated in the combination treatment group, the total macrophage number (F4/80) was reduced with anti-VEGF and COX-2 inhibition (Supplementary Fig. S4A and S4B).

Our data support that VEGF production by tumor cells is independent of COX-2, especially following COX-2 inhibition, and the data also strongly support that COX-2 activity on tumor cells is linked closely to the induction and/or maintenance of a less differentiated tumor cell phenotype. Epithelial plasticity is a common pathway exploited by tumors to resist therapeutic interventions, including chemotherapy and targeted therapy. Our data demonstrate that reducing hypoxia-induced epithelial plasticity by blocking COX-2 enhances the therapeutic activity of anti-VEGF in PDA. We have shown previously that anti-VEGF therapy (mcr84) of PDA induces hypoxia, which drives an increase in TGFβ and subsequent increase in collagen deposition. Furthermore, we found that collagen and TGFβ in the tumor microenvironment stimulated tumor cell EMT (9). Additionally, we previously reported that COX-2 inhibition (apricoxib) reduces EMT in models of GI cancer in vivo and TGFβ-induced EMT in vitro (28, 29). Therefore, we further investigated the effect of COX-2 inhibition on the level of active TGFβ in orthotopic Colo357 pancreatic tumors. We found that anti-VEGF (mcr84) increased active TGFβ levels, as anticipated but that this increase was blunted by COX-2 inhibition (data not shown), which suggests that COX-2 inhibition reduces EMT and immune suppression in part by reducing hypoxia-induced TGFβ expression (Supplementary Fig. S5). TGFβ, a multifunctional cytokine, can drive tumor cell EMT and is also a potent immunosuppressive factor produced by tumor cells, fibroblasts, and tumor-infiltrating lymphocytes (42). TGFβ can inhibit innate and adaptive immune responses in the tumor microenvironment. For example, TGFβ can polarize macrophages toward an immunosuppressive phenotype, support regulatory T-cell differentiation and directly inhibit effector T-cell activity (43). In addition, our results are consistent with reports that celecoxib, another selective COX-2 inhibitor, reduces hepatic expression of TGFβ, thereby attenuating EMT of hepatocytes (44). Furthermore, COX-2 has been shown to participate in TGFβ-driven EMT in human hepatocellular carcinoma (45). Thus, there are multiple examples of a connection between COX-2 activity and TGFβ-driven tumor progression.

We also found that COX-2 inhibition might reduce immune suppression in PDA. The immunosuppressive microenvironment is a major limitation for the efficacy of cancer immune therapy (46). Our data are consistent with other studies that have shown that antiangiogenic agents and COX-2 inhibitors have the potential to reduce the immunosuppressive tumor microenvironment and enhance immunotherapy (47–49). Our results support the findings of Motz and colleagues (30), who found that pharmacologic blockade of VEGF and COX-2 resulted in a significant increase in infiltrating CD8⁺ T cells and a reduction in FoxP3⁺ Tregs by downregulating FasL expression on tumor endothelial cells in multiple murine cancer models. Our data indicate that in KIC tumors, VEGF blockade or COX-2 inhibition alone could reduce FasL expression on the tumor endothelium, but combination therapy resulted in higher T effector cell recruitment and lower Treg infiltration than single-agent therapy.

In summary, our data support the rationale of a combination of anti-VEGF and COX-2 inhibition in PDA patients and also provide evidence that this combination might prime PDA or other tumors for increased efficacy with immune therapy.

R.A. Brekken is a consultant for Peregrine Pharmaceuticals and reports receiving a commercial research grant from Tragara Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Zhang, A. Kirane, H. Huang, F.J. Burrows, R.A. Brekken

Development of methodology: Y. Zhang, A. Kirane, H. Huang, N.B. Sorrelle

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Zhang, A. Kirane, H. Huang, N.B. Sorrelle, M.T. Dellinger

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Zhang, A. Kirane, H. Huang, N.B. Sorrelle, R.A. Brekken

Writing, review, and/or revision of the manuscript: Y. Zhang, A. Kirane, H. Huang, F.J. Burrows, R.A. Brekken

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Zhang, R.A. Brekken

Study supervision: Y. Zhang, R.A. Brekken

We acknowledge helpful discussions from members of the Brekken laboratory and editorial assistance by Dave Primm in the Department of Surgery at UT Southwestern.

The work was supported by NIH grants R01 CA192381 and U54 CA210181 (PI: M. Ferrari) to R.A. Brekken, T32 CA136515 (PI: J. Schiller) to A. Kirane, F31 CA19603301 to N.B. Sorrelle, a sponsored research agreement from Tragara Pharmaceuticals, Inc., to R.A. Brekken, and the Effie Marie Cain Scholarship in Angiogenesis Research to R.A. Brekken. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.