High COX‐2 expression in cancer‐associated fibiroblasts contributes to poor survival and promotes migration and invasiveness in nasopharyngeal carcinoma

Abstract Nasopharyngeal carcinoma (NPC) has the highest rate of metastasis among head and neck cancers, and distant metastasis is the major reason for treatment failure. We have previously shown that high cyclooxygenase‐2 (COX‐2) expression is associated with a poor prognosis of patients with NPC and inhibits chemotherapy‐induced senescence in NPC cells. In this study, we found that COX‐2 was upregulated in cancer‐associated fibroblasts (CAFs) derived from NPC by RNA‐Seq. Furthermore, elevated COX‐2 expression in CAF was detected in NPC patients with poor survival and distant metastasis by using immunohistochemistry. Then, we identified that COX‐2 is highly expressed in CAF at the distant metastasis site in seven paired NPC patients. High expression of COX‐2 and secretion of prostaglandin E2, a major product catalyzed by COX‐2 in fibroblasts, promotes migration and invasiveness of NPC cells in vitro. On the contrary, inhibition of COX‐2 has the opposite effect in vitro as well as in the COX‐2−/− mouse with the lung metastasis model in vivo. Mechanistically, we discovered that COX‐2 elevates tumor necrosis factor‐α expression in CAF to promote NPC cell migration and invasiveness. Overall, our results identified a novel target in CAF promoting NPC metastasis. Our findings suggested that high expression of COX‐2 in CAF may serve as a new prognostic indicator for NPC metastasis and provide the possibility of targeting CAF for treating advanced NPC.


| INTRODUCTION
Nasopharyngeal carcinoma (NPC) is a type of head and neck cancer that exhibits an endemic distribution with a high prevalence in Southern China and Southeast Asia. 1,2 The etiologic factors for NPC include Epstein-Barr virus (EBV) infection, ethnics, genetic susceptibility, and environmental factors, including consumption of food with volatile nitrosamines. [3][4][5] Upon diagnosis, most patients present with metastasis to the regional lymph nodes or even distant organs. The common sites of distant metastasis of NPC are the bone, lung, liver, and retroperitoneal lymph nodes. [6][7][8] Most distant metastasis occurs within 3 years after radiotherapy completion, with distant metastasis occurring in 52% of patients in the 1st year, 23% in the 2nd year, and 20% in the 3rd year. 8 To date, although NPC is sensitive to radiotherapy, distant metastasis is the primary cause of treatment failure. 9 Tumor metastasis is closely related to tumor microenvironment (TME). The TME has cellular components and noncellular extracellular matrix (ECM). [10][11][12] There are evidence suggesting that EBV-infected NPC cells interacted with TME components to facilitate metastasis. An increased presence of Foxp3 + Treg cells and CD68 + tumor-associated macrophages (TAMs) has been found in EBV-positive NPC specimens and associated with poor prognosis. 13 Another recent study has revealed an interacting loop between NPC cells and TAMs in driving NPC metastasis. In plethora of tumor microenvironmental cell types, cancer-associated fibroblasts (CAFs) have involved as one of the most promising targets owing to their abundant presence and functional significance in various tumor entities, including multiple myeloma, oral cancer, and gastric cancer. 14-16 CAF can be phenotypically identified based on markers such as fibroblast activation protein α (FAP), α-smooth muscle actin (SMA), and FSP-1. 12,[17][18][19] Several studies have shown that CAF can be used as an important prognostic factor in a variety of tumors. [20][21][22] Chen 23 reported that overexpression of α-SMA-positive fibroblasts (CAFs) in NPC predicts poor prognosis. As a major and important component of tumor matrices, CAF plays an important role in tumor invasiveness and metastasis. CAF can also promote dissemination and metastasis through engaging in heterotypic interactions with tumor cells in ovarian cancer. 24 Considering CAF is a major component in TME of NPC, however, its clinical significance in the invasiveness and metastasis of NPC has rarely been reported.
Cyclooxygenase (COX) is a rate-limiting enzyme in prostaglandin biosynthesis. There are two isoforms of COX-COX-1 and COX-2.
COX-1 is constitutively expressed in a number of tissues and mainly plays a role in tissue homeostasis. By contrast, COX-2 is an inducible enzyme responsible for the production of prostaglandins at sites of inflammation and wound-healing. 25 Of note, COX-2 is highly expressed in numerous types of human cancer, such as breast, ovarian, colorectal cancer, and NPC. [26][27][28][29] Our previous studies reported that COX-2 serves as a marker of poor prognosis in NPC, and COX-2 expression induces proliferation and chemoresistance of NPC cells. 30 However, stromal expression of COX-2 has not been specifically evaluated in NPC to date.
In this study, we first found that COX-2 is highly expressed in CAF from patients with NPC by RNA-seq analysis. Subsequently, we observed clinical significance of the expression of COX-2 in CAF correlated with lymph-node (N) stage, metastasis (M) stage, relapse, and survival in patients with NPC. Further functional study of COX-2 in CAF will be explored on NPC metastasis in vitro and in vivo with CAF derived from primary NPC patients, NPC cell lines, and COX-2 knockout Human normal fibroblast (NF) and CAF were derived from opposite normal nasopharynx and nasopharyngeal carcinoma tissues from patients with newly diagnosed NPC. Fibroblasts isolation procedure was as described previously. 31 Briefly, the fresh tissues were cut into pieces and isolated using Type I collagenase digestion for 30 minutes at 37°C and were thereafter cultured in low-glucose DMEM medium for about 1 week until formation of fibroblasts.
Specimens were obtained with written informed consent from patients with NPC enrolled in the Xiangya Hospital of Central South University (CSU). The NF and CAF used for further functional studies were less than third passages.
The following reagents were used in this study: NS398 (a selective COX-2 inhibitor) and prostaglandin E2 (PGE2) were purchased from Cayman Chemicals (Cayman, MI). The recombinant human protein tumor necrosis factor-α (TNF-α) and TNF-α neutralizing antibody were purchased from Sino Biological (SB Inc, Beijing, China).
2.2 | Conditioned medium derived from human normal fibroblast and cancer-associated fibroblasts NF and CAF (2 × 10 5 /mL), mouse fibroblast (2 × 10 5 /mL), and WI38 cells (2 × 10 5 /mL) were plated into six-well culture plates in low-glucose DMEM supplemented with 10% FBS and cultured overnight, and subsequently refreshed with 0.5 mL of serum-free high-glucose DMEM, and then 2 mL of serum-free high-glucose DMEM was added to six-well culture plates. The culture supernatants were harvested after 48 hours.
Then, cell debris was removed with centrifugation and stored at −80°C until experimentation. To obtain conditioned medium (CM) from fibroblast cells with NS398 or PGE2, CAF derived from patients with NPC, WI38-COX-2Ctr cells, and fibroblasts from COX-2 +/+ mice were cultured in serum-free high-glucose DMEM with 20 μM NS398, or NF derived from patients with NPC, WI38-COX-2sh cells and fibroblasts from COX-2 −/− mice were cultured in serum-free high-glucose DMEM with 10 μM PGE2. The CM from these cultures was assayed for woundhealing assay and transwell of NPC cells as described below.

| Enzyme-linked immunosorbent assay
To measure PGE2 release, serum samples were obtained from peripheral blood, CM was collected 48 hours after fibroblasts or WI38 cells were cultured in serum-free high-glucose DMEM. PGE2 levels in the supernatants were measured by using PGE2 enzyme-   Amplification of specific targets was performed to determine the messenger RNA (mRNA) levels using Bio-Rad iCycler iQ Real-Time PCR Detection System (Bio-Rad, CA). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as a loading control. The specific primers used for amplification are summarized in Table 2. The relative mRNA levels were calculated as the value of Δ 2 C t normalized to the control.

| Western blot analysis
Western blot was performed as previously described. 34

| Immunofluorescence
Cells were seeded onto glass coverslips with an appropriate confluent overnight, and were washed with phosphate-buffered saline (PBS) and fixed with 4% paraformaldehyde, permeabilized with 0.5% Triton X-100, and blocked with 4% bovine serum albumin. Subsequently, the samples were incubated with primary antibodies for α-SMA

| Statistical analysis
Analyses were performed using GraphPad Prism 7 software (Graph-Pad Prism Inc, CA). The two-tailed t-test was utilized for the comparison of two conditions. The statistical tests were analyzed as paired where appropriate. P < .05 was considered statistically significant and marked as P < .05*, P < .01**, and P < .001***. Graphical results are presented as mean ± standard error of the mean (SEM).

| COX-2 was upregulated in CAF and correlates with metastasis in nasopharyngeal carcinoma
Paired NF and CAF (n = 2) derived from patients with NPC were stained with α-SMA, an established marker of fibroblasts ( Figure S1A). RNAsequencing was applied to examine the critical gene expression between primary NF and CAF. We discovered that the major type of gene signature is the inflammatory response pathway ( Figure 1A),

| 269
among which PTGS2 is ranked at second ( Figure 1B). Interestingly, PTGS2, also known as COX-2, a key molecule in inflammatory response, was the most upregulated gene in CAF compared with NF.
We next verified whether COX-2 was upregulated in CAF by quantitative reverse-transcriptase polymerase chain reaction (qRT-PCR). The mRNA levels of COX-2 were indeed elevated in CAF compared with NF in three paired NPC patients ( Figure 1C).  (Table 1). In addition, we also showed that a high expression of COX-2 in CAF was positively correlated with relapse (P = .02) and poor survival (P = .034) in patients with NPC, suggesting that the high expression of COX-2 in Abbreviations: COX-2, cyclooxygenase-2; GAPDH, glyceraldehyde 3-phosphate dehydrogenase; IL-6, interleukin 6; PCR, polymerase chain reaction; SMA, smooth muscle actin; TNF-α, tumor necrosis factor-α.

| 271
CAF confers poor outcome in NPC (Table 1). To further confirm the relevance of COX-2 and metastasis in NPC, seven paired NPC sequential samples with primary site and distant metastasis site that include two lung metastasis (P5 and P6) and five cervical lymph-node metastasis (P1, P2, P3, P4, and P7) were explored to examine the COX-2 expression in CAF. Interestingly, we found that COX-2 was significantly upregulated in CAF derived from distant metastasis sites compared with primary sites ( Figure 1E). Taken together, these results indicate that altered COX-2 levels in CAF may promote metastasis in NPC.

| Increased COX-2 and PGE2 secretion from CAF promotes migration of NPC cell lines
Schematic diagram for the treatment of NPC cells with CM collected from NF and CAF (Figure 2A). To detect the function of COX-2 in CAF, first, we applied IF and found that COX-2 was upregulated in CAF ( Figure 2B). Then, considering that PGE2 is the major product catalyzed by COX-2, we tested PGE2 level in CM from three paired NF and CAF by ELISA. Consistent with COX-2 expression, CAF secreted more PGE2 than NF ( Figure 2C). Importantly, we confirmed the presence of elevated PGE2 levels in serum from 18 primary NPC patients, especially in the 14 NPC with metastatic group compared with 14 healthy donors by using ELISA ( Figure 2D).
Considering the relevance of PGE2 in NPC patients with metastasis, cell migration and invasiveness assays were applied to examine the migration and invasiveness capacities of COX-2 in NPC cells in vitro.
First, CNE1 was treated by CM from NF and CAF, we found migration index of CNE1 was higher in the CAF group ( Figure 2E,F) by woundhealing assay. Meanwhile, we also revealed CM from CAF contributed to invasiveness of CNE1 and CNE2 ( Figure 2G) by invasiveness assay.
Consistently, exogenous PGE2 was used to mimic COX-2 overexpression in NF. The invasiveness of CNE1 and CNE2 was enhanced when incubated with CM from NF with PGE2 treatment compared with CM from NF alone ( Figure 2H). In contrast, opposite trends were found in NPC cells cultured with CM from CAF with NS398, a selective COX-2 inhibitor, treatment, compared with CM from CAF alone ( Figure 2H). To characterize the epithelial to mesenchymal transition features of NPC cells during migration and invasiveness, we also found higher expression of Vimentin and lower expression of E-cadherin in CNE1 treatment with CM from CAF by Immunofluorescence (IF) ( Figure S2). These data suggest that COX-2/PGE2 in CAF results in enhanced migration and invasiveness of NPC cells in vitro.

| COX-2 positively correlated with TNF-α expression in CAF
Gene set enrichment analysis (GSEA) showed the inflammatory response was significantly enriched in CAF ( Figure 1A), leading us to speculate that certain inflammatory cytokines secreted by CAF may be responsible for migration and invasiveness of NPC cells. We first examined the expression of 26 inflammatory cytokines, COX-2-related genes, and six CAF markers by using qRT-PCR assay (Primer sequence; Table 2). Most inflammatory cytokines were upregulated in CAF compared with NF, including CXCL12, TNF-α, and interleukin 6 (IL-6; Figure 5A). Next, we investigated whether COX-2 regulates the expression of those cytokines, we detected the expression of those cytokines in WI38-COX-2sh cells also by qRT-PCR. Among these cytokines, CXCL12, TNF-α, and IL-6 were significantly reduced in the WI38-COX-2sh cells ( Figure 5B).
Considering that TNF-α is a signaling cytokine of NF-κB pathway and COX-2 acts as a target gene of NF-κB signaling. We next investigated whether TNF-α was indeed regulated by COX-2. Enhanced TNF-α expression was found in CAF compared with NF. Conversely, TNF-α was downregulated either in WI38-COX-2sh cells or COX-2 −/− -SF and COX-2 −/− -LF ( Figure 5C). Then, we assessed TNF-α in seven paired NPC patients by IHC. Consistent with increased expression of COX-2, the expression of TNF-α was elevated in fibroblasts from metastatic NPC compared with primary NPC ( Figure 5D). Interestingly, COX-2 expression was found to significantly positively correlate with TNF-α expression in seven paired NPC patients ( Figure 5E). These results indicate that COX-2 might be positively correlated with the expression of TNF-α in NPC.

| CAF promotes NPC cell migration and invasiveness through COX-2-PGE2-TNF-α axis
The previous study demonstrated that TNF-α serves as a prognosis factor for NPC cells. However, whether and how COX-2 induced TNF-α expression in CAF to promote NPC metastasis is still unclear.
To investigate whether COX-2 promotes cell migration and invasiveness through TNF-α, we examined the migration and invasiveness of CNE1 with treatment of TNF-α recombinant protein or TNF-α neutralizing antibody. As expected, TNF-α rescued the inhibitory

| Host COX-2 modulates lung metastasis of LLC cells correlated the expression of TNF-α in vivo
A detailed delineation of the group distribution for in vivo experiment ( Figure 7A). To explore the COX-2 function in vivo, LLC lung metastasis assay was applied in COX-2 −/− mice. Briefly, COX-2 +/+ and COX-2 −/− mice were injected with 10 6 LLC cells intravenously. We found that COX-2 +/+ mice dramatically enhanced LLC metastasis to lung with increased metastatic nodules compared with COX-2 −/− mice ( Figure 7B). Then, we detected the COX-2 and TNF-α expression in the lung of fibroblasts, and we found that COX-2 and TNF-α are significantly high expressed in COX-2 +/+ mice ( Figure 7C). Interestingly, COX-2 expression was found significantly positively correlate with TNF-α expression ( Figure 7D). Taken together, these results showed that high COX-2 in host fibroblasts affects lung metastasis of LLC cells.

| DISCUSSION
Metastasis is the major cause of treatment failure in NPC and thus, preventing, predicting, and inhibiting metastasis is critical to improve treatment outcomes. In the current study, we reported the high expression of COX-2 in CAF promotes NPC cells metastasis.
Moreover, we observed that high COX-2 expression in CAF was positively correlated with N stage, M stage, relapse, and survival in patients with NPC. Previous studies reported that CAF markers include, but are not limited to, α-SMA, FSP-1, FAP, podoplanin (PDPN), PDGFR-α/β, and NG2. 12,18 However, most of these markers are also expressed in other cell compartments and hence lack specificity for CAF/FSCs.
For example, PDPN is also expressed in lymphatic endothelial cell. 36,37 NG2 and PDGFR-β are commonly used to identify pericytes. 38 In the stroma of pancreatic cancer, distinct populations of CAF differentially contribute to desmoplasia and inflammation and are molecularly distinguishable through α-SMA expression and IL-6 secretion. 39 In this study, we found CAF derived from NPC represents the high expression of α-SMA, FAP, PDGFR-α/β, and  COX-2 expression in CAF was positively correlated with NPC metastasis. Interestingly, we obtained seven paired patients that primary and metastatic NPC tissues from the same patients, and we confirmed that low expression of COX-2 in primary NPC tissues of fibroblast, but high expression of COX-2 in metastatic site of CAF. Then, by applying migration and invasiveness assay, high expression of COX-2 in CAF and PGE2 was produced and released from CAF facilitate metastasis in NPC in vitro. In this study, we demonstrated that COX-2 induces PGE2 secretion in CAF and subsequently increases metastasis of NPC cells.
Our studies and other groups demonstrated that high expression of COX-2 contributes to tumor cell proliferation, metastasis, and drug resistance through regulating several oncogenes or cellcycle-related molecules such as p53, β-catenin, Snail1, etc, in NPC and other cancers. 30,41,42 However, in our study, we found TNF-α, another new molecular positively correlated with COX-2 by RNA-Seq. Bourouba 43 showed that TNF-α promotes tumor growth via a NOS2-dependent mechanism in NPC. 43 This is the first time to demonstrate TNF-α was upregulated in CAF in NPC. We detected TNF-α expression in the paired NPC patients, and we found that high expression of TNF-α and COX-2 in metastatic site of CAF in NPC.
Then, we found TNF-α was decreased in COX-2 −/− LF and WI38-COX-2sh and also has impaired invasiveness abilities in vitro. Finally, we employed a LLC lung metastasis assay in COX-2 −/− mouse models, and we confirmed that the COX-2 in host fibroblasts affect lung metastasis of LLC cells and correlated with the expression of TNF-α in vivo. These results suggested that high expression of COX-2 in fibroblasts promotes NPC metastasis through COX-2-PGE2-TNF-α axis. NS398 and anti-TNF-α significantly decreased the invasiveness abilities in vitro, and also suggested the potential therapeutic effect on CAF in NPC.
In summary, our study is the first to elucidate the critical role of COX-2 in CAF in promoting NPC metastasis and predicting poor prognosis. Our results suggested that high expression of COX-2 in CAF may serve as a new prognostic indicator for predicting NPC metastasis and provide the possibility of targeting CAF for treating advanced NPC.