Gastrin-Induced Cyclooxygenase-2 Expression in Barrett’s Carcinogenesis

The association between cyclooxygenase (COX)-2 and human cancer is becoming well established. The best site-specific evidence comes from colon cancer, in which population-based studies (1, 2, 3) and animal models (4, 5, 6) strongly suggest a causal relationship. COXs are membrane-associated proteins that catalyze the rate-limiting step in the prostaglandin (PG) production pathway (7, 8). COX-2 and derived PGs may influence carcinogenesis in various ways. They can promote angiogenesis, inhibit immune surveillance, increase cell proliferation, reduce apoptosis and cell adhesion, and bind to the nuclear peroxisome proliferation activator receptors that act directly as transcription factors on ligand binding (9, 10, 11). It has also been observed that COX-2 is coexpressed with gastrin in cancers of the stomach, intestine, and lung (12, 13, 14). Gastrin is an intestinal peptide that stimulates gastric parietal cells to secrete acid and is a potent growth factor for epithelial cells (15). Gastrin has been shown to induce proliferation and COX-2 expression in gastric and intestinal cells through the activation of the cholecystokinin 2 (CCK₂) receptor (16, 17, 18, 19). In addition, its trophic effects in colon cancer can be blocked using COX-2 inhibitors (20). Many cancer cells are capable of expressing their own gastrin, which can act in an autocrine or paracrine manner without contributing to systemic hypergastrinemia (15).

With regard to the upper gastrointestinal tract, the incidence of esophageal adenocarcinoma (AC) has increased in Western countries for the past three decades (21, 22, 23, 24). The known predisposing factors for this malignancy are gastroesophageal reflux disease and Barrett’s esophageal epithelium (BE; Refs. 25 and 26). BE increases the risk of esophageal AC at least 30-fold compared with the general population, with an estimated incidence of 1 in 180 patient-years (27, 28, 29). Hence, the overall cancer risk in patients with BE is small, although when it occurs, esophageal AC carries a dismal prognosis (30). As a result, screening and surveillance programs have been introduced to detect BE and associated dysplasia before the development of invasive, incurable cancer (31). In view of the unproven efficacy of these programs, there has been an increasing interest in understanding the biological mechanisms responsible for cancer development and progression, so that chemoprevention strategies can be used.

COX-2 levels have been found to be increased in BE compared with the squamous epithelium above the Barrett’s segment (32, 33, 34, 35). However, reports are conflicting regarding whether COX-2 expression increases with progression to dysplasia and cancer (32, 33, 34, 35, 36). There are several possible explanations for these conflicting reports. First, there may be interpatient variability in expression levels (32). This could be addressed by examining COX-2 expression levels in the same patients during their progression from metaplasia to dysplasia and cancer. Second, variations in tumor differentiation status may be important because in gastric cancer studies it has been demonstrated that well-differentiated, intestinal-type tumors express more COX-2 compared with poorly differentiated tumors (37). There are no data comparing COX-2 expression with the degree of differentiation in Barrett’s-associated AC. Third, COX-2 expression may be affected by local or systemic gastrin levels because most of these patients use acid suppressants to control their symptoms (31, 38, 39), which may lead to a moderate increase in gastrin levels (40). The effects of exogenous gastrin on COX-2 expression in Barrett’s esophagus have not been examined.

Hence, the first aim of the study was to determine the degree of COX-2 expression in BE and other upper gastrointestinal tissues and whether COX-2 expression changes during progression to cancer in a longitudinal case-control study. The second aim of the study was to investigate whether the degree of differentiation affects COX-2 expression in BE-associated cancer. The final aim of our study was to examine the functional relationship between COX-2 and gastrin in BE and in esophageal cancer cell lines.

Eighty-four patients with an endoscopic and histological diagnosis of BE (columnar epithelium extending ≥3 cm from gastroesophageal junction containing intestinal metaplasia) and 45 patients with esophageal AC were recruited. Normal esophageal squamous epithelium and duodenal biopsies were obtained from 40 control patients who were endoscoped for a range of clinical indications and in whom there was no endoscopic or biopsy evidence of Barrett’s esophagus or inflammation. The dose and type of any acid-suppressant medication were recorded for all patients. In each case, biopsies were taken for routine histopathological diagnosis, and additional biopsies were taken for research. In the case of Barrett’s esophagus, quadrantic biopsies were taken every 2 cm from the gastroesophageal junction according to international surveillance guidelines (31). Biopsies were either fixed in formaldehyde and embedded in paraffin for histology and immunohistochemistry, snap frozen in liquid nitrogen and stored at −80°C until used for PCR and Western blotting, or placed in media and used immediately for organ culture (41). The number of patients used for each experiment varied and is stated in each figure legend.

For the retrospective case-control study, serial biopsies from nine patients whose esophageal AC or high-grade dysplasia (HGD) was detected through a surveillance program (followed up for a median of 6 years; range, 3–13 years) were compared with control patients with BE who did not progress. Two controls for each case were matched for age and length of follow-up. The cancer patients had at least two endoscopies before the development of HGD, and they all progressed through the metaplasia−low-grade dysplasia−HGD sequence. The control cases had no more than a diagnosis of focal low-grade dysplasia in one of their surveillance endoscopies.

The study was approved by the research ethics committees of Addenbrookes Hospitals, Barts & The London, and University College Hospitals National Health Service Trusts.

All dysplasia and cancer diagnoses were verified by a second pathologist, and the grade of dysplasia was determined using the criteria established by Axon et al. (42). Sections from sequential levels in the AC blocks were used for histopathology and immunohistochemistry. As well as the routine clinical histopathological diagnosis, all of the slides were also reviewed by a single consultant pathologist (M. R. N.) to determine the degree of tumor differentiation (43). For two tumors, there was disagreement between the histopathologists, and hence villin staining was used as a further differentiation marker [see “Immunohistochemistry” (44)]. Some of the tumors contained areas of both moderate and poor differentiation, and the degree of differentiation was determined according to the predominant type. All 30 tumors were found to be either moderately or poorly differentiated; therefore, no well-differentiated tumors were available for inclusion in the study.

Protein was extracted from snap-frozen biopsies and cells using lysis buffer containing protease inhibitors (Rouche). Protein content was measured using the BCA protein assay kit (Sigma; Ref. 45), and 25 μg of total protein were separated using 10% SDS-PAGE and blotted onto Hybond-P membranes (Amersham). The membranes were probed using monoclonal antibodies for COX-1 and COX-2 (1:1,000; Cayman Chemicals) and β-actin (1:5,000; Sigma). Visualization was achieved using biotinylated antimouse IgG, horseradish streptavidin (Vector Laboratories, Peterborough, United Kingdom), and chemiluminescence (enhanced chemiluminescence; Amersham). Band intensity was determined using Kodak Electrophoresis Documentation and Analysis System 120 software (Eastman Kodak Co., Rochester, NY). The arbitrary densitometry units were converted into ng/μg total protein values using serial dilutions of the recombinant COX-1 and COX-2 proteins, as described previously (34). A full-length representative gel is shown in Fig. 5 A.

Ninety slides from 27 BE patients for the case-control study were immunostained for COX-2. In addition, slides from 30 esophageal ACs were immunostained for COX-2 and the differentiation marker villin. Tissue sections were processed for immunohistochemistry as described previously (46). The primary antibodies were COX-2 (1:150; Cayman Chemical) and villin (1:1000; Chemicon, Harrow, United Kingdom). Visualization was performed using the 3,3′-diaminobenzidine method (Vector Laboratories). The specificity of positive antibody staining was verified by an absorption test using blocking peptide. For negative controls, PBS was substituted for the primary antibody to exclude nonspecific binding from the secondary antibody. COX-2 expression was scored independently by two investigators (S. I. A. and P. L-S.) using a modification of a semiquantitative scale described previously (34). The overall distribution and strength of COX-2 staining were examined under low power magnification (×10). Further cell count was performed in at least three high-power fields in each available section to determine the percentage of the positive cells, especially in the lamina propria. Staining intensity (1 = weak staining, 2 = moderate staining, 3 = strong staining, and 4 = very strong staining) and distribution (1, ≤25% of cells stained; 2, 25–50% of cells stained; 3, 50–75% of cells stained; and 4, >75% of cells stained) were added together to generate a mean score.

Reverse transcription-PCRwas performed as described previously (47). Total RNA was isolated from tissue biopsies and cell lines using TRIzol (Life Technologies, Inc., Paisley, United Kingdom), and 2 μg of total RNA were transcribed into cDNA. Two μl of cDNA were amplified in 50 μl of reaction volume [10 mm Tris (pH 8.3), 50 mm KCl, 1.5 mm MgCl₂, 200 μm deoxynucleotide triphosphates, 20 pmol each of 5′ and 3′ primer, and 0.5 unit of Taq polymerase]. Primers for minichromosome maintenance protein 2 (Mcm2) were 5′-AGTTGCGTATTCAGGCTGCT-3′ (forward primer) and 5′-AAACAGACAGGGAGCAATGG-3′ (reverse primer). Primers for glyceraldehyde-3-phosphate dehydrogenase, gastrin, and CCK₂ receptor have been published previously (48). The amplification consisted of initial denaturation at 95°C for 5 min followed by 40 cycles of 45-s denaturation at 95°C, 90-s annealing at 60°C (64°C for Mcm2), and 90-s extension at 72°C. This was followed by a 3-min final extension at 72°C. PCR products were analyzed in a 1.5% agarose gel containing ethidium bromide. All reagents for reverse transcription-PCR were purchased from Life Technologies, Inc. The bands were visualized using Kodak Electrophoresis Documentation and Analysis System 120 software (Eastman Kodak Co.).

For organ culture, endoscopic biopsies were weighed before incubation in medium 199 supplemented with 10% FCS, 1 μg/ml insulin, 500 units/ml streptomycin, and 250 units/ml penicillin as described previously (47). In addition, a human esophageal squamous cell line (OE21; European Collection of Cell Cultures, Wiltshire, United Kingdom) and two BE AC cell lines [SEG-1, a gift from D. Beer (University of Michigan; Ref. 49) and OE33 (European Collection of Cell Cultures)] were used. Cells were seeded into 24-well plates at a concentration of 25 × 10³ cells/well and maintained in original medium (RPMI 1640 for OE21 and OE33; DMEM for SEG-1) supplemented with 10% FCS, 100 units/ml penicillin, 100 μg/ml streptomycin, and 1 mm glutamine.

To test the effect of gastrin on COX-2 and cell proliferation in organ culture and cell lines, cultures were assigned to one of the following groups for 24 h: controls (media alone); gastrin (10 nm G-17; Sigma Aldrich); gastrin and CCK₂ inhibitor (100 nm YM022; Tocris); or gastrin and COX-2 inhibitor [1 μm NS-398 (50)]. These concentrations were derived from preliminary experiments using a range of concentrations and were consistent with previous reports (19, 51). Cell proliferation was determined in biopsies and cell lines by using Mcm2 (a marker expressed only in proliferating cells) and a 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide-based cell growth determination kit (Sigma) according to the manufacturer’s recommendations (52). All experiments were performed in triplicate.

A specific enzyme-linked immune assay kit (Amersham) was used to quantify the PGE₂ levels in supernatant from organ and tissue culture experiments. PGE₂ was extracted and purified using C18 Ampi-columns (Amersham) and evaporated to dryness under a steam of nitrogen. Enzyme-linked immune assay was performed in triplicate according to the manufacturer’s instructions, and PGE₂ concentration was expressed as pg/ml/mg tissue.

The data are presented as a mean ± SE. The Jonckheere-Terpstra test was used to identify trends in the longitudinal cohort study. ANOVA and χ² test were used to detect variability, and the median test and Student’s t test were used to identify specific differences between groups. P < 0.05 was required for significance.

The study included a total of 169 patients from three groups (Table 1). The demographics of the BE patients are in keeping with previous studies (53). The mean length of the BE segment was 5.5 cm (range, 3–12 cm). Sixteen of 84 (19%) of the BE patients had low-grade dysplasia, and 5 of 84 (6%) of the BE patients had HGD. Fifty-five percent of the esophageal AC patients had endoscopic evidence of BE. All squamous and duodenal control samples were histologically normal.

COX-1 and COX-2 proteins were constitutively expressed in all of the diagnostic groups evaluated, including the control tissues (Fig. 1, A−C). However, there were marked differences in COX-2 expression levels between the tissue groups. COX-2 expression was invariably increased in all BE and duodenal samples compared with normal squamous esophageal epithelium (NE; Fig. 1, A and C). COX-2 expression in dysplastic BE samples was variable (Fig. 1,B), and no overall difference was found between expression levels in patients with and without dysplasia when analyzed separately (data not shown). Once AC had developed, expression levels were extremely variable, with an overall reduction in expression compared with BE (Fig. 1, B and C). Overall, COX-2 is expressed in BE (both dysplastic and nondysplastic) > duodenum > AC > NE (P < 0.0004, BE and duodenum versus NE; Fig. 1,C), with a corresponding increase in PGE₂ in all glandular tissue types (duodenum, BE, and esophageal AC) compared with NE (P < 0.05, BE versus NE; Fig. 1 D).

In view of the marked variability of COX-2 expression in AC samples, the relationship between protein expression and tumor grade was examined. Using standard histopathology criteria, 12 of 30 tumors were classed as moderately differentiated, and 18 of 30 were poorly differentiated. The brush border protein villin was increased in tumors graded as moderately differentiated using the standard histopathological criteria and very low or absent in poorly differentiated tumors (P < 0.05; Fig. 2, A and B), with 67% sensitivity (95% confidence interval, 35–90%) and 88% specificity (95% confidence interval, 62–98%). Using a combined score for the degree of differentiation, there was a trend toward increased COX-2 expression in moderately differentiated tumors compared with poorly differentiated tumors, although this did not reach statistical significance (P = 0.09; Fig. 2, C and D). This may be explained by the heterogeneous nature of the degree of cellular differentiation as well as the immunohistochemical COX-2 staining.

The longitudinal study was performed to eliminate variation between patients and to determine whether there was any correlation between COX-2 expression levels and the likelihood for progression to AC. Considering all of the time points together, there was no overall difference in COX-2 expression levels between the nine patients who progressed to HGD and esophageal AC and the control group (P = 0.35). However, there was more variability within the control group, despite the larger sample size (Fig. 3,A). Interestingly, we observed that all patients expressed more COX-2 over time during the follow-up period, regardless of whether or not they progressed to cancer (P = 0.017 and 0.002 for the case and the control groups, respectively; Fig. 3 B). Because the majority (85%) of patients undergoing surveillance were on proton pump inhibitors (PPIs), we speculated that this temporal trend might be related to gastrin-induced COX-2 levels.

We first determined whether esophageal tissues expressed the CCK₂ receptor and whether they are capable of producing their own gastrin hormone (51). Duodenal biopsies were used as a positive control because they contain gastrin-producing enterochromaffin cells (54). All NE, BE, and esophageal AC biopsies expressed variable amounts of endogenous gastrin mRNA, with a 2-fold increase in BE samples compared with NE, dysplastic BE, and esophageal AC (Fig. 4, A and B; P = nonsignificant). The CCK₂ receptor was expressed in the majority of samples studied (6 of 8 NE samples, 8 of 10 BE samples, 6 of 10 dysplastic BE samples, 3 of 4 esophageal AC samples, and 4 of 4 duodenal samples), in keeping with previous studies (Ref. 51; Fig. 4,A). A 24-h stimulation with 10 nm gastrin (G-17) in an organ culture induced COX-2 expression in all tissue types, and this effect was reduced by the CCK₂ inhibitor YM022 (Fig. 5,A). Although the gastrin-induced increase in COX-2 protein levels did not reach statistical significance when the data were pooled (data not shown), PGE₂ levels were significantly increased in NE (P = 0.008), BE (P < 0.04), and esophageal AC biopsies (P < 0.04). This stimulatory effect was eliminated by addition of the CCK₂ inhibitor YM022 (P < 0.05 for NE, BE, and esophageal AC; Fig. 5 B).

To extend the functional studies, we looked at the effect of exogenous gastrin on cell proliferation. Biopsies from BE and control tissues (n = 3) were stimulated with gastrin in organ culture in the presence or absence of CCK₂ and COX-2 blockers. Gastrin induced the proliferation marker Mcm2 in some BE samples, but not in the control tissues (NE and duodenum). The gastrin-induced proliferation in BE was abrogated by COX-2 inhibitor NS-398 as well as by the specific CCK₂ inhibitor YM022 (Fig. 6,A). Due to the heterogeneity of proliferation within BE, a panel of cell lines was also used to complement the ex vivo results. Gastrin mRNA was expressed in all cell lines, with very low levels in SEG-1 cells (Fig. 6,B). Both Barrett’s cell lines (SEG-1 and OE33) expressed the CCK₂ receptor, whereas the squamous OE21 carcinoma cell line did not. Gastrin induced COX-2 and PGE₂ expression levels in SEG-1 and OE33 Barrett’s cell lines, but not in OE21 as expected (data not shown). Ten nm gastrin (G-17) induced proliferation in SEG-1 and OE33 Barrett’s cell lines, which was most significant in the OE33 cells (220% compared with the control; P < 0.04). This gastrin-induced proliferation was reversed by the COX-2 inhibitor NS-398 as well as by the specific CCK₂ inhibitor YM022 (Fig. 6 C; P < 0.04).

We have demonstrated that expression of the COX-2 enzyme has phenotypic specificity with increased expression levels in both benign and malignant glandular epithelia (BE > duodenum > AC) compared with squamous esophagus samples. The degree of COX-2 expression is highly variable within dysplastic and AC samples, with no consistent changes in expression levels within patients as they progressed through the metaplasia−dysplasia−cancer sequence compared with control patients who did not progress. The degree of tumor differentiation may contribute to the variable COX-2 status, although this does not seem to be a major factor. Because all patients undergoing surveillance were found to have increasing levels of COX-2 expression in BE over time, we hypothesized that this may be related to PPI-induced gastrin production. Hence, we have demonstrated for the first time that gastrin stimulation can increase COX-2, PGE₂ expression, and proliferation in BE and associated cell line tissues via the CCK₂ receptor. Furthermore, the mitogenic effect of gastrin could be overcome by inhibition of COX-2 or the CCK₂ receptor. Therefore, it is likely that the gastrin-induced proliferation occurs as a result of COX-2 induction. Hence, these data add support to the concept of COX-2 inhibitors as part of a chemoprevention strategy that involves PPIs in Barrett’s esophagus.

For the first time, we have systematically compared the COX-2 expression levels in a large number of samples with both squamous and duodenal samples from unrelated control patients. We have confirmed previous reports that COX-2 expression levels are greater in BE than in NE (34). However, we have also demonstrated that COX-2 expression is increased in duodenal samples compared with the squamous epithelium. This is in keeping with previous reports suggesting that glandular epithelia have a propensity to express increased amounts of COX-2 (55, 56). It is also possible that the increased COX-2 levels in the duodenum are secondary to the high levels of gastrin found in the proximal small intestine.

The degree of variability of COX-2 expression within cancer samples was unexpected. Some previous reports have described increased COX-2 expression in HGD and esophageal AC (34) that correlated with a reduction in patient survival (57). In contrast, other studies have demonstrated reduced COX-2 expression with progression to dysplasia (35) and cancer (58). For other tumor types, such as gastric AC, there is evidence that differentiation status may affect COX-2 expression (37). Our data suggest a trend toward an increased COX-2 expression in more differentiated tumors. Unfortunately well-differentiated tumors are uncommon and were not found in this cohort, and the classification of tumor differentiation histologically is a highly subjective procedure. Hence, a larger data set may be required to achieve a statistically significant correlation, and it is likely that other clinical and histological variables may also play a role.

Plausible explanations for the increases in COX-2 expression seen in both the cases and control groups of the longitudinal study include a worsening of gastroesophageal reflux disease over time, leading to more damage in the esophageal mucosa. However, this is very unlikely because most patients do not have worsening inflammatory changes noted during BE endoscopic surveillance. Another possible explanation is the effect of medications used to relieve reflux symptoms. A high percentage of BE patients use PPIs that cause a moderate increase in gastrin levels. This hormone has previously been shown to induce COX-2 expression in epithelial cells (17, 18), although this has not been studied specifically in the context of the esophagus. In this study, we have demonstrated that in an ex vivo culture system, gastrin is able to induce a modest increase in COX-2 in all of the epithelial cell types studied, and PGE₂ was significantly induced by gastrin in both squamous and glandular esophageal tissues. The significant increase in PGE₂ is not surprising because the enzyme-linked immune assay method used to quantify PGE₂ is much more sensitive than the Western blotting used for COX-2. The increased COX-2 expression and PGE₂ activity were reversed by the specific CCK₂ inhibitor YM022, which suggests that gastrin mediated its effects through the activation of the CCK₂ receptor.

As well as the ability of gastrin to induce COX-2, a recent study has shown that gastrin can stimulate proliferation in BE biopsies ex vivo, although there was no increase in serum gastrin levels in BE patients in that study (51). PPIs are associated with a 2–4-fold increase in gastrin levels over the basal levels in one third of the patients (40). Therefore, it has been assumed that elevated serum fasting gastrin levels in patients on PPIs are an indicator of adequate acid suppression (59, 60). In addition to serum hypergastrinemia, previous studies have demonstrated that some tumors express their own gastrin (15). Here we have shown that there are increased levels of gastrin within nondysplastic BE samples compared with other tissues studied that may signal in an autocrine or paracrine manner without contributing significantly to hypergastrinemia. Interestingly, this increased gastrin production diminishes once dysplasia and cancer have developed. This suggests that gastrin could play an early role in the development of Barrett’s metaplasia rather than driving the metaplasia−dysplasia−carcinoma sequence.

The finding that inhibition of COX-2 as well as CCK₂ receptors decreases gastrin-induced proliferation is in keeping with previous data that demonstrated that acid- and bile-induced hyperproliferation in BE was COX-2 dependent (61). In addition, treating patients with the COX-2 inhibitor Rofecoxib inhibited COX-2 expression and proliferation in BE patients (62). The COX-2 inhibitor NS-398 was shown to reduce proliferation (63) and induce apoptosis (64) in pancreatic cell lines, and it arrested G₁ phase of the cell cycle in lung cancer cells (65). COX-2 inhibitor may block proliferation either by reducing the availability of the peroxisome proliferation activator receptor ligands (COX-dependant mechanism) or by reducing the binding of the peroxisome proliferation activator receptor and ligand complex to the nuclear DNA (COX independent; Refs. 66, 67, 68). We have shown that gastrin stimulated COX-2 and PGE₂ in all tissue types; however, the effect of gastrin on proliferation was restricted to BE samples.

In conclusion, our results do not support an independent role for COX-2 in the malignant progression in Barrett’s esophagus. However, this study does suggest that the autocrine production of gastrin may be involved in the pathogenesis of Barrett’s esophagus before the development of dysplasia, partly through inducing COX-2 expression. Furthermore, inhibitors of COX-2 and the CCK₂ receptor may ameliorate the epithelial hyperproliferation induced by multiple stimuli in the lower esophagus including acid, bile, and gastrin. These complex interactions between COX-2 and gastrin need to be considered when evaluating a chemoprevention strategy for Barrett’s AC.

We are grateful to Sarah Vowler for her advice on statistics and to the staff and patients in the Endoscopy Department for their cooper-ation.