Four of five epidemiological studies have indicated that the use of nonsteroid anti-inflammatory drugs (NSAID) is associated with reduced risk of esophageal cancer.1, 2, 3, 4, 5, 6 The best-known target of NSAID is the cyclooxygenase (COX) enzyme and it is primarily the inducible COX-2 isoenzyme that has been linked to human carcinogenesis.7, 8, 9, 10, 11 Genetic deletion of COX-2 leads to reduced polyp burden in mouse models of familial adenomatous polyposis (FAP),12, 13 and selective COX-2 inhibitors were shown to reduce polyp burden in patients with FAP.14, 15 Furthermore, we showed recently that expression of COX-2 is an independent factor of poor prognosis in patients with adenocarcinoma of the esophagus.16
Elevated COX-2 expression has been associated with resistance to treatment and to reduced overall survival for patients who received (neo-) adjuvant chemo- or radiotherapy.17, 18, 19, 20 Experimental models have demonstrated that selective COX-2 inhibitors can enhance the efficacy of radiotherapy or chemotherapy.21, 22, 23, 24, 25, 26, 27 For patients with squamous cell carcinoma of the esophagus, neoadjuvant radiotherapy alone seems to have no role in the treatment,28 although for neoadjuvant chemoradiotherapy an improvement in locoregional control has been demonstrated.29 In this context, it would be desirable to identify subgroups of patients who might benefit from (neo-) adjuvant therapy. The aim of our study was to assess the expression of COX-2 by immunohistochemistry in patients with esophageal squamous cell carcinomas. The analysis was done in patients who underwent potentially curative resection with or without neoadjuvant chemotherapy, and the results were correlated with clinicopathological parameters and overall survival.
Material and methods
Between January 1 1993 and December 31 2000, 504 patients underwent esophageal resection with proximal gastrectomy for an esophageal malignancy with curative intent, i.e., locally resectable disease without distant metastases. The data from these patients were prospectively collected in a database. The pathology reports of all patients were reviewed to identify 138 patients diagnosed with a squamous cell carcinoma of the esophagus. Thirteen tumors with adenosquamous characteristics were excluded during review of the slides, and another 8 patients were excluded during immunohistochemical analyses due to non-representative archival slides compared to the original tumor. Thus, 117 patients remained for further analysis. In 71 patients resection was carried out by a transhiatal approach without thoracotomy and extended lymph-node dissection. Forty-six patients underwent esophagectomy through a right-sided thoracotomy followed by a laparotomy in combination with 2-field lymph node dissection. Patients were followed until death or December 2003, ensuring a minimal potential follow-up of 3 years. The median actual follow-up was 33.0 months (range = 7 days to 8.5 years), which includes those 4 patients who died in the hospital within 30 days after the surgery. The patients were seen on a regular basis for 5 years in the outpatient clinic. For the first 2 years they were seen at 3–4-month intervals, afterward at 6-month intervals. For the present study, patients or their family practitioners were contacted by phone to assess their current status when they had been discharged by the surgeon after 5 years. None of the patients were lost to follow-up.
Thirty-six patients received chemotherapy preoperatively as part of a randomized controlled trial30: cisplatin (CddP; 80 mg/m2 i.v. Day 1) and etoposide (E; 100 mg i.v. Day 1+2; 200 mg/m2 orally Day 3+5). After 2 cycles of chemotherapy, a clinical response was evaluated based on clinical findings (e.g., decrease of dysphagia) and repeated endoscopy and CT-scanning. Patients with tumor response received another 2 cycles followed by surgery, whereas non-responding patients underwent surgery after the second cycle. No adjuvant treatment was administered postoperatively. Presence of necrosis was evaluated from H&E-stained samples.
Immunohistochemistry and scoring
Formalin-fixed and paraffin-embedded specimens representing the deepest tumor infiltration were sectioned (5 μm), deparaffinized, microwaved and immunostained using a COX-2 specific mouse anti-human monoclonal antibody (160112, Cayman Chemical Co., Ann Arbor, MI) in a dilution of 1:200 (2.5 μg/ml) as described previously.16, 31 Every 20th sample of the trial series was a known colon adenocarcinoma specimen, in which stromal cells at an area of ulceration were scored 3, cancer cells from 2–3 and adjacent nonneoplastic epithelium 1 (for scoring criteria see below). This procedure confirmed that there was no significant intra- and interassay variability of the staining intensity, and helped us to score the trial specimens. Specificity of the antibody was confirmed as described previously.16 COX-2 immunohistochemical staining was scored independently and in a blinded manner by 2 investigators (A.S. and C.J.B.) without knowledge of the status of the patients. The following scoring criteria of the tumor cells were agreed upon before the analysis: 0, no staining; 1, weak diffuse cytoplasmic staining (may contain stronger intensity in less than 10% of the cancer cells); 2, moderate to strong granular cytoplasmic staining in 10–90% of the cancer cells; 3, >90% of the tumor cells stained with strong intensity. Scores 0 and 1 were categorized as “COX-2 low” and scores 2 and 3 as “COX-2 high” for the statistical analyses (see below). The allocation of tumors to the “COX-2 low” vs. the “COX-2 high” category by the 2 investigators was similar (93% of the specimens were categorized identically). In cases of disagreement (n = 9) the slides were re-evaluated by a group of investigators (A.S., C.J.B. and G.J.A.O.) using a multiheaded microscope. Immunohistochemistry of p53 was carried out using an established procedure,32 and staining for p53 was considered weakly positive if 0–50% of the tumor cells were immunopositive and strongly positive if 50–100% of the tumor cells were immunopositive.
The associations between factors were analyzed using Student's t-test (continuous data) and χ2 test (categorical data). Overall survival was estimated according to the Kaplan-Meier method and compared using the log-rank test. The Cox proportional hazard model was used to examine the role of COX-2 expression and clinicopathological parameters as prognostic factors. Variables with multiple categories were recoded into dichotomous variables by combining categories with comparable prognosis (differentiation grade: well vs. moderate and poor; radicality of resection: microscopically radical [R0] vs. microscopically nonradical [R1] and macroscopically nonradical [R2]; tumor stage: Stage I and IIa vs. IIb, III and IV). p-Values of 0.05 or below were considered statistically significant. All statistical analyses were carried out using the Statistical Software Package version 11.0 (SPSS Inc., Chicago, IL).
In the patient group who did not receive neoadjuvant chemotherapy, there were 45 males (55.6%) and 36 females (44.4%), with a median age of 60 years (range = 33–79 years). For the 36 patients who received neoadjuvant chemotherapy the age and gender distribution was similar (19 males, 52.8%; 17 females, 47.2%; median age = 59 years, range = 38–78 years).
COX-2 protein expression in esophageal squamous cell carcinoma
Expression of COX-2 protein was evaluated in the 117 esophageal squamous cell carcinoma specimens using immunohistochemistry. COX-2 expression was localized mainly in the neoplastic cells, whereas nonneoplastic squamous epithelium was consistently negative or only weakly positive. Immunoreactivity in the cancer cells was relatively heterogeneous, however, when compared to the esophageal adenocarcinoma specimens analyzed previously.16 Among the 81 patients who did not receive neoadjuvant chemotherapy, immunoreactivity was negative in 34 (42.0%), weakly positive in 27 (33.3%), moderately positive in 19 (23.5%) and strongly positive in one (1.2%) of the specimens. Of 36 patients who received neoadjuvant treatment, no immunoreactivity was observed in 14 (38.9%), weak in 11 (30.6%), moderate in 11 (30.6%) and strong in none of the samples. There was no statistically significant difference in COX-2 expression between the groups with or without neoadjuvant chemotherapy (p = 0.8).
Correlation between COX-2 expression and clinicopathological parameters
In the patient group who did not receive neoadjuvant treatment, COX-2 expression was associated with the location of the tumor with a higher COX-2 expression in distal tumors (p = 0.02). For the other clinicopathological parameters no correlation with COX-2 expression was demonstrated (Table I). In the group of 36 patients who received neoadjuvant chemotherapy, high COX-2 expression was associated with the distal location of the tumor (p = 0.03), with male gender (p = 0.02), and with response to neoadjuvant chemotherapy (p = 0.04). Furthermore, a significant association was found between low COX-2 expression and the development of distant metastases (p = 0.03), with all of the eight distant metastases developing in the COX-2 low group. Although 13 of 16 locoregional recurrences developed in the COX-2 low group, this was not statistically significant (p = 0.1). For all other parameters, there existed no significant association with COX-2 expression (Table II). Notably, there were no statistically significant differences in the clinicopathological parameters between esophageal squamous cell carcinoma patients with or without chemotherapy treatment.
Table I. Correlation of Clinicopathological Parameters and COX-2 Expression of 81 Patients with A Squamous Cell Carcinoma of The Esophagus (Patients Did Not Receive Chemotherapy Before Operation)
Low (n = 61)n (%)
High (n = 20)n (%)
T1, tumor limited to the submucosa; T2, tumor infiltrates muscularis propria, but not adventitia; T3, tumor infiltrates adventitia; T4, tumor infiltrates adjacent structures.
N0: no lymph node metastases, N1, lymph node metastases present.
I, T1N0M0; IIa, T2-3N0M0; IIb, T1-2N1M0; III, T3N1M0, T4anyNM0; IV, anyTanyNM1.
Association of COX-2 expression with overall survival
COX-2 expression did not associate with overall survival in the patient group who did not receive neoadjuvant treatment (p = 0.4). The 5-year survival was 32.0% (95% confidence interval [CI] = 20.0–43.9) in the COX-2 low group and 36.0% (95% CI = 13.9–58.1) in the COX-2 high group. In the 36 patients who received neoadjuvant chemotherapy, a significant association was found between low COX-2 expression and reduced overall survival (p = 0.02). The 5-year survival was 19.4% (95% CI = 2.9–35.9) in the COX-2 low group and 72.7% (95% CI = 46.4–98.9) in the COX-2 high group. Kaplan-Meier curves for patient survival are depicted in Figure 1. There was no difference, however, in overall survival between the combined therapy group and the surgery alone group (p = 0.5).
Univariate and multivariate analysis
To evaluate the role of COX-2 as an independent prognostic factor in the group of neoadjuvant treated patients a multivariate analysis was carried out. In the univariate Cox regression analysis, a significant prognostic effect was found for tumor stage, radicality of resection and low COX-2 expression. The impact of low COX-2 expression on decreased overall survival persisted after adjustment for the other possibly confounding variables in the multivariate analysis (Table III).
Table III. Univariate and Multivariate Analysis of Prognostic Factors in The 36 Patients with A Squamous Cell Carcinoma of The Esophagus Who Received Neoadjuvant Chemotherapy
Presence of necrosis and p53 expression in the neoadjuvant treated group
It has been shown previously that COX-2 expression can associate with tumor necrosis and p53 immunopositivity.33, 34, 35, 36, 37 To address this issue, COX-2 expression was re-analyzed and correlated to presence of necrosis and p53 immunopositivity. Tumoral necrosis was demonstrated in 16 (44.4%) samples derived from the 36 patients with neoadjuvant treatment. Although necrosis did not significantly correlate with COX-2 expression (p = 0.5) or overall survival (p = 0.4), there was an association between the presence of necrosis and clinical response to the neoadjuvant chemotherapy (p = 0.03). High expression of p53 protein was detected in 24 of the 36 samples (66.7%). There was no association between high COX-2 expression and p53 expression (p = 0.8), nor was there any correlation between p53 expression and tumor response to chemotherapy (p = 0.7) or overall survival (p = 0.8). It seems that COX-2 expression is independent of necrotic processes and loss of p53 function in this material.
Our results show that COX-2 expression does not associate with survival of esophageal squamous cell carcinoma patients who were treated with surgery alone. These data are in line with previous publications,37, 38 and one study demonstrated a significant correlation between low expression of COX-2 and advanced tumor stage and presence of distant metastases.39 Similarly to our study, a heterogenic intratumoral staining pattern of COX-2 was found38 with a significantly higher expression in tumors located in the distal esophagus than in more proximal areas.37 The value of our series is that it originated from the same cohort of patients from which COX-2 has been analyzed previously from esophageal and gastric cardia adenocarcinomas using identical staining protocol and scoring criteria. In previous studies we found COX-2 expression to be an independent prognostic variable for reduced overall survival in esophageal adenocarcinoma,16 but it did not show any prognostic value in adenocarcinoma of the gastric cardia.40 The prognostic significance of COX-2 is apparently distinct in different histological types and locations of digestive tract carcinomas.
In esophageal squamous carcinoma patients who received neoadjuvant chemotherapy, low COX-2 expression associated with the development of distant metastases and it was an independent prognostic variable for poor outcome. This was an unexpected result, which was investigated further. Because we have shown previously that COX-2 expression is associated with necrotic areas in lung and bladder carcinomas33, 34 and COX-2 is induced in an experimental necrosis model,41 we correlated the presence of necrosis and COX-2 expression. Although COX-2 expression associated with the clinical response to chemotherapy, it did not correlate with histologically detectable necrosis. We also investigated a possible correlation between p53 overexpression as an indicator of mutation of p53 gene and elevated COX-2 expression, because there seems to be a direct link between a defective p53 pathway and induction of COX-2 expression in esophageal squamous cell carcinomas37, 42 as well as in other cancers.35, 36, 43 The COX-2 gene has been shown to be induced in p53 defective cells and downregulated by wild-type p53.44, 45 We found no association between p53 immunoreactivity and COX-2 expression. We were unable to correlate COX-2 expression to other markers that might be related to the response to treatment.
Several in vitro studies and one in vivo study show that radiation and chemotherapeutic agents can enhance expression of COX-2.46, 47, 48, 49, 50 Although the frequency of COX-2 expression did not differ between non-treated and neoadjuvant treated patients, we cannot exclude the possibility that COX-2 expression was induced by the chemotherapy. If this was the case, it could be hypothesized that COX-2 signals a favorable response to the neoadjuvant treatment. Ferrandina et al.51 demonstrated that in locally advanced cervical cancer patients receiving neoadjuvant chemotherapy, COX-2 expression increased progressively from responders to partial responders to non-responders and COX-2 positivity was a prognostic variable of shorter overall survival. Kulke et al.52 reported recently that in esophageal squamous cell carcinoma patients low pretreatment COX-2 expression correlated with response to neoadjuvant chemoradiation, but no correlation was found between pretreatment COX-2 expression and survival of the patients. We cannot directly compare these results to our data because in both studies the biopsy specimens were obtained before neoadjuvant treatment and in our study the samples were obtained from surgical resection material after the chemotherapy. We decided not to include an analysis of the preoperative biopsy specimens because the COX-2 staining pattern was relatively heterogeneous in the histological tumor samples, and it was assumed that staining results of preoperative biopsies might only reflect sampling artifacts.
Our study demonstrates that low COX-2 expression is associated with poor prognosis in patients with esophageal squamous cell carcinomas who received neoadjuvant chemotherapy. For patients who did not receive neoadjuvant therapy, COX-2 expression was not associated with overall survival. This is in contrast to a previous study in which we demonstrated that elevated COX-2 expression is an independent prognostic variable for reduced overall survival in patients with esophageal adenocarcinoma. Our results suggest that the prognostic significance of COX-2 depends on the histologic type of esophageal carcinoma and on the treatment applied. In addition, COX-2 may serve as a possible marker for favorable response to neoadjuvant treatment.
We thank Ms. E. Laitinen and Ms. P. Peltokangas from the Helsinki University Central Hospital and E. Caspers, F. Morsink and A. Musler from the AMC Department of Pathology for excellent technical assistance.