Cyclooxygenase-2 Expression in Human Colorectal Cancer Is Unrelated to Overall Patient Survival

Colorectal cancer is among the most frequent malignant diseases worldwide, and is one of the leading causes of cancer-related deaths (1). In the United States, 150,000 newly diagnosed cases and 50,000 deaths from colorectal cancer are diagnosed every year (2). With 57,000 new cases yearly, colorectal cancer is the most frequent type of cancer in Germany, ranking above breast cancer (46,000) and lung cancer (37,000; ref. 3). Most colorectal tumors arise from precursor adenomatous polyps that develop into invasive adenocarcinoma, typically over a 5- to 10-year or longer period, and only about 10% of all adenomas proceed to cancer (4). Despite of the recognition of the adenoma-to-carcinoma transition in the pathogenesis of colorectal cancer, the etiology of this cancer remains unknown (5).

Established risk factors for colorectal cancer are familial history of colorectal and other tumors as well as lifestyle factors, such as a high-fat diet, obesity, inactivity, and smoking (3). Identification of such risk factors has led to emphasis on primary prevention (5), which involves dietary modifications and chemoprevention (1). Experimental and epidemiologic data suggest that inhibitors of the cyclooxygenase (COX, prostaglandin synthetase) system (6), such as nonsteroidal anti-inflammatory drugs (NSAID), including aspirin and selective COX-2 inhibitors, suppress the growth of intestinal tumors (7–12).

Several studies have shown that COX-2 expression is elevated in human colon cancers and adenomas compared with normal mucosa (13–16), thus making COX-2 a potential target for chemoprevention. A possible relationship between COX-2 expression in colorectal cancer tissue and patient survival was examined on middle-sized numbers of specimens and patient data (17–23). These data are as yet inconclusive. However, most of the studies confirm that COX-2 expression is not related to patient survival.

The aim of this study was to examine a large number of human colorectal cancer specimens and to test the hypothesis whether COX-2 expression is related to patient survival.

Patients and tissues. Specimens of 946 consecutive patients with primary colorectal cancer who received surgery in the years 1987 until 1997 in the Robert Bosch Hospital Stuttgart were available. None of the patients were pretreated with chemotherapy or radiotherapy. The specimens were retrieved from the files of the Department of Surgical Pathology. From these 946 patients, 199 patients (21%) were excluded because of the following reasons: (a) no follow-up or incomplete data (n = 65), (b) no tumor tissue contained in specimen (n = 27), or (c) no archival blocks or insufficient tumor material (n = 107). The clinical data were retrieved from the doctors' discharge letters and the reports of the Department of Surgical Pathology.

Survival of patients was recorded from the time of surgery until death. Patients who died because of cancer or because of unknown origin were analyzed as dead. Patients who died of other causes than cancer were analyzed as survivors with a survival time lasting from surgery until death. The mean follow-up of all patients was 48.7 ± 53.6 (SD) months (median: 24.6 months). The mean follow-up of survivors was 59.8 ± 58.8 (SD) months (median: 45.5 months). The survival time for nonsurvivors was 29.4 ± 35.6 (SD) months (median: 16.1 months).

All specimens and connected data were analyzed in anonymized form without knowledge of individual data. All analyses were done after omitting the personal data of the patients.

Adenocarcinoma with >50% glands with extracellular mucus were termed mucinous adenocarcinoma.

Paraffin embedding. All specimens were fixed on 4% formalin immediately after surgical removal of the tumor. In the first years (up to 1993), we used unbuffered formalin; later on, buffered formalin was used. A bias introduced by fixation or storage time was excluded by comparing cases before with cases after 1993 with concern to COX-2 expression. After formalin fixation (for 24-72 hours), H&E sections were done by standard procedures.

Dot-blot investigation of cyclooxygenase-2 antibody. We investigated several COX-2 antibodies (Santa Cruz, Heidelberg, Germany; IBL, Hamburg, Germany; Cayman Chemical, Ann Arbor, MI) for specificity and sensitivity with the dot blot method by blotting COX-1, COX-2, and albumin on nitrocellulose sheets (1 μL). The stock solution was 500 ng/μL of each protein diluted down in six steps to 15.6 ng/μL. The antigens were detected by a two-step antibody method (primary antibody 1:50, secondary alkaline phosphatase–labeled antibody 1:1,000). The final reaction product was formed by incubation in nitroblue tetrazolium chloride and 5-bromo-4-chloro-3-indolyl phosphate, buffered with TRIS.

Cyclooxygenase-2 immunohistochemistry. For demonstration of COX-2, we used an Envision system (DAKO, Germany). After applying an antigen retrieval system with steam heater and citrate buffer and after blocking of endogenous peroxidase by H₂O₂/methanol, we applied a COX-2 goat polyclonal antibody (diluted 1:50; sc-1745, Santa Cruz; refs. 24, 25), followed by a rabbit antibody (1:10,000) and a dextran polymer coupled with peroxidase. Controls omitted every single step (i.e., peroxidase blocking medium, antigoat antibody, primary antibody, and Envision system, respectively). As positive control, we used an intestinal metaplasia and normal kidney tissue, which yielded similar results. The antibody sc-1745 by Santa Cruz showed less unspecific background staining than another COX-2 antibody tested (4H12, Novocastra, Newcastle, United Kingdom) and was, therefore, chosen for the staining procedure in this study.

Evaluation of cyclooxygenase-2 immunostaining. Seven hundred forty-seven specimens immunostained for COX-2 were evaluated under a transmission light microscope by two investigators who were unaware of the patients' background and of the other investigator's result. The scoring of COX-2 expression in tumor epithelial cells was done according to the method of Remmele and Stegner (26). Intensity of staining was scored as 0 (negative), 1 (weak), 2 (medium), and 3 (strong; Fig. 1). Extent of staining was scored as 0 (0%), 1 (1-20%), 2 (21-50%), 3 (51-80%), and 4 (81-100%), indicating the percentage of positive staining in carcinoma tissue. Multiplication of intensity score (0-3) and extent score (0-4) resulted in the COX-2 immunoreactivity score (IRS-COX2), which ranges between 0 and 12. All possible cutoff points for IRS-COX2 were analyzed concerning the relationship of the respective COX-2–positive versus COX-2–negative specimens to patient survival.

Statistical analyses. Comparison of included and excluded patients was calculated using an n × k Kruskal-Wallis test to rule out selection bias. Survival analysis was done using the Kaplan-Meier method for univariable analysis and with a Cox regression model for multivariable analysis. The proportional hazards assumption of the Cox model was verified for each of the prognostic factors. Interobserver and intraobserver variability were calculated by κ statistic, which was done as described by Landis and Koch (27). κ values of 0.2 to 0.4 indicate “fair,” of 0.4 to 0.6 “moderate,” and of >0.6 “excellent” results.

Patients. Included (n = 747) and excluded (n = 199) patients did not differ significantly with regard to tumor (P = 0.99), node (P = 0.69), metastasis (P = 1.0), grading (P = 0.67), stage (P = 0.82), localization (P = 0.80), IRS-COX2 (P = 0.11-0.47, depending on the cutoff value chosen for analysis), ratio of mucinous and nonmucinous adenocarcinoma (P = 0.74), and mean age (P = 0.32). Thus, included patients are considered representative of the cohort.

Clinical data and well-accepted conventional prognosis factors (such as tumor, node, metastasis, grading, and tumor stage) of the patients under study are given in Table 1. The survival rates for 1, 3, 5, and 10 years were 81.0%, 66.8%, 60.2%, and 49.8% (median: 117.3 months; 95% confidence interval, 102.3-132.0), respectively, for all stages (I-IV) of colorectal cancer or 83.0%, 69.2%, 62.2%, and 52.3% (median: 139.0 months; 95% confidence interval, 122.6-155.4) for stage I to III cases without rectal cancer. Stage was a good univariable and node a good univariable and multivariable predictor of disease outcome (data not shown).

Cyclooxygenase-2 immunostaining. COX-2 immunostaining was not influenced neither by storage time nor by kind of fixation as specimens embedded before and after 1993 did not reveal any statistically relevant differences in COX-2–positive staining (χ² = 7.09; P = 0.53). The results of the COX-2 immunostaining are shown in Table 2.

COX-2 expression in colorectal cancer was predominantly found in the epithelial compartment of cancer tissue. COX-2 staining was localized mainly to the cytoplasm and occasionally to the perinuclear region and nuclei. COX-2 expression was detectable in interstitial cells as well. Inflammatory cells of the tumor area and normal colonic mucosa adjacent to the cancer tissue showed variable COX-2 staining from negative (0) to strong (3; data not shown) without any staining category dominating significantly. Concerning the distribution of COX-2 immunostaining, only 14.1% of the tumors (with IRS-COX2 values of 8, 9, or 12; Table 2) showed homogeneous staining, whereas most tumors were heterogeneously stained. COX-2 expression in tumor epithelial cells was not related to the histologic subtypes of human adenocarcinoma. Particularly, mucinous adenocarcinoma did not differ in COX-2 expression from nonmucinous adenocarcinoma. The pattern and localization of COX-2 immunoreactivity found using the second antibody (4H12, Novocastra) was similar to that with the first antibody (sc-1745, Santa Cruz).

Cyclooxygenase-2 immunostaining and prognosis. According to our immunohistochemical assessment, COX-2 expression in tumor epithelial cells was not related to overall patient survival in both univariable and multivariable analysis, irrespectively of the cutoff point that was chosen for analysis (Table 3; Fig. 2). Similar results were obtained when overall survival was replaced by analysis of disease-free survival (data not shown). When the different variables being significant in a univariable Kaplan Meier analysis (stage, tumor, node, metastasis, and grade; Table 1) were entered in a Cox regression model for overall survival (forward log likelihood method), only node and metastasis were significantly related to overall patient survival. Comparing mucinous adenocarcinoma versus all other forms, there was no significant difference in overall survival (log rank 1.53; P = 0.272).

We also did an analysis of stage I to III colon cancer patients only (n = 478; i.e., stage IV and rectal cancer patients were excluded). Even here, COX-2 expression was unrelated to overall patient survival (Table 3).

κ statistics. Depending on the number of classes in which COX-2 expression was measured, we retained interobserver κ values between 0.44 and 0.56 and intraobserver κ values between 0.46 and 0.72.

The purpose of this study was to evaluate the relationship between COX-2 expression in human colorectal cancer and overall patient survival. The large sample size of consecutive patients and use of a rigorous scoring system are strengths of this trial. Univariable and multivariable analysis of 747 colorectal cancer specimens have shown that COX-2 expression in tumor epithelial cells is not a significant prognostic factor for overall patient survival, irrespectively of the cutoff point (Table 3). Thus, our findings are consistent with other studies (Table 4) that did not find any relationship between COX-2 expression and survival by multivariable analysis (17–21). In contrast, few studies have claimed that there is a relation between COX-2 expression and patient survival. This difference is based upon (a) univariable analysis (18, 19, 21) or (b) investigation of disease-free (and not overall) patient survival (22), with further possible explanations being (c) the smaller amount of specimens examined (mostly n < 100), (d) a biased selection of patients, (e) different scoring systems, or (f) different antibodies used.

Interestingly, all of the known studies on COX-2 expression in colorectal cancer (Table 4) used different scoring systems. We used the scoring system recommended by Remmele and Stegner (26). It is established internationally for the evaluation of breast cancer, and it is very similar to scoring systems used in other studies about COX-2 in colorectal cancer (17, 19). Our staining method (avidin-biotin complex method of Hsu et al.; ref. 28) is well established and has been used in many published studies of COX-2 immunostaining (17, 18).

Patients with stage IV carcinoma or rectal cancer have a poorer prognosis due to the pathology of the tumor and due to impaired surgical curability of the tumor, respectively. Exclusion of these patients led to a number of n = 478 patients with stage I to III colon cancer (Table 3) whose COX-2 expression was also unrelated to overall survival, similar to the entire study population.

It is believed that conventional NSAIDs and selective COX-2 inhibitors suppress the formation of intestinal tumors (29–34) and that they prevent colon carcinogenesis mainly in adenomas (35, 36). However, there is no evidence that NSAIDs regress or cure established colorectal cancer (36). The enhancement of angiogenesis, tumor invasion, and metastasis (37) by COX-2 is most likely due to the stimulation of vascular endothelial growth factor production (38, 39) by COX-2–synthesized prostaglandins. The recognition of vascular endothelial growth factor as a prognostic marker for human colorectal cancer (40, 41) has led to the development of anti–vascular endothelial growth factor antibodies for treatment of metastatic colorectal cancer (42, 43). Bevacizumab, as one of these, has been approved by the Food and Drug Administration (44) for this indication (45, 46).

Recent findings suggest that the chemoprotection by aspirin and other NSAIDs may, among others (47), include peroxisome proliferator–activated receptor δ/γ (5). By peroxisome proliferator–activated receptor γ stimulation, NSAIDs may stimulate apoptosis and inhibit cancer growth (48). However, human data about peroxisome proliferator–activated receptor γ activation with troglitazone in human breast cancer have not been very promising (49).

In summary, COX-2 expression in colorectal cancer epithelial cells is not related to overall patient survival. Besides COX-2, several NSAID-responsive targets, such as the peroxisome proliferator–activated receptors, are involved in carcinogenesis. Their prognostic relevance has yet to be determined.

We thank McClellan for excellent technical assistance and Lüttgen for help with the follow-up.