Cyclooxygenases regulate the production of prostaglandins and play a role in tumor development and progression. The authors investigated the prognostic impact of expression of the cyclooxygenase (COX) isoforms, COX-1 and COX-2, on disease-free survival and progression-free survival in patients with primary breast carcinoma as well as the association between COX expression and other clinicopathologic parameters.
In this study COX isoform expression was determined by immunohistochemistry in a cohort of 221 patients with primary breast carcinoma.
Expression of COX-2 was detected in 36% of breast carcinoma samples and was associated significantly with several clinicopathologic parameters, including positive lymph node status (P < 0.0005), larger tumor size (P < 0.0005), poor differentiation (P < 0.0005), vascular invasion (P = 0.03), and negative estrogen receptor status (P = 0.04). In contrast, COX-1 was expressed in 45% of tumors and was associated with smaller tumor size (P = 0.02) and with negative lymph node status (P = 0.01). In a univariate survival analysis, a significant association was observed between elevated COX-2 expression and decreases in disease-free survival (P = 0.0007) and overall survival (P = 0.02). In a multivariate analysis, expression of COX-2 was of borderline significance for disease-free survival (relative risk, 1.90; 95% confidence interval, 1.00–3.59), adjusting for tumor size, histologic grade, number of positive lymph nodes, and patient age. Elevated expression of COX-1 in tumor tissue had no statistically significant influence on patient prognosis.
Breast carcinoma is the most common malignancy in women. In 2002, it is estimated that there will be 200,000 new diagnoses of breast carcinoma in the U.S., and approximately 40,000 patients will die of the disease.1 Accepted prognostic and predictive factors for breast carcinoma include lymph node status, tumor size, tumor grade, and (to some extent) patient age.2, 3
Epidemiologic data show that nonsteroidal antiinflammatory drugs (NSAIDs) reduce the risk of several types of malignant disease. For breast carcinoma, at least 14 epidemiologic studies have investigated the role of NSAIDs.4 Twelve of those studies observed a reduction in the risk of breast carcinoma with the use of NSAIDs. In a meta-analysis of those studies, regular use of NSAIDs was associated significantly with an 18% reduction in breast carcinoma.4 Cyclooxygenases are the major cellular target of NSAIDs and regulate the synthesis of prostaglandins. There are two isoforms of cyclooxygenase (COX), COX-1 and COX-2. Whereas COX-1 is expressed constitutively in many tissues, COX-2 is inducible by growth factors and inflammatory stimuli. Cyclooxygenases, particularly COX-2, are involved in tumor development and progression.5 In animal models, treatment with selective COX-2 inhibitors reduced growth and metastasis formation of experimental tumors. Thus, a role for COX-2 inhibitors in the prevention and in the treatment of malignant tumors is emerging.
Several studies6–11 have shown that COX-2 is expressed in a subset of breast carcinomas, but most studies have investigated only a comparably small number of patients. To our knowledge, only a few studies9, 11 have investigated a possible prognostic role for COX-2 in breast carcinoma, whereas the role of COX-1 has not been studied to date. To clarify further the prognostic role of COX-2 in breast carcinoma and to evaluate the possible contribution of COX-1, we studied expression of both COX isoforms in a in cohort of 221 patients with primary breast carcinoma who were diagnosed at our breast cancer center.
MATERIALS AND METHODS
Immunohistochemical analysis was performed retrospectively on tissue samples that were taken for routine diagnostic purposes. Based on the availability of tissue samples in the archives of the Institute of Pathology of the Charité University Hospital, patients with primary breast carcinoma who were diagnosed between June 1991 and June 1996 were identified. Patients who had distant metastasis at the time of diagnosis or bilateral breast carcinomas were not included in this study. A total of 286 patients were identified. All patients who were residents of the city of Berlin (221 patients, representing 77% of tissue samples) were included in the study. For all 221 of these patients, overall survival data were extracted from the data base of the local registration office. Overall survival was defined as the time between diagnosis and death. The median follow-up of all patients who were still alive at the time of analysis was 87 months (range, 2–132 months). Within the follow-up period, 67 patients (30%) died. Clinical data regarding disease recurrence were extracted from the files of the breast cancer center and were available for 169 of 221 patients (76%). Disease-free survival was defined as the time between diagnosis and the first clinical or pathologic evidence of locoregional or distant recurrent disease. Forty-eight of 169 patients (28%) experienced disease recurrence. The median follow-up for all patients who were without recurrent disease was 60 months (range, 1–120 months).
Baseline patient characteristics are outlined in Table 1. The median age of all patients at the time of diagnosis was 60 years (range, 28–88 years). One hundred seventy-nine patients (81%) had ductal carcinoma, 28 patients (13%) had lobular carcinoma, and 14 patients (6%) had carcinoma of other histology. One hundred sixty-four of 221 patients (74%) underwent surgery with the same surgeon (K.-J. W.). Fifty patients received hormone therapy.
To evaluate COX expression in samples of benign breast disease, seven samples from patients with fibrocystic disease also were stained for COX-1 and COX-2. These patients were not included in the statistical evaluation.
Tissue samples were fixed in 4% neutral buffered formaldehyde and embedded in paraffin. Routine hematoxylin and eosin staining was performed on the sections for histopathologic evaluation. Tumor histology and tumor grade were evaluated at primary diagnosis and were extracted from pathology reports. Tumors were graded according to the Bloom–Richardson grading modified by Elston and Ellis.12 Lymph node status was assessed at primary diagnosis, and a median of 18 lymph nodes were examined (range, 1–55 lymph nodes). For the 104 patients who were diagnosed with pN0 disease, a minimum of 6 lymph nodes were examined (median, 17 lymph nodes; range, 6–37 lymph nodes). In 96 of 104 patients (92%), at least 10 lymph nodes were examined. For this study, lymph node status was reclassified according to the sixth edition of the TNM classification system.13 Lymph nodes with micrometastasis were classified as pN1mic. For evaluation of c-erbB2 reactivity the DAKO scoring system was used (0 = negative; + = partially membranous; ++ = complete memebranous, weak; +++ = complete membranous, strong).
Immunohistochemical staining was performed according to standard procedures. We used the mouse antihuman COX-2 monoclonal antibody from Cayman Chemical Company (Ann Arbor, MI), which has been used widely for immunohistochemical staining of COX-2. For control of antibody specificity, we performed blocking experiments with the specific peptide that resulted in complete inhibition of COX-2 staining (data not shown). For investigation of COX-1, a mouse antihuman COX-1 monoclonal antibody was used (Cayman Chemical). Briefly, slides were boiled in citrate buffer in a pressure cooker for 10 minutes and incubated with the monoclonal COX-1 antibody (1:200 dilution) or COX-2 antibody (1:1000 dilution) overnight at 4 °C, followed by incubation with a biotinylated antimouse secondary antibody and the multilink biotin-streptavidin-amplified detection system (Biogenex, San Ramon, CA). Staining was visualized using a fast-red chromogen system (Sigma, St. Louis, MO).
Quantification of Immunohistochemical Staining
In this study, we evaluated the hypothesis that increased expression of COX-2 in tumor tissue may be associated with poor survival. This hypothesis was suggested in several previous studies on the expression of COX-2 in malignant tumors. The majority of studies on the prognostic impact of COX-2 expression used a semiquantitative evaluation of COX-2 staining and defined cut-off points to distinguish between COX-2 positive tumors and COX-2 negative tumors. In many studies, only tumors with at least moderate expression of COX-2 were regarded as COX-2 positive. In other studies, a defined percentage of positive cells was required for a tumor to be defined as COX-2 positive.14
Based on the published data regarding possible cut-off points, we employed a simple scoring system in our study that was identical to the scoring system used for the evaluation of hormone receptor status in patients with breast carcinoma15 in which COX-2 expression was evaluated according to the percentage of positive cells and the intensity of staining. The percentage of positive cells was scored as 0 (0% positive cells), 1 (< 10% positive cells), 2 (10–50% positive cells), 3 (50–80% positive cells), or 4 (> 80%). The staining intensity was scored as 0 (negative), 1 (weak), 2 (moderate), or 3 (strong). For the immunoreactive score (IRS), the percentage of positive cells and the staining intensity were multiplied, resulting in a value between 0 and 12. To separate tumors with weak COX expression or strong COX expression and to define a cut-off point that might be reproducible for future studies, we combined patients who had an IRS of 0–6 into one group with negative to weak COX expression (the “COX negative” group) and combined patients who had an IRS of 7–12 a “COX positive” group. The minimum requirement for a positive IRS was either moderate expression in > 80% of cells or strong expression in > 50% of cells. These requirements are similar to the cut-off points that have been used in previous studies on the prognostic role of COX-2.9, 10, 16, 17
The intensity of COX-1 or COX-2 immunostaining in tumor cells was evaluated independently by two investigators (W. W. and C. D.) who were blinded to patient outcome. Samples that resulted in a disagreement on the IRS between the two investigators were discussed with a third investigator (S. H.) using a multiheaded microscope until consensus of at least two investigators was achieved.
The correlations between expression of COX-1 or COX-2 and several clinicopathologic parameters were assessed with a two-sided Fisher exact test or a chi-square test, as indicated. In a univariate survival analysis, the probability of disease-free survival or overall survival was determined by the Kaplan–Meier method. Because the median survival was not reached in several subgroups, Kaplan–Meier estimates of the 5-year disease free survival rate and the 5-year overall survival rate are shown. Different survival curves were compared using the log-rank test. A multivariate survival analysis was performed using Cox proportional hazards regression modeling. In the multivariate analysis, all variables that were associated significantly with patient survival in the univariate analysis were included in the final analysis. Patient age was included as an additional parameter in the multivariate analysis. Generally, P values < 0.05 were considered significant. For statistical evaluation, we used the SPSS software package (version 10.0; SPSS, Inc., Chicago, IL).
Expression of COX-1 and COX-2 in Human Breast Carcinoma
Using immunohistochemical analysis of COX-1 and COX-2 in human breast carcinoma, positive immunoreactivity was detected in a subset of tumors. COX-2 was expressed in a granular cytoplasmic pattern that was enhanced in the perinuclear area in some tumors. We detected positive COX-2 expression (IRS, 7–12) in 36% of breast carcinoma samples, whereas COX-1 was expressed in 45% of tumors (Table 1, Fig. 1). Forty-three tumors (19.7%) were positive for both COX-1 and COX-2. In 41% of tumors for COX-2 and in 28% of tumors for COX-1, a focal expression pattern was observed in 10–80% of the tumor area. In the remaining tumors, staining was homogenous throughout the tumor or was absent.
In addition, seven tissue samples from patients with fibrocystic disease were stained for expression of the COX isoforms. Whereas four of seven samples were positive for COX-1, all samples were negative for COX-2. In contrast, increased expression of COX-2 often was observed in morphologically unremarkable breast tissue adjacent to invasive carcinomas (not shown).
Correlation with Clinicopathologic Parameters
We observed a significant correlation between COX-2 expression and several established clinicopathologic tumor parameters (Table 2). COX-2 was expressed in 41% of invasive ductal carcinomas but in only 14% of lobular carcinomas and 21% of other carcinomas (P = 0.012). COX-2 expression was correlated significantly with tumor size: Positive COX-2 expression was found in 58% of tumors that measured > 20 mm in greatest dimension, compared with 24% of tumors that measured < 20 mm (P < 0.0005). COX-2 expression also was correlated significantly with the presence of lymph node metastasis: Only 16% of tumors without lymph node metastasis were positive for COX-2 expression, whereas > 50% of lymph node positive tumors were positive for COX-2 expression (P < 0.0005). Similarly, there was a significant correlation between histologically detectable vascular invasion and COX-2 expression (P = 0.03). Other significant correlations were observed between COX-2 expression and poor differentiation (P < 0.0005) as well as negative estrogen receptor status (P = 0.04).
Table 2. Correlation between Cyclooxygenase-2 Expression and Various Clinicopathologic Factors
For COX-1 expression, an inverse correlation was observed with both tumor size and lymph node status. COX-1 was expressed in 52% of tumors that measured < 20 mm but in only 34% of tumors that measured > 20 mm (P = 0.02). Fifty-seven percent of tumors without lymph node metastasis were positive for COX-1, and approximately 30% of pN2a tumors and pN3a tumors were positive for COX-1 (P = 0.01). In contrast, there was no correlation between COX-1 expression and histologic tumor grade and other tumor parameters (histologic type, vascular invasion, estrogen receptor status, c-erb B2 score, proliferation, and patient age). Table 3 shows the correlation of COX-1 expression with selected clinicopathologic parameters.
Table 3. Correlation of Cyclooxygenase-1 Expression and Selected Clinicopathologic Factorsa
In the univariate survival analysis, we observed a significant association between COX-2 expression and decreases in both disease-free survival and overall survival (Table 4, Fig. 2). The estimated 5-year disease-free survival rates were 65% for patients with COX-2 positive tumors and 82% for patients with COS-2 negative tumors (P = 0.0007). In parallel, the 5-year overall survival rates were 65% for patients with COX-2 positive tumors and 78% for patients with COX-2 negative tumors (P = 0.02). For the subgroup of patients who received hormone therapy, a similar correlation was observed between increased COX-2 expression and reduced disease-free survival (P = 0.01) and overall survival (P = 0.04).
Table 4. Univariate Kaplan–Meier Survival Analysis: Mean Disease-Free Survival and Mean Overall Survival of All Patients According to Clinicopathologic Factors and Cyclooxygenase-1 or Cyclooxygenase-2 Expression
We did not observe a correlation between the expression of COX-1 and disease-free or overall survival (Fig. 2). Other prognostic factors for disease-free survival and overall survival were tumor size, lymph node status, and histologic grade. In addition, vascular invasion and tumor proliferation both were prognostic parameters for disease-free survival, but not for overall survival (Table 4).
Multivariate Survival Analysis
In the multivariate survival analysis, histologic grade, tumor size, patient age, and the absolute number of positive lymph nodes were included in the Cox regression model. In this analysis, the most important predictors of disease-free survival were the number of positive lymph nodes and histologic grade (Table 5). Increased COX-2 expression was of borderline significance, with a P value of 0.049 and a relative risk of 1.9 (95% confidence interval, 1.004–3.59).
Table 5. Multivariate Survival Analysis: Cox Regression Model
For overall survival, the number of positive lymph nodes, tumor size, and patient age at diagnosis were identified as independent prognostic factors that reached statistical significance. Other parameters (COX-2 expression and histologic grade) did not reach statistical significance (Table 5).
The current study is the first to investigate the possible prognostic impact of both isoforms of cyclooxygenase, COX-1 and COX-2, in patients with primary breast carcinoma. We found that, although both isoforms were expressed in breast tumors, only COX-2 (and not COX-1) was associated with patient prognosis. COX-2 expression was correlated significantly with tumor size > 20 mm, histologic grade, positive lymph node status, and the presence of angioinvasion. In contrast, COX-1 expression was increased significantly in smaller tumors (< 20 mm) and in tumors without lymph node metastasis. Our data on COX isoform expression suggest that there is a shift from the expression of COX-1 in smaller and nonmetastatic tumors to the preferential COX-2 expression in larger and metastasizing tumors.
The expression of COX-2 protein in human breast carcinoma also has been observed by other authors. Soslow et al.8 showed that 56% of 60 breast tumors (invasive carcinomas and ductal carcinomas in situ) expressed COX-2. Similarly, Half et al.9 found expression of COX-2 at intermediate or high levels in 43% of 57 invasive breast carcinomas. A correlation between COX-2 expression, angiogenesis, and lymph node metastasis was observed by Costa et al.11 in a study of 46 patients with invasive ductal breast carcinoma. In the same study, an association between COX-2 expression and short disease-free survival was observed for a subgroup of 26 patients who had survival data available.
In a recent study by Ristimäki et al.,10 the impact of COX-2 expression on patient prognosis was assessed in a large, population-based, multicenter study in Finland using cancer registry data from multiple centers and regions and a tissue microarray for COX-2 expression analysis. This approach is different from that used in the current study: We recruited our cohort from a single center and included only patients who were diagnosed at the breast cancer center at our university hospital. Most of our patients underwent surgery with the same surgeon. Thus, potential regional or institutional differences in the care of patients with breast carcinoma that may have had an additional influence on their outcome were minimized. For the 221 tumor samples in our study, it was possible to use large paraffin sections for immunohistochemistry, so that heterogeneity of the staining in different areas of the tumor could be controlled. However, despite the differences in study design, there are remarkable similarities between the results of Ristimäkis group and the current study. Both studies observed moderate-to-strong expression of COX-2 in one-third of invasive breast carcinomas. Furthermore, both studies found a significant correlation between COX-2 expression and lymph node status, tumor size, histologic grade, and negative estrogen receptor status. Similar to our results, the Ristimäki group showed that COX-2 expression was a negative survival factor for distant disease free survival, which they defined as the time between diagnosis and the occurrence of metastases outside the locoregional area or death from breast carcinoma.
In our study, we provided additional data on the prognostic role of COX-2 in patients with primary breast carcinoma that, to our knowledge, have not been reported previously. We used two different outcomes for survival analysis: disease-free survival (which included both locoregional recurrent disease and distant metastasis) and overall survival. In the univariate survival analysis, we showed that the prognostic impact of COX-2 expression was particularly strong for disease-free survival and translated into overall survival as well. In the multivariate analysis, we found that COX-2 expression was related to disease-free survival, with an estimated relative risk of 1.9 and 95% confidence limits ranging from 1.00 to 3.59. Despite the lack in precision of the point estimate due to the small sample size, the current results suggest that COX-2 expression was an independent prognostic factor for unfavorable outcome in patients with primary breast carcinoma in our study cohort. Taken together, the data from our study as well as other studies provide a basis for the translation of preclinical research into prospective clinical studies investigating the role of COX-2 as a prognostic factor as well as the impact of inhibition of COX-2 in adjuvant treatment for patients with breast carcinoma.
It has been shown previously that treatment with COX-2 inhibitors reduces incidence and growth of breast carcinomas in animal tumor models.18 In a rat mammary carcinogenesis model, celecoxib reduced mammary tumor development19 and reduced the volume of preexisting tumors.20 In animal models, selective COX-2 inhibitors were more effective for the prevention of experimental mammary tumors compared with the nonspecific COX inhibitor ibuprofen,19 which is consistent with our results showing a prognostic role for COX-2, but not for COX-1, in human breast carcinoma. These results suggest that selective COX-2 inhibitors may be more useful in adjuvant tumor therapy compared with COX-1 specific inhibitors.
We did not detect any expression of COX-2 in samples of benign breast disease. In contrast, expression of COX-2 was found in benign breast tissue adjacent to breast carcinoma tissue (data not shown). In two previous studies, Half et al. observed the expression of COX-2 in 81% of normal breast tissue adjacent to carcinoma tissue,9 whereas Soslow et al. did not detect the expression of COX-2 in normal breast tissues that were sampled at least 1 cm from breast carcinoma tissues.8 Based on these results, it will be interesting for further studies to analyze the interaction of stromal cells, tumor cells, and adjacent normal tissue and their contribution to the regulation COX-2 expression in the tumor microenvironment. It has been shown in colon carcinoma cells that c-erb B2 (HER2) is involved in the regulation of COX-2 overexpression.21 For breast carcinoma, the results still are controversial. In one study, a significant correlation between COX-2 and c-erb B2 was found.10 In contrast, we did not observe this correlation in our study. Similar to our results, two other studies did not find an association between c-erb B2 and COX-2.9, 11
The mechanisms by which COX-2 contributes to the poor prognosis of patients with breast carcinoma have not been elucidated completely to date. COX-2 expression has been linked to tumor invasion. This is supported by our observation that both lymph node metastasis and histologically detectable angioinvasion are correlated with increased expression of COX-2. Other factors involved may be the induction of angiogenesis as well as the suppression of apoptosis by COX-2-related pathways.11 Further studies will be required to determine whether COX-2 inhibitors may be useful for therapy or for the prevention of breast carcinoma. Possible indications may include primary chemoprevention of breast carcinoma (e.g., in patients with familial breast disease), adjuvant therapy for patients with breast carcinoma, or secondary prevention of recurrent disease.