Cyclooxygenase-1 (COX-1) and cyclooxygenase-2 (COX-2) are prostaglandin endoperoxide synthases, which are rate limiting enzymes in the conversion of the fatty acid arachidonic acid into physiologically active eicosanoids such as prostaglandin, thromboxane and prostacyclin.1 COX-2 is the inducible form of COX that is frequently elevated in cancer tissues.2, 3, 4 Although the mechanism of COX-2 action in carcinogenesis or progression is not well established, COX-2 inhibitors have been shown to be chemopreventive in cancer, specifically in colorectal cancer.5, 6, 7
The expression and role of COX-2 in prostate cancer has been a topic of several reports over the decade.8, 9 There exists some controversy over the expression, detection and the putative role of COX-2 in prostate carcinogenesis and progression. While early studies reported high-level expression of COX-2 in prostate, findings from subsequent studies have been mixed.10, 11, 12 Typically, later studies reported low or no expression in normal, benign prostatic hyperplasia or low-grade cancer tissues but elevated levels in prostatic intraepithelial neoplasia (PIN) and high-grade cancer.13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26 A prominent study, however, discounted COX-2 specific staining in all epithelial tissues and claimed that COX-2 is expressed only in infiltrating lymphocytes and proliferative inflammatory atrophy.27 Nonetheless, several studies, using experimental models and established prostate cancer cell lines, have established a chemopreventive and antitumor activity of COX-2 inhibitors, implying the putative role of COX-2 in prostate carcinogenesis.28, 29, 30, 31, 32, 33, 34, 35, 36, 37
A direct demonstration of potential role of COX-2 in advanced prostate cancer was recently demonstrated by one of us.38 Using an antisense COX-2 cDNA construct (tetracycline-inducible model) to suppress COX-2 expression in PC-3ML cell line (a highly metastatic androgen unresponsive prostate cancer line), we demonstrated tumor-enhancing function of COX-2. In a related study, we also showed chemosensitization of prostate tumor xenografts by a COX-2 inhibitor celecoxib, indicating a potential role of COX-2 in both carcinogenesis and resistance to treatment.39
Until recently, elevated COX-2 levels were shown to be related to carcinogenesis but not progression as illustrated by its elevated levels in preneoplastic colon polyps, inflammatory atrophy and PIN. Moreover, the significance of the elevated expression of COX-2 was believed to be secondary to the elevation of several COX-2 inducing signaling mechanisms, notably the activated levels of Akt (phospho-AKT)40, 41 and nuclear-factor kappa B (NFkB) that activates COX-2 promoter and transcription.40, 41, 42 The significance of COX-2 overexpression in disease progression, especially in prostate cancer, has evaded rigorous testing until now. This ambivalence is due to earlier studies that showed uneven labeling of COX-2 in prostate cancer tissues in archival samples. However, if COX-2 expression in prostate cancer specimens correlates with disease recurrence/progression, it may explain the conflicting results reported in various studies regarding COX-2 expression.10, 28 Investigating such a correlation is also important to understand prostate cancer progression because at least in experimental models of prostate cancer, COX-2 overexpression leads to increase in aggression of tumor cells, higher secretion of angiogenic factors and metastasis.43, 44, 45, 46 In this study, we investigated COX-2 expression in archival radical prostatectomy specimens from random samples of 60 prostate cancer patients on whom a minimum 62-month follow-up was available. Our results show that COX-2 expression in prostate cancer cells is an independent prognostic indicator for predicting biochemical recurrence.
Specimens and study patients
Sixty prostate cancer specimens were obtained from patients who underwent radical retropubic prostatectomy between 1992 and 1995 at the University of Miami Medical Center, Miami, Florida. These patients did not receive neoadjuvant hormonal therapy, and 58 patients were node negative. The minimum available follow-up on all patients was 62 months. The study was conducted under a protocol approved by the University of Miami's Institutional Review Board for human subjects' research. Of the 60 patients, 23 patients had biochemical or clinical recurrence (mean time to recurrence: 38.2 months; range: 1–121 months), and 37 patients were free of disease recurrence (mean follow-up: 95 months; range: 62–142 months). Biochemical recurrence was defined as a PSA level ≥ 0.4 ng/ml in 2 successive measurements after radical prostatectomy, in which case the first date of elevated PSA level was considered as the date of recurrence. The patient characteristics with respect to age, preoperative PSA and tumor (i.e., Gleason sum, stage, margin, extraprostatic extension (EPE) and seminal vesicle (SV) invasion) are shown in Table I. The specimens chosen for staining were those that represent true major Gleason score recorded by the pathologist on service. Thus, if the pathologist recorded Gleason score as 3+4, paraffin block representing 3+4 was chosen.
Table I. Pre- and Postoperative Parameters of the Study Patients
EPE, extraprostatic extension of tumor; SV, seminal vesicle invasion.
Two patients who had biochemical recurrence had positive lymph nodes.
Three of the patients had only designation of tumor stage as T2.
For all specimens, paraffin embedded blocks containing prostate cancer tissues representing the major Gleason score was selected. The block selection was made by one of the authors, a certified pathologist (FC). From each block, 6 slides were prepared and 2 were used for COX-2 staining. The other slides were used to either optimize COX-2 antibody concentration for staining, for determining nonspecific staining, for repeating the staining in case of any discrepancy or for comparing COX-2 staining, using antibodies from different companies (as discussed later). The specimen slides were deparaffinized, rehydrated and treated with an antigen retrieval solution (Dako USA, Carpentaria, CA). The slides were incubated with a rabbit polyclonal antihuman COX-2 IgG, PG-27b (200-fold dilution of the antiserum; Oxford Biomedical Res., Oxford, MI) at 4°C for 16 hr. Following incubation, the slides were washed and incubated with a linking solution containing a biotinylated swine antirabbit IgG (Dako LSAB kit, Dako USA) at room temperature for 30 min. The slides were then treated with streptavidin peroxidase and DAB chromogen. The slides were counterstained with hematoxylin, dehydrated and mounted.
Initially, about 20 specimens were stained with anti-COX-2 antibodies from Santa Cruz Biotechnology (Santa Cruz, CA) and Cayman Chemical (Ann Arbor, MI) along with the PG-27b antibody to evaluate the specificity of different antibodies. We also evaluated 10 specimens for COX-2 staining, using a nonbiotin-based Envision kit, as per the manufacturer's protocol (DAKO Laboratories).
COX-2 staining of tumor cells in each slide was initially graded as 0 to 3+. To account for heterogeneity in staining, the overall staining grade for each slide was assigned based on the staining intensity of the majority of the tumor tissue in the specimen. However, if ˜ 50% of the tumor cells in the section were assigned +1 staining, and the other 50% as 3+, the overall staining grade was 2+. If the staining distribution was, ˜ 50% of the tumor cells staining 2+ and the remaining staining as 3+, the overall staining inference assigned were 3+. The staining was later grouped as low- and high-grade staining. High-grade staining represented 2+ and 3+ staining, whereas low-grade staining included 0 and 1+ staining intensities. Normal-benign tissues near and away from tumor cells were also evaluated for staining intensity and graded as low- and high-staining. Two readers independently evaluated all slides in a blinded fashion. Out of the total 60 stained slides, there was discrepancy in 8 slides. These discrepancies were resolved by both readers reexamining those slides simultaneously. In addition, to check for the repeatability of the evaluation system, a 3rd reader, after familiarizing with the grading system, randomly picked 40 slides and graded them for staining intensity. The discrepancy in slide evaluations by the 3rd reader was <10%.
For COX-2 staining, high-grade staining was considered as a true positive if the patient had biochemical recurrence. Consequently, low-grade staining was considered as a true negative, if the patient had no biochemical recurrence. The sensitivity, specificity, accuracy, positive predictive value (PPV) and negative predictive value (NPV) for COX-2 staining inferences were calculated using the 2 × 2 contingency table (high-grade/low-grade staining and progressed/nonprogressed CaP patients) at 62, 72 and 100 months cut-off limits.47 The data on various biochemical, surgical and pathologic parameters, as well as COX-2 staining inferences, were analyzed by Cox proportional hazard model, using single variable analysis (univariate analysis) or step-wise selection (multivariate) analysis. Stratified Kaplan-Meier analyses were performed on the variables that were found to be significant in the multivariate Cox proportional hazard model. For PSA subset analysis, Mantael-Haenszel χ2 analysis or student's t test were used to determine statistical significance. Statistical analysis was carried out using the SAS software program (version 8.02; SAS Institute, Cary, NC).
Localization of COX-2 in prostate cancer tissues
Previous studies on the localization of COX-2 in prostate cancer cells or tissues have used anti-COX-2 antibodies from at least 3 different commercial sources i.e., Oxford Biomedical Research, Cayman Chemical and Santa Cruz Biotechnology.13, 16, 18 Among these antibodies, a 1:200 dilution of the PG-27b anti-COX-2 antibody from Oxford Biomedical Research was specific in localizing COX-2 in prostate cancer tissues and yielded optimal staining intensity. As we have reported previously, the level of COX-2 detected by this antibody (P27b) in PC-3ML cells transfected with antisense COX-2 cDNA or empty vector correlated with activity levels, using several techniques, including Immunofluorescence, western blotting and COX-2 activity ELISA.38 Also, as reported previously, we also determined the ability of PG27b to localize COX-2 in archival prostate tissues, using a nonbiotin-based immunohistochemical method (Envision Plus47). However, this method yielded only weak to no staining in any prostate glands (i.e., normal-benign or cancer) from recurred or nonrecurred patients. Therefore, we used PG27b antibody at 1: 200 dilution to stain all of the 60 archival prostate cancer tissues.
The 60 archival tissues used in this study were from patients with prostate cancer, who either had biochemical recurrence or did not, in a mean follow-up of up to 95 months. As shown in Figure 1a, very little COX-2 staining is observed in prostate cancer cells from 2 patients with either Gleason 7 (panel a) or Gleason 8 (panel c) prostate cancer, and who had no biochemical recurrence within 62 months. Among 43 patients with prostate cancer, who did not have biochemical recurrence in 62 months, specimens from 35 patients showed low-grade staining. Figure 1a also shows high-grade staining in prostate cancer cells of 2 different patients with either Gleason 7 (panel b) or Gleason 8 (panel d) prostate cancer, and who had biochemical recurrence within 62 months (mean time to recur 38 months). Among the 17 patients who had biochemical recurrence within 62 months, 14 had high-grade staining in their specimens.
We next examined COX-2 staining in normal-benign glands. As shown in Figure 1b, normal-benign glands that are away from prostate cancer cells have low-grade staining, regardless of whether the specimen was obtained from a patient who did not recur (panel a) or recurred within 62 months (panel b). However, the normal-benign gland that is adjacent to tumor cells shows high-grade staining, if the specimen was obtained from a patient who had recurred (panel d). The normal-benign glands adjacent to tumor cells do not stain for COX-2, if the specimen was obtained from a patient who did not recur (panel c). Out of the 37 specimens from nonrecurred patients, staining of normal glands was observed in 4 patients, whereas normal glands in 14 out of 23 recurred patients' specimens were stained. These results show that COX-2 expression in prostate cancer cells is related to disease progression and that normal prostate epithelial cells also express COX-2, if they are adjacent to prostate cancer cells that have malignant potential.
Determination of sensitivity, specificity, accuracy, PPV and NPV of COX-2 staining
On all patients, a minimum follow-up of 62 months was available (mean follow-up: 95 months; range: 62–142 months). We, therefore, determined the sensitivity, specificity, accuracy, PPV and NPV of COX-2 staining at 62 months, 72 months and 100 months of follow-up. As shown in Table II, the sensitivity of COX-2 staining ranges from 82 to 86% between 62 and 100 months of follow-up. Out of the 23 patients who recurred, one recurred after 121 months and the COX-2 staining in the specimen was low-grade. Thus, the overall sensitivity for all 23 patients at 121 months is 82.6%. The specificity of COX-2 staining at 62 months is 81% but it increases to 84% and 86.7% at 72 and 100 months, respectively. The increased specificity is because, 5 of the patients who were false positive at 62-months follow-up had a biochemical recurrence within 62–100 months of follow-up (Table II). Consequently, the accuracy of COX-2 staining also increases from 81.6% at 62 months to 86.5% at 100 months of follow-up. Similarly, the PPV of the COX-staining increases significantly from 62 months (PPV, 63.6%) to 100 months (PPV, 90.5%) of follow-up. The NPV of COX-2 staining decreases by about 10% at 100 months of follow-up, which might be due to the low number of patients in the nonrecurred category for that follow-up.
Table II. Sensitivity, Specificity, Accuracy, PPV and NPV of Cox-2 Staining Inferences
62 Months (%)
72 Months (%)
100 Months (%)
Note that 62 months, 72 months and 100 months were used as a cut point for determining biochemical recurrence.
Evaluating the prognostic potential of pre and postoperative parameters and COX-2 staining in prostate specimens
Since the patients in this cohort had variable follow-up between 62 and 142 months, we used the Cox proportional hazards model and single-parameter analysis to determine the prognostic significance of each of the preoperative (i.e., age, PSA and clinical stage) and postoperative (i.e., Gleason sum, margin, EPE and SV invasion) parameters, as well as staining inferences of COX-2. As shown in Table III, age and clinical stage are not significant in predicting biochemical recurrence. However, preoperative PSA, Gleason sum, margin status, EPE, SV invasion and COX-2 staining inferences significantly predict biochemical recurrence (Table III). Patients with Gleason sum ≥7 are at a greater risk for progression. In the univariate analysis, the hazard of developing biochemical recurrence in Gleason ≥7 patients (p = 0.0215; χ2: 5.288; hazard ratio = 1.41) was not higher than when all Gleason sums were analyzed together (Table III).
Table III. Univariate Analysis of Pre- and Postoperative Prognostic Parameters and IHC Staining Inferences
Statistically significant. CI, confidence interval. Cox proportional hazard model and single parameter analysis was used to determine the prognostic significance of preoperative (age, preoperative PSA and clinical stage) and postoperative (Gleason sum, margin +/−, EPE +/−, SV invasion +/−) parameters and COX-2 staining inferences.
To determine the smallest number of variables that could jointly predict biochemical recurrence in this cohort of patients, we used the Cox proportional hazards model and step-wise selection analysis. When age, preoperative PSA, clinical stage, Gleason sum (either overall or Gleason ≥ 7), EPE, SV invasion and COX-2 staining inference were included in the model, only preoperative PSA (p = 0.0036, hazard ratio/unit PSA change = 1.08) and COX-2 staining inference (p < 0.0001, hazard ratio = 16.442) reached statistical significance in predicting biochemical recurrence (Table IV). Furthermore, COX-2 inclusion in the analysis further classified the PSA group into high and low risk patients. It is noteworthy that although preoperative PSA distribution was relatively narrow in both groups, still preoperative PSA was selected as an independent prognostic indicator in the multivariate analysis.
Table IV. Multivariate Analysis of Pre- and Postoperative Prognostic Parameters and IHC Staining Inferences
Cox proportional hazard model and step-wise selection was used to determine which of the preoperative (i.e., age, PSA and clinical stage) and postoperative (i.e., Gleason sum, EPE, margin +/−, and SV invasion) parameters and Cox-2 staining inferences have independent prognostic significance. The significant parameters (p > 0.05) selected by the model are shown.
To demonstrate the joint effect of COX-2 and preoperative PSA on biochemical recurrence, we performed Kaplan-Meier analysis. Since PSA was a continuous estimate, with the median PSA level for the entire cohort of patients (n = 60) as 6.9 ng/ml, we divided the cohort into those with PSA levels ≤ 7 and > 7 ng/ml. As shown in Figure 2, individuals with COX-2 staining high and PSA > 7 ng/ml had the highest probability of recurrence, followed by those with COX-2 high and PSA ≤ 7 ng/ml. Individuals with low COX-2 staining and either PSA > 7 ng/ml or ≤ 7 ng/ml had the lowest probability of recurrence.
PSA subgroup analysis.
It has been suggested that biochemical recurrence before 24 months indicates systemic disease, whereas biochemical recurrence beyond 24 months suggests local recurrence.47 By Mantel-Haenszel χ2 analysis, we found that COX-2 staining inferences along with Gleason sum, preoperative PSA, EPE, SV invasion and margin status could distinguish between PSA recurrence before and after 24 months in a statistically significant manner (p < 0.0001; χ2, 16.7812), see Table V.
Table V. PSA Subgroup Analysis
The ability of age, PSA and Gleason grade to predict PSA recurrence within 24 months was determined using the t test for these continuous variables.
Mantel-Haenszel χ2 analysis was used to evaluate the ability of clinical stage, postoperative parameters and COX-2 expression to predict PSA recurrence within 24 months.
There are few reliable markers for accurate prediction of prostate cancer recurrence besides preoperative PSA and Gleason sum.48 However, nearly two thirds of patients diagnosed with prostate cancer have the preoperative PSA range 4–10 ng/ml, with T1C disease and a biopsy Gleason score of 5–7.47, 48, 49, 50, 51, 52 For such patients, one or a combination of accurate prognostic indicators could improve the physicians' ability to identify aggressive disease, so that more individualized treatments could be offered. In this study, preoperative PSA level and COX-2 expression were found to provide independent prognostic information.
In specimens from prostate cancer patients who later progressed, normal-benign prostate glands adjacent to highly stained tumor cells also showed high-level COX-2 expression, indicating a paracrine mechanism of COX-2 induction. Therefore, it is more likely that factors that induce COX-2 enzyme may also be potential markers for predicting progression. Identifying such a factor or factors is quite likely to be daunting because of the promiscuity of COX-2 induction. Several cytokines, chemokines and inflammatory enzymes are known to induce COX-2 expression.53, 54, 55
The data analysis presented above for COX-2 staining intensity show for the first time that COX-2 expression can independently predict prostate cancer recurrence. The sensitivity, specificity and accuracy of COX-2 staining to predict biochemical recurrence exceeded 80% at 62–100 months follow-up after radical prostatectomy. Furthermore, both univariate and multivariate analyses showed significant association of COX-2 overexpression and biochemical recurrence. Kaplan-Meier analysis shows that prostate cancer patients with high COX-2 expression and PSA > 7 ng/ml have the highest probability of disease progression. This explains why the multivariate analysis selected COX-2 staining inference and preoperative PSA as independent prognostic indicators. The data also show that prostate cancer patients with PSA > 7 ng/ml could be further classified based on the COX-2 staining in their prostate cancer specimens to predict biochemical recurrence. In this study, biochemical recurrence indicates the possibility of developing metastatic disease, since all of the study patients underwent radical prostatectomy, rising PSA levels would indicate either local or distant metastases. The high sensitivity, specificity and PPV (63–93%) of COX-2 expression point to the possibility that future treatment should consider inhibition of inflammatory pathways either at the time of initial treatment or together with androgen ablation during the treatment of residual disease.56
The findings presented in this report are consistent with studies conducted on preclinical models of prostate cancer. For example, Fujita et al45 reported that forced expression of COX-2 in a relatively indolent LNCaP tumor model increases tumor growth, angiogenesis and PSA secretion, without increasing androgen receptor levels or activity. Since, increased angiogenesis results in increased invasion and metastasis, increased COX-2 expression may be predictive of disease progression in prostate cancer patients. Our earlier work also showed that suppression of COX-2 by antisense cDNA transfection in an aggressive tumor model (PC-3ML) reduces tumor growth and increases apoptosis in vivo.38
Earlier studies on immunohistochemistry localization and semiquantitative estimation of COX-2 expression concluded that COX-2 expression has limited association with stage and disease severity (local invasion) but it strongly associates with tumor grade.10, 11, 12, 13, 14, 15 On contrary, some reports have dismissed any association with tumor grade and have suggested that COX-2 is expressed by infiltrating lymphocytes and macrophage, which occurs during inflammatory atrophy of the prostate.27 However, none of these reports analyzed COX-2 expression with respect to biochemical recurrence (i.e., disease progression). Consistent with earlier studies, we also did not find any correlation between COX-2 expression and tumor grade and stage (p > 0.05, χ2 analysis). In fact, the inclusion of COX-2 staining inferences in the multivariate model eliminated the prognostic significance of all clinical and pathologic parameters except preoperative PSA.
Although the expression of COX-2 with prostate cancer is controversial in several cancers, COX-2 expression has been shown to associate with disease progression, poor response to treatment and survival. For example, in non-Hodgkin's lymphoma, high COX-2 expression correlates with increased stage, lower survival and lower complete response rate to therapy.55 In breast cancer, COX-2 overexpression occurs in 43% of human invasive breast cancers and 63% of ductal carcinomas in situ.57, 58 Furthermore, high expression of COX-2 associates with decreased disease specific survival.61 In gastric cancer, COX-2 is expressed not only in cancer cells but also in precancerous lesions such as metaplastic and adenomatous cells, suggesting that COX-2 overexpression is an early event in gastric cancer development.59 In bladder cancer, COX-2 expression in carcinoma in situ significantly associates with disease recurrence and progression but it is not associated with survival.60, 61 However, among bladder cancer patients who receive systemic chemotherapy, strong COX-2 expression significantly correlates with poor overall survival.61 These studies support the notion that COX-2 expression and inflammatory processes are involved in cancer progression.
In summary, this study with up to 10-year follow-up shows that COX-2 expression in prostate cancer correlates with aggressive disease and is an independent prognostic indicator of biochemical recurrence. Validation of this study involving relatively small number of patients (n = 60) will establish the independent prognostic potential of COX-2 in predicting biochemical recurrence. The results of our study also suggest that the inhibition of inflammatory processes may be an effective means of controlling prostate cancer progression.