COX-2 overexpression in resected pancreatic head adenocarcinomas correlates with favourable prognosis

The study included 230 consecutive patients (103 women and 127 men) undergoing a standard Whipple’s procedure for adenocarcinoma with curative intent 1998 -2011 at Oslo University Hospital, Rikshospitalet. The study was approved by the Regional Committee for Medical and Health Research Ethical for Southern Norway.

Standard demographic, clinicopathological, and tumour-specific data were collected retrospectively from hospital records. Overall survival data was obtained from the Norwegian Population Registry, updated June 20, 2013. Since all Norwegian inhabitants receive a unique personal identification number, no patients were lost to follow-up in the present study. Patients were followed until death or censored after maximum five years (60 months). By the end of the study 177 patients were dead. Median follow-up for the remaining 53 patients was 62 months (interquartile range 29 -119 months). Perioperative death (defined as death within 30 days of operation) was included in the analyses (four patients). Analysis excluding perioperative death gave similar results. None of the patients received preoperative chemotherapy or chemoradiotherapy. From 2008, adjuvant chemotherapy with 5-fluororuracil was recommended for eligible patients operated for pancreatic cancer. Thirty-nine percent of the patients (13 of 33) operated in this period received adjuvant chemotherapy (5-FU-based in 11 patients, 2 patients received gemcitabine).

The resection specimens were examined according to a standardized protocol as described previously [1, 35]. All registered parameters of the prospectively collected data base, including anatomic site of tumour origin, where later reevaluated by slide review [1]. The histological type of differentiation was evaluated and all tumours were classified either as intestinal or pancreatobiliary type [3, 36]. In brief, pancreatobiliary tumours typically have simple or branching glands and small solid nests of cells surrounded by a desmoplastic stroma, and have cuboideal to low columnar epithelium arranged in a single layer and the nuclei are rounded but with marked variation in size and shape from one cell to the next. Intestinal tumours typically resembled colon cancer, have tall and often pseudostratified columnar epithelium with oval nuclei located in the more basal aspect of the cytoplasm, and there may also often be presence of mucin [36, 37].

Formalin-fixed, paraffin-embedded tissue was sectioned (3 μm), dried at 60°C, and processed in a Ventana BenchMark Ultra machine (Ventana Medical Systems Inc. (Tucson Arizona USA). Slides were incubated with monoclonal anti-COX-2 antibodies (Thermo Fischer Scientific rabbit), Universal Alkaline Phosphatase Red Detection Kit (Ultra View 760-501) and αSMA (Dako M.0851), DAB (Ultra View 760-500). Additional immunostaining on duplicates of twenty slides was performed with monoclonal COX-2 mouse antibody Invitrogen (Camarillo, CA, USA). Slides were counterstained with haematoxylin, fixed, mounted and analyzed using an inverted light microscope (Olympus, Center Valley, PA, USA).

Immunohistochemistry was performed on whole tumour slices, which were assessed without prior knowledge of the clinical and pathological parameters. In each section, five different representative high-power fields (100×) with tumour infiltration were selected and examined by light microscopy. The intensity of staining was estimated on a scale from 1-3 (1-negative, 2-moderate, 3-strong). Cells were considered positive only if COX-2 intensity was moderate or strong. The extent of the immunolabeling was assessed as the percentage of positively stained tumour cells and was expressed on the scale from 1-3 where 1 represented less than 10% cells stained, 2 represented 10-50% and 3 over 50%. Since COX-2 demonstrated considerable heterogeneity within individual cases, the final immunoscore was obtained as the average of the numeric scores for five high-power fields of each case considered positive in intensity scoring. Based on histograms of the staining for all tumours, the optimal cut-off value for discrimination between negative and positive staining was found to be 1.4. Islets of Langerhans and mucosa of the duodenum were moderately to strongly positive for COX-2, including those tumours with no COX-2 expression, and served as internal controls. Identical sections with omission of the primary antibody were used as negative controls. To test the validity of the Thermo antibody used for the study cohort, we performed additional immunostaining with a different monoclonal COX-2 mouse antibody, Invitrogen (Camarillo, CA, USA), on duplicates of twenty pancreatic cancer slides from the study cohort. The results were identical (Figure 1a and e). As Thermo antibody was not suitable for western blotting (producer recommendation), only the Invitrogen antibody was subjected to analysis by western blotting. The results showed a highly specific bond for COX-2 (Figure 1f).

Almost half of study specimens (44%) were evaluated independently by two examiners (EP and AS) and kappa interobserver was 0.73, indicating substantial agreement (95% CI 0.6-0.9).

Associations between variables were examined using Chi-square test, Fisher’s exact test and Mann-Whitney test. Continuous variables were reported as median with corresponding range or interquartile range (IQR). Unadjusted survival analysis was performed using the Kaplan-Meier method, comparing curves using log-rank test. Multivariable Cox regression analysis was used for adjusted survival analysis. Possible interactions were evaluated by inclusion of an interaction term in the models. For all tests, a two-sided p < 0.05 was considered statistically significant. Statistical analyses were performed in SPSS 19 for Windows (SPSS Inc., Chicago, IL).

COX-2 staining was very similar in all three tumour types, with a positivity rate of 71% in PC, 72% in DBC, and 77% in AC. The COX-2 expression was detected in the cytoplasm of cancer cells in all three types of adenocarcinoma. No COX-2 immunostaining was detected in the stroma cells (Figure 1a,b, and e). The expression pattern showed heterogeneity both among different tumours and within the individual tumour, as areas with moderate to strong staining coexisted with negative areas within the same tumour (Figure 1c). Islet cells expressed moderately to strong COX-2 staining in all cases including those with no COX-2 expression in the tumour (Figure 1d). Irrespective of tumour origin, overall patient survival was more favourable in COX-2 positive than COX-2 negative tumours (Figure 2a-c). This was particularly prominent in AC (p = 0.011) and PC (p = 0.043) whereas the same tendency was seen in DBC although not reaching significance (p = 0.06). COX-2 expression varied according to the type of histological differentiation. In tumours with pancreatobiliary type of differentiation, two thirds of the tumours were COX-2 positive irrespective of anatomical origin (67%, 69%, and 68% in AC, DBC and PC, respectively). However there was no significant difference in overall survival when comparing COX-2 positive and negative tumours in this group (Figure 2d-f). All PC and DBC tumours with intestinal type of differentiation were COX-2 positive whereas 84% of the intestinal AC tumours expressed COX-2. The survival data of the intestinal AC tumours showed a favourable prognosis for patients with tumours expressing COX-2 (p = 0.003) (Figure 2g-i).

COX-2 expression status was compared across clinical parameters associated with survival in the subgroup consisting of the 92 patients resected for pancreatic adenocarcinoma. The median survival for patients with COX-2 positive tumours was 18 months (95% CI 14-22) as compared to 11 months (95% CI 9.6-12) for patients with COX-2 negative tumours (p = 0.043). COX-2 positive tumours were more likely associated with high degree of differentiation (p = 0.002) and with intestinal type of differentiation, although, the latter did not reach significance (p = 0.099) (Table 1) probably due to the low number of tumours of the intestinal differentiation type.

There was no association with COX-2 positivity and R-status, lymph node ratio (LNR), lymph node status, tumour diameter, T classification, and vascular or perineural infiltration (Table 1). Since tumours expressing COX-2 were significantly more likely to be highly differentiated than COX-2 negative tumours, the joint effects of COX-2 status and differentiation grade on survival were assessed by Kaplan-Meier analysis, stratifying for COX-2 status (positive vs. negative) and differentiation grade (grade 1 and 2 vs. grade 3 and 4) (Figure 3a). Patients whose tumours did not express COX-2 and had a low differentiation grade (grade 3 and 4) had significantly poorer survival than the other three groups (p = 0.006).

In a previous report we found that LNR independently predicted prognosis in a multivariate model for survival in resected pancreatic cancer [38]. We thus also examined the joint effects of COX-2 status and LNR, and found that patients with COX-2 negative tumours and LNR >0.2 had significantly worst prognosis (p < 0.001) (Figure 3b).

In a multivariate analysis model including COX-2 expression, LNR, tumour size, margin status, vascular and perineural infiltration, COX-2 negative tumours and LNR > 0.2 independently predicted poor prognosis (Table 2). Since there was a strong correlation between COX-2 expression and differentiation grade (p = 0.002) it was not possible to include differentiation grade in the same model.

Only a minority of the patients received adjuvant chemotherapy. Although the numbers are small, there was no difference in survival between patients with COX-2 positive and COX-2 negative tumours who received adjuvant treatment.