Cyclin E Expression and Outcome in Pancreatic Ductal Adenocarcinoma

Pancreatic cancer is the fourth leading cause of cancer death in men and women in Western societies, with a 5-year survival rate of <10% (1). Pancreatic cancer presents at an advanced stage, and, as a result, only 10% to 20% of patients are suitable for surgical treatment at the time of presentation (1). Clinical management of these patients is complicated by inconsistencies in the influence of conventional clinicopathologic variables on outcome, suggesting that some of these variables lack accuracy. In addition, preoperative assessment of some variables, such as lymph node metastases, is difficult. Whereas in other cancers, assessment of aberrations in gene expression that cosegregate with therapeutic response and outcome is being adopted to increase predictive power (e.g., estrogen receptor and HER2/neu, and potentially cyclin E in breast cancer), there remain no molecular markers of proven clinical utility in pancreatic cancer. This highlights the need to identify novel molecular markers of prognosis relevant to pancreatic cancer that may also have diagnostic and therapeutic utility.

Dysregulation of the normal cell cycle machinery is integral to the neoplastic process and there is now compelling evidence that the development and progression of most human cancers is associated with a high frequency of abnormalities in the retinoblastoma pathway that controls G₁-S phase progression (2). The elements in the retinoblastoma pathway that have been implicated in cancer include retinoblastoma itself, a tumor suppressor that in its underphosphorylated state represses the transcription of genes necessary for cell cycle progression, the cyclin-dependent kinases (CDK) that phosphorylate retinoblastoma during G₁ phase, and the cyclins and CDK inhibitors that regulate CDK activity. Cdk4 and Cdk6 are activated by the D-type cyclins, cyclins D1, D2, and D3, and initially phosphorylate retinoblastoma during G₁. Subsequent retinoblastoma phosphorylation by cyclin E-Cdk2 then relieves retinoblastoma inhibition of cell cycle progression and allows initiation of DNA synthesis. Recent studies have identified functions of cyclin E in addition to Cdk2 activation and retinoblastoma phosphorylation that may contribute to tumorigenesis; these include initiation of DNA replication, genomic instability, and centrosome amplification (3). The CDKs are regulated at multiple levels, including regulation of the abundance of the activating cyclins and two families of endogenous low molecular weight inhibitors. The INK4 family of CDK inhibitors includes p16^INK4A and specifically targets cyclin D–associated CDKs. The CIP/KIP family includes p27^Kip1 and p21^WAF1/Cip1, which preferentially inhibit the activity of cyclin E-Cdk2 complexes. Many common cancers display deletion or mutation of retinoblastoma, overexpression of cyclins D1 or E, or reduced expression of p16^INK4A or p27^Kip1 (2). In pancreatic ductal adenocarcinomas, alterations in cell cycle regulators (cyclins D1 and D3, p16^INK4A, p21^WAF1/CIP1, and p27^Kip1) have been reported at high frequencies (4) and are apparent in the precursor lesions of pancreatic cancer—pancreatic intraepithelial neoplasia (5-8) and intraductal papillary mucinous neoplasms (9, 10)—but do not reliably cosegregate with survival. Despite the association between aberrant cyclin E expression and outcome in other cancers (11-13), in particular the strong association of high cyclin E expression and poor survival in breast cancer (12, 13), the influence of cyclin E on outcome in pancreatic cancer has not been addressed.

Here, we report a strong association between high cyclin E expression and poor outcome in pancreatic cancer. Furthermore, high cyclin E expression was an independent prognostic factor in resected pancreatic cancer that was a better indicator of patient outcome than traditional clinicopathologic variables.

We identified a cohort of patients with a diagnosis of pancreatic ductal adenocarcinoma that underwent pancreatic resection or biopsy at Westmead Hospital, Concord Hospital, Royal Prince Alfred Hospital, or the St Vincent's Hospital Campus in Sydney, Australia, between 1972 and 2001. Multicenter ethical approval for data collection and tissue use was granted by the St Vincent's Hospital/University of New South Wales, and Royal Prince Alfred Hospital Ethics Committees, as well as specific approval from the Westmead Hospital and Concord Hospital Ethics Committees. Cyclin E and p27^Kip1 expression were evaluated in 118 patients. The clinicopathologic and survival data presented in Table 1 are confined to the whole cohort of 118 patients who had either pancreatic resection or biopsy, and a subgroup of 75 patients treated with operative resection. Eighty-two of this cohort were included in a previously described group of 139 patients examining prognostic factors in pancreatic cancer (14), whereas the remaining 36 have been accrued since July 1999.

Clinical variables, including sex, age at diagnosis, preoperative assessment of disease state, and type of operative procedure, were gathered retrospectively from patient records. Pathologic findings, including tumor size, Unio Internationale Contra Cancrum T stage, and lymph node status were obtained from the pathologists' original report. In addition, microscopic findings (tumor type, degree of differentiation, perineural invasion, and margin status) were independently reassessed by a pathologist (J.G.K.). Date and cause of death was obtained on a return-to-notifier basis from The Cancer Council NSW.

Seven pancreatic cancer tissue arrays consisting of 2-mm-diameter tissue core biopsies were constructed from the archival tissue. Four-micrometer sections of the tissue arrays were dewaxed in xylene and rehydrated through graded alcohol concentrations. Antigen retrieval was achieved using EDTA buffer solution (pH 8.0) at high pressure in a microwave for 30 minutes. Slides were then transferred to a DAKO autostainer (DAKO Corporation, Carpinteria, CA) and endogenous peroxidase activity was quenched by 3% hydrogen peroxide treatment before incubation with mouse monoclonal antibody against cyclin E (clone 13A3, Novocastra, Newcastle-upon-Tyne, United Kingdom) or p27^Kip1 (clone 57, Transduction Laboratories, Lexington, KY). For evaluation of cyclin E protein expression, MDA-MB-157 human breast cancer cells (which display amplification and overexpression of cyclin E; ref. 15) and normal placenta were used as positive controls, and normal testis as a negative tissue control. Epithelial nuclei at the bases of normal duodenal crypts stained strongly, providing internal positive controls within the tissue arrays. IgG2a-negative mouse serum was used as a technical control to show antibody specificity. The positive control for p27^Kip1 staining was normal skeletal muscle, whereas normal spleen was used as a negative control. IgG1-negative mouse serum served as a technical control to show antibody specificity. The primary antibody was visualized using the DAKO Envision+ secondary detection system followed by color development using 3,3-diaminobenzidine (DAKO). Sections were counterstained using hematoxylin.

Up to four separate cores of pancreatic cancer were examined per patient. Staining was assessed by two independent blinded observers for each case (D.A.S. and J.G.K.). Standardization of scoring was achieved by comparison of scores between observers and by multiviewer microscope conferencing, where any discrepancies were resolved by consensus. The following criteria were used to dichotomize each antigen: Cyclin E expression was considered high if >10% of nuclei were positive; p27^Kip1 expression was considered high if ≥5% of nuclei were stained. Cyclin E expression scores clustered in two groups with the majority of scores ≤7%, and the remainder >15%. The bimodal distribution had a mean of 14% and a median of 5%. Based on this distribution of scores, a cut point of 10% was selected to discriminate the two groups. For p27^Kip1, published studies in pancreatic cancer have used either >1% or >5% (16, 17); we chose ≥5% as the threshold because this was closer to the median score for this cohort. When >1 core was present for a particular cancer, the highest scoring core was used in the analysis.

The association between cyclin E or p27^Kip1 protein expression and clinicopathologic variables, including clinical stage, degree of differentiation, operative resection, perineural invasion, tumor size, lymph node invasion, and surgical margin involvement were evaluated using nonparametric statistical tests (Kruskal-Wallis or Mann-Whitney).

Both univariate and multivariate analyses were done to assess the association of cyclin E and survival in relation to covariates using Statview 5.0 Software (Abacus Systems, Berkeley, CA). P < 0.05 was considered to be statistically significant. The outcome variables were assessed as time-to-event, which was defined as the difference between the time of diagnosis and the time of death.

For univariate analysis, event-survival curves were constructed using the Kaplan-Meier method for each cyclin E category. The differences in survival times between categories were compared using the two-tailed, log-rank statistic. In addition, survival curves for the coexpression of cyclin E with p27^Kip1 were plotted. Cox proportional hazards models were used to estimate hazard ratio (and its 95% confidence interval) associated with each risk factor and covariate. Those factors that were prognostic on univariate analysis were then assessed in multivariable models to identify factors that were independently prognostic.

The cohort consisted of 118 patients with a histologic diagnosis of pancreatic ductal adenocarcinoma (64 males and 54 females). Clinicopathologic and survival data for these patients are presented in Table 1. The mean age at diagnosis was 64.8 years (range 34.4-89.7 years). Tissue was available from 75 patients who had undergone pancreatic resections and 43 who had intraoperative biopsies. A large proportion of the cancers were either moderately (66 of 118, 56%) or poorly (43 of 118, 37%) differentiated with only 9 (7%) well-differentiated tumors. Well-differentiated and moderately differentiated tumors were grouped together for analysis. The majority of patients had advanced stage disease (23% Unio Internationale Contra Cancrum stage I, 9% stage II, 59% stage III, and 9% stage IV). Stage I and II tumors were grouped together for analysis and compared with stage III and IV tumors, which were also grouped together.

The median follow-up for the cohort was 8.5 months (range 0-117.4 months). Of the 118 patients, seven patients were alive at the census date (September 21, 2002), 105 had died from pancreatic cancer, 2 due to other causes (postoperative deaths), and 4 were lost to follow-up. The overall median survival was 8.5 months and the disease-specific survival was 8.7 months. The overall disease-specific 1-year survival rate was 35%, whereas only one patient survived longer than 5 years. The actuarial 5-year survival rate for patients who underwent resection was 10%. On univariate analysis for the whole cohort, those patients that underwent pancreatectomy survived longer (log-rank P < 0.0001) as did those with a lower clinical stage (log-rank P < 0.0001) and those with non–poorly differentiated tumors (log-rank P = 0.008; Table 1; Fig. 1A-C). For those patients who underwent pancreatectomy, those with a tumor size of <20 mm survived longer (log-rank P = 0.04), as did those with surgical margin clear of tumor (log-rank P = 0.002) and those without lymph node involvement (log-rank P = 0.02; Table 1; Fig. 2A-C). However, neither degree of differentiation (log-rank P = 0.053) nor perineural invasion (log-rank P = 0.11) influenced survival.

Nuclear cyclin E immunostaining was identified in 101 of 118 patients; the remainder had no detectable staining (Fig. 3A-B). Beyond a threshold of 10%, high expression of cyclin E was detected in 39 of 118 (33%) of the whole cohort and 24 of 75 (32%) of the resected cohort. On univariate survival analysis, high cyclin E expression was associated with shortened survival in the whole cohort (log-rank P = 0.005; hazard ratio, 1.8; 95% confidence interval, 1.2-2.7) with a median survival of 6.4 months for patients with high cyclin E expression, compared with 9.8 months for those with low cyclin E expression. Similarly, high cyclin E expression was associated with reduced survival in the patients who underwent resection (median survival 8.5 versus 14.2 months, log-rank P = 0.03; hazard ratio, 1.8; 95% confidence interval, 1.1-3.1; Table 1; Figs. 1D and 2D).

Cyclin E did not cosegregate with any clinicopathologic variables assessed (data not shown). We next did multivariate analysis on the whole and resected cohorts, including all clinicopathologic variables that were significant predictors of survival on univariate analysis. Cyclin E overexpression was an independent marker of prognosis in both the whole cohort and the resected subgroup (Table 2). Other independent predictors of poor outcome in the whole cohort were poor differentiation, advanced clinical stage (stages III or IV versus stages I or II), and operative biopsy versus resection. In the resected subgroup, in addition to high cyclin E expression, tumor size >20 mm and positive surgical margins were independent markers of poor prognosis. In Table 2, group A shows the resolved model for the whole cohort, whereas group B shows the multivariate model for resected tumors before resolution to the final model shown in group C.

Because cyclin E functionally interacts with p27^Kip1 and this interaction seems to be prognostically significant in breast cancer (11-13), we evaluated the influence of cyclin E and p27^Kip1 coexpression on outcome. Low p27^Kip1 expression alone, i.e., <5% of nuclei positive for p27^Kip1, did not correlate with any clinical or pathologic variables, nor was it associated with survival in either the whole cohort (P = 0.42) or in those patients that underwent resection (P = 0.38). Differential expression of p27^Kip1 did not significantly alter the association of cyclin E with survival in patients with high cyclin E expression (log-rank P = 0.73) or low cyclin E expression (log-rank P = 0.93).

In this study, we present the first report concerning cyclin E expression and outcome in patients with pancreatic cancer. High cyclin E expression was associated with a poor prognosis on univariate analysis, and was the most influential prognostic factor on multivariate analysis compared with clinicopathologic variables in patients that underwent pancreatectomy. These findings suggest a potentially important role for cyclin E in the clinical behavior of pancreatic cancer supporting similar data seen in breast cancer where high cyclin E expression is a powerful predictor of outcome (12, 13). Hence, cyclin E expression is a potentially clinically useful prognostic marker for pancreatic cancer creating scope for the application in pancreatic cancer of novel therapeutic strategies being developed that target aberrant cyclin E.

Numerous studies have assessed the prognostic value of clinicopathologic variables in pancreatic cancer (18-22). Only involved surgical margins and large tumor size are consistently associated with poor patient outcome (14). These factors are either difficult to identify or indeterminable preoperatively. The prognostic significance of molecular markers in pancreatic cancer, including growth factors and their receptors, cell cycle regulators, oncogenes, tumor-suppressor genes, apoptotic factors, angiogenic factors, and stromal factors, have been assessed (14, 23, 24), but at present have limited clinical utility. Given the importance of cell cycle deregulation in carcinogenesis, and that aberrant expression of any other cell cycle regulatory molecules important in pancreatic cancer (cyclin D1, p53, p16^INK4A, p21^WAF1/CIP1, and p27^Kip1) has not been reliably associated with outcome to date (reviewed in ref. 23),⁵

Oncogenic effects of cyclin E are usually attributed to its ability to promote cell cycle progression through phosphorylation of retinoblastoma but recent data provide increasing evidence for induction of genetic instability as a likely contributor (3). Cyclin E overexpression in cultured cells leads to chromosome loss, apparently via generalized chromosomal instability (25). Potential mechanisms for this effect include interference with the assembly of a prereplication complex at origins of replication (26), and centrosome amplification, which can contribute to aneuploidy by induction of mitotic spindle defects (27). Aneuploidy has been reported as a poor prognostic factor in some studies (28), and it would be of interest to examine the relationship between aneuploidy and cyclin E expression.

Three previous studies have examined cyclin E expression in pancreatic cancer. The different cut points for staining scores between all the studies make comparisons difficult. Schraml et al. (29) examined the expression of cyclin E using immunohistochemistry in 128 tumor types and identified high expression of cyclin E in one of eight pancreatic cancer samples. Although amplification of cyclin E using fluorescence in situ hybridization was seen in some tumor types, it was not detected in their samples of pancreatic cancer. Other studies, both using a cut point of 5%, reported higher rates of cyclin E overexpression in smaller cohorts: Al-Aynati et al. (7) identified cyclin E overexpression in 75% of 38 pancreatic cancer samples, and Yue et al. (30) reported cyclin E overexpression in 69% of 32 pancreatic cancers. At a cut point of 10%, we identified high cyclin E expression in 33% of 118 pancreatic cancers. It is likely that these differences in incidence are due to the different cut points used, rather than a true difference in cyclin E expression between cohorts, with our large cohort containing sufficient numbers to detect a bimodal distribution, which, in turn, cosegregated with outcome. Only one study has assessed cyclin E expression in pancreatic intraepithelial neoplasia (8), the precursor lesion of pancreatic cancer, where cyclin E overexpression was seen in 25% of advanced (PanIN-3) lesions (7).

In summary, we have shown that high cyclin E expression is an independent prognostic factor in pancreatic cancer. In patients who underwent pancreatectomy, high cyclin E expression was a superior predictor of survival than clinicopathologic variables, including tumor size, margin involvement, and lymph node status. Hence, cyclin E is a potentially clinically useful prognostic marker in pancreatic cancer, suggesting a simple, well-described test that can be implemented in the routine pathology laboratory in a disease where there are few useful markers of prognosis. If confirmed in other independent cohorts, these data create scope for the potential application of novel therapies targeting aberrant cyclin E, a marker of poor prognosis cancers, in pancreatic cancer.

We thank Dr. Catherine Langusch for her help with the maintenance of the Garvan Institute of Medical Research Pancreatic Cancer tissue bank and clinical database.