Overexpression of the Cell Cycle Inhibitor p16^INK4A in High-grade Prostatic Intraepithelial Neoplasia Predicts Early Relapse in Prostate Cancer Patients¹

Age-standardized incidence rates of PC³ have increased significantly as a result of widespread PSA testing. Of the men who are diagnosed with PC as a result of screening, a significant proportion may not benefit from aggressive treatment with its associated morbidity. However, identifying patients with PCs that have the potential to progress is difficult with current diagnostic tools that rely on clinical stage, biopsy Gleason grade, and pretreatment PSA levels (1, 2). The clinical management of patients diagnosed with HGPIN on prostate biopsy is also problematic. HGPIN is believed to be a precursor of invasive PC, and previous reports indicate that when HGPIN alone is diagnosed at initial biopsy, ∼50% of repeat biopsies contain carcinoma (3, 4). Because autopsy studies have demonstrated that PIN may precede cancer by several decades, the potential exists to identify molecular markers of outcome in HGPIN that would facilitate stratification of patients into discrete risk groups very early in the disease process and prior to the onset of carcinoma (5).

Considerable evidence now implicates aberrations in the expression and function of genes controlling G₁ to S phase progression of the cell cycle in the development and progression of almost all human cancers (6). The G₁ to S phase transition is controlled by the sequential transcriptional activation of cyclin genes and the consequent transient accumulation and activation of different cyclin/cyclin-dependent kinase complexes resulting in hyperphosphorylation of the retinoblastoma gene product, pRB (6). The cell cycle inhibitor p16^INK4A binds to Cdk4 and Cdk6 and inhibits their catalytic activity leading to reduced phosphorylation of pRB and G₁ cell cycle arrest (7, 8). p16^INK4A is frequently inactivated in human cancers (9, 10). In PC, loss of p16^INK4A function via homozygous deletion and hypermethylation of the INK4A promoter appear to be rare events (11, 12). However, a recent report found that overexpression of p16^INK4A protein in radical prostatectomies was associated with an earlier time to PSA relapse (13).

We studied a group of 206 patients, treated with RP for clinically localized PC at St. Vincent’s Hospital, Sydney between 1989 and 1996, for whom sufficient archival tissue and clinicopathological data were available. The demographic and clinical features of these patients are listed in Tables 1 and 2. Follow-up data, including outcome and time to recurrence were available for all of the 206 patients (mean follow-up, 73.4 months; median, 72.1 months; range, 4.9–139.3 months). Recurrence was defined by the following criteria: biochemical disease progression with a serum PSA concentration at or above 0.4 ng/ml rising over a 3-month period or local recurrence on digital rectal examination, confirmed by biopsy or by subsequent rise in PSA. Thirty-eight of the 206 patients in this group received NHT prior to surgery.

Paraffin blocks of surgical specimens were obtained from the archives of the Department of Anatomical Pathology at St. Vincent’s Hospital, Douglass Hanly Moir Pathology (Sydney, Australia) and Sugerman Hampson Macquarie Pathology (Sydney, Australia). The RP specimens were painted with ink and fixed in neutral-buffered formalin before 3-mm step sectioning and complete paraffin embedding. Pathological evaluation of all of the blocks from each prostate was performed by one of three histopathologists. Tumor stage was classified according to the TNM (tumor, nodes, metastases) staging system and Gleason grade. A block most representative of each cancer was selected for sectioning and diagnosis of each block was confirmed by examination of a routinely H&E-stained section juxtaposed to the section used for p16^INK4A immunostaining. Because the classification of HGPIN was not uniformly reported throughout the tissue evaluation period, the presence of HGPIN was reviewed for each slide after immunostaining for p16^INK4A by an independent histopathologist (C. S. L.) according to the criteria of McNeal and Bostwick (14). Of 206 specimens, 154 (75%) contained areas of HGPIN concurrent with carcinoma.

The reliability of the immunohistochemical assay was established by demonstrating comparable levels of p16^INK4A protein on immunoblotting and immunohistochemistry in a panel of breast cancer cell lines with known p16^INK4A status. The cell lines analyzed were: MDA-MB-468, SK-BR-3 and MDA-MB-157, which express high levels of p16^INK4A protein; MDA-MB-330 and MDA-MB-231, which contain homozygous deletion of the INK4A gene; and T-47D, which exhibits complete INK4A promoter methylation (15, 16).

Immunohistochemical staining for p16^INK4A was performed on routinely processed, paraffin-embedded tissue specimens. As a positive tissue control, a squamous cell carcinoma of the head and neck that has high levels of p16^INK4A expression was included (17). In addition, paraffin-embedded cell pellets of the breast cancer cell lines MDA-MB-157 and MDA-MB-231 were used as positive and negative controls, respectively. Four-μm sections of the tissue and control specimens were cut and mounted on Superfrost Plus adhesion slides (Lomb Scientific, Australia). Prior to staining, the sections were dewaxed and rehydrated and then unmasked in 0.01 m citrate buffer (pH 6.0). The mouse monoclonal anti-p16^INK4A antibody (Clone ZJ11; Neomarkers, CA) was applied to the sections, and the sections were incubated overnight at 4°C. A streptavidin-biotin peroxidase detection system was used according to the manufacturer’s instructions (Vectastain Elite kit; Vector laboratories, CA), with 3,3′-diaminobenzidine as substrate. Counterstaining was performed with Whitlock’s hematoxylin followed by Light green (BDH Laboratory Supplies, Poole, United Kingdom).

Immunostaining was assessed initially by one of the primary investigators (S. M. H., D. I. Q.) and subsequently reviewed independently by a histopathologist (C. S. L.) without knowledge of the patients’ outcome. For each slide, the percentage of nuclear p16^INK4A immunostaining within areas of HGPIN and carcinoma was expressed as the ratio of p16^INK4A-positive cells:total number of cells counted. Levels of p16^INK4A expression were classified as high (>1% nuclear immunoreactivity) or low (≤1% nuclear immunoreactivity). This cutoff was based on the mean levels of p16^INK4A expression detected in areas of normal prostate tissue, which was calculated as 1.5% in 154 cases in which normal prostate tissue was observed concurrent with areas of HGPIN and cancer. This compared with a mean score of 7.6% immunoreactivity in areas of HGPIN and 17.2% in areas of cancer. A 1% cutoff has been used previously in other published studies examining the prognostic potential of p16^INK4a expression in human cancers (Ref. 17; Fig. 1). Interobserver variability was tested using a paired t test and showed no difference between the scores (P = 0.128). Because a 5% cutoff was used in previous studies to stratify p16^INK4a expression in PCs (13), we also assessed relapse-free survival in patients with HGPIN stratified on the basis of ≤5% versus >5% p16^INK4A expression.

The primary outcome, disease specific relapse, was measured from the date of RP. Data were evaluated for disease relapse using the Kaplan-Meier method and log-rank test and by univariate and bivariate analyses in a Cox proportional hazards model for p16^INK4A status and other clinical and pathological predictors of outcome (18, 19). The multivariate model was produced by assessing p16^INK4A status in areas of cancer and HGPIN in the same specimens with other baseline covariates of clinical relevance: Gleason grade, pathological stage, margin status, NHT, and preoperative PSA, which were modeled as dichotomous or continuous variables as appropriate. The associations between p16^INK4A expression and discrete categorical variables were tested using the χ² test. Interobserver variability was tested using a paired t test. A P of <0.05 was required for significance. All the reported Ps are two-sided. All of the statistical analyses were performed using Statview 4.5 software (Abacus Systems, Berkeley, CA).

The levels of nuclear p16^INK4A expression were evaluated by immunohistochemistry in 206 PCs and classified as high (staining in >1% of cells) or low (staining ≤1% of cells) (Fig. 1). HGPIN was present in 154 of these patients, and p16^INK4A levels were high in 47 (31%) of these (Table 1). In the cancer, high levels of p16^INK4A expression were detected in 134 (65%) of 206 cases (Table 2).

Levels of p16^INK4A expression in areas of HGPIN and cancer were correlated with other tumor characteristics and risk factors (Tables 1 and 2). High p16^INK4A in HGPIN was positively correlated with pathological stage of concurrent carcinoma (P = 0.01), with 14 (22%) of 65 pT₂N₀ tumors showing elevated p16^INK4A expression compared with 19 (30%) of 63 pT₃N₀ tumors and 14 (54%) of 26 cases staged greater than pT_3CN₀ (Table 1). Increased p16^INK4A expression in HGPIN was also strongly associated with clinical stage (P = <0.001) with 7 (13%) of 53 T₁ tumors, 34 (37%) of 91 T₂ cases, and 6 (67%) of 9 T₃ tumors demonstrating p16^INK4A overexpression. Levels of p16^INK4A expression in HGPIN were not associated with pretreatment serum PSA levels or Gleason grade.

In the cancers, high p16^INK4A was associated with pathological stage (P = 0.005) with 100% of pT_XN₊ cases showing elevated p16^INK4A levels (Table 2). Although there was a trend toward increased p16^INK4A expression in the cancer and increased clinical stage, the association was not statistically significant (P = 0.11). High p16^INK4A in the cancer was not associated with age, pretreatment PSA concentration, or Gleason grade. In this cohort, 81 (39%) of 206 patients had relapsed including 2 who died from PC. High p16^INK4A in the cancer (P = <0.0001; Table 2) was strongly associated with disease relapse.

In agreement with other published series (20, 21) , significant predictors of relapse on univariate analyses in this cohort were pretreatment serum PSA levels, pathological stage, Gleason grade, clinical stage, NHT, seminal vesicle involvement, and surgical margin involvement (Table 3). High p16^INK4A in either HGPIN or cancer alone was a significant predictor of relapse on univariate analysis (P < 0.001 and < 0.001, respectively; data not shown). Bivariate analysis in a Cox’s proportional hazards model for p16^INK4A status and pathological stage showed that high p16^INK4A predicts earlier relapse independently of pathological stage (P < 0.001).

The diagnosis of HGPIN at RP was not a significant (P = 0.196) predictor of outcome on univariate analysis (Table 3). In agreement with this observation, Kaplan-Meier product limit analysis showed no survival difference for men based on the presence or absence of HGPIN at the time of RP (log-rank, P = 0.194; Fig. 2,A). In contrast, Kaplan-Meier curves of disease-free survival in patients with HGPIN, stratified on the basis of low (≤1%) versus high (>1%) p16^INK4A expression, demonstrated significant differences in relapse between the two strata (log-rank, P = <0.001; Fig. 2,B). This result was robust when Kaplan-Meier curves of disease-free survival in patients with HGPIN stratified on the basis of ≤5% versus >5% p16^INK4A expression were performed (log-rank, P = <0.0001). Kaplan-Meier analysis of p16^INK4A expression in the coexistent cancer also showed significant differences in relapse based on high (>1%) or low (≤1%) p16^INK4A expression (log-rank, P < 0.001; Fig. 2 C).

The combined assessment of p16^INK4A levels in patients diagnosed with coincident HGPIN and cancer demonstrated three predominant patterns of expression in 149 patients: Group A, low p16^INK4A expression (≤1%) in HGPIN and cancer in 51 (34%) cases; Group B, high p16^INK4A expression (>1%) in cancer and low in HGPIN in 56 (38%) cases; and Group C, high p16^INK4A expression (>1%) in HGPIN and cancer in 42 (28%) cases. On the basis of these criteria, patients in Groups B and C had a significantly higher chance of relapse compared with patients in Group A on univariate analysis (Table 3). Kaplan-Meier product limit analysis for relapse demonstrated significant differences in relapse between these three groups of patients based on the pattern of p16^INK4A expression (Fig. 2,D; overall log-rank, P < 0.001). Differences between the three groups were significant by log-rank calculation between Group A and B: P = 0.032; and Group B and C: P = 0.018 (Fig. 2 D).

In a multivariate analysis that examined p16^INK4a expression in HGPIN with pretreatment PSA, pathological stage, Gleason score, and surgical margins, independent predictors of outcome were p16^INK4a expression in HGPIN, Gleason score, and pathological stage, whereas pretreatment PSA and margin status were not (Table 4). When p16^INK4a expression in HGPIN was excluded, preoperative PSA was independently predictive of relapse, in agreement with other published series (20). When the multivariate analysis was repeated with Gleason score dichotomized at ≤6 and ≥7, the results were unchanged.

Thirty-eight of the 206 patients in this group received NHT prior to surgery. Of the 38 specimens, 25 contained areas of HGPIN concurrent with carcinoma. After exclusion of the patients treated with NHT, Kaplan-Meier analysis of p16^INK4a expression in cancer in 168 patients showed significant differences in relapse based on high (>1%) or low (≤1%) p16^INK4a expression (log-rank, P < 0.001; Fig. 2,E). Similarly, Kaplan-Meier curves of disease-free survival in the 129 “hormone-naïve” patients with HGPIN, stratified on the basis of high (>1%) or low (≤1%) p16^INK4a expression, demonstrated significant differences between the two strata (log-rank, P = 0.013; Fig. 2 F). In a multivariate model adjusted for Gleason score, pathological stage, margin status, pretreatment PSA, and NHT, p16^INK4a expression in HGPIN remained an independent predictor of relapse (hazards ratio, 2.24; 95% CI, 1.28–3.92).

In the present study, we have shown that overexpression of the cell cycle inhibitor p16^INK4A in HGPIN is correlated with, but independent of, pathological stage in the coincident cancer and is associated with early relapse in PC patients treated with RP. Notably, overexpression of p16^INK4A in HGPIN and cancer is an independent predictor of disease relapse in a multivariate analysis adjusted for pretreatment PSA, Gleason grade, margin status, and pathological stage. On the basis of the interpretation of genetic progression models published previously (22, 23), these data suggest that p16^INK4a accumulation may be an early event in PC progression and provide the first evidence that overexpression of p16^INK4a in HGPIN is an independent predictor of clinical failure after surgery.

Diagnostic procedures used to categorize PC patients into discrete risk groups for disease recurrence based on preoperative prognostic factors are known to be inaccurate, because clinical stage and biopsy Gleason grade only approximate postoperative pathological stage (24). Although models are now available that incorporate postoperative data there remains no accurate marker of pathological stage prior to treatment (24). Our finding that overexpression of p16^INK4A correlates significantly with pathological stage raises the possibility of using biopsy p16^INK4A status as a surrogate marker for pathological stage before treatment decisions are taken. For the 14% of men who present with HGPIN only, in the absence of clinical stage and biopsy Gleason grade, accumulation of p16^INK4A may represent the most accurate marker of disease recurrence described to date. Although the scope of this study did not allow us to investigate p16^INK4A expression in biopsy specimens of HGPIN in which cancer was not evident, our findings strongly support the initiation of prospective trials to further evaluate the prognostic value of p16^INK4A levels in biopsy specimens in which HGPIN is diagnosed with or without concurrent carcinoma.

Because the normal physiological function of p16^INK4A is inhibition of Cdk4 and Cdk6, the INK4A gene is a tumor suppressor gene, and the loss of p16^INK4A expression due to homologous deletion, point mutation, or methylation-induced promoter silencing is well documented in a broad range of human cancers (8). More recently, high p16^INK4A expression has also been associated with disease progression and poor prognosis in several tumor types. In agreement with our data in PC, p16^INK4A overexpression in the presence of pRB expression was noted in early-stage ovarian cancer and was correlated with poor disease outcome (25). Similarly, high levels of p16^INK4A were associated with an adverse prognosis in breast (26) and prostate carcinoma (13). These new data raise questions as to the mechanisms responsible for p16^INK4A overexpression and its functional consequences in terms of loss of normal growth control.

Although loss of transcriptional repression in the presence of inactivating mutations in the RB gene is the most well-defined mechanism of p16^INK4A overexpression (8), we found no evidence of a correlation between high p16^INK4A and aberrant pRB expression as determined by immunohistochemistry (data not shown). These data imply that PC cells are resistant to the growth inhibitory effects of p16^INK4A in the presence of normal pRB. Although this is contrary to present dogma, data from recent studies of p16^INK4A complex formation and activity suggest that p16^INK4A-mediated growth inhibition may occur only when cyclin E/Cdk2 complexes are inactivated concurrently by the cell cycle inhibitor p27^KIP1 (27). Thus, cells may become resistant to the growth-inhibitory effects of p16^INK4A when (a) cyclin E levels are elevated, or (b) p27^KIP1 is unavailable (27, 28). The MDA-MB-157 breast cancer cell line is one example of a tumor cell line that retains functional pRB but can proliferate in the presence of p16^INK4A overexpression. This cell line also expresses high levels of cyclin E (29).

Several of these potential scenarios exist in carcinoma of the prostate. Although there are no published studies that have addressed the potential role of overexpression of cyclin E in PC outcome, preliminary data in a subset of our cohort show that a statistically significant number of tumors that overexpress p16^INK4A also overexpress cyclin E. In comparison, none of the tumors with low levels of p16^INK4A overexpressed cyclin E. This is consistent with a model in which aggressive disease may involve cooperation of high p16^INK4A and high cyclin E. Furthermore, others have demonstrated that loss of, or decreased expression of, p27^KIP1 is a common event in PC and is a predictor of treatment failure after prostatectomy (30, 31).

Irrespective of whether or not p16^INK4A is involved mechanistically in PC progression, our data identify overexpression of p16^INK4A in HGPIN as a marker of disease outcome. This may be of considerable clinical utility as a pretreatment diagnostic tool to stratify patients for aggressive treatment very early in the disease process, potentially several years prior to the onset of invasive disease.

We thank Drs. Jennifer Turner and John Finlayson for providing material and Dr. Liz Musgrove for her critical comments.