p14ARF expression in invasive breast cancers and ductal carcinoma in situ – relationships to p53 and Hdm2

Introduction p14ARF stabilises nuclear p53, with a variable expression of p14ARF mRNA in breast cancers. In vitro, nuclear p14ARF binds Hdm2 to block Hdm2-dependent nucleocytoplasmic shuttling of p53, which is required before cytoplasmic degradation of p53. p14ARF is negatively regulated by p53 and through p53-independent pathways. No studies have yet examined levels of p14ARF protein expression in breast cancer and their relationship to Hdm2/p53 immunoreactivity or subcellular localisation. Previously, immunohistochemical expression of cytoplasmic p14ARF, p53 and Hdm2 has been described. HER-2 (c-erbB2/neu) predicts prognosis and interacts with the p14ARF/Hdm2 pathway to inactivate p14ARF and to influence Hdm2 activity and localisation. This study examined p14ARF and p53/Hdm2 expression and subcellular localisation by using immunohistochemistry in a series of invasive ductal breast cancers (IDCs) with concomitant ductal carcinoma in situ (DCIS), to evaluate whether findings in vitro were related to clinicopathological parameters such as HER-2 and their effect on patient outcome. Methods The 4C6 anti-p14ARF monoclonal antibody and Dako Envision Plus system were used to evaluate p14ARF expression in 103 patients; p53/Hdm2 staining was performed. Results p14ARF was evaluable in 96 patients, with nuclear p14ARF expression (modified Quick-score ≥ 3) in 79% (n = 76) of IDCs and in associated DCIS in 74 patients. Cytoplasmic p14ARF was detectable in 23 breast cancers. Nuclear and cytoplasmic p14ARF showed no correlation with p53 subcellular immunoreactivity. Increasing levels of cytoplasmic p14ARF were associated with nuclear and cytoplasmic Hdm2 expression (P < 0.001). Subcellular ARF expression was not associated with clinicopathological parameters, and although not an independent prognosticator, these preliminary findings suggest that cytoplasmic p14ARF might be associated with a better overall survival (P = 0.09; log rank). The association between HER-2 positivity and nuclear p14ARF (P = 0.038), as well as nuclear Hdm2 (P = 0.019), reflects the in vitro findings of HER-2 interaction with the ARF/Hdm2 pathway. Cytoplasmic p53 and Hdm2 expression might have biological implications, through an association of cytoplasmic p53 with increased tumour proliferation (P = 0.005), and an improved overall survival (P = 0.002, log rank) in cytoplasmic Hdm2-expressing tumours, that independently predict favourable overall survival (P = 0.02) and disease-free survival (P = 0.03). Conclusions Nuclear p14ARF expression is similar in IDCs and DCIS and is associated with Hdm2 immunoreactivity. Nuclear p14ARF and Hdm2 might be regulated by HER-2. Clearly, our findings in vivo suggest a complexity of p14ARF/Hdm2 and p53 pathways in which consideration of cytoplasmic p14ARF and Hdm2 might have tumorigenic implications.

Introduction p14 ARF stabilises nuclear p53, with a variable expression of p14 ARF mRNA in breast cancers. In vitro, nuclear p14 ARF binds Hdm2 to block Hdm2-dependent nucleocytoplasmic shuttling of p53, which is required before cytoplasmic degradation of p53. p14 ARF is negatively regulated by p53 and through p53-independent pathways. No studies have yet examined levels of p14 ARF protein expression in breast cancer and their relationship to Hdm2/p53 immunoreactivity or subcellular localisation. Previously, immunohistochemical expression of cytoplasmic p14 ARF , p53 and Hdm2 has been described. HER-2 (c-erbB2/neu) predicts prognosis and interacts with the p14 ARF /Hdm2 pathway to inactivate p14 ARF and to influence Hdm2 activity and localisation. This study examined p14 ARF and p53/Hdm2 expression and subcellular localisation by using immunohistochemistry in a series of invasive ductal breast cancers (IDCs) with concomitant ductal carcinoma in situ (DCIS), to evaluate whether findings in vitro were related to clinicopathological parameters such as HER-2 and their effect on patient outcome.

Methods
The 4C6 anti-p14 ARF monoclonal antibody and Dako Envision Plus system were used to evaluate p14 ARF expression in 103 patients; p53/Hdm2 staining was performed.
Results p14 ARF was evaluable in 96 patients, with nuclear p14 ARF expression (modified Quick-score ≥ 3) in 79% (n = 76) of IDCs and in associated DCIS in 74 patients. Cytoplasmic p14 ARF was detectable in 23 breast cancers. Nuclear and cytoplasmic p14 ARF showed no correlation with p53 subcellular immunoreactivity. Increasing levels of cytoplasmic p14 ARF were associated with nuclear and cytoplasmic Hdm2 expression (P < 0.001). Subcellular ARF expression was not associated with clinicopathological parameters, and although not an independent prognosticator, these preliminary findings suggest that cytoplasmic p14 ARF might be associated with a better overall survival (P = 0.09; log rank). The association between HER-2 positivity and nuclear p14 ARF (P = 0.038), as well as nuclear Hdm2 (P = 0.019), reflects the in vitro findings of HER-2 interaction with the ARF/Hdm2 pathway. Cytoplasmic p53 and Hdm2 expression might have biological implications, through an association of cytoplasmic p53 with increased tumour proliferation (P = 0.005), and an improved overall survival (P = 0.002, log rank) in cytoplasmic Hdm2-expressing tumours, that independently predict favourable overall survival (P = 0.02) and disease-free survival (P = 0.03).

Introduction
The development and growth of breast cancers result from the inactivation of p53 or retinoblastoma (pRb) tumour suppressor proteins that regulate cell cycle control. Such regulatory pathways trigger cell cycle arrest or apoptosis in response to intracellular challenges such as DNA damage, hypoxia and oncogene activation, with preservation of genomic stability [1]. pRb maintains checkpoint integrity through binding and blocking E2F transcription factors, a process reinforced by the G1 cyclin-dependent kinase inhibitor p16 INK4a [2]. Similarly, p53 accumulates to transcriptionally activate the cyclin-dependent kinase inhibitor cc = Spearman correlation coefficient; DCIS = ductal carcinoma in situ; DFS = disease-free survival; ER = oestrogen receptor; GPG = good prognostic group; HRP = horseradish peroxidase; IDC = invasive ductal cancer; MPG = moderate prognostic group; MQS = modified Quick-score; NPI = Nottingham Prognostic Index; OS = overall survival; PPG = poor prognostic group; pRb = retinoblastoma protein; TBS = Tris-buffered saline. p21 WAFI/CIPI , as well as its own negative regulator Hdm2 (human Mdm2; murine double minute 2), which terminates the p53 response. Nuclear retention of p53 underlines its tumour suppressor response and suggests an additional pathway for disabling p53 over and above its mutation in more than 50% of human cancers [3][4][5]. In an autoregulatory feedback loop, Hdm2 maintains low levels of p53 in normal non-stressed cells and inhibits nuclear p53 through multiple and diverse mechanisms. Hdm2 binds p53 to inhibit its transactivation function and shuttles p53 from the nucleus to the cytoplasm to facilitate its degradation [6][7][8].
Hdm2 is also an E3 ubiquitin ligase that targets p53 for the ubiquitin-dependent 26S proteosome in the cytoplasm [9].
ARF expression in vitro and in vivo is associated with a G1 or G2 cell cycle arrest or apoptosis after inappropriate mitogenic stimuli or DNA damage [10][11][12][13]22,23]. Although these effects in large part relate to activation of the p53 pathway, recent reports suggest that p14 ARF inhibits growth independently of p53 [12,17,23,24]. Low or undetectable p14 ARF expression in normal tissues and its function independently of p53 in highly proliferating, homeostatic tissues, in comparison with tumours, suggests that p14 ARF function has yet to be completely established [17,24]. Subcellular localisation of p14 ARF is preferentially nucleolar or nuclear, where it binds Hdm2 to inhibit the latter's activities towards p53 [15,[25][26][27]. ARF-bound Hdm2 blocks Hdm2-dependent nucleocytoplasmic shuttling of p53, to produce nuclear retention and activation of p53 [13,27]. Further, ARF averts degradation of p53 by inhibiting Hdm2-dependent ubiquitination of p53 in vivo [13,28,29]. p14 ARF acts upstream of p53 and is subject to its negative feedback regulation, suggesting that p53 mutations or its inactivation by HDM2 amplification are often accompanied by overexpression of ARF [10][11][12][13][14]30]. p53-positive tumours are also likely to have sustained epistatic mutations such as HDM2 amplification or ARF loss [11]. Inactivation of ARF might occur through the overexpression of several ARF repressors, including Twist and Tbx-2 in breast cancers and Bmi-1 in other tumours [31,32].
Nuclear import and export is a feature of both p53 and Hdm2 with implications for their functional regulation, such that cytoplasmic p53 is associated with tumours with a poor prognosis [3][4][5][6]8]. Just as Hdm2 regulates the nuclear export of p53, in vivo evidence suggests that ARF influences the subcellular localisation of Hdm2 [21]. Cytoplasmic ARF occurs through binding to Pex19p with evidence of weak to moderate cytoplasmic staining in human tumours, whereas two studies disregarded cytoplasmic p14 ARF expression [33][34][35][36][37]. In a single in vitro study, mouse p19 ARF binds a cytoplasmic protein Pex19p (cloned from mouse testis) in normal cells (NIH 3T3) [33]. Consequently, the subcellular localisation of these proteins and the relationship between their levels of expression are likely to be important in evaluating breast cancers and their development from pre-invasive ductal carcinoma in situ (DCIS) to invasive disease.
Previous studies have examined p14 ARF mRNA expression in breast cancers, with evidence suggesting altered expression and an association with p53 [38,39]. The aim of this study was to evaluate the levels of protein expression for p14 ARF in relation to Hdm2 and p53 using immunohistochemistry in DCIS and invasive ductal breast cancers (IDCs), including the study of their subcellular localisation. So far, no studies have examined p14 ARF /Hdm2 and p53 in relation to clinicopathological parameters and prognosis in breast cancer. We have shown nuclear p14 ARF expression in 79% of IDCs and in associated DCIS in 74 patients. Cytoplasmic p14 ARF was detectable in 23 breast cancers. Levels of expression and subcellular localisation for p14 ARF , Hdm2 and p53 were similar in IDCs and DCIS. ARF expression showed no correlation with p53 immunoreactivity and was associated with Hdm2 nuclear and cytoplasmic expression. HER-2-positive breast cancers were associated with nuclear p14 ARF and nuclear Hdm2 immunoreactivity. Our preliminary findings suggest that cytoplasmic p14 ARF and cytoplasmic Hdm2-expressing breast cancers might be associated with a better outcome.

Patients
The study included 103 patients aged 26-88 years (median 59 years) with IDC of the breast, in association with concomitant DCIS diagnosed between 1996 and 2000 at the Bristol Royal Infirmary, Bristol, UK (Table 1). Eligibility criteria were histological diagnosis of IDC and DCIS, no distant metastases, and unilateral tumour. Regional Ethics Committee approval was granted before commencement of the study. Axillary lymphadenopathy was evaluable in 90 patients: 37 (41%) of the patients were lymph node negative, and 53 (59%) were node positive (N1 [mobile ipsilateral] or N2 [fixed ipsilateral]). No axillary surgery was undertaken in the remaining 13 patients because of age-related co-morbidity.
Clinicopathological subgroups were analysed in accordance with the Nottingham Prognostic Index (NPI) and divided into good (GPG), moderate (MPG) and poor (PPG) prognostic groups as described, with a modification that included no assessment of the internal mammary lymph nodes [40]. Evaluation of the NPI was precluded in 15 patients because of non-evaluable regional lymphadenopathy and tumour size. Similarly, subgroups of DCIS were analysed in accordance with the Van Nuys Pathologic Classification (Table 1), which was not assessable in six patients [41]. The design of the study to include representative samples of synchronous IDC and DCIS precluded analysis by the Van Nuys Prognostic Index [41]. Adjuvant treatment groups comprised tamoxifen in 60 patients (27 GPG, 16 MPG, 4 PPG and 13 no NPI), and cyclophosphamide-methotrexate-5-fluorouracil-containing and anthracycline-containing regimes in 17 and 21 patients, respectively (4 GPG, 19 MPG, 13 PPG and 2 no NPI). Five patients received no adjuvant treatment. The median follow-up duration was 51 months (range 6-120 months). All were primary tumours with the exception of six local tumour recurrences, which were excluded from the analysis of patient outcome (see Table 4 and Fig. 3).
Tumour samples were collected shortly after surgery and were fixed in buffered formalin for 24-48 hours at room temperature (20-22°C). Tumours were classified in accordance with the guidelines of the UK National Health Service Breast Screening Programme [42] and were graded by the modified Bloom's grading system described by Elston and Ellis [43]. Oestrogen receptor (ER) immunostaining was performed with a standard three-layered streptavidin-avidin-biotin horseradish peroxidase (HRP) method with a mouse anti-human ER primary antibody (M0747, 1:100 dilution; Dako, Ely, UK) and a biotinylated rabbit anti-mouse secondary antibody (E0354 1:350 dilution; Dako). Expression of ER was assessed with Quickscore (0-8) and classified as positive (4-8; more than 3) or negative (0-3; 3 or less) in five high-power fields [44].

Clinicopathological variables in the patients entered into this study
used as a secondary antibody. Tonsillar tissue was used as a positive control and primary antibody was replaced with Tris-buffered saline (TBS) as a negative control. Ki67 staining was evaluated as positive or negative, with low proliferation indicative of less than 10% of positive-staining cells, compared with high proliferation with at least 10% positiv-ity [45]. HER-2 immunostaining was performed with the mouse monoclonal anti-HER-2 antibody (RTU-CB11; NovaCastra/Vector, Newcastle upon Tyne, UK) and the Envision Plus HRP system (K4006; Dako). HER-2 expression was scored according to the degree and proportion of membrane staining, with a score of 0 or 1+ defined as neg-

Figure 1
Examples of p14 ARF , p53 and Hdm2 immunoreactivity in infiltrating ductal carcinoma of the breast Examples of p14 ARF , p53 and Hdm2 immunoreactivity in infiltrating ductal carcinoma of the breast. Immunostaining was performed as described in the Methods section, and nuclei were counterstained with haematoxylin. ative, and 2+ or 3+ as HER-2 positive [46]. Lymphovascular invasion was assessed as present or not, and was analysed together with ER, HER-2 and Ki67 in the Department of Pathology (by CS and CJC).  Relationships between subcellular localisation of p14 ARF and Hdm2 Relationships between subcellular localisation of p14 ARF and Hdm2. Scatter plots according to nuclear p14 ARF modified Quick-score (MQS) (a, b) and cytoplasmic p14 ARF MQS (c, d). p14 ARF and Hdm2 staining was analysed as described in the Methods section. (a, b) Distribution of nuclear p14 ARF and nuclear Hdm2 (a), and also cytoplasmic Hdm2 (b), showed no significant correlations. (c, d) Distribution of cytoplasmic p14 ARF and nuclear Hdm2 (cC), and also cytoplasmic Hdm2 (d), showed significant correlations (P < 0.001, respectively). P values in bold (P < 0.05) indicate significance.

Nuclear Hdm2
Nuclear ARF
Hdm2 expression was assessed with the mouse monoclonal antibody OP46 (Oncogene Research, CN Biosciences, Nottingham, UK) that detects C-terminal Hdm2 with specificity for the 90 kDa isoform [47]. OP46 was selected after testing with other antibodies including 2A10 and 3G5 (kindly given by AJ Levine, Rockefeller University, New York) both of which had technical or other limitations [48]. OP46 was used at a 1:80 dilution of 100 µg/ml and evaluated in the context of Hdm2 overexpressing A375 malignant melanoma cells, and negatively controlled with IgG2b, which was substituted for the primary antibody. Biotinylated rabbit anti-mouse IgG secondary antibody (E0354, 1:300 dilution of 1.26 mg/ml; Dako), together with the Strept-AB Complex/HRP (0377; Dako, Glostrup, Denmark), was used for p14 ARF and Hdm2. Technical limitations precluded dual staining for ARF and Hdm2. p53 immunostaining was performed as described [49] by using the DO7 monoclonal antibody (NovaCastra/Vector). Breast carcinomas known to overexpress p53 with known TP53 gene mutations and protein accumulation were used as positive controls. Negative controls were obtained by omitting the primary antibodies.
Formalin-fixed paraffin sections of breast cancer tissue and cell pellets (controls) were mounted on 3-aminopropyl-triethoxysilane-coated (Sigma, Poole, Dorset, UK) glass slides and were baked for 30 min at 56-60°C before being dewaxed in Clearene (Surgipath Europe, Peterborough, UK). The tissue was rehydrated by sequential immersion in 100% and 50% ethanol to distilled water. Tissue sections were subjected to heat retrieval of antigens in citrate buffer (pH 6) in a pressure cooker for 3 min; after cooling they were incubated in 0.3% (v/v) hydrogen peroxide for 5 min. Subsequently, sections were washed in tapwater and TBS (pH 7.45). Before incubation with p14 ARF and Hdm2 primary antibodies, sections were exposed to avidin and biotin blocking solutions (Vector Laboratories, Burlingame, California) for 15 min each. Further blocking was achieved through exposure to normal rabbit serum (diluted with TBS) for 30 min at room temperature. Primary antibody was applied and incubated at 4°C overnight (18 hours) for p14 ARF and Hdm2, except for p53, which was incubated for 1 hour at room temperature. After washing with TBS, biotinylated rabbit anti-mouse secondary antibody was applied for 30 min at room temperature. The Strept-AB Complex/ HRP was used for the detection of p14 ARF and Hdm2, and involved an additional 30 min incubation with TBS/biotinylated streptavidin/HRP. Staining was revealed by development in the chromogen 3,3-diaminobenzidine tetrahydrochloride for 5-10 min before counterstaining with haematoxylin before mounting.
Immunostaining was assessed with a Zeiss Axioskop microscope with a 40 × Achrostig mat lens (×400 overall magnification) and a field diameter of 0.46 mm. In the neoplastic cell population for IDC and DCIS, the degree of staining intensity and the proportion of cells with p14 ARF , Hdm2 and p53 immunoreactivity in the nucleus and cytoplasm were individually graded semiquantitatively to produce an intensity distribution score or H-score for each localisation, with invasive and pre-invasive components given separate scores [50]. Initial scoring was of 10 highpower fields; however, in view of the homogeneous staining this was reduced to 5 high-power fields. A modified Quick-score (MQS) was used to define a score of 0-3 for intensity (0, no staining; 1, weak staining; 2, moderate staining; 3, strong staining) and 0-5 for distribution (0, no staining; 1, 1% or less; 2, 1-10%; 3, 11-33%; 4, 34-66%; 5, 67-100% of cells staining), giving an overall score of 0-8 [44]. Sections were scored independently by two pathologists (CS and CJC) in the Department of Pathology. Unless otherwise stated, scores were assessed as a continuum for the purposes of statistical correlation. For the purposes of describing p14 ARF , Hdm2 and p53 immunoreactivity, tumour positivity was defined as a modified Quickscore of 3 or more (1+/2+) (see Table 2).

Statistical analysis
Data were analysed with the SPSS 10.0 for Windows (SPSS Incorporated, Chicago, Illinois) statistics software and summarised with descriptive statistics. The associations between p14 ARF , p53 and Hdm2 and patient characteristics were assessed with the Spearman non-parametric test for continuous variables and the χ 2 test and Fisher's exact test for categorical factors. The relationship between p14 ARF localisation and the subcellular localisation of p53 and Hdm2 was analysed with the Spearman correlation coefficient (cc) for non-parametric data, with the p14 ARF and p53/Hdm2 MQS used as continuous variables. Analyses of survival data were performed with the log-rank test and the Cox regression model, with survival curves computed with the Kaplan-Meier method. For p14 ARF , p53 and Hdm2, univariate and multivariate analyses were performed and adjusted for the NPI score and treatment received (tamoxifen/chemotherapy/none). Because the NPI is based on nodal involvement, on tumour size and on grade, patients (n = 11) with non-evaluable lymphadenopathy and tumour size were excluded from the multivariate regression analyses (see Table 4).

p14 ARF levels of expression and subcellular localisation in IDC and DCIS
p14 ARF immunoreactivity was evaluable in the nucleus and cytoplasm in 96 (93%) cases of IDC and in associated DCIS in 91 (88%) cases ( Fig. 1 and Table 2 (Table 2). Weak to moderate (MQS 3-5) nuclear p53 expression was observed in 17 cases of IDC (18%), with similar cytoplasmic staining (18 cases). There was no nuclear or cytoplasmic p53 expression (MQS 0-2) in most IDCs (67% and 75%, respectively). Strong nuclear p53 expression (MQS 6-8) was evident in 15% of IDCs, with 5% of these expressing intense cytoplasmic staining. A comparison of the levels of p53 expression and subcellular localisation in DCIS were similar to those in IDC ( Table 2).

Association of p14 ARF , p53 and Hdm2 with clinicopathological factors
We investigated the relationships between nuclear and cytoplasmic levels of p14 ARF expression and clinicopathological parameters (listed in  Table 2) in relation to clinicopathological parameters (data not shown) showed no significant associations.
Nuclear and cytoplasmic levels of p53 expression (assessed as a MQS continuum) were analysed in relation to clinicopathological criteria as for p14 ARF . Nuclear p53 levels were significantly associated with increased proliferation (Ki67) (cc 0.275, P = 0.008) in IDCs, with cytoplasmic p53 showing a similar association in DCIS (cc 0.295, P = 0.005). There was a possible significant trend in IDCs between nuclear p53 and the NPI (cc 0.210, P = 0.063). Table 3 Relationships between nuclear expression of p14ARF with nuclear p53 and Hdm2 in invasive breast cancers We further analysed the relationship between nuclear and cytoplasmic Hdm2 expression and established parameters as described (Table 1), using the MQS as a continuum. In invasive tumours, increasing levels of nuclear and cytoplas-mic Hdm2 expression were significantly associated with increased proliferation (Ki67) (cc 0.281, P = 0.007, and cc 0.224, P = 0.032, respectively). In DCIS nuclear Hdm2 was associated with HER-2 overexpression (2+/3+) (cc 0.249, P = 0.019), with no similar association with cytoplasmic Hdm2 (cc 0.181, P = 0.09). A further evaluation of Hdm2 positivity (MQS ≥ 3) to described parameters showed no other significant associations.

Relationship of clinicopathological factors to prognosis and the predictive potential of p14 ARF in relation to p53 and Hdm2 expression
Overall survival (OS) and disease-free survival (DFS) were determined in 89 and 97 patients, respectively (primary tumours), with a median follow-up of 51 months (range 4-120 months). Disease relapses (local or distant recurrences) occurred in 30 women; of these, death was confirmed in 23 patients, with 8 suspected deaths in the absence of a recorded mortality date. Locoregional recurrence occurred at a median duration of 28.5 months (range 3-120 months) from diagnosis. Breast cancer-related mortality occurred at a median of 26 months (range 8-98 months) from presentation. The mean duration of OS and DFS were 93 months and 87 months, respectively. Fouryear DFS and OS were 70% and 77%, respectively. The relationship of established clinicopathological features to OS and DFS were analysed with Cox's regression analysis (Table 4). Generally poor prognostic factors such as large tumour size, high tumour grade, lymph node metastases, ER negativity, HER-2 overexpression and NPI were significantly associated with decreased OS and DFS. High tumour proliferation (Ki67 on immunohistochemistry), although associated with a smaller percentage of patients remaining disease-free (DFS) and alive (OS) at 4 years, did not reach statistical significance on univariate or multivariate analysis.
Univariate and multivariate analysis with the continuous MQS variables were used to investigate possible relationships between patient outcome data and levels of subcellular expression for p14 ARF , Hdm2 and p53. When scores were measured as a continuum, neither nuclear nor cytoplasmic ARF expression was predictive of OS or DFS after univariate or multivariate analysis (data not shown). Similarly, a Kaplan-Meier analysis for OS and DFS of the nuclear p14 ARF subgroups (negative [0-2], weak/moderate [3][4][5] and strong [6][7][8]) showed no significant relationship (data not shown). The small number of patients with strong cytoplasmic p14 ARF precluded a similar analysis (see Table  2). Kaplan-Meier survival curves showed that the presence of cytoplasmic p14 ARF (MQS ≥ 3) might be associated with a better OS compared with ARF negativity (MQS < 3) (P = 0.09, Fig. 3b), with a similar separation of DFS curves, although not statistically significant (data not shown). The presence or absence of nuclear p14 ARF (MQS ≥ 3 compared with MQS < 3) was not significantly associated with patient outcome (Fig. 3a).
Subcellular levels of p53 expression were not shown to be independent prognosticators for OS and for DFS on univariate or multivariate analyses (data not shown). Kaplan-Meier analyses of nuclear p53 subgroups (negative [0-2], weak/moderate [3][4][5] and strong [6][7][8]) showed a significant association between strong nuclear p53 expression Relationships between p14 ARF and Hdm2 localisation and overall survival Relationships between p14 ARF and Hdm2 localisation and overall survival. Overall survival curves according to nuclear p14 ARF (a), cytoplasmic p14 ARF (b) and cytoplasmic Hdm2 (c), with patient numbers and deaths in parentheses. The classification of 'present' or 'absent' p14 ARF or Hdm2 expression was determined by an MQS of 3 or more to define positivity, compared with an MQS of <3 to define absent expression. Patient numbers reflect the exclusion of six local tumour recurrences from the analysis of patient outcome and include only recorded deaths in 23 patients. P values are given for the log ranks that reached significance.

R582
and a reduced DFS (P = 0.05; log rank), although patient numbers were small for this group (data not shown). Similarly, nuclear p53 expression (MQS ≥ 3 compared with MQS < 3) was associated with a reduced OS and DFS, although this was not statistically significant (data not shown). Nuclear Hdm2 expression was not an independent prognosticator for OS and for DFS on univariate or multivariate analyses. Cytoplasmic Hdm2 expression (analysed as a continuous MQS) was associated with improved OS (P = 0.01, univariate analysis) and DFS (P = 0.05, univariate) ( Table 4). Cytoplasmic Hdm2 independently predicted for improved OS (P = 0.02, multivariate) and DFS (P = 0.03, multivariate) ( Table 4). Kaplan-Meier survival curves showed that the presence of cytoplasmic Hdm2 (MQS ≥ 3 compared with MQS < 3) was associated with improved OS (P = 0.002) (Fig. 3c).

Discussion
p14 ARF suppresses cancer cell growth, and several in vitro studies have reported its inhibition of Hdm2 to stabilise and activate p53, with a loss of ARF increasing p53 degradation. The relevance of these mechanisms in vivo requires further investigation, in view of ARF's proposed function as a tumour suppressor and the suggestion that p14 ARF overexpression is a surrogate marker of dysregulation of the pRb and p53 pathways [30]. p14 ARF mRNA expression in breast cancers underlines a variability with overexpression in 17-19% and underexpression in 24-26% [38,39]. There are few studies clarifying p14 ARF protein levels in vivo, with low ARF expression in non-neoplastic epithelium [17,34,35]. The present study suggests an increased frequency of nuclear p14 ARF overexpression in 47% of invasive and non-invasive breast cancers. Nuclear p14 ARF overexpression has been shown by immunohistochemistry in 22% of B cell lymphomas to predict tumour aggression and outcome [30]. Possible abnormalities in the p53 pathways might be implicated [10][11][12][13][14]30,34]. p53-independent mechanisms might also contribute through oncogenic stimulation of ARF [16,17,30].
The lack of p14 ARF expression in 21% of tumours is not dissimilar to other studies of ARF mRNA levels in breast cancers [38,39]. B cell lymphomas lacked nuclear p14 ARF in 11% of cases, in association in large part with promoter hypermethylation, as reported for breast cancers [21,30,38]. Predominant nucleolar localisation of p14 ARF is determined by amino-terminus and exon 2 carboxy-terminus, with evidence of this after immunofluorescence of a number of cancer cell lines [25,26] and exclusive nucleolar p14 ARF in a subset of lymphomas and non-tumour tissue [25,26,30]. The 4C6 monoclonal antibody and other p14 ARF monoclonal antibodies have verified the nuclear/ nucleoplasmic localisation of ARF with intactness of its functional pathway, as well as its implications as a prognostic surrogate marker compared with nucleolar ARF [27,30].
This study shows nuclear p14 ARF in the majority of breast cancers, with 24% cytoplasmic detection. Few other studies have attempted to analyse cytoplasmic p14 ARF , although detectable on immunohistochemistry, with in vitro evidence of an ARF-binding cytoplasmic protein [33][34][35][36][37]. Cytoplasmic p14 ARF has been observed in non-small cell lung cancers, in oral squamous carcinomas and in another study of lung and pancreatic tumours with the use of several monoclonal and polyclonal antibodies [34][35][36][37]. This study analysed cytoplasmic ARF in the context of positive and sero-matched negative controls as described, with no evidence of ARF expression in surrounding normal breast epithelial cells. Such observations might invoke an additional pathway of ARF regulation through changes in its subcellular localisation as for other tumour suppressors, such as p53 and p21 WAF1/CIP1 [5,8,51,52]. Similar genomic aberrations might occur in DCIS and IDC, highlighting possible similarities in protein expression [53,54]. This study shows a consistent similarity in levels of protein expression and subcellular localisation for p14 ARF , p53 and Hdm2 in DCIS and IDCs.
It is suggested that the upregulation of nuclear p14 ARF expression is a consequence of cell cycle malfunction involving p53 and Hdm2 [30]. Increasing levels of nuclear ARF on immunohistochemistry are a measure of p53 inactivation by mutation, or by Hdm2 overexpression resulting in disruption of the p53-p14 ARF negative feedback loop [30]. The implication is that p14 ARF expression is associated with P53-deficient cell lines, suggesting a p53-mediated downregulation of ARF [10][11][12][13][14]. Although p53 immunostaining is not necessarily tightly correlated with TP53 gene function, 57% (17 cases) of p53-positive tumours in this study strongly expressed ARF, reflecting similar in vivo evidence showing concomitant p14 ARF mRNA expression and p53 immunostaining in breast cancers, with a similar relationship between increasing levels of ARF protein and p53 mutations in B cell lymphomas (Table 3) [30,34,39]. Others have shown a lack of an inverse relationship between the two genes, suggesting that the ARF-p53 pathways are not strictly linear and that decreased ARF expression and TP53 mutations are not mutually exclusive [39,55]. ARF stabilises nuclear p53 and is postulated to congregate with Hdm2 and p53 in ternary complexes or 'nuclear bodies' that have yet to be demonstrated in vivo [10,25,27,30].
Double-fluorescent immunolabelling in B cell lymphomas revealed a partial co-localisation of nuclear p14 ARF and p53, with no Hdm2 association, suggesting that the mechanisms underpinning ARF's role in inhibiting the nuclear export of p53 by Hdm2 inhibition remains to be defined [30]. Studies in vivo have not previously examined a possible association between ARF expression and p53 subcellular localisation in breast cancer. In large part, this interpretation depends on the accumulation of p53 that is likely to reflect mutant p53 as inferred from immunohistochemistry [30,34,49]. Proportionately more (55%) p53expressing breast cancers showed exclusive nuclear p53 in association with nuclear ARF expression (data not shown).
Hdm2 expression is a feature of tumorigenesis, with overexpression attributed in part to gene amplifications in the minority of breast cancers [56][57][58][59]. Other contributing mechanisms include enhanced transcription and translation and an extended protein half-life. This study demonstrates a minority (11%) of Hdm2-overexpressing (MQS 6-8) invasive breast cancers and 16% of non-invasive breast cancers. Recent evidence shows that p14 ARF overexpression is associated with increased levels of Hdm2 in cancer cells, with a similar finding in oral cancers and B cell lymphomas that indicate a direct association between p14 ARF and Hdm2 expression determined by immunohistochemistry [27,30,35]. Similar findings are reflected in this series of breast cancers (Table 3).
An additional mechanism of Hdm2-mediated regulation by p14 ARF involves intracellular compartmentation [21]. The subcellular localisation of Hdm2 involves its nucleolar localisation signal, as well as both nuclear import and export receptors [59]. In vitro immunofluorescent studies verify the ARF-mediated localisation of Hdm2 in the nucleolus and nucleoplasm [27]. Not uncommonly in colorectal cell lines and primary tumours, in vivo inactivation of ARF by methylation is associated with increasing cytoplasmic Hdm2 expression, emphasising the importance of p14 ARF in nuclear Hdm2 localisation [21]. ARF-negative cancers, implying inactivation in this series, show no specific association with increased cytoplasmic Hdm2 expression (Fig. 2). There is a further suggestion in this series of breast cancers that nuclear ARF expression is associated with both nuclear and cytoplasmic Hdm2-expressing breast cancers, underlining the absence of a preferential nuclear localisation of Hdm2 (data not shown). Similarly, a study of colorectal cancers shows the lack of an exclusive relationship between ARF function and the subcellular localisation of Hdm2, in comparison with cell lines [21]. HER-2/neuneu might also regulate the subcellular localisation of Hdm2, with better prognostic HER-2-negative breast cancers associated with Hdm2 in both the nucleus and cytoplasm, in contrast to HER-2-mediated phosphorylation of Hdm2 to produce its nuclear localisation and the degradation of p53 [60]. Such findings could imply that the relationship between p14 ARF and the subcellular localisation of Hdm2 is not as exclusive as that found in cell lines.
Clinically, the implications of p14 ARF levels of expression in breast cancer are unknown; two previous studies examined p14 ARF mRNA expression in relation to prognostic parameters [38,39]. Variable p14 ARF mRNA expression including both overexpression as well as decreased mRNA levels are reported to be significantly associated with poor prognostic criteria including p53 mutational status, peritumoural vessel invasion, lymph node metastases and negative progesterone receptors [38,39]. By comparison, nuclear p14 ARF overexpression in B cell lymphomas with the use of immunohistochemistry predicts tumour aggression and reduced overall survival [30]. We find no clear associations between ARF levels and prognostic parameters or outcome, although the presence of cytoplasmic p14 ARF as opposed to its absence suggests a better outcome (Fig. 3b). At present, a poor understanding of the implications of cytoplasmic ARF expression in tumours requires further study [35][36][37]. In vitro, HER-2/neuneu interacts with the p14 ARF /Hdm2 pathway; some in vivo studies show that p14 ARF mRNA overexpression in breast cancers is correlated with HER-2 negativity [38,39,61]. In vitro, HER-2 promotes Hdm2-mediated p53 degradation through the inactivation of ARF, and HER-2 further enhances mammary tumorigenesis in ARF heterozygous mice [60,61]. Our finding of an association between ARF and HER-2 expression is preliminary and suggests the importance of HER-2 in the ARF pathway.
Inactivation of p53 might occur through pathways other than mutation, involving p14 ARF /Hdm2 and its degradation through subcellular localisation. Immunohistochemical p53 expression might detect up to 89% of TP53 point mutations in breast carcinoma specimens, although it does not always correlate with specific mutations in exons 5 to 9 [34,49]. Despite an established body of literature regarding the implications of p53 nuclear immunoreactivity, few or no in vivo studies have evaluated cytoplasmic p53. The cytoplasmic localisation of p53 is a prerequisite for its proteosomal degradation and has been implicated in patient prognosis [5,62]. A study of inflammatory breast cancers showed the presence of wild-type cytoplasmic p53 in 37% of cases [5]. We found a significant association of cytoplasmic p53 with increased tumour proliferation in DCIS, suggesting the value of its further investigation in future studies. Similarly, Hdm2 undergoes nucleocytoplasmic shuttling; highly proliferative, invasive breast cancers were associated with increasing levels of nuclear or cytoplasmic Hdm2 expression in this study. Immunohistochemical p53 and Hdm2 expression were also shown to be correlated with ki67 in B cell lymphomas [30]. A possible functional relevance of subcellular Hdm2 expression is suggested, although Ki67 was not an independent prognosticator in this study.
Our findings of an association between HER-2 and nuclear Hdm2 expression might support in vivo findings in which HER-2, through phosphoinositide 3-kinase/Akt-mediated phosphorylation, is preferentially associated with nuclear Hdm2 [60]. In the present study the association between the presence of cytoplasmic Hdm2 and improved outcome would initially seem counterintuitive, yet it is substantiated by recent evidence that HER-2-negative and Akt-negative cell lines are associated predominantly with cytoplasmic Hdm2 [60]. In vivo, the study of a small subset of breast cancers (21 tumours) confirmed these in vitro findings, suggesting an important relationship between HER-2 activation and the p14 ARF /Hdm2 and p53 pathways [60].

Conclusion
Defining relevant pathways regulating p53 function are fundamental to an understanding of its tumour suppressor role and mechanisms of p53 inactivation, which might be independent of a p53 mutation. p14 ARF and Hdm2 are key upstream and downstream regulators of p53, although in part they might be independent of p53. We have demonstrated a spectrum of tumour-specific p14 ARF upregulation with predominant nuclear ARF expression in the absence of exclusive nucleolar staining in invasive and non-invasive breast cancers; we suggest that the presence of cytoplasmic p14 ARF might be favourable in terms of breast cancer outcome. Predominant nuclear ARF levels showed no clear clinical implications about the degree of ARF expression versus its negativity. Clearly our in vivo findings do not completely mirror in vitro relationships between p14 ARF /Hdm2 and p53 to suggest a greater complexity of these pathways in human solid tumours. Increasingly, evidence suggests the importance of evaluating the subcellular localisation of checkpoint proteins to include p53, Hdm2 and p14 ARF [51,52]. Potentially, changes in the cellular distribution of these proteins might be driven by key tumorigenic processes to which HER-2 and other growth factor pathways might contribute.