p14^ARF nuclear overexpression in aggressive B-cell lymphomas is a sensor of malfunction of the common tumor suppressor pathways

The 2 major tumor suppressor pathways, represented by the proteins p16^INK4a-CDK4/6-Rb and p14^ARF-Hdm2-p53,1 are inactivated in most human cancers. p53 is a transcription factor that induces cell cycle arrest and/or apoptosis in response to a variety of stimuli (eg, DNA damage, hyperproliferative signals).2 p53 is negatively regulated by Hdm2 through a multiple mechanism: Hdm2 binds to the transactivation domain of the p53 tetramer, inhibiting p53 transcriptional activity3,4; in addition, Hdm2 functions as an E3-ubiquitin ligase which targets p53 for nuclear export and proteasomal degradation.5-8 Hdm2 is itself a p53-responsive gene, thus establishing a feedback loop through which p53 regulates its own activity and turnover.9 

The atypical structure of the INK4a/ARF locus in 9p21 encodes 2 unrelated tumor suppressor proteins, p16^INK4a and p14^ARF (the human counterpart of murine p19^ARF).10,11 These are specified by different first exons that are spliced to a common second exon translated in alternative reading frames, their expression being controlled by independent promoters. p16^INK4a, a specific inhibitor of cyclin D–dependent kinases, contributes to G1 arrest by blocking Rb phosphorylation.1 On the other hand, p14^ARFinterferes with all of the known functions of Hdm2 (eg, direct inhibition of p53-mediated transactivation,12,13 ubiquitin ligase activity,14 and nuclear export of p5315,16), possibly through induction of Hdm2 degradation,17 indirectly leading to an increase in the activity and stability of p53. p14^ARF is induced by inappropriate hyperproliferative signals (such asmyc,18 E2F-1,19,ras,20 E1A,21v-abl22) and mediates p53 activation in response to oncogenic stimuli. Specifically, responsiveness of the p14^ARF promoter to E2F-1 makes p14^ARF a nexus between the Rb and p53 pathways. p53 suppresses p14^ARFexpression through a poorly understood mechanism, which generates an additional regulatory circuitry.13 23 

p14^ARF is a highly basic protein that localizes to the nucleolus.10,12,24 When induced, p14^ARF binds to Hdm2, thereby allowing p53 to stabilize and accumulate in the nucleoplasm. Classically, this p14^ARF-Hdm2 binding has been assumed to take place in the nucleolus,16 although antagonization of Hdm2 by p14^ARF independently of nucleolar localization has recently been reported.25 In addition, it has been suggested that p14^ARF-p53 direct binding without requirement for Hdm2 as a bridging molecule is also possible,23 although the functional implications of this interaction remain unknown.

Both human and murine p14^ARF contact the central acidic domain of Hdm2 through 2 independent binding sites.26-28Additionally, nucleolar localization sequences (NrLS) have been mapped in exons 1β and 2 of p14^ARF. These motifs are required for the localization of p14^ARF to the nucleolus; mutations in the p14^ARF NrLS have been described as impeding the correct localization of this protein, resulting in its nucleoplasmic accumulation and the consequent loss of its ability to stabilize p53.29 

In human tumors, the p53 gene is inactivated by mutation in more than 50% of cases; in a high proportion of the rest, the p53 pathway would be expected to be disrupted by Hdm2 amplification or p14^ARF loss. In some cancers, the frequency of p14^ARF alteration is remarkably high; deletions affecting the 9p21 region (eg, in the cases of glioblastoma and astrocytoma30,31) and hypermethylation of CpG islands in the p14^ARF promoter (eg, in the case of gastric cancer32) are the main inactivation mechanisms. Point mutations are infrequent and usually also affect p16^INK4a. In other neoplasias, p14^ARF loss appears to be a rarer event. Nevertheless, most of the information available is derived from analysis at the gene level, whereas little is known about the expression level and distribution of the protein in different tumors.

In aggressive B-cell non-Hodgkin lymphomas (NHLs), the frequency of p53 mutation is modest (∼20%),33,34 whereas amplification of the 12q14 region (where the Hdm2 gene is located) has not been detected.35 Therefore, our purpose was to investigate the expression pattern of the p14^ARF protein and its subcellular localization, the possible presence of genetic or epigenetic alterations in p14^ARF, and the relationship between the status of p14^ARF and the major tumor suppressor pathways in a group of large B-cell lymphoma and Burkitt lymphoma.

We obtained 74 samples of tumor specimens from aggressive NHLs from the routine files of the Virgen de la Salud Hospital (Toledo, Spain) and the Spanish National Cancer Center (CNIO) Tumor Bank (Madrid, Spain) and diagnosed using the criteria of the Revised European-American Lymphoma (REAL) classification.36 The samples included 55 cases of diffuse large B-cell lymphoma (DLBCL), 7 cases of follicular lymphoma grade 3 (FL-3), and 12 cases of Burkitt lymphoma (BL). Frozen tissue (for performing molecular studies) and clinical follow-up information were available for 51 of these samples (36 DLBCL, 3 FL-3, and 12 BL).

The human lymphoid cell lines RAJI, NAMALWA, MOLT-4, GRANTA-519, KARPAS-422, WSU-NHL, RPMI 8226, and HuT 78 were obtained from ATCC (Manassas, VA). Cells were cultured in Dulbecco modified Eagle medium (DMEM) (GRANTA-519) or RPMI-1640 medium (all other cell lines) supplemented with 10% fetal bovine serum, 2 mM L-glutamine, penicillin, and streptomycin. Cells were kept in cell culture flasks in a humidified incubator at 37°C and 5% CO₂.

For immunostaining, cells were harvested by centrifugation, washed with cold phosphate buffered saline (PBS), cytospun onto poly-L-lysine–coated slides, and fixed in ethanol/acetone 1:1.

Exons 5 to 8 of the p53 gene and exons 1β and 2 of the p14^ARF gene were amplified from genomic DNA extracted from tissue samples and cell lines, using previously described primers and conditions.31,34,37 Polymerase chain reaction (PCR) products were purified using the Microcon PCR system (Millipore, Bedford, MA). Direct sequencing of purified PCR products was performed with an automated DNA Sequencer ABI PRISM 3700 Genetic Analyzer (Applied Biosystems, Weiterstadt, Germany). Results for p53 and p14^ARF exon 2 mutations in tumor samples have been previously published.33 38 

Homozygous deletions affecting p14^ARF in cell lines were confirmed by simultaneous amplification of p14^ARF exon 2 and p53 exon 8 (multiplex PCR). Results for homozygous or hemizygous deletions in the 9p21 region in the series of NHLs have been published as part of previous studies concerning p16^INK4astatus.33 38 

Methylation-specific PCR (MSP)39 assays were performed to determine the methylation status of the CpG islands of the p14^ARF promoter in tissue samples and cell lines. Briefly, 1 μg of denatured genomic DNA was modified by reaction with sodium bisulfite under conditions that convert unmethylated cytosines to uracils. Modified DNA was purified using the Wizard DNA Clean-Up system (Promega, Madison, WI). Modification was completed by NaOH 0.3 M treatment for 5 minutes at room temperature, followed by ethanol precipitation.

A quantity of 50 ng bisulfite modified DNA was amplified using p14^ARF unmethylated-specific and methylated-specific primers.40 The Hodgkin disease–derived L-540 cell line was used as a positive control of p14^ARFmethylation41; DNA from nontumoral samples was included as a negative control. Methylation of the p14^ARF promoter was detected by the amplification of a 122-bp fragment with the methylated-specific primers.

The following primary antibodies were used for immunohistochemistry: goat polyclonal anti-p14^ARF C-18 (Santa Cruz Biotechnology, Santa Cruz, CA); mouse monoclonal anti-Hdm2 Ab-1 2IF-2 (Oncogene, Darmstadt, Germany); mouse monoclonal anti-p53 DO-7 (Novocastra, Newcastle upon Tyne, United Kingdom). Proliferation index was evaluated by the expression of the nuclear antigen Ki67, detected with the MIB1 monoclonal antibody (Novocastra).

Immunofluorescence was performed with the same antibodies, with the exception of Hdm2, which was detected with the SMP14 monoclonal antibody (Santa Cruz Biotechnology). Mouse monoclonal anti-C23 MS-3 (Santa Cruz Biotechnology) was used as a nucleolar marker.42 43 

Immunohistochemical techniques were performed on paraffin-embedded tissue sections or cytospin preparations of the cell lines using an initial heat-induced antigen retrieval step (slides were heated in a pressure cooker for 3 minutes in a 0.01 M solution of sodium citrate prior to incubation with the antibodies).

After incubation with the primary antibodies, immunodetection was performed with biotinylated antimouse or antigoat secondary antibodies as appropriate, followed by peroxidase-conjugated streptavidine (DAKO, Carpinteria, CA) with diaminobenzidine chromogen as substrate. Immunostaining was performed with the Techmate 500 (DAKO) automatic immunostaining device. Incubation omitting the specific antibody was used as a control of the technique.

p53 and Hdm2 expression were scored semiquantitatively and expressed as the nearest tenth percentile. Ki67 expression was quantified by scoring up to 200 tumoral cells in representative areas.

Double fluorescent immunolabeling was performed on 3-μm–thick tissue sections mounted on poly-L-lysine–coated slides. After antigen retrieval and simultaneous incubation with the 2 primary antibodies, slides were washed and incubated with the secondary antibodies Cy3-conjugated donkey anti–goat (Jackson Immunoresearch, Baltimore, MD) and Alexa 488–conjugated rabbit anti–mouse (Molecular Probes, Eugene, OR). Nuclei were stained with TO-PRO-3 (Molecular Probes). The slides were mounted with VectaShield (Vector, Burlingame, CA) and examined with a TCS NT laser scanning confocal microscopy system (Leica Microsystems, Wetzlar, Germany). The series of images was processed with the software package provided (Leica Microsystems) and Adobe Photoshop 5.5.

The relationship between p14^ARF and other variables (p53, Hdm2, Ki67) was examined using the Kruskal-Wallis test. In order to assess the specific association between p14^ARF and Ki67, a multivariate analysis was performed including those markers significantly related to Ki67 levels as revealed by the univariate analysis. Since neither Ki67 nor its log-transformed values were normally distributed, as assumed by standard multiple regression techniques, a median regression analysis44 was performed. This method compares the median instead of the mean of each category.

The prognostic value of p14^ARF and all other markers was evaluated by standard survival analysis, using Kaplan-Meier and Cox regression. Hazard ratios were computed for each marker, adjusting for International Prognostic Index (IPI) and histology.

Statistical analyses were performed using the STATA and SPSS software packages.

p14^ARF expression was analyzed by immunostaining in several human lymphoid cell lines in which p53 and p14^ARF gene status were also determined (Figure 1; Table 1).

Inactivation of the p53 pathway was observed in the vast majority (7/8) of cell lines: 3 out of 8 lines had only p53 mutation; 1 out of 8 had only p14^ARF deletion; 2 out of 8 featured both alterations simultaneously; 1 out of 8 had both p53 mutation and a heterozygous mutation within p14^ARF exon 2 (R120L). Only in one of the cell lines were no alterations detected. None of the cell lines analyzed showed p14^ARF promoter hypermethylation.

Loss of p14^ARF expression was observed exclusively in the cell lines displaying p14^ARF deletion. In the cell lines expressing p14^ARF the signal was predominantly nucleolar, although additional nucleoplasmic staining was evident in the RPMI 8226 cell line (Figure 1).

In nontumoral lymphoid tissue (tonsils and reactive lymphadenitis), reactive cells showed distinct nucleolar p14^ARF immunostaining (Figure2A). Staining for Hdm2 (Figure 2B) and p53 (Figure 2C) revealed weak nucleoplasmic expression of both proteins in proliferating cells within the germinal centers. No Hdm2 nucleolar staining was observed in normal cells.

p14^ARF expression was analyzed in a series of 74 cases of NHLs including 55 DLBCL, 7 FL-3, and 12 BL. In these samples, an internal control of the technique was provided by the nucleolar staining of small lymphocytes, macrophages, and some endothelial cells.

Concerning the p14^ARF expression pattern in tumoral cells, the following situations were observed: (1) p14^ARF loss in a subset of NHLs. A small number of cases (8/74; Table2) showed no appreciable p14^ARF staining in large cells (Figure 2D), whereas nucleolar staining was observed in benign lymphocytes. The predominant phenotype in this subset of cases was characterized by a low level of expression of both Hdm2 (Figure 2E) and p53 (Figure 2F). Of these cases, 3 could be sequenced for p14^ARF, but no mutations were found. In 1 of 3 cases (case 19), silencing of the p14^ARF gene was associated with promoter hypermethylation coupled with an LOH (loss of heterozygosity), which could affect the unmethylated allele. A homozygous deletion within the 9p21 region was detected in another case (case 47). The p53 gene was wild-type in the 4 cases sequenced. (2) Nucleolar p14^ARFexpression. Nucleolar p14^ARF staining in tumoral cells was the most frequently observed expression pattern (50/74 cases). The intensity of staining was variable, ranging from weak to strong nucleolar staining in most large cells (Figure 2G). The staining for Hdm2 (Figure 2H) and p53 (Figure 2I) was always nucleoplasmic and of variable intensity, often with distinct nucleolar exclusion. Nucleolar accumulation of Hdm2 was never observed. No alterations in the p14^ARF sequence (a nucleotide change in exon 2, detected in one case,33 does not affect the p14^ARFprotein) or promoter hypermethylation were detected in any of the 42 cases analyzed in this group; 4 cases showed LOH in 9p21. p53 mutations were detected in a minority of cases (1 nonsense and 3 missense mutations out of 42 samples). (3) p14^ARF nuclear overexpression. A subset of NHLs (16/74) showed an atypical p14^ARF expression pattern characterized by intense nuclear staining that was not confined to the nucleolus (Figure 2J). Since mutations affecting the NrLS of p14^ARF have been described as partially or totally impeding the correct localization of this protein,29 both exons of the p14^ARF gene were sequenced in 9 of 16 cases. However, no alterations were detected. Allelic loss and promoter hypermethylation analyses, performed in the same samples, also yielded negative results. This group of cases (Table3) was characterized by high p53 and Hdm2 levels of expression (Figure 2K-L). When compared with the rest of the series, a statistically significant relationship was found between p14^ARF nuclear overexpression and higher expression of both p53 and Hdm2 (Kruskal-Wallis test, P = .001 for p53 and P = .010 for Hdm2). p53 was either wild-type or mutated in this subset of cases, but the frequency of p53 mutation was significantly higher (5/9 sequenced cases) when compared with the rest of the series (Fisher exact test, P = .005). There were no significant differences in the distribution of the cases with p14^ARF nuclear overexpression between the categories defined by histologic diagnosis (BL vs non-BL; Fisher exact test,P = .715) or IPI (Pearson chi square: 0.360,P = .948).

In order to characterize more precisely the subcellular localization of the p14^ARF, Hdm2, and p53 proteins, 5 tumor samples with different patterns of expression of p14^ARFwere analyzed by confocal microscopy. Double fluorescent immunolabeling was performed for p14^ARF and either C23 (nucleolar marker), p53, or Hdm2.

Existence of a nucleoplasmic fraction of p14^ARF was effectively observed in cases with abnormal p14^ARF nuclear overexpression, as demonstrated by double labeling with C23 (Figure3A). Although nucleolar exclusion was not observed, part of the signal was clearly extranucleolar. Nucleoplasmic p14^ARF was restricted to the large tumoral cells, whereas small lymphocytes showed distinct nucleolar p14^ARFstaining. This pattern contrasts with the more typically observed situation in both normal and tumoral cells, in which p14^ARFwas confined to the nucleolus (Figure 3B). C23 characteristic nucleolar staining demonstrates that the integrity of the nucleolus was always preserved, and anomalous localization of p14^ARF was not a consequence of structural alterations of the nucleolus.

Double labeling for p14^ARF-p53 revealed a partial colocalization of these proteins in the nucleoplasm, suggesting the existence of a fraction of nucleoplasmic p14^ARF not bound to p53 (Figure 3C). This colocalization was absent in samples exhibiting p14^ARF nucleolar localization (Figure 3D).

A striking pattern was observed for Hdm2, in that it was not found to colocalize with nucleoplasmic or nucleolar p14^ARF; in fact, these signals usually appeared in alternative cells, and Hdm2 never accumulated in the nucleolus (Figure 3E-F).

Since p14^ARF has been reported as accumulating in response to oncogenic stimuli,18-22 we examined whether p14^ARF nuclear overexpression had any correspondence with the biology of the tumor or clinical outcome of the patients.

To this end, Ki67 expression was quantified in 70 of 74 cases (range: 32%-100%, median: 82%) as a measure of proliferation index. Univariate analysis showed a strong association between p14^ARF nuclear overexpression and higher Ki67 levels (Kruskal-Wallis test, P = .002). Histology (BL vs non-BL) and p53 and Hdm2 expression were also significantly related to Ki67 (Table 4). However, after performing a multivariate analysis that included the variables found to be significant in the univariate analysis, the only variable that retained statistical significance, in addition to histology, was p14^ARF nuclear overexpression (P = .035).

To test this possible association between p14^ARF nuclear overexpression and tumor aggressivity further, an overall survival analysis was performed for the subset of cases (51/74) for which clinical follow-up information during at least 60 months was available. A worse prognosis was observed for the cases with nuclear p14^ARF when compared with the rest of the series (P = .060, crude analysis) (Table 5); this relationship proved to be statistically significant after adjusting the results by IPI and histology (P = .043; Figure4A). Since the small group of cases in which p14^ARF expression was not detected may include aggressive tumors in which p14^ARF overexpression is impeded by alterations at the gene level, the cases lacking p14^ARFexpression were excluded from the analysis and overall survival of the patients overexpressing p14^ARF was compared with that of the cases showing a nucleolar expression of the protein; again, differences were found to be significant (P = .037).

p14^ARF overexpression was also a negative predictor of survival when only DLBCLs were considered (P = .047, crude analysis); however, the reduced number of BL cases prevented us from performing informative statistical analyses for this histologic class as a separate group. When cases were grouped according to IPI and overall survival was analyzed separately for each group, a similar association between p14^ARF nuclear overexpression and prognosis was found for the category of medium IPI (2-3) (P = .055); analyses were not considered informative for the groups of low (0-1) or high (4-5) IPI due to the reduced number of events in the first case and the small number of patients with p14^ARF nuclear overexpression (2 patients) in the second.

In a previous report33 we analyzed different alterations in tumor suppressor pathways in a group of NHLs that included 51 of the samples described here. For these cases, information was available concerning p53 mutations, p16^INK4a inactivation by promoter hypermethylation, deletion, or mutation, and p27^KIP1overexpression, presumably reflecting its inactivation by CDK4-cyclin D3 binding.45 

First, the relationship between p14^ARF nuclear overexpression and each of these individual alterations was analyzed. As mentioned above, there was a strong association between p53 mutation and p14^ARF nuclear overexpression (P = .005). Presence of nuclear p14^ARF was also more frequent in tumors overexpressing p27^KIP1, although this did not reach statistical significance (Fisher exact test, P = .118).

An association between p16^INK4a alterations and p14^ARF overexpression was not found when all the possible mechanisms of inactivation of p16^INK4a were considered as a whole. However, this may be due to the fact that LOHs or homozygous deletions at 9p21 are likely to affect both genes simultaneously, preventing p14^ARF from being overexpressed. Effectively, when LOHs were excluded and only p16^INK4a promoter hypermethylation—plus a single mutation that did not affect p14^ARF—was considered, an almost significant association was obtained between these alterations and higher frequency of p14^ARF nuclear overexpression (Fisher exact test,P = .053). It was also observed that hypermethylation of the p14^ARF and p16^INK4a promoters were independent events, as could be inferred from their remarkably different incidence in this group of lymphomas (17/51 cases displayed p16^INK4a hypermethylation whereas only 1/51 cases had p14^ARF hypermethylation).

Finally, we analyzed the frequency of p14^ARF abnormal overexpression as a function of the number of accumulated alterations in p53, p16^INK4a, and p27^KIP1. As shown in Figure 4B, the proportion of cases with anomalous p14^ARFexpression increases with the number of cell cycle defects, this relationship being statistically significant (Pearson chi square: 10.233, P = .017). Interestingly, p14^ARFnuclear overexpression was not observed in any of the 19 cases where no alterations in the p53 and Rb pathways were found.

Information concerning regulation of the p14^ARF-Hdm2-p53 pathway is mainly derived from in vitro studies and animal models, whereas a comprehensive analysis of its significance in human tumors has been hampered by technical considerations such as the lack of appropriate antibodies for detecting the p14^ARF protein. We have chosen NHL as a model, since it features tumors with a high growth fraction in which p53 mutations are present in only a small proportion of cases, and therefore may be useful for revealing the complexity of molecular alterations taking place in this pathway.

Unlike what has been found in other neoplasias, molecular alterations resulting in loss of p14^ARF expression are rare in NHLs. Hypermethylation of p14^ARF promoter and deletions within the 9p21 region, the 2 main mechanisms for p14^ARFsilencing, have been detected in our series of cases only exceptionally. This situation sharply contrasts with that observed in lymphoid cell lines, in which cancellation of the p53 pathway by alternative—and occasionally simultaneous—inactivation of p14^ARF or p53 is a very frequent finding (Table 1). Tumors lacking detectable p14^ARF are characterized by low levels of p53 and Hdm2 proteins, which is consistent with the proposed role of p14^ARF as an inhibitor of Hdm2; thus, in the absence of p14^ARF, active Hdm2 promotes a rapid p53 degradation, which in turn leads to a low level of p53-induced Hdm2 expression.

In contrast with the exclusively nucleolar localization of p14^ARF observed in normal cells and most in vitro models, abnormal p14^ARF nucleoplasmic accumulation has been found in a significant number of aggressive NHLs. It does not seem likely that this is a consequence of nonspecific staining, as a distinct nucleolar signal has been obtained with the same antibody and under the same conditions in nontumoral tissue, many NHL samples, and several lymphoid cell lines. Moreover, no staining (either nucleolar or nucleoplasmic) has been detected in cell lines with p14^ARFsilencing by homozygous deletion (Figure 1). p14^ARF-C23 double immunolabeling has demonstrated that our observations are not an artifact related to loss of nucleolar integrity. We have also ruled out the possibility that nucleolar distribution is impeded by mutations affecting the p14^ARF NrLS, since the p14^ARFgene has been found to be wild-type in all the cases that have been sequenced.

The significance of this findings is enhanced by the observation that p14^ARF atypical overexpression defines a group of lymphomas characterized by higher aggresivity. A strong association exists between p14^ARF nuclear overexpression and higher proliferation index, which remains significant even in a multivariate analysis including relevant variables such as p53, Hdm2, and histologic diagnosis. A similar correlation existed at the prognostic level, since cases with abnormal p14^ARF overexpression were also characterized by a shorter overall survival.

An explanation for these findings can be found in the consideration of the central role of p14^ARF in cell cycle control, as a nexus between the major tumor suppressor pathways. Thus, the p14^ARF promoter has an E2F-1 binding site which “senses” oncogenic stimuli transduced through the Rb pathway19; p14^ARF is also induced by typical oncogenes (myc, ras, viral genes), and negatively regulated by p53 (Figure 5A). With the exception of alterations affecting p14^ARF itself, which we have shown to be very rare in NHLs, virtually every cancer-related defect in these pathways (Figure 5B) should result in p14^ARF upregulation: alterations of the Rb pathway such as p16^INK4a inactivation or cyclin D overexpression, deregulation of oncogenes (eg, myc in BL), and p53 inactivation by mutation or by Hdm2 overexpression (resulting in disruption of the p53-p14^ARF negative feedback loop). Therefore p14^ARF should integrate all these stimuli, its level of expression being a measure of the accumulation of alterations in different points of the cell cycle, and consequently a marker of tumor aggressivity. Consistent with this hypothesis, p14^ARFnuclear overexpression is a more frequent finding in tumors displaying simultaneous inactivation of several major tumor suppressors (p53, p16^INK4a, p27^KIP1), in addition to correlate with some of these alterations, taken individually. The postulation of an overexpressed nuclear p14^ARF as a surrogate of a highly deregulated cell cycle is consistent with the higher aggressivity (as measured by proliferation index and worse prognosis) observed in this group of lymphomas.

Even if p14^ARF overexpression is a consequence of cell cycle malfunction, the atypical nuclear localization of the protein remains an intringuing finding. Results of confocal microscopy suggest that p14^ARF nucleoplasmic accumulation is only partially dependent on p53 binding and probably independent of Hdm2 binding. A recent report has suggested that the predominant p14^ARFnucleolar accumulation is accompanied by an usually undetectable nucleoplasmic fraction which could be responsible for p53 activation.25 Consistent with this model, it would be expected that under conditions of massive p14^ARF induction the nucleoplasmic fraction would become detectable, as could be the case for a group of aggressive lymphomas.

Our findings may help to resolve the controversy concerning Hdm2 and p14^ARF subcellular localization, both in normal and tumoral cells. Thus, the presence of Hdm2 and p14^ARF seems to be mutually exclusive, as shown by the lack of nucleolar Hdm2 staining in cells expressing nucleolar p14^ARF, and the absence of Hdm2 in cells overexpressing nuclear p14^ARF. This suggests that p14^ARF-Hdm2 complexes, if they exist, should have a short half-life, dependent on a rapid Hdm2 degradation induced after p14^ARF binding.

We are indebted to A. I. Sáez for her invaluable help with the statistical analysis, and to I. Fernández and M. J. Acuña for their expertise and excellent technical assistance with molecular and immunohistochemical assays, respectively. We also thank Dr Juan Carlos Martı́nez-Montero for his kind collaboration in the initial stages of the immunohistochemical analysis.