p14^ARF Silencing by Promoter Hypermethylation Mediates Abnormal Intracellular Localization of MDM2¹

In a normal cell, microanatomical localization of a protein determines, many times, how it works. For example, the election between nuclear or cytosolic localization is tightly regulated. However, in cancer cells, this organized system often fails. The recent case of the oncoprotein MDM2 is an excellent example. MDM2 binds to and abrogates the tumor suppressor function of p53 by either targeting p53 for degradation in the cytoplasm or by repressing p53-mediated transcriptional activity in the nucleus (1). MDM2 at the same time is also under stringent regulation by the putative tumor suppressor p14^ARF. The p14^ARF protein is encoded by the INK4a/ARF locus on chromosome 9p21, a region with a high rate of loss of heterozygosity in human cancer. p14^ARF has a separate first exon (exon 1B) that splices into common exons 2 and 3, shared with the tumor suppressor gene p16^INK4a in a different reading frame, resulting in a protein bearing no amino acid sequence similarity to p16^INK4a (2). Showing its contribution to the tumorigenic process, the p14^ARF null mice (where only the exon 1B is lost) develop spontaneous tumors at an early age (3). p14^ARF physically interacts with MDM2, inducing stabilization of p53 (4, 5, 6, 7). Furthermore, recent elegant experiments indicate that the mechanism of MDM2-mediated regulation by p14^ARF involves intracellular compartmentalization (8, 9, 10). p14^ARF would stabilize p53 by sequestering MDM2 from the cytosol and confining it to the nucleus, mainly in the nucleolus (8, 9, 10).

Despite this important function for p14^ARF in tumorigenesis, p14^ARF specific genetic lesions in human cancer are very rare. The homozygous deletions and the mutations at the exon 2 at the INK4a/ARF locus also affect p16^INK4a. Mutations in p14^ARF exon 2, as expected, impair the delicate nucleocytoplasmic shuttling of MDM2 (10). However, no spontaneous mutations in the unique exon 1B of p14^ARF have been described to our knowledge. An alternative and rapidly growing area of study for gene inactivation involves epigenetic mechanisms. Transcriptional silencing by promoter hypermethylation has been demonstrated affecting bona fide tumor suppressor genes and DNA repair genes including p16^INK4a, VHL, BRCA1, and hMLH1 (11, 12). In this field, we and others (13, 14) have demonstrated that the promoter region of p14^ARF possesses a CpG island that undergoes dense hypermethylation in some colorectal cancer lines (13, 14) and primary tumors (14, 15). In vitro treatment with the demethylating drug 5-aza-2-deoxicytidine is able to restore p14^ARF expression (13, 14). The cellular functional consequences of p14^ARF epigenetic inactivation, however, remains unknown. In this study, we demonstrate that in cancer cells the methylation-mediated silencing of p14^ARF plays a central role in MDM2 localization.

The primary tumor samples used in this study were obtained at the time of the surgery from the Johns Hopkins Hospital and The Greenbaum Cancer Center in Baltimore and the Hospitals Sant Pau and Duran y Reynals in Barcelona as described previously (16, 17). The Ethics Committee in each institution approved specimen collection procedures. The colorectal cancer cell lines Colo 205, LoVo, SW480, HT29, and DLD1 used in this study were obtained from the American Type Culture Collection. Cell lines were maintained in appropriate medium, and LoVo cells were treated with the demethylating agent 5-aza-2′-deoxycytidine (Sigma) at a concentration of 1 μm for 3–5 days to achieve demethylation as described previously (14).

DNA methylation patterns in the CpG island of the p14^ARF gene were determined by methylation-specific PCR (14). Methylation-specific PCR distinguishes unmethylated from methylated alleles in a given gene based on sequence changes produced after bisulfite treatment of DNA, which converts unmethylated, but not methylated, cytosines to uracil, and subsequent PCR using primers designed for either methylated or unmethylated DNA (18). The primer sequences designed for p14^ARF spanned six CpGs within the 5′ region of the gene. Primer sequences of p14^ARF for the unmethylated reaction were 5′-TTT TTG GTG TTA AAG GGT GGT GTA GT-3′ (sense) and 5′-CAC AAA AAC CCT CAC TCA CAA CAA-3′ (antisense), which amplify a 132-bp product; and for the methylated reaction, 5′-GTG TTA AAG GGC GGC GTA GC-3′ (sense) and 5′-AAA ACC CTC ACT CGC GAC GA-3′ (antisense), which amplify a 122-bp product. The 5′ position of the sense unmethylated and methylated primers corresponds to bp 195 and 201 of GenBank sequence no. L41934. Both antisense primers originate from bp 303 of this sequence. The annealing temperature for both the unmethylated and methylated reactions was 60°C. Placental DNA treated in vitro with SssI methyltransferase was used as a positive control for methylated alleles. DNA from normal lymphocytes was used as negative control for methylated genes. Ten μl of each PCR reaction was directly loaded onto nondenaturing 6% polyacrylamide gels, stained with ethidium bromide, and visualized under UV illumination.

Reverse transcription (RT)-PCR was performed as described previously (14), using 3 μg of total cellular RNA to generate cDNA. One hundred ng of this cDNA were amplified by PCR with primers for exon 1β (5′-GGT TTT CGT GGT TCA CAT CCC GCG-3′) and exon 2 (5′-CAG GAA GCC CTC CCG GGC AGC-3′) of p14^ARF, which amplify a 254-bp product spanning sequence 204–437 from GenBank S78535. RT-PCR for GAPDH served as a positive control. Ten μl of each PCR reaction were directly loaded onto nondenaturing 6% polyacrylamide gels, stained with ethidium bromide, and visualized under UV illumination. Cell lysates for protein analysis were prepared and analyzed by Western blotting by using three anti-p14^ARF antibodies, Ab-1 (rabbit polyclonal antibody IgG; Neomarkers, Fremont, CA), Ab-2 (mouse monoclonal antibody IgG; Neomarkers), and C-18 (goat polyclonal IgG; Santa Cruz Biotechnology, Santa Cruz, CA). Equal loading was tested by reprobing with a polyclonal antibody against human topoisomerase I (Topogen, Inc., Columbus, OH). The gels were cast using the XCell SureLock Mini-Cell system (Invitrogen Corp./NOVEX, Carlsbad, CA) and developed using ECL immunodetection reagents (Amersham Pharmacia Biotech, Piscataway, NJ).

Immunoperoxidase staining of deparaffinized sections was performed as described previously (19). Briefly, deparaffinized sections were treated with 1% H₂O₂ to block endogenous peroxidase activity, subsequently immersed in boiling 0.01% citric acid in a microwave oven to enhance antigen retrieval, allowed to cool, and incubated with 10% normal horse serum to block nonspecific tissue immunoreactivities. Primary antibodies were incubated overnight at 4°C. Biotinylated horse antimouse IgG antibodies (Vector Laboratories, Burlingame, CA; 1:500 dilution) were applied for 30 min, followed by avidin-biotin-peroxidase complexes (Vector Laboratories; dilution, 1:25). Diaminobenzidine was used as the final chromogen. Cytospins of the cells were prepared using a Cytospin 2 (Shandon, Cheshire, United Kingdom) at 1400 rpm for 4 min and fixed in acetone:methanol for 1 min and stained using a similar protocol as described above. A panel of well-characterized antibodies were used, which included mouse monoclonal antibodies to MDM2 (clone 2A10, to the human MDM2 product; kindly supplied by Dr. A. Levine, Rockefeller University, New York, NY; dilution, 1:500), p53 (clone DO-7, to the human p53 product; DAKO Corp., Carpinteria, CA), and Rb³ (clone 1F8, to the human Rb product; Neomarkers). Staining was evaluated in coded slides by three authors (C.C-C., K. S. P., and D. N. W.), who had no knowledge of the results of the molecular analyses.

To address first how p14^ARF aberrant methylation affects the patterns of MDM2 expression, we analyzed the p14^ARF methylation and expression status of several colorectal cell lines and the corresponding MDM2 protein localization. Methylation-specific PCR analysis showed that the colorectal cancer cell lines Colo205, SW837, and SW480 were unmethylated at the p14^ARF CpG island, whereas LoVo was fully methylated at this locus (Fig. 1). Concordant with this analysis, p14^ARF expression at the mRNA level was abundant in the unmethylated cell lines Colo205, SW837, and SW480, but no p14^ARF transcript was detected in LoVo, a p14ARF hypermethylated cell line, as described previously (data not shown; Ref. 14). Western blot analysis corroborated this data showing undetectable p14^ARF protein in LoVo cells, whereas the unmethylated cell line SW480 demonstrated abundant p14^ARF expression (Fig. 1). When LoVo cells were treated with the demethylating agent 5-aza-2-deoxycytidine, reexpression of p14^ARF was observed at the transcript and protein levels (Fig. 1), coincident with the appearance of unmethylated alleles (Fig. 1).

These cell lines were then cytospinned, fixed in histological slides, and studied for MDM2 protein status by immunostaining. The p14^ARF unmethylated and expressing cell lines Colo205, SW837, and SW480 demonstrated exclusive nuclear staining as shown in Fig. 2,a. However, LoVo cells (p14^ARF hypermethylated and nonexpressor), in addition to the nuclear staining, showed strong cytosolic staining (Fig. 2 a). This would be the pattern expected if p14^ARF function was impaired; MDM2 cannot be sequestered in the nucleus and may “leak” to the cytosol. A careful examination of the MDM2 protein staining revealed that the nuclear staining in the p14^ARF unmethylated cell lines is compatible with a nucleolar localization. Interestingly, the nuclear staining that remained in the methylated cell line LoVo is more homogeneously distributed in the nucleus, suggesting a nucleoplasmic nonnucleolar localization.

Unlike mutational inactivation, epigenetic changes offer the possibility of reversibility. For other genes, such as the mismatch repair gene hMLH1 and the apoptotic gene DAPK, cell treatment with demethylating agents not only restores gene expression but also gene function in those respective cases where DNA repair activity and IFN-γ induced apoptosis (20, 21). We examined the MDM2 pattern of expression in the LoVo cell line after 5-aza-2-deoxycytidine treatment. LoVo cells treated with a demethylating agent resulted in MDM2 cytosolic staining no longer being observed, and the only demonstrable MDM2 staining was limited to the nucleus (Fig. 2 a). Furthermore, its pattern is clearly suggestive of a nucleolar localization. Thus, these data support that the reexpression of p14^ARF was able to reinternalize the “leaked” cytosolic MDM2 to the nucleus.

We next examined whether the reexpression of p14^ARF and the resultant return of MDM2 to the nucleus affect targets downstream in this pathway. We studied the expression and localization of two described MDM2 interactors, p53 and Rb (22, 23). We observed recently that in cancer cell lines with a mutant p53, the p14^ARF promoter remains unmethylated (14), following the reasoning that only one “hit” in the same pathway is necessary. Thus, in the cell lines SW480, Colo205, and SW837, which all harbor a mutant p53 gene, a strong nuclear p53 expression was observed (Fig. 2,b). In addition, in LoVo, a cell line with wild-type p53 and a methylated p14^ARF promoter, p53 expression was not detectable (Fig. 2 b). However, after p14^ARF expression was induced by the treatment with the demethylating drug, not only MDM2 was now normally confined in the nucleus as described above, and some nuclear p53 expression was detectable, consistent with the p14^ARF function of triggering p53 stabilization (6, 24, 25). In contrast, no difference in Rb expression or localization was observed in these cell lines independently of their p14^ARF methylation status, and no change was observed upon reexpression of p14^ARF after the treatment with 5-aza-2′-deoxycytidine (data not shown).

The above data indicating that epigenetic silencing of p14^ARF disrupts MDM2 function impelled us to examine the prevalence of the p14^ARF promoter hypermethylation alteration in human cancer. Previously, this question has only been addressed in primary colorectal and gastric tumors (14, 15). The study of an extensive panel of 559 human primary tumors revealed that p14^ARF aberrant methylation has a tumor type-specific pattern. p14^ARF promoter hypermethylation was particularly common in gastrointestinal tumors, being found in 28% of colorectal, 26% of gastric, and 22% of esophageal tumors (Table 1). It was also found in 15% of endometrial and 13% of renal tumors. However, other tumor types, including lung and ovary, show that p14ARF methylation in <5% of cases and other tumor types such as breast, pancreas, or liver are devoid of this epigenetic lesion. No evidence of p14^ARF methylation was found in any normal tissue analyzed. Table 1 summarizes all of the data obtained in primary tumors, and Fig. 3 illustrates several examples of the tumors analyzed for p14^ARF methylation.

Similar to the study developed in the cell lines, we wondered about the putative relation between p14^ARF promoter hypermethylation and MDM2 intracellular distribution in primary neoplasms. Thus, we analyzed, blinded to molecular data concerning the p14^ARF methylation status, the localization of the MDM2 protein in histological sections of 45 primary tumors (33 colorectal and 12 endometrial), where the p14^ARF methylation status had been studied. Exclusive nuclear staining was observed in the vast majority, 81% (22 of 27) of the tumors unmethylated at p14^ARF, whereas only 19% (5 of 27) presented both nuclear and cytosolic positivity. In comparison, the tumors with hypermethylation at p14^ARF demonstrated a higher rate of cytosolic staining, being present in a 44% of the cases (8 of 18). A representative example of the methylation analysis and the two patterns of MDM2 protein localization observed are illustrated in Fig. 4. Thus, the trend between p14^ARF epigenetic inactivation and MDM2 cytosolic localization was present but without the exclusivity demonstrated in the cell lines. We observed a similar phenomena previously when comparing p14^ARF methylation and p53 mutational status. In vitro, all of the cell lines hypermethylated at p14^ARF are p53 wild type; in primary tumors, only a trend toward this relation was observed (14).

In human cancer, the frequency that the INK4a/ARF locus is disrupted is almost as prevalent as p53 mutations (26). Loss of heterozygosity, homozygous deletions, and point mutations occur at this site, but a major mechanism for inactivation of both genes is epigenetic silencing by promoter hypermethylation (12, 14). Aberrant methylation of the cell cycle inhibitor p16^INK4a in cancer cells has been described extensively, and recently the other occupant of the INKa/ARF locus, p14^ARF, has also been shown to undergo hypermethylation-associated inactivation in cultured cancer cells and primary tumors (13, 14, 15). This last alteration is particularly intriguing in view of the recently recognized interplay between p14^ARF and MDM2 and because aberrant methylation of p14^ARF can have important consequences in the p53 pathway. Supporting this postulate, we now provide data that the transcriptional inactivation of p14^ARF by the methylation of its CpG island has important effects for the compartmentalization of the MDM2 protein in cancer cells.

The tumor suppressor function of p14^ARF is thought to be related to binding to MDM2, thereby blocking p53 degradation by inhibiting the E3 ubiquitin-ligase activity associated with MDM2 (27) and preventing nuclear export of MDM2 (8, 10). A proposed step in this process is that a fully functional p14^ARF should sequester MDM2 in the nucleolus. p14^ARF has nucleolar localization signals and is predominantly localized in nucleoli but also can be noted in the nucleoplasm (28). Consistent with these data, our results show that in colorectal cancer cells with an intact and unmethylated p14^ARF gene, MDM2 is confined to the nucleus with a pattern compatible with nucleolar localization. However, if p14^ARF expression is shut-down by promoter hypermethylation, MDM2 became predominantly nucleoplasmic and even cytosolic. These data agree with several studies (9, 10, 29) that also show that in the absence of p14^ARF, MDM2 was localized mainly in the nucleoplasm of the transfected cells, with clear evidence of nucleolar exclusion. In our case, because p14^ARF was only silent but genetically intact, and taking advantage of the well-recognized capacity of releasing the epigenetic silencing by the use of the demethylating agent 5-aza-2′-deoxycytidine, we were able to “rescue” p14^ARF function. The induced reexpression of p14^ARF now relocalized MDM2 to the nucleus in a pattern that closely resembled that observed in the unmethylated cancer cell lines. Similar experiments restoring the function of a gene silenced by methylation has also been done in the past with the mismatch repair gene hMLH1 and the apoptotic gene DAPK (20, 21). Furthermore, the reactivation of p14^ARF, coincident with the redistribution of MDM2, was able to induce nuclear p53 expression. Supporting our data, the transient transfection with a cDNA encoding p14^ARF also causes a nuclear p53 accumulation (30). We did not observe p53 expression in all cells treated with 5-aza-2′-deoxycytidine, most likely because of incomplete demethylation of p14^ARF (Fig. 1).

The relation between the p14^ARF methylation status and MDM2 localization is not a phenomenon restricted to cancer cell lines. After a screening of more than 500 primary human tumors of different cell types, p14^ARF promoter hypermethylation was found as a relatively common event in several neoplasms, including colorectal, gastric, and uterine tumors. The patterns of MDM2 expression in a set of these tumors showed a trend between the presence of p14^ARF epigenetic silencing and abnormal localization of MDM2. This association between both events was not as exclusive as we found with the cell lines, but it agrees with other studies that show that p14^ARF loss of function, either by epigenetic (14) or genetic (31) mechanisms, is not either restricted to primary tumors with wild-type p53. Mouse models also provide similar examples of overlapping between p14^ARF, MDM2, and p53 defects, such as the presence of some p53 mutations in the tumors arising in the p19^ARF knock-out (3). These emergent data may illustrate the fact that the relation of p14^ARF, MDM2, and p53 is not always linear. Each component may have other potential partners such as Rb, p300, E2F1, and other p53 family members for MDM2 (26); each protein may receive cross-talk from other cell networks, i.e., Ras regulates MDM2 and p19^ARF (32); and each gene may suffer a wide spectrum of concomitant lesions (epigenetic silencing, gene amplification, homozygous deletion, and point mutation) in a primary tumor, that is more heterogeneous than in a cell line. For example, a primary tumor with p14^ARF methylation may also harbor a p53 mutation for the p14^ARF functions unrelated to its binding to MDM2 or the MDM2 functions unrelated to its binding to p53. Finally, there are complex feed-back loops, i.e., p53 has been shown to repress the p14^ARF promoter (13), MDM2 is itself a p53-responsive gene (33), and MDM2-p53 levels oscillate in a tightly regulated manner (34). Nevertheless, considering the vast amount of literature generated in recent years about the presence of altered expression of MDM2 in a wide variety of human tumors, the loss of p14^ARF function by aberrant methylation seems to be a central factor behind the observed abnormalities.

In summary, the data compiled in this current work support an important role for the putative tumor suppressor gene p14^ARF in delimiting where the oncoprotein MDM2 is localized in the cell. Furthermore, it shows how the epigenetic silencing of p14^ARF by promoter hypermethylation has profound consequences in that process, allowing an aberrant MDM2 compartmentalization in the transformed cell.

We thank Kornel Schuebel for technical advice and helpful comments.