|Home | About | Journals | Submit | Contact Us | Français|
The ARF tumor suppressor is a nucleolar protein that activates p53-dependent checkpoints by binding Mdm2, a p53 antagonist. Despite persuasive evidence that ARF can bind and inactivate Mdm2 in the nucleoplasm, the prevailing view is that ARF exerts its growth-inhibitory activities from within the nucleolus. We suggest ARF primarily functions outside the nucleolus and provide evidence that it is sequestered and held inactive in that compartment by a nucleolar phosphoprotein, nucleophosmin (NPM). Most cellular ARF is bound to NPM regardless of whether cells are proliferating or growth arrested, indicating that ARF-NPM association does not correlate with growth suppression. Notably, ARF binds NPM through the same domains that mediate nucleolar localization and Mdm2 binding, suggesting that NPM could control ARF localization and compete with Mdm2 for ARF association. Indeed, NPM knockdown markedly enhanced ARF-Mdm2 association and diminished ARF nucleolar localization. Those events correlated with greater ARF-mediated growth suppression and p53 activation. Conversely, NPM overexpression antagonized ARF function while increasing its nucleolar localization. These data suggest that NPM inhibits ARF's p53-dependent activity by targeting it to nucleoli and impairing ARF-Mdm2 association.
INK4a/ARF is the second most frequently inactivated gene locus in human cancers (65). It encodes two unrelated tumor suppressor proteins, p16INK4a and ARF (alternative reading frame), from overlapping reading frames within its second exon (14, 44, 59, 68). Whereas p16 functions specifically in the retinoblastoma protein tumor suppressor pathway, ARF is a nucleolar protein that protects cells from oncogenic transformation through partially defined p53-dependent and p53-independent pathways (67). The major pathway activated by ARF is p53-dependent cell cycle arrest or apoptosis, which requires inactivation of the Mdm2 oncogene (13, 30, 49, 58, 69, 82, 84). p53 is a transcription factor that activates numerous growth suppressive genes in response to cellular stresses (34), whereas Mdm2 is a p53-responsive gene that encodes an E3 ubiquitin ligase that plays a critical role in restricting p53 activity (50, 51). ARF stabilizes and activates p53 by binding Mdm2, thereby blocking its ability to inhibit p53 function.
The molecular mechanisms underlying Mdm2 inhibition by ARF are presently unclear. While Mdm2 localizes to the nucleoplasm, the majority of ARF resides in nucleoli (39, 40, 59, 78), subnuclear compartments that are sites of ribosomal assembly (17). It was originally proposed that ARF physically sequesters Mdm2 in nucleoli, thus relieving nucleoplasmic p53 from Mdm2 control (74, 78). Despite its elegant simplicity, that model has been challenged by work showing that endogenous Mdm2 is not relocalized from the nucleoplasm to nucleoli during ARF-induced growth arrest (31, 38, 40) and that nonnucleolar forms of exogenous ARF can activate p53 and suppress growth (40, 48, 62). Together, those findings indicate that nucleolar localization is nonessential for ARF function.
Perhaps because of its distinctive localization to nucleoli, there nevertheless remains a strong perception that ARF signaling emanates from within the nucleolus. The predominant idea is that ARF inactivates a variety of cellular proteins, besides Mdm2, by physically sequestering them in nucleoli. In that regard, ARF associates with numerous proteins, many of which relocalize to nucleoli upon overexpression with ARF (10, 16, 21, 26, 28, 45, 57, 63, 83). Unfortunately, there has been no demonstration that nucleolar relocalization of those proteins occurs or even correlates with ARF function under physiologic conditions. Other proteins may collaborate with ARF in nucleoli (21, 83), or conversely, abolish ARF function by drawing it out of the nucleolus into the cytoplasm (71). Finally, some work suggests that ARF may exert p53-independent activities within the nucleolus. ARF can inhibit growth in the absence of p53 and Mdm2 (5, 15, 46, 75), and recent work reveals that it retards rRNA processing independently of p53 (72). It was proposed that ARF might impair ribosome biogenesis and cellular growth by associating with and promoting the degradation of the nucleolar phosphoprotein, nucleophosmin (NPM; also called B23, NO38, or numatrin) (2, 25).
NPM is an extremely abundant, highly conserved protein that resides most prominently in nucleoli, although it shuttles rapidly between the nucleus and cytoplasm (4). It is typically upregulated by mitogenic signals and is a Myc transcriptional target (20, 81), consistent with the idea that NPM promotes cell growth and survival. Indeed, malignant and actively dividing cells express elevated levels of NPM (12, 52, 70), and cells expressing large amounts of NPM are resistant to apoptosis induced by either UV damage or hypoxia (35, 79). Conversely, NPM downregulation delays the cell cycle and M-phase entry (27). NPM is a multifunctional protein primarily associated with ribosomal protein assembly and transport, although it is also involved in centrosome duplication, targeting proteins to nucleoli, and preventing protein aggregation (36, 37, 53, 73, 75). It is also frequently targeted in chromosomal translocations associated with leukemias, which results in the expression of highly oncogenic NPM fusion proteins (7, 61, 80). Similar to other proto-oncogenes, NPM overexpression in primary cells stabilizes and activates p53 (8, 25). Taken together, the cumulative evidence suggests NPM is a proto-oncogene that is required for growth, but its deregulated expression in normal cells may activate p53-dependent checkpoints.
In this study, we show that NPM negatively regulates ARF, providing a new twist on recent work suggesting that ARF inhibits NPM (2, 25). Specifically, we discovered that NPM targets ARF to nucleoli and blocks ARF-mediated p53 activation and growth suppression in a dose-dependent manner. When NPM expression levels are reduced, ARF is released from its nucleolar constraints and exhibits significantly greater Mdm2 association, p53 activation, and growth-inhibitory activity. We propose that NPM normally retains ARF in nucleoli, thereby impairing its ability to interact with nucleoplasmic Mdm2 and stimulate p53. These findings challenge the perception that ARF's primary site of action is in the nucleolus.
NIH 3T3 fibroblasts, monkey COS cells, human 293T and U2OS osteosarcoma cells, and Narf6 cells (provided by Gordon Peters, Imperial Cancer Research Fund), were maintained in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum, 2 mM glutamine, and 100 μg/ml each of penicillin and streptomycin. Primary mouse embryo fibroblasts (p21−/−, provided by Phil Leder, Harvard University; ARF−/−, provided by Martine Roussel and Chuck Sherr, St. Jude Children's Research Hospital; and p53/Mdm2/ARF−/−, provided by Gerry Zambetti, St. Jude Children's Research Hospital) were grown in the same medium supplemented with 0.1 mM nonessential amino acids and 55 μM 2-mercaptoethanol. Production of ecotropic retroviruses and infections into fibroblasts was performed with pSRalpha-MSCV-tkneo plasmids containing hemagglutinin (HA)-tagged wild-type mouse p19ARF or its mutants, as described (59). Cells were transfected with Lipofectamine (Gibco-BRL) or by modified calcium phosphate precipitation (6).
Expression plasmids encoding various forms of mouse and human ARF as well as glutathione S-transferase (GST)-tagged NPM fusion proteins have been described previously (8, 31, 77). HA-tagged human NPM and its mutants were generated by PCR amplification with a GST-tagged full-length NPM cDNA template. Wild-type NPM residues 1 to 295 (forward primer 5′-CCATCGATGAAGATTCGATGGACATGGACATGAG-3′ and reverse primer 5′-GGAATTCTAAAGAGACTTCCTCCACTGCCAG-3′), NPM.117 to 259 (forward primer 5′-CCATCGATGCTGTGGAGGAAGATGCAGAGTC-3′ and reverse primer 5′-GGAATTCAACCACCTTTTTCTATACTTGCTTGC-3′), NPM.187-259 (forward primer 5′-CCATCGATGAAGAAAAAGCGCCAGTGAAGAAATC-3′ and reverse primer 5′-GGAATTCAACCACCTTTTTCTATACTTGCTTGC-3′), NPM.117-295 (forward primer 5′-CATCGATGCTGTGGAGGAAGATGCAGAGTC-3′ and reverse primer 5′-GGAATTCTAAAGAGACTTCCTCCACTGCCAG-3′), and NPM.187-295 (forward primer 5′-CATCGATGAAGAAAAAGCGCCAGTGAAGAAATC-3′ and reverse primer 5′-GGAATTCTAAAGAGACTTCCTCCACTGCCAG-3′). The PCR products were digested with EcoRI and ClaI restriction enzymes and subcloned into similarly digested pXM-HA vector (22).
For RNA interference-mediated knockdown of NPM, double-stranded oligonucleotides targeting NPM for silencing (5′-GAATTGCTTCCGGATGACT-3′) or a point mutant control (5′-GAATTGCTTACGGATGACT-3′) were subcloned into the pSUPER-neo plasmid (OligoEngine), according to the manufacturer's specifications. Dilutional subcloning and selection in neomycin (0.6 mg/ml) yielded U2OS-derived monoclonal cell lines stably expressing the pSUPER vector, point mutant small hairpin RNA control, or the knockdown small hairpin RNA construct. Clones with specific NPM knockdown compared to parental or vector control cells were identified by immunoblotting for NPM and nonspecific targets (tubulin, nucleolin, p53, Stat5, and Mdm2).
Two days after treatment with isopropylthiogalactopyranoside (IPTG), Narf6 cells were labeled with 0.75 to 1 mCi of 32P per ml for 3 h at 37°C in 5% CO2, rinsed once with phosphate-buffered saline, and lysed on the dish on ice with high-salt buffer (50 mM Tris, pH 8.0, 500 mM NaCl, 1 mM EDTA, 0.5% NP-40). Lysates were incubated on ice for 1 h and clarified by centrifugation at 15,000 rpm for 10 min at 4°C. To reduce nonspecific background, lysates were precleared with isotype-matched immunoglobulin G prior to immunoprecipitations for human p14ARF (10 μg/immunoprecipitation, Novus; 6 μg/immunoprecipitation, DCS-240, Sigma) or NPM (1.5 μg/immunoprecipitation, Zymed). Immunoprecipitations were carried out overnight at 4°C with protein A or protein G Sepharose, and resin was washed four times with high-salt buffer prior to resolution of the protein complexes by sodium dodecyl sulfate (SDS)-polyacrylamide gel electrophoresis (PAGE) and transfer to polyvinylidene difluoride membranes (Millipore). ARF-associated phosphoproteins were detected by autoradiography and immunoblotting.
Frozen cell pellets were lysed on ice at approximately 107 cells/ml in NP-40 buffer (50 mM Tris, pH 7.5, 120 mM NaCl, 1 mM EDTA, 0.5% NP-40) supplemented with 0.1 μg of protease inhibitor cocktail (Sigma) per ml and 30 μM phenylmethylsulfonyl fluoride. Lysates were briefly vortexed and sonicated (5-s pulse) and incubated on ice for 1 h prior to clarification by microcentrifugation at 14,000 rpm for 10 min at 4°C. Immunoprecipitations were performed with protein A or G Sepharose plus antibodies against mouse p19ARF (59), human p14ARF, NPM, Mdm2 (2A10 mouse monoclonal antibody, 200 μl/immunoprecipitation), or the HA epitope (3F10 rat monoclonal conjugated to agarose, Roche). Immune complexes were washed, separated by SDS-PAGE, and analyzed by Coomassie blue staining of fixed gels or immunoblotting.
For Western blot analyses, equivalent amounts of total cellular protein were separated by SDS-PAGE and transferred to polyvinylidene difluoride membranes. Proteins were detected by enhanced chemiluminescence (ECL, Amersham) with the following antibodies: anti-mouse p19ARF (Novus NB 200-106, 1 μg/ml; Calbiochem, Ab-1 polyclonal, 1 μg/ml), anti-human p14ARF (Sigma, clone DCS-240, 2 μg/ml; Novus, 1 μg/ml), anti-Mdm2 (2A10, 1:75 dilution; Ab-1, Oncogene Research Products, 1 μg/ml), anti-NPM (Zymed Laboratories Inc., clone FC-61991, 0.5 μg/ml), antinucleolin (Santa Cruz Biotechnology, clone MS-3, 1 μg/ml), anti-human p21 (BD Pharmingen, 2 μg/ml), anti-p53 (Oncogene Research Products, Ab-7, 1:2,500 dilution), and anti-gamma tubulin (Sigma, 1:10,000 dilution).
Coupled in vitro transcription and translation of plasmids containing human p14ARF, mouse p19ARF, or mouse ARF mutants was performed with the TNT kit (Promega). Radiolabeled proteins were incubated for 2 h at 4°C on a rotator with equivalent amounts of GST, GST-NPM, or GST-NPM mutant proteins previously bound to GST-Sepharose. Complexes were washed four times with NP-40 buffer and separated by SDS-PAGE. Gels were fixed (30% methanol, 10% acetic acid) and dried, and binding was detected and quantified with a PhosphorImager (Molecular Dynamics).
For localization assays, cells were transfected with pSRalpha-MSCV-(human ARF)-tkCD8 or pSRalpha-MSV-(d2-14)-tkCD8 plasmid plus pXM-HA vector, pXM-HA-NPM (wild type), or pXM-HA-NPM.117-295. One day posttransfection, cells were trypsinized and seeded (104 cells/well) onto eight-well poly-l-lysine-coated chamber slides (Becton Dickinson). The next day, cells were fixed for 10 min with 4% paraformaldehyde, permeabilized for 15 min with 0.2% Triton X-100, and human p14ARF was detected by staining with anti-ARF antibodies (Sigma, clone DCS-240, 1 μg/ml), biotinylated anti-mouse IgG (Amersham, 1:200), and streptavidin-conjugated Texas Red (Amersham, 1:200). Mouse p19ARF was detected after fixing cells in methanol-acetone (1:1) for 10 min at −20°C and staining with a rat monoclonal antibody (5-C3-1; dilution 1:100; kindly provided by Martine Roussel, St. Jude Children's Research Hospital) (1) followed by fluorescein-conjugated anti-rat antibodies (Amersham, 1:100).
To assess nucleolar integrity, paraformaldehyde-fixed cells were stained with antibodies to endogenous fibrillarin (1:5 dilution, ANA-N; Sigma) followed by incubation with biotinylated anti-human IgG (Amersham, 1:200) and streptavidin-Texas Red, as above. Exogenous HA-NPM was detected with HA-fluorescein isothiocyanate-conjugated IgG (Roche) at 1:50. Nuclei were visualized by staining with 4′,6′-diamidino-2-phenylindole (DAPI) at 1 μg/ml for 1 min, and immunofluorescence was analyzed by confocal microscopy (Bio-Rad).
Cell cycle progression into S phase was measured by bromodeoxyuridine incorporation into replicating DNA (31). Briefly, cells were transfected with various ARF constructs or with empty vector alone and reseeded onto eight-well chamber slides 48 h after transfection. Bromodeoxyuridine (10 μM) was added to the culture medium 22 to 24 h prior to fixation in methanol-acetone (1:1) for 10 min at −20°C, and cells were stained for human ARF as described above. Cells were treated with 1.5 N HCl for 10 min and stained with a sheep polyclonal antibody to bromodeoxyuridine (Abcam) for 1 h, and bromodeoxyuridine incorporation was detected with a fluorescein isothiocyanate-conjugated anti-sheep IgG (Amersham). Nuclei were stained with DAPI, and ARF-positive cells were visualized by fluorescence microscopy (Zeiss).
U2OS cells were plated in six-well dishes at 105 cells/well and transfected the following day with a total of 12 μg of DNA. A p53 reporter construct, p53-luc (Stratagene; 800 ng), containing p53-responsive enhancer elements fused to the firefly luciferase gene, was cotransfected with a plasmid encoding human ARF (pSRalpha-MSV-humanARF-tkCD8; 1 μg) and empty vector plus various amounts of pXM-HA.NPM. A pRL-SV40 construct containing Renilla luciferase (Promega; 100 ng) was included in all transfections as an internal control to normalize for transfection efficiency. For all experiments, luciferase activity was measured in triplicate samples 48 h after transfection with the dual-luciferase reporter assay system (Promega) and a Sirius Luminometer V3.1 machine (Berthold Detection Systems).
In the ARF-Mdm2-p53 pathway, phosphorylation plays an important role in regulating the functions of Mdm2 and p53 (18, 50). ARF contains numerous potential sites of phosphorylation, and one study suggested that it could be a target of death-associated protein kinase (60). Since little is known about the posttranslational regulation of ARF, we examined whether it is a phosphoprotein. Human Narf6 (ARF-null, p53 wild type) cells, a derivative of U2OS osteosarcoma cells, express IPTG-inducible ARF and undergo p53-dependent growth arrest (69). After treatment of Narf6 cells with IPTG for 2 days, cells were transiently labeled with 32P-labeled inorganic phosphate. Human ARF complexes were then separately immunoprecipitated from cell lysates with antibodies raised against different regions of human ARF, and proteins were separated by denaturing polyacrylamide gel electrophoresis and transferred to membranes (Fig. (Fig.1).1). Western blotting confirmed efficient precipitation of human ARF by each antibody, yet repeated experiments showed no phosphorylation of ARF by autoradiography (Fig. (Fig.1A).1A). Phospholabeling analyses performed in NIH 3T3 fibroblasts arrested by mouse ARF similarly failed to show ARF phosphorylation (data not shown). These data strongly suggest that, at least under these conditions, ARF does not exist as a phosphoprotein in vivo.
Our analyses revealed that ARF specifically associates with several phosphoproteins in growth-arrested cells. Autoradiograms of ARF complexes recovered from phospholabeled Narf6 cells showed prominent interactions between ARF and three distinct phosphoproteins of approximately 37, 90, and 110 kDa (Fig. (Fig.1A).1A). Mdm2 is the first-identified and best-characterized binding partner of ARF (29, 58, 69, 82). It is a highly phosphorylated protein that typically migrates between 70 and 90 kDa on reducing gels (55). We probed the same membrane for Mdm2 and confirmed that the 90-kDa phosphoprotein in the ARF complexes comigrated with human Mdm2 present in the complex.
Immunoblotting was also used to test whether the 37-kDa phosphoprotein was nucleophosmin (NPM). We suspected that it was NPM for several reasons: NPM is a 37-kDa phosphoprotein predominantly located in nucleoli (4), it colocalizes with ARF (39, 40), and observations shared with us from Jason Weber's laboratory suggested that ARF interacted with NPM (personal communication). Western blotting with NPM antibodies confirmed that the ARF-associated 37-kDa phosphoprotein comigrated with NPM (Fig. (Fig.1A).1A). Consistent with that finding were observations that ARF coprecipitated with nucleolin/C23 (Fig. (Fig.1A1A and and1B),1B), an abundant, 110-kDa nucleolar phosphoprotein known to associate with NPM (19, 37). We initially tested whether the 110-kDa phosphoprotein was nucleolin based upon our previous identification of nucleolin as an ARF-interacting protein in a yeast two-hybrid screen (J. Hagen, X. Luo, and D. E. Quelle, unpublished data).
Association between ARF and NPM was supported by immunofluorescence studies showing that ARF and NPM colocalize within the nucleoli of IPTG-treated Narf6 cells (Fig. (Fig.1C).1C). The staining pattern for ARF and NPM indicated that both proteins resided within the granular region of nucleoli, in keeping with previous work assigning localization of each protein to that compartment (3, 39, 78). Reciprocal immunoprecipitation-Western blot analyses with antibodies to both proteins in IPTG-stimulated Narf6 cells further established the existence of ARF-NPM complexes (Fig. (Fig.1D).1D). Although the nonquantitative nature of the NPM immunoprecipitations limits conclusive interpretation of the data, most of the immunoprecipitated ARF appeared to complex with NPM, while only a minor amount of NPM coprecipitated with ARF.
NPM overexpression was recently shown to inhibit proliferation in primary fibroblasts via stabilization and activation of p53 (8, 25). Since ARF inhibits growth through p53-dependent and p53-independent pathways, we tested whether its association with NPM correlated with proliferation and/or was affected by p53. For this purpose, NPM-ARF interactions were examined in p53-negative BALB/c 3T3 mouse fibroblasts (designated 10-1) versus p53-positive NIH 3T3 cells infected with vector control or ARF retroviruses (Fig. (Fig.2).2). The 10-1 cells proliferate rapidly in culture despite high expression levels of endogenous ARF (30, 59), whereas NIH 3T3 fibroblasts undergo a complete G1 and G2 cell cycle arrest in response to ARF (59).
Reciprocal immunoprecipitation-Western blots showed that endogenous ARF and NPM are associated in rapidly dividing 10-1 cells (Fig. (Fig.2A),2A), and the efficiency of complex formation was equivalent to that observed in ARF-arrested NIH 3T3 cells (Fig. (Fig.2B).2B). ARF-NPM complexes were also detected in G1-arrested ARF- Mdm2- p53-null primary mouse embryo fibroblasts expressing exogenous ARF (data not shown). These results revealed that ARF-NPM association does not require p53, in agreement with other recent studies (2, 25). More importantly, however, the data from 10-1 cells show that ARF-NPM association does not correlate with growth suppression.
To identify the regions within mouse ARF required for association with NPM, reciprocal immunoprecipitation-Western blotting was performed in both COS-7 and NIH 3T3 cells ectopically expressing the vector control, wild-type mouse ARF, or various ARF mutants (Fig. (Fig.3A).3A). Wild-type ARF effectively associated with endogenous NPM even when expressed at relatively low levels compared to ARF mutants in COS cells. In distinct contrast, the ARF double mutant lacking conserved amino-terminal residues 1 to 14 and 26 to 37 failed to interact with NPM. Those domains were previously shown to mediate ARF's nucleolar localization and cooperative binding to Mdm2, and they are also essential for its growth-inhibitory activity (77). Other mutants bearing at least one intact binding domain retained the ability to associate with NPM, although loss of residues 1 to 14 (D1-14) or 29 to 34 (D29-34) consistently reduced the efficiency of complex formation. Notably, several ARF mutants that lack growth-inhibitory activity (D6-10 and D21-25) (31) nonetheless interacted with NPM. Such data further support the conclusion that ARF-NPM association is not sufficient to suppress growth.
NPM is predominantly nucleolar, whereas the ARF double mutant is strictly nucleoplasmic (31, 77). Potential reasons for why those proteins fail to associate in vivo are that they reside in different subnuclear compartments or they lack intrinsic binding ability due to loss of NPM association domains in the double mutant. To assess the latter possibility, in vitro binding assays were performed (Fig. (Fig.3B).3B). Radiolabeled in vitro translation products of ARF and some of its mutants were mixed with GST-tagged wild-type NPM fusion protein or the GST control, and complexes were analyzed by SDS-PAGE and autoradiography. GST failed to associate with ARF or its mutants, whereas GST-NPM showed direct and efficient association with wild-type ARF. Importantly, and in agreement with our earlier observations, slightly weaker association with NPM was observed for ARF mutants D1-5 (1.3-fold decrease) and D1-14 (1.4-fold decrease), whereas binding was significantly compromised for the double mutant (3-fold decrease). These results indicated that mouse ARF residues 1 to 14 and 26 to 37 directly mediate the interaction between NPM and ARF. Since those residues are critical for ARF nucleolar localization, Mdm2 binding, and ARF-induced growth arrest, the data implied that NPM might regulate those processes.
NPM contains several distinct domains that are directly relevant to its cellular functions (Fig. (Fig.4A)4A) (23). This includes an N-terminal homo-oligomerization domain (HoD; residues 1 to 117) required for formation of NPM dimers and hexamers, a C-terminal nucleic acid binding domain (NBD; residues 260 to 295) essential for association with RNA, and a heterodimerization domain (HeD; residues 187 to 259) implicated in targeting other proteins, such as nucleolin, to nucleoli (36, 37, 75).
To identify domains within NPM required for association with ARF, in vivo binding assays with NPM mutants were performed (Fig. (Fig.4B).4B). Influenza virus hemagglutinin (HA) epitope-tagged NPM and its mutants were coexpressed with ARF in COS cells, and ARF-NPM binding was assessed by Western blotting of anti-HA immunoprecipitates. Notably, NPM residues 1 to 117 were intentionally excluded from our mutants since that domain mediates homodimerization with endogenous NPM. We reasoned that any observed coprecipitation of ARF with NPM mutants retaining residues 1 to 117 might only reflect indirect complex formation between the proteins with endogenous NPM serving as a bridge, making the binding data uninterpretable. As shown in Fig. Fig.4B,4B, all HA-tagged forms of NPM were expressed and efficiently immunoprecipitated by anti-HA agarose, but only full-length HA-NPM and HA-NPM.187-295 coprecipitated with ARF (left panel). The fact that others recently failed to detect complexes between ARF and NPM.187-295 (2) likely reflects the use of different immunoprecipitation conditions (i.e., anti-ARF immunoprecipitations as opposed to the anti-HA immunoprecipitations used in this study), since we similarly found that ARF antibodies were unable to precipitate a complex with mutant NPM (data not shown).
Unfortunately, several observations complicated interpretation of our in vivo binding studies. Most notably, ARF-NPM association always coincided with coprecipitation of endogenous NPM and nucleolin (Fig. (Fig.4B4B and Table Table1).1). Although it was anticipated that full-length HA-NPM would dimerize with endogenous NPM, complexes containing HA-NPM.187-295 and endogenous NPM were not expected since the known homodimerization sequences are absent from that mutant. However, the observation that endogenous NPM and the 187 to 295 mutant coprecipitate in cells lacking ARF (Fig. (Fig.4B,4B, right panel) suggests that residues 187 to 295 dimerize with endogenous NPM and thus indirectly coprecipitate ARF. Alternatively, because nucleolin is also found in the same complexes, nucleolin might bridge the association between endogenous and mutant NPM. Indeed, NPM residues 187 to 259 govern association with nucleolin (37), and the proteins colocalize within nucleoli (Table (Table1).1). The in vivo binding results similarly raise the possibility that nucleolin could be required for ARF-NPM association. Curiously, none of these possibilities explain why NPM mutant 117 to 295 behaves differently from the 187 to 295 mutant and fails to associate with ARF and only associates with endogenous NPM and nucleolin in the absence of ARF.
Our earlier finding that in vitro-translated ARF binds to bacterially produced GST-NPM (Fig. (Fig.3B)3B) proved that the ARF-NPM association is direct and does not require nucleolin. To extend those findings and unequivocally identify the ARF interaction domain(s) in NPM, additional in vitro binding assays were performed with purified GST-NPM mutants and in vitro-translated wild-type ARF (Fig. (Fig.4C).4C). We found that NPM residues 187 to 295 bound to ARF as efficiently as full-length NPM, whereas residues 1 to 260 and 260 to 295 were insufficient for ARF binding on their own. Overall, the data suggest that the nucleolar targeting domain of NPM (residues 187 to 259) governs association with ARF, with residues 260 to 295 somehow contributing to the interaction. That result could imply that RNA binding is important for ARF-NPM association because residues 260 to 295 constitute the RNA binding domain. However, since RNA binding is likely minimal under in vitro binding conditions, and since others recently showed that RNase treatment did not reduce the ARF-NPM interaction (2, 25), it seems more likely that those residues favorably impact NPM structure.
Our binding data suggested that NPM targets ARF to nucleoli. To test that hypothesis, we generated stable U2OS-derived cell lines with reduced expression of NPM by RNA interference with small hairpin RNAs. As shown in Fig. Fig.5A,5A, two clonal populations were identified that exhibited approximately 1.5- to 2-fold lower levels of NPM (knockdown cells kd.1 and kd.2) compared to control cells expressing either the empty small hairpin RNA vector or the NPM small hairpin RNA with a single point mutation. Neither control exhibited knockdown of NPM, and both yielded identical data for all experiments throughout this study. Stable NPM knockdown lines exhibited marginally slower growth compared to control populations, with no apparent apoptosis (data not shown). The best knockdown obtained by RNA interference was twofold (kd.1), which was highly comparable to the modest reduction of NPM associated with ARF-induced growth arrest in various cell types (Fig. (Fig.5B5B).
We transfected NPM knockdown or control cells with plasmids containing wild-type ARF and assessed its localization by immunofluorescence (Fig. (Fig.5C).5C). Importantly, ARF expression was driven by a relatively weak promoter (murine sarcoma virus long terminal repeat), which yielded expression levels that closely approximated the lower levels of endogenous ARF observed in wild-type mouse embryo fibroblasts (data not shown). ARF exhibited its usual punctate nucleolar staining pattern in control cell populations. In contrast, ARF became distinctly more nucleoplasmic in NPM knockdown cells. That effect was readily observed in 100% of ARF expressers throughout the population, and it was specific for ARF because reduction of NPM did not alter the localization of fibrillarin, another nucleolar protein (Fig. (Fig.5D5D).
The increased nucleoplasmic localization of wild-type ARF in NPM-deficient cells was dramatic but incomplete since a large fraction of ARF remained in nucleoli. Such partial relocalization of ARF was consistent with only a twofold reduction of NPM by knockdown. To more easily quantify NPM's influence on ARF localization, we took advantage of a human ARF mutant (d2-14) that lacks one of the NPM binding domains and is already partially compromised in nucleolar localization (77). We hypothesized that localization of this mutant would consequently be more sensitive to changes in cellular NPM levels than wild-type ARF. In control cells containing normal levels of NPM, d2-14 exhibited either cytoplasmic (50.1% of cells) or both cytoplasmic and nucleolar (49.9% of cells) distribution (Fig. (Fig.6A6A and and6B).6B). Strikingly, NPM knockdown resulted in nearly complete loss of d2-14 nucleolar localization, with a more significant effect observed in cells expressing lower levels of NPM. In contrast to control cells, only 13.6% (P = 0.0002) of kd.1 cells and 26.3% (P = 0.0013) of kd.2 cells expressed the ARF mutant in nucleoli. Conversely, exogenous HA-NPM dramatically altered d2-14 subcellular distribution, directing it to nucleoli in 95% (P = 0.0001) of cells. As a control, the HA.NPM117-295 mutant, which lacks ARF binding, failed to alter the localization pattern of d2-14 in control cells (Fig. (Fig.6B).6B). These findings demonstrate that NPM targets ARF to nucleoli in a dose-dependent manner.
We speculated that NPM-mediated nucleolar targeting would limit ARF-human Mdm2 association in the nucleoplasm and thus impair ARF's ability to activate p53 and inhibit growth. We began testing that idea by examining the efficiency with which ARF associated with human Mdm2 in cells that expressed low versus normal levels of NPM. Consistent with the fact that more ARF resides in the nucleoplasm of NPM-deficient cells, a much greater fraction of ARF associated with human Mdm2 in knockdown cells than in controls (Fig. (Fig.7A).7A). This was most evident in the human Mdm2 immunoprecipitations from knockdown cell lysates, where significantly more ARF was coprecipitated despite the reduced expression of ARF in those cells compared to controls. Notably, the lower expression of ARF in NPM knockdown cells reflects the fact that NPM normally stabilizes ARF by blocking its ubiquitination (32).
We next tested the ability of NPM to override ARF-mediated activation of p53 with reporter assays in ARF-null U2OS cells. In support of our hypothesis, moderate overexpression of NPM (2.5 or 5 μg of plasmid) nearly abolished p53 transcriptional activation by ARF (Fig. (Fig.7B).7B). At those concentrations, NPM alone showed no ability to activate p53. By comparison, higher expression levels of exogenous NPM (10 μg of plasmid) resulted in a fourfold stimulation of p53 activity in the absence of ARF. These results show that the magnitude of NPM expression impacts its regulation of ARF and p53, which may help explain contradictory findings that NPM can either inhibit p53 function (35) or activate p53 (8, 25). Specifically, our data suggest that normal to moderately high levels of NPM suppress ARF's ability to activate p53, whereas excessive expression of NPM causes p53 activation independently of ARF.
To test the effect of NPM on ARF-mediated growth suppression, ARF's ability to inhibit DNA synthesis in the presence of various levels of NPM was measured. Cells were transfected with either empty vector, ARF, or ARF plus NPM and subsequently labeled with bromodeoxyuridine for 24 h. The incorporation of bromodeoxyuridine into newly synthesized DNA was determined by immunofluorescence (Fig. (Fig.7C).7C). Equivalent and nearly complete bromodeoxyuridine incorporation was observed in both knockdown (89.5% positive) and control (90.5% positive) cells expressing empty vector, demonstrating similar rates of proliferation in both populations. ARF blocked DNA synthesis in both cell types, although it exerted greater growth-suppressive activity in cells expressing less NPM. Specifically, ARF exhibited the strongest growth-inhibitory activity in NPM knockdown cells (12% bromodeoxyuridine positive), moderate activity in control cells expressing normal levels of NPM (29% bromodeoxyuridine positive), and the least activity in cells overexpressing exogenous NPM (51% bromodeoxyuridine positive).
Although NPM overexpression did not fully override ARF's ability to suppress growth, it repressed ARF-mediated induction of the p53 targets p21Cip1 and human Mdm2 (Fig. (Fig.7D).7D). Significantly lower transfection efficiencies in control cells for this experiment preclude comparison of protein expression levels between control and knockdown cells. Nonetheless, within each cell type, exogenous NPM downregulated ARF-induced expression of p21. NPM effects on ARF-induced human Mdm2 expression were less marked, possibly because overexpressed ARF and NPM can both stabilize human Mdm2 (24, 33). Still, these data provide additional evidence that NPM inhibits ARF-mediated p53 activation, which is all the more remarkable given that NPM simultaneously blocks ARF degradation (32) and enhances ARF expression (Fig. (Fig.7D;7D; compare lanes 3 to 2 and 6 to 5).
Taken together, our findings showed that the targeting of ARF to nucleoli by NPM correlates with significantly reduced ARF function. Therefore, we asked whether a normally inactive, nucleolar form of ARF would acquire growth-inhibitory activity in NPM-deficient cells (Fig. (Fig.8).8). We examined a mouse ARF mutant, D21-25, which localizes to nucleoli and lacks growth-inhibitory activity despite retaining Mdm2 binding ability (31). As expected, D21-25 resided within the nucleoli of control cells containing normal levels of NPM, whereas NPM knockdown caused a significant redistribution of the protein throughout the nucleus (Fig. (Fig.8A).8A). Notably, that disruption in nucleolar localization correlated directly with enhanced growth-inhibitory activity of D21-25, as shown by a nearly twofold decrease in bromodeoxyuridine incorporation within knockdown cells (Fig. (Fig.8B).8B). These data strongly support the notion that NPM inhibits ARF activity by nucleolar sequestration.
A prevailing concept observed throughout the literature is that ARF acts from within the nucleolus, where it normally resides. This study challenges our current understanding of ARF function by suggesting that ARF primarily exerts its tumor-suppressive activity outside the nucleolus, in agreement with ideas originally proposed by Llanos et al. (40). Our data suggest that ARF is sequestered in nucleoli by association with NPM and is consequently restricted from activating p53. Two of the most important observations in our study are (i) that NPM knockdown specifically reduces nucleolar localization of ARF yet significantly enhances ARF-human Mdm2 association, ARF-mediated p53 activation, and ARF-induced growth arrest and (ii) that NPM overexpression antagonizes ARF function while concurrently increasing ARF nucleolar localization. We propose that ARF is normally tethered to nucleoli by NPM and that its ability to activate p53 is manifested only after it escapes into the nucleoplasm, where it can efficiently bind and antagonize human Mdm2 (Fig. (Fig.99).
The unexpected and novel conclusion that NPM negatively regulates ARF by targeting it to nucleoli is supported by several observations. First, our in vitro and in vivo binding assays showed that the nucleolar targeting domain of NPM interacts with the nucleolar localization domains of ARF and that ARF mutants lacking NPM association failed to localize to nucleoli. Those findings pinpointed the discrete functional domains in ARF and NPM responsible for their association, thus extending initial observations regarding ARF-NPM association (2, 25). Second, endogenous NPM targets ARF to nucleoli in a dose-dependent manner, a finding that provides new insight into the mechanisms controlling ARF localization. Previous work suggested that a cryptic nucleolar localization signal in human Mdm2 facilitates ARF nucleolar localization (41, 42). However, the fact that ARF resides in nucleoli in cells lacking Mdm2 argued that other mechanisms normally govern ARF localization. We suggest that NPM regulates that process, in agreement with the virtually superimposable subnuclear distributions of ARF and NPM throughout the cell cycle or in response to various drugs (11). Third, and most importantly, ARF's ability to activate p53 and inhibit growth was inversely related to cellular NPM levels and ARF's localization to nucleoli. Whereas enhanced nucleolar localization of ARF by NPM coincided with reduced ARF function, the redistribution of ARF into the nucleoplasm in NPM-deficient cells significantly enhanced its growth-suppressive activity. That was particularly evident for an inactive, nucleolar mutant of ARF that acquired growth-inhibitory activity when its nucleolar localization was disrupted by NPM silencing.
It is noteworthy that we concluded that the nucleolar targeting domain of NPM (residues 187 to 295) mediates interaction with ARF since that finding contrasts with equally divergent results obtained by two other groups. Whereas Itahana et al. concluded that NPM residues 1 to 117 mediate ARF binding (25), Bertwistle et al. suggested that residues 117 to 187 were essential, although that region was not sufficient for binding and required some contribution from residues 1 to 117 (2). The key difference between the studies is that our findings were derived from both in vivo and in vitro binding analyses, while the other investigations relied exclusively upon in vivo binding assays. We showed there are inherent complications with the in vivo binding assays that preclude their interpretation, namely, homodimerization of NPM mutants with endogenous NPM and association with other cellular factors like nucleolin. Consequently, any observed binding between ARF and NPM mutants might only reflect indirect complex formation with endogenous NPM or nucleolin inside cells. The import of those problems was overlooked in earlier studies, where it was nonetheless recognized that the only NPM mutants capable of binding ARF also interacted with endogenous NPM, regardless of ARF status (2). Such complications are absent from our in vitro binding assay, which provides the first evidence that NPM associates directly with ARF and unambiguously identifies NPM residues 187 to 295 as the ARF interaction domain.
Our discovery that NPM inhibits ARF complements recent findings that ARF downregulates NPM (2, 25), and we propose a model wherein ARF and NPM function in a negative autoregulatory feedback loop to regulate cellular proliferation (Fig. (Fig.9A9A and and9B).9B). At low levels of expression, as seen in presenescent primary cells, ARF would be directed to the nucleolus and retained in a nonfunctional state by NPM, allowing unabated cellular growth. That is consistent with observations that most cellular ARF protein appears bound to NPM, whereas only a small fraction of NPM associates with ARF (2; this study). During senescence or unrestricted oncogenic signaling, however, ARF levels would rise and more efficiently promote the degradation of NPM by triggering its ubiquitination (25). Indeed, we observed a modest decrease in NPM levels upon ARF-induced growth arrest in various cell types. Reduction of NPM would then enable some fraction of ARF to escape into the nucleoplasm, bind human Mdm2, and induce p53-dependent apoptosis or cell cycle arrest (Fig. (Fig.9B).9B). These findings are consistent with the discovery that ARF binds and inactivates human Mdm2 in the nucleoplasm under physiological conditions (38, 40) and our observation that lower levels of NPM enhance ARF-human Mdm2 association and ARF-mediated p53 activation.
We speculate that the true purpose of ARF-mediated NPM degradation is not to impair ribosomal assembly but rather to enable ARF to antagonize human Mdm2 and stimulate p53. That idea is highly consistent with ARF's tumor-suppressive function, since it is well established that p53 activation will effectively kill cells or permanently arrest cell proliferation. By comparison, we show that the modest reduction of NPM caused by ARF has little effect on cellular proliferation. That was demonstrated by our ability to generate and maintain stable NPM knockdown cell lines with 50% reduced expression of NPM and the related observation that no more than 50% diminishment (and often less) of NPM results from ARF-induced growth arrest in a variety of cell types. Although Itahana et al. reported that NPM degradation can cause apoptosis (25), the robust downregulation of NPM induced by high-level expression of adenoviral constructs and transient RNA interference in that study does not seem to reflect the more subtle reduction of NPM invoked by physiologic levels of ARF. Thus, while ARF can impact NPM expression and impair rRNA processing (2, 25, 72), our findings suggest that those activities would not be sufficient to mediate tumor suppression by ARF.
It may seem remarkable that a twofold reduction in expression of NPM, which is an extremely abundant protein, can influence ARF function so dramatically. One of the most salient points in this regard is that ARF and NPM associate with many other proteins, and they exist together in 2- to 5-MDa supramolecular complexes within cells (2). Thus, small changes in the stoichiometry of those complexes could have profound effects. Specifically, 50% reduction of NPM expression levels would be expected to yield a similar decrease in the pool of NPM available to associate with ARF, which would have a tremendous impact on ARF function, particularly since most ARF protein normally interacts with NPM. Indeed, even minor increases in nucleoplasmic ARF levels and ARF-human Mdm2 complexes within NPM knockdown cells should markedly enhance ARF's p53-dependent signaling, since others showed that only a small fraction of ARF needs to be associated with human Mdm2 to activate p53 pathways (40).
The initial hypothesis tested in this work was that ARF function might be regulated by phosphorylation. Little is currently known about posttranslational events that modify ARF and control its activity, with the exception of disulfide bonding facilitating ARF homo-oligomerization (47). Mouse and human ARF contain a conserved threonine (Thr8) and potential protein kinase C phosphorylation site within their essential amino-terminal domain, and an earlier study suggested that death-associated protein kinase could phosphorylate ARF in vitro (60). Our experiments provide the first assessment of ARF phosphorylation in vivo, and they consistently showed that it is not phosphorylated. This was true regardless of whether cells were growth arrested or actively proliferating (10-1 cells; data not shown) or whether ARF was expressed in p53-positive or p53-negative cell types. It remains possible, however, that particular stresses or upstream regulators of ARF, such as oncogenic Ras or Myc, might induce ARF phosphorylation. We did find that ARF associates with phosphorylated forms of NPM, Mdm2, and nucleolin. Nucleolin is a relative newcomer to an expanding list of ARF-associated proteins (2). Interestingly, like NPM, nucleolin is an abundant nucleolar phosphoprotein that contributes to multiple steps in the biosynthesis of ribosomes (19). Also, it is similarly induced by cellular stress to enter the nucleoplasm and bind p53 (9). Additional studies are ongoing to characterize the functional significance of the ARF-nucleolin association.
A valid question that remains is whether ARF has a nucleolar function(s). Although we propose that ARF's p53-dependent activities occur outside the nucleolus, our findings do not preclude that ARF participates in p53-independent signaling pathways inside the nucleolus. In fact, we observed that NPM overexpression drives ARF into nucleoli yet does not fully override its growth-inhibitory activity. While one interpretation is that excessive amounts of NPM in some cells can activate p53 and suppress growth independently of ARF, it is also possible that ARF has some nucleolar function(s). In addition to its ability to inhibit rRNA processing (2, 25, 72), there is evidence that ARF may block E2F activation from within the nucleolus (45; A. Datta et al., unpublished data). How much these and other p53-independent functions of ARF contribute to tumor suppression remains to be determined, although our finding that NPM knockdown and ARF-NPM association are not sufficient for growth inhibition suggests that NPM deficiency and impaired rRNA processing play a limited role. Another possibility is that nucleoli play an important role as ARF depositories, as suggested by several studies (40, 64).
Overall, our findings are in complete accordance with an emerging concept that disruption of the nucleolus contributes to p53 signaling and growth suppression (11, 56, 66). The idea proposed by Rubbi and Milner is that high levels of DNA damage or other cellular stresses perturb the nucleolus, causing release of nucleolar proteins (such as ARF) into the nucleoplasm, which then bind Mdm2 and activate p53. Given that many oncogenic signals activate p53 through ARF (43), whereas UV damage stabilizes p53 via nucleoplasmic redistribution of NPM (33), it is possible that different stresses modulate cell proliferation and survival by disrupting distinct nucleolar proteins. A separate implication of the model is that nucleolar sequestration of ARF would be critical for maintenance of low p53 activity and normal cell growth. Examination of these and related questions will be important for validating and extending the Rubbi-Milner model.
It is important to emphasize that ARF is a tightly repressed gene whose expression is largely undetectable during embryogenesis and postnatal development (43, 46, 85, 86). Thus, as suggested by recent studies in mice (54), it is unlikely that ARF mediates homeostatic responses to most cellular stresses. Rather, most of the evidence suggests that ARF specifically counters the development of oncogenically transformed cancer cells (43, 86). In that situation, we postulate that NPM acts as a critical rheostat within the cell, preventing ARF function until growth suppression is warranted by a marked elevation in ARF levels. Conversely, the overexpression of NPM that is associated with cancer and deregulated growth (12, 52, 70) may block ARF-mediated tumor suppression by preventing its mobilization into the nucleoplasm. One exciting possibility is that treatment of cancer cells with small-molecule inhibitors designed to specifically disrupt the ARF-NPM interaction would release ARF into the nucleoplasm and initiate p53-dependent cell death. Although that approach would only work in cancers bearing wild-type p53 and ARF (perhaps 25 to 40% of cancers), the beauty of the strategy is that normal cells, which lack ARF, should be spared.
We thank Pier Giuseppe Pelicci, Chuck Sherr, Martine Roussel, Gerry Zambetti, and Gordon Peters for reagents. We are also grateful to Jason Weber, Emanuela Colombo, Pier Giuseppe Pelicci, Martine Roussel, and Chuck Sherr for sharing findings prior to publication. These studies were performed with assistance from the University of Iowa Flow Cytometry Facility, the Holden Comprehensive Cancer Center, and core facilities of the Diabetes and Endocrinology Research Center at the University of Iowa.
This work was supported by grants to F.W.Q. (RO1-CA79889) and D.E.Q. (RO1-CA90367).