Arf Induces p53-dependent and -independent Antiproliferative Genes

INTRODUCTION

Alternatively spliced transcripts arising from the Ink4a/Arf locus encode two tumor suppressor proteins, p16^Ink4a and p19^Arf (p14^ARF in humans; Refs. 1 , 2 ) that regulate the activities of the Rb 4 protein and the p53 transcription factor, respectively (3) . The p16^Ink4a protein inhibits the activities of cyclin D-dependent kinases, Cdk4 and Cdk6, to prevent their ability to phosphorylate and inactivate Rb and other Rb-family proteins (p130 and p107), thereby arresting cells in the G₁ phase of the cell cycle. By contrast, p19^Arf blocks various activities of the p53-negative regulator Mdm2, leading to p53 stabilization and a p53-dependent transcriptional response (3 , 4) .

Although the p53-inducible Cdk inhibitor p21^Cip1 contributes to Arf-induced growth arrest, p19^Arf can prevent proliferation of Cip1-null primary MEFs indicating that other p19^Arf-inducible genes, whether p53-dependent or not, can compensate (5 , 6) . Moreover, p19^Arf halts proliferation of MEFs lacking both Mdm2 and p53, albeit more slowly than wt MEFs, implying that p19^Arf can interact with targets other than Mdm2 (7) . Consistent with these findings, TKO mice lacking Arf, p53, and Mdm2 develop multiple and more aggressive tumors per animal than mice lacking either gene alone (7) .

We have now used oligonucleotide and cDNA arrays to study gene expression after p19^Arf induction. Here, we provide evidence that, when overexpressed in primary MEFs, p19^Arf induces many genes that can block cell proliferation. Some of these genes are also induced in a p53- and Mdm2-independent manner, and can inhibit cell proliferation in TKO cells lacking Arf, p53, and Mdm2.

MATERIALS AND METHODS

Cell Culture, Gene Transduction, and Colony Formation.

A cDNA encoding HA-tagged mouse p19^Arf (2) was inserted into the NotI and ClaI sites of the pMTCB6+ vector (8) , thereby placing Arf under control of the sheep MT1 (9) . NIH3T3 cells were transfected with the pMTArf plasmid and selected with 800 μg/ml G418 (Life Technologies, Inc., Rockville, MD). G418-resistant colonies were isolated, treated with 100 μm zinc sulfate, tested for regulated expression of HA-p19^Arf, and subcloned by end point dilution. pMT-Arf clone 3–4 (Fig. 1) ⇓ was used for microarray analyses.

Ink4a-Arf null NIH3T3 fibroblasts and pMT-Arf derivatives were cultured in DMEM (BioWhittaker, Walkersville, MD) containing 10% fetal bovine serum (HyClone, Logan, UT), 4 mm l-glutamine, and penicillin/streptomycin (both from Life Technologies, Inc.) in an 8% CO₂ humidified incubator. Primary MEFs were explanted from E13.5 embryos and grown as described previously (10) . High titer retroviral vectors were produced transiently in human kidney 293T cells and used to infect primary cells as described previously (10) .

For colony formation assays, cells (1 × 10⁵) were seeded on 100-mm diameter dishes and allowed to attach overnight. Where indicated, 100 μm ZnSO₄ was added to the culture medium the next day, and cells were maintained in the presence or absence of zinc for 2 weeks with medium changes twice weekly. Alternatively, naïve cells were infected with MSCV-IRES-GFP retroviruses (11) , and, 48 h later, 100 μm ZnSO₄ was added with fresh medium. Cultures were maintained in the presence of zinc for 2 weeks, stained with HEMA-QUIKII Wright Giemsa stain solution (Biochemical Sciences Inc., Swedesboro, NJ) for 15–30 s, destained in water, and air-dried.

Protein and RNA Expression.

Procedures for immunoblotting were described previously (10) . Briefly, cells were lysed in buffer containing Tween 20, and protein was quantified by BCA assay (Pierce, Rockford, IL). Protein (200 μg/lane) was separated on 10% denaturing polyacrylamide gels and transferred to nitrocellulose membranes (Osmonics, Inc., Westborough, MA). HA-p19^Arf, Mdm2, p53 and p21, and Flag-tagged proteins were detected using affinity-purified rabbit polyclonal antisera against the COOH terminus of p19^Arf (2) ; mouse monoclonal antibodies to Mdm2 (2A10, provided by Dr. Arnold Levine, Rockefeller University, New York, NY), p21^Cip1 (F-5; Santa Cruz Biotechnology, Santa Cruz, CA), and to the Flag tag (M2; Sigma Chemicals, St. Louis, MO); and sheep polyclonal antiserum to p53 (Ab-7; Oncogene Research, Cambridge, MA), respectively.

Total cellular RNA (20 μg) was loaded and separated on 1% formaldehyde-agarose gels and transferred onto Nylon Hybond-N+ membranes (Amersham Pharmacia Biotech, Piscataway, NJ) by a rapid downward transfer system, TURBOBLOTTER (Schleicher & Schuell, Inc., Keene, NH). Prehybridization, hybridization, and washing were carried out as described (10) .

Cell Cycle Analyses.

pMTArf clone 3–4 cells (5 × 10⁵) were seeded in 100-mm diameter dishes, allowed to attach overnight, and treated with 0.1 mm ZnS0_4. Cells were trypsinized from the dishes 24 h later, and cell pellets were resuspended in 1 ml of propidium iodide solution (0.05 mg/ml in 0.1% sodium citrate and 0.1% Triton) for flow cytometric analysis of DNA content as described (2) . Alternatively, cells (2 × 10⁴) seeded in 35-mm diameter dishes were induced with ZnSO₄ for 24 h, incubated in medium containing 10 μm BrdUrd for another 24 h, and then fixed with methanol:acetone (1:1) at −20°C for 10 min. Immunofluorescence staining was used to simultaneously detect mouse p19^Arf (nucleolar staining in ∼95% of cells) and BrdUrd incorporated into replicated DNA as described previously (12) .

RNA Extraction for Microarray Analysis.

Two × 10⁶ NIH3T3 or pMTArf clone 3–4 cells were seeded in 150-mm diameter dishes, allowed to attach overnight, and treated with 100 μm ZnSO₄. At 0, 2, 4, 8, and 24 h after zinc treatment, cell cultures were washed with PBS twice and lysed in TRIzol reagent (Life Technologies, Inc.). Total RNA was isolated according to the manufacturer’s instructions. Kinetic induction experiments were repeated three times each for profiling with cDNA microarrays and Affymetrix GeneChip Probe Arrays.

DNA Arrays and RNA Labeling.

Gene probes used for microarray construction included PCR products of 5376 IMAGE consortium clones included in the Ready-to-Spot Mouse GEM 1 library (Incyte Genomics, St. Louis, MO). This library was created using known mouse genes from the Institute of Genomic Research Mus.ET database as well as ESTs from the GenBank mouse database. Additional elements in the microarray included hybridization controls and PCR products representing Cip1, Kip1, Cdk2, Cdk4, Ink4c, Ink4d, Ink4a-Arf, nucleophosmin, Mdm2, and p53. PCR products were printed using an Omnigrid (GeneMachines) arrayer on glass slides coated with poly-l-lysine. Microarrays were postprocessed according to Eisen and Brown (13) and prehybridized according to Hegde et al. (14) .

Purified total RNA (13 μg) was converted to cDNA in the presence of amino-allyl dUTP by reverse transcriptase by a method developed at Rosetta Inpharmatics, Kirkland, WA. 5 The resultant Cy3- and Cy5-labeled cDNAs were hybridized to the array, scanned with a Packard ScanArray5000, and analyzed with ScanAlyze software. 6

Affymetrix GeneChip Microarrays.

Microarrays were probed and analyzed as described elsewhere (15) . Briefly, double-stranded cDNA was synthesized from total RNA, and biotin-labeled cRNA was generated by in vitro transcription reactions, mixed with other components of the hybridization mixture, and applied onto the probe arrays. Arrays were stained with phycoerythrin-streptavidin and scanned. Expression levels of targets were extrapolated from the 570 nm excitation intensity of bound phycoerythrin. The intensity difference between perfect match and mismatch probes (perfect match-mismatch) in each pair was taken for expression analysis. Expression data were analyzed by Affymetrix Microarray Suite v.5 (Affymetrix, Santa Clara, CA; Ref. 16 ).

Statistical Analysis.

Analysis of the Affymetrix GeneChip data consisted of three parts: data preprocessing, gene selection, and gene clustering. The expression data consisted of the values of the signal, detection call, and detection P computed for the probe sets by the Affymetrix Gene Suite v.5 program (see Appendices C–E of Ref. 16 ). The signal intensity value of a probe set reflects the abundance of the gene in the sample. Hence, if the detection call designates a probe set as absent, the actual value of the signal is not meaningful. For this reason, the signal value of a probe set was reset to zero if the corresponding detection call was absent. The differences between experimental and control signals from three pairs of arrays at each time point were created to mimic the experimental to control ratio (Cy5:Cy3) in the cDNA microarrays, the control being the zinc-treated NIH3T3 cells that do not contain pMTArf. The gene selection was guided by the least statistical significance of several known genes on the pathway (Mdm2, Cip1, Btg2, and Arf). Cip1 turned out to be the least significant among Mdm2, Btg2, and Arf, with P = 0.0173. Thus, all of the genes (probe sets) with P ≤ 0.0173 were selected. The selected genes were then organized into clusters according to their expression patterns by the supervised clustering algorithm.

The statistical methods used to analyze the cDNA array data also consist of three parts: normalization, gene selection, and gene clustering. The influence of extraneous values in the Cy3 and Cy5 intensities were reduced by a logarithm transformation of all of the channel intensity values. A mixed effects ANOVA method (17) was modified to normalize the Cy3 and Cy5 channel raw log intensity values separately (See Supplementary Data). On the basis of the experimental design and the 16-pin configuration of the robot printer, for normalization we modeled two sources of variations: Time (0, 2, 4, and 8 h) and Grid (1–16; each chip consists of 16 grids, and gene spots in the same grid are printed by the same pin). Grid was regarded as a random effect to account for the intra-grid correlation of the intensity values. The Cy3 and Cy5 intensity values are paired for each gene; thus, the difference of normalized log Cy5 intensity and the normalized log Cy3 intensity was used as the expression data for each gene. ANOVA with time as the factor of interest was used in gene screening and selection. The gene selection was guided by the least statistical significance of several known genes on the pathway (Mdm2, Cip1, Btg2, and Arf). All of the genes with a P ≤ 0.0147 (P associated with Arf) were then selected for the clustering step. The selected genes were then organized into clusters according to their expression patterns by the supervised clustering algorithm.

The statistical analyses were implemented using SAS (SAS Institute, Cary, NC) and are detailed in the Supplementary Data. Additional information on complete data for Affymetrix GeneChip and cDNA microarrays, and complete references are available on line. 7

Wilcoxon rank sum tests (18) were used to compare colony formation in infected versus control cells using the method of Benjamini and Hochberg (19) to adjust Ps.

RESULTS

Conditional Arf Induction in NIH3T3 Fibroblasts Arrests Cell Cycle Progression.

A cDNA encoding HA-tagged mouse p19^Arf (2) was expressed under the control of a sheep metallothionein promoter in NIH3T3 cells, which lack the Ink4a/Arf locus, and retain functional p53 and Mdm2 genes. We selected subclones with minimal basal p19^Arf expression in which induction blocked cell cycle progression within 24 h after zinc addition. Expression of HA-p19^Arf, p53, and the p53-responsive proteins, p21^Cip1 and Mdm2 was induced by zinc treatment (Fig. 1A) ⇓ . In clone 3–4, which was used for gene expression profiling, significant accumulation of HA-p19^Arf occurred within 4 h of zinc addition and increased to maximum levels by 24 h (Fig. 1B) ⇓ . By this time, cells underwent arrest in G₁ and G₂ phases, as judged by analysis of DNA content (Fig. 1C) ⇓ . Labeling with BrdUrd for an additional 24 h indicated that only 1% of cell nuclei incorporated the precursor as compared with 94% of uninduced clone 3–4 cells (data not shown). When maintained for 2 weeks in medium containing ZnS0₄, clone 3–4 cells expressing HA-p19^Arf failed to form colonies (<2 colonies/10⁴ cells plated), whereas zinc-treated parental NIH3T3 cells formed hundreds of colonies (compare Fig. 5B ⇓ , below).

Arf Induction Triggers Diverse Temporal Patterns of Gene Expression.

To identify genes regulated by p19^Arf, we performed oligonucleotide arrays using total RNA extracted from zinc-treated NIH3T3 and p19^Arf expressing clone 3–4. We used the Affymetrix murine genome array U74A version 2 chips, which detect ∼12,500 mouse genes, including ∼6,000 functionally characterized genes from the Mouse UniGene database (Build 74) and ∼6,000 EST clusters. Total RNAs isolated at various times after ZnSO₄ addition were used as probes in three independent time course experiments performed to reduce random fluctuations in signal intensities.

A total of 275 genes were selected based on significant changes (P < 0.0173) in temporal expression patterns (Fig. 2 ⇓ ; Supplementary Data; Table 1 ⇓ ). Mdm2 was used as a prototype to group genes with similar or opposite expression patterns in clusters 1a (70 genes) and 1b (40 genes), respectively (Fig. 2, A and B) ⇓ . Arf was used as a core in cluster 2a for positive correlation (8 genes) and in cluster 2b for negative correlation (15 genes). Arf and Ink4a were not distinguished in this analysis because they are encoded in part by alternative reading frames (CdkN2A, cluster 2a). Among other selected clones, clone AI172920 was used as a prototype for cluster 3a, yielding genes that correlated positively (7 genes) and negatively (2 genes, cluster 3b), respectively. The remaining 133 genes were placed in cluster 4.

Apart from Mdm2, cluster 1a contained many p53-responsive genes with antiproliferative functions, including Cip1, Btg2, (B-cell translocation gene-2) an antiproliferative gene (20) , Tob1 (Transducer of ErbB-2; see below), Wig1, B99, and cyclin G. Cyclin D1 was also induced by p19^Arf in agreement with a previous report (21) . Besides these, p19^Arf induced many genes implicated in pathways other than cell cycle regulation (Fig. 2 ⇓ C; Supplementary Data; Table 1 ⇓ ), including two proapoptotic genes (Apaf1 and Fas antigen) and many genes involved in transcriptional regulation. Genes down-regulated by Arf (cluster 1b) included several involved in metabolism, rRNA processing, and ribosome biogenesis.

To validate the p19^Arf-induced genes identified by oligonucleotide assays, we performed cDNA microarray analyses using the same RNAs as probes. We used a chip printed from the Incyte M2 library containing a total of 5376 murine ESTs together with 206 known cDNAs spotted as controls, including Arf itself, Cip1, Mdm2, p53, Ink4a, Ink4c, Ink4d, Kip1, Cdk2, Cdk4, and nucleophosmin. RNAs from induced pMT-Arf cells were labeled with Cy5 (emission wavelength: 670 nm, red), and those from parental NIH3T3 cells were labeled with Cy3 (emission wavelength: 570 nm, green) and hybridized onto the chips. Normalized Cy5:Cy3 ratios were used to identify genes of which the expression patterns changed significantly (P ≤ 0.015).

We clustered 201 genes into six groups according to their expression patterns using a logarithmic scale (Fig. 3 ⇓ ; Table 2; Supplementary Data). The first five clusters were assembled based on correlation analysis (Pearson’s correlation coefficient ≥0.77), relegating the remaining genes together with Arf to cluster 6 (Fig. 3, A and B) ⇓ . To identify p19^Arf-inducible genes that might logically represent p53-dependent targets, Mdm2 was again used as the prototype (or “core/seed”) in the first cluster (Fig. 3B) ⇓ . Thirty-one genes with kinetics of induction similar to that of Mdm2 were assigned to cluster 1. These again included Cip1 (Fig. 3C) ⇓ and Btg2 (20) . Cluster 2 contained 66 genes of which the expression oscillated after Arf induction. Cluster 3 contained 68 genes that were down-regulated at both 4 and 8 h, including several that encode proteins involved in S phase entry and DNA replication, such as DNA polymerase-α, the DNA polymerase-δ small subunit, replication factor C, and Cdc25a phosphatase (Fig. 3C) ⇓ . Cluster 4 contained 14 genes that were transiently down-regulated from 2 to 4 h but whose expression returned to baseline after 8 h of zinc treatment. Cluster 5, in contrast, contained 12 genes that were down-regulated from 2 to 8 h but did not return to their initial levels of expression. After five rounds of clustering, 10 genes remained ungrouped because of lower correlation coefficient values with all of the cores and were placed by default into cluster 6.

Ink4a/Arf was the only gene registering extremely high signals on the arrays because of its overexpression from the inducible promoter. In the oligonucleotide arrays, expression of Arf was evident by 2 h and gradually increased. In the cDNA microarrays, the rate of induction in the first 4 h was similar, but the Cy5:Cy3 ratio did not additionally increase. The different kinetics might simply be because of the stability of hybrids formed with oligonucleotides versus longer cDNA fragments. Additionally, the extremely high level of expression of Arf at 8 h might have saturated the hybridization capacity of the printed target cDNA. All of the other detected genes were within the range of induction expected by this type of analysis.

Validating Expression of p19^Arf-inducible Genes.

A selected number of Arf-activated genes known to have antiproliferative or tumor suppressive functions were additionally studied by Northern blotting analysis. HA-Arf transcripts appeared within 2 h of zinc addition and increased throughout the 24-h time course (Fig. 4) ⇓ . As expected, p53-responsive Mdm2, Cip1, Cyclin G, and Wig1 were induced by HA-p19^Arf in clone 3–4 cells, whereas addition of zinc to parental NIH3T3 cells had minimal effect. Their expression correlated with p53 induction tested in parallel by immunoblotting.

One group of Arf-inducible genes included the Btg (B-cells translocation gene) family. Of six known members, Arf induced four of them, including Btg1, Btg2, Btg3, and Tob1, each with different kinetics (Fig. 4) ⇓ . Unlike Btg1-3, Tob1 was strongly induced only between 8 and 24 h of zinc treatment. Despite its assignment using cDNA microarrays to cluster 2 (down-regulated by Arf), expression of Tob2, another family member, was difficult to detect in uninduced cells and remained low after induction (Fig. 4) ⇓ . Btg4 transcripts were undetectable (data not shown).

Btg Proteins Inhibit Proliferation of Primary MEFs, Including Those Lacking Both Mdm2 and p53.

Given the identification of four Btg family members as p19^Arf-responsive genes, we used retroviral gene transfer to enforce their expression in different genetically defined MEF strains. These included primary MEFs derived from embryos of wt, Arf-null, p53-null, and Arf/p53/Mdm2 triple null (TKO) animals. Cells were infected with retroviral vectors encoding GFP together with HA-Arf, or with FLAG-tagged Btg1, Btg2, or Tob1. Infected cells expressed comparable levels of HA-Arf or FLAG-tagged Btg proteins (Fig. 5A) ⇓ . We were unable to enforce expression of Btg3 raising the possibility that overexpression was incompatible with cell growth.

Unlike the response to p19^Arf, enforced overexpression of Btg1, Btg2, or Tob1 proteins in NIH3T3 cells or in primary wt MEFs had no immediate effect on the ability of cells to enter S phase, as judged by their ability to incorporate BrdUrd during intervals between 24 and 48 h after infection (data not shown). However, when infected cells were maintained in culture for several weeks, overexpression of Btg1, Btg2, or Tob1 in NIH3T3 cells, wt MEFs, or Arf-null MEFs (all of which retain functional p53) reduced their rates of cell proliferation without affecting cell viability. This ultimately led to a reduction in the number and size of colonies, as compared with what was observed with cells expressing GFP alone (Table 1) ⇓ , although the individual effects of the Btg proteins were far less profound than that of p19^Arf (Fig. 5B) ⇓ . By contrast, infection of these cells with a virus encoding Mdm2 increased their efficiency of colony formation (data not shown), consistent with the ability of Mdm2 to counter the effects of p53 in primary MEFs and increase their proliferative rate (22) .

Although cells lacking p53 are acutely resistant to cell cycle arrest by p19^Arf (23) , p53-null primary MEFs (24) or those lacking Arf, Mdm2, and p53 (TKO cells; Ref. 7 ) are inhibited in forming colonies from single cells when high levels of p19^Arf expression are enforced. Infection with the Arf retrovirus limited colony formation in cells lacking p53 alone or in TKO cells, albeit less well than in cells retaining p53 function (Fig. 5B) ⇓ . Enforced expression of Btg1, Btg2, and Tob1 in the latter cell strains reduced not only the number of colonies but significantly limited the size of the colonies that grew out (Fig. 6) ⇓ . The diminution in colony size was not because of apoptosis, as tested by trypan blue exclusion, fluorescence-activated cell sorter analysis of subdiploid DNA content, and terminal deoxynucleotidyl transferase-mediated nick end-labeling assays, indicating that it reflected diminished cell proliferation. Analysis of the morphology of the cells did not reveal significant changes either (data not shown). Therefore, the antiproliferative effects of these Btg family members can be mediated through both p53-dependent and p53-independent mechanisms.

To determine whether p19^Arf could induce Btg expression in cells lacking both Mdm2 and p53, we infected TKO cells with the Arf retrovirus and performed Northern blotting analyses at various times after infection (Fig. 7) ⇓ . Under these conditions, >95% of cells underwent arrest by 72 h after infection (data not shown; Ref. 7 ). Btg family members responded to Arf overexpression but in a subtler manner than in pMT-Arf cells and also more slowly. Btg protein expression could not be confirmed because of the lack of antibodies that can specifically recognize each family member. 8

To confirm that p19^Arf induces antiproliferative genes besides the Btgs in a p53-independent manner, we performed gene profiling by Affymetrix GeneChip arrays in TKO cells infected with an MSCV-Arf-IRES-GFP retrovirus (11) . We analyzed gene expression at 72 h after infection, when all of the cells were growth-arrested. Many of the genes induced by p19^Arf in pMTArf cells were also induced in TKO cells. We noted, in particular, the induction of all four of the Btg genes (Btg1, Bg2, Btg3, and Tob1) and other antiproliferative genes, including cyclin G (25) , B99 (26) , Zac1 (27) , Litaf (28) , and Gas7 (29) . Many of the genes down-regulated by p19^Arf in pMT-Arf cells responded similarly in TKO cells, including genes encoding metabolic enzymes, such as Impdh2 (inosine 5′-phosphate dehydrogenase) and Amd2 (S-adenosyl methionine decarboxylase 2; see Supplementary Data; Table 1 ⇓ ). Therefore, not only can overexpression of individual Btg proteins limit cell proliferation in cells lacking an intact Arf-Mdm2-p53 pathway (Fig. 5B ⇓ and Fig. 6 ⇓ ), but p19^Arf can induce Btgs and other antiproliferative genes in this setting. Although Btg proteins exhibit antiproliferative activity both in the presence or absence of Mdm2 and p53, no single Btg protein was as effective as p19^Arf in limiting cell growth (Fig. 5B) ⇓ . In turn, MEFs lacking both Btg1 and Btg2 or lacking Tob1 alone (but retaining functional Mdm2 and p53) remained as susceptible as wt MEFs to p19^Arf-induced arrest, exhibiting p53 stabilization and induction of both Mdm2 and p21^Cip1 (data not shown). At face value, then, loss of individual Btg family members appears not to compromise the Arf-induced p53 response overtly. We conclude that induction of multiple antiproliferative genes acting in concert may account for growth arrest by p19^Arf in cells lacking both p53 and Mdm2.

DISCUSSION

Use of microarrays to study the transcriptional response to Arf induction has additionally helped to identify genes that likely contribute to p19^Arf-induced cell cycle arrest. By clustering genes that showed a temporal pattern of induction similar to that of Mdm2, we pinpointed several known p53-responsive genes, including Cip1, cyclin G, B99, Wig1, Btg2, and Tob1, which have antiproliferative functions. Whereas Cip1 and members of the Btg family (see below) inhibit the G₁ to S transition, both B99 and cyclin G facilitate p53-dependent G₂ phase arrest after DNA damage (25 , 26) . Conversely, the genes that were rapidly down-regulated included several involved in S phase entry and DNA replication (Cdc25a, DNA polymerase-α, DNA polymerase-δ small subunit, and replication factor C large subunit). Its ability to induce multiple antiproliferative genes and to inhibit DNA synthesis helps to explain how p19^Arf can efficiently induce p53-dependent growth arrest in both G₁ and G₂ (2) , as well as in cells lacking Cip1 (5 , 6) .

Cells lacking p53 (24) or both p53 and Mdm2 (7) can undergo p19^Arf-induced arrest, albeit only in response to relatively high levels of p19^Arf and with slower kinetics than those observed in cells exhibiting normal p53 function. Therefore, p19^Arf can interact with targets other than Mdm2 and p53. Arf induced four of six genes of the Btg family, including Btg1 (APRO 2), Btg2 (APRO 1, Pc3, and Tis21), Btg3 (APRO 4, Ana, and Tob5) and Tob1 (APRO 6), all of which can inhibit cell cycle progression from G₁ to S phase (30 , 31) . Whereas overexpression of Btg proteins did not significantly inhibit cell cycle progression immediately after gene transduction, their enforced expression in MEFs containing or lacking functional p53 (and Mdm2) suppressed colony formation.

Although the functions of Btg family members are not well understood, proteins that physically associate with them include transcriptional regulators and chromatin-modifying factors, such as CAF-1, hPOP2, HOXb9, and Smads (30) . Both Btg1 and Btg2 interact with Hoxb9, a homeodomain-containing protein, to enhance its DNA binding and transcriptional activity (32) . Interestingly, p19^Arf was seen to induce Hoxb9 in our studies. Tob1 associates with Smad proteins and regulates BMP2-induced, Smad-mediated transcription in pluripotent mesenchymal precursor cells and T cells (33 , 34) .

Btg2 also binds to PRMT1 (an arginine methyltransferase) to positively regulate its enzymatic activity (35) . Arginine methylation can affect the functions of proteins involved in signal transduction, such as STAT1 (36) , but another substrate is fibrillarin (37) , a nucleolar protein (like p19^Arf) that binds to small nucleolar RNAs necessary for rRNA processing. Expression of genes encoding fibrillarin and small nucleolar RNAs (U22 host gene) was suppressed by p19^Arf. Moreover, arginine methylation of fibrillarin may block RNA binding (38) and contribute independently to reduced ribosome production. Given that p53 inhibits PolI-dependent transcription of ribosome RNA precursors (39) , Arf, PRMT1, and Btg2 might collaborate in limiting ribosome synthesis.

The BTG1 locus was first identified in association with a chromosomal translocation involving c-MYC in B-cell lymphocytic leukemia; t(8;12)(q24;q22; Ref. 40 ). Mice lacking Btg1 or Btg2 exhibit no overt anomalies and do not spontaneously develop tumors, 9 but Tob1-null mice exhibit hyperproliferation of bone because of relief of Tob1 suppression of BMP-induced, Smad-mediated gene transcription (33) . Loss of Tob1 in mice leads to tumor formation after a long latency and prevents Ras-mediated cell transformation and proliferation, suggesting that it is a tumor suppressor (41) . 10

Expression of genes encoding many other transcriptional regulators was modulated in response to p19^Arf induction (see Supplementary Data for complete listing). This fuels the concept that the effects of Arf on cell proliferation, senescence, and tumor suppression are likely to be far more complex than presently thought. For example, whereas Arf-null mice are born at the expected Mendelian ratio and seem to develop normally (23) , the animals become blind soon after birth. Arf is induced postnatally in the vitreous of the mouse eye just before regression of the hyaloid vascular system, but the hyaloid vascular system does not regress in Arf-null mice and ultimately leads to blindness, a phenotype not observed in p53-null animals (42) . Arf-induced cDNAs associated with cell adhesion, angiogenesis, and metastasis were identified, including genes similar to DRIM (down-regulated in metastasis), cortactin, integrin α 5, and others. Comparison of p19^Arf-induced genes with p53 target genes reported previously (43) using identical gene chips identified similarly regulated genes. Most of the common targets of p19^Arf and p53 regulation are involved in cell cycle control, consistent with their growth-inhibitory properties. Mining of these data and comparison with results of other gene profiling efforts should facilitate an additional understanding of both Arf-p53 interactions and p53-independent roles for Arf.