INTRODUCTION

The protein products of the immediate-early/primary response genes are thought to act as ``third messengers,'' mediating phenotypic alterations in cells in response to ligands such as growth factors, hormones, neurotransmitters, cytokines, and neurotrophins. Many immediate-early genes encode transcription factors (e.g. Fos, Jun, Egr-1) that initiate transcriptional cascades required for proliferation or differentiation (Herschman, 1991). Other ligand-induced immediate-early genes encode paracrine mediators of cellular communication whose products (e.g. prostaglandin synthase-2, inducible nitric- oxide synthase, and cytokines such as MCP-1) modulate the behavior of neighboring cells (Smith and Herschman, 1995).

Because immediate-early/primary response genes have been cloned on the basis of their induction characteristics, rather than the functions of their protein products, a number of these genes encode proteins whose biological roles have not yet been determined. One such immediate-early gene is TIS21. The TIS21 cDNA was cloned by differential screening, both from a cDNA library prepared from mitogen-treated, quiescent murine Swiss 3T3 cells (Fletcher et al., 1991) and from a cDNA library prepared from nerve growth factor-treated rat PC12 pheochromocytoma cells (Bradbury et al., 1991). The predicted rat and mouse TIS21 proteins differ at only four out of 158 amino acid residues. We demonstrated, by metabolic labeling followed by immunoprecipitation, that maximal TIS21 protein synthesis occurs within the first hour after exposure to ligand, both in mitogen-stimulated Swiss 3T3 cells and in nerve growth factor-stimulated PC12 cells (Varnum et al., 1994). Moreover, the half-life of both mitogen- and nerve growth factor-induced TIS21 protein is less than 15 min (Varnum et al., 1994). Despite substantial investigation into both the structure of the TIS21 gene and the induced expression of the TIS21 message and protein, no function has been identified for this protein.

The human BTG1 gene was cloned and characterized (Rimokh et al., 1991) because of its proximity to a chromosome translocation breakpoint (8;12) associated with a chronic lymphocytic leukemia. BTG1 overexpression in murine 3T3 cells decreases growth rate and reduces colony formation, leading to the suggestion that BTG1 is an ``antiproliferative gene'' whose loss of function contributes to uncontrolled growth (Rouault et al., 1992). The human BTG1 and rodent TIS21 open reading frames share 59% identity and 75% conserved sequence similarity at the amino acid level. Moreover, the exon/intron structures of the TIS21 and BTG1 genes are remarkably similar (Varnum et al., 1991; Rimokh et al., 1991). However, BTG1 is not the human TIS21 orthologue; the murine BTG1 cDNA has been cloned and identified as a distinct entity with an amino acid sequence identical to human BTG1 (Rouault et al., 1993). TIS21 and BTG1 are, therefore, members of a gene family.

The stringent regulation of TIS21 gene expression, the short time window for the presence of TIS21 protein following ligand stimulation, and the similarities between the TIS21 and BTG1 proteins suggest that these proteins play important roles in mediating ligand-induced biological responses. We previously speculated that, during its brief availability in mitogen-stimulated cells, TIS21 might reverse a BTG1 antiproliferative barrier by formation of a TIS21-BTG1 heterodimer (Varnum et al., 1994). Alternatively, TIS21 and BTG1 might serve as regulatory partners that modulate the function of a common catalytic subunit. To test these hypotheses, we used the yeast two-hybrid interaction system (Fields and Song, 1989; Fields and Sternglanz, 1994) to determine (i) whether TIS21 and BTG1 proteins can form heterodimers and/or homodimers, and (ii) whether there exist proteins that interact with both TIS21 and BTG1.

EXPERIMENTAL PROCEDURES

The open reading frame from TIS21 cDNA (Fletcher et al., 1991) was subcloned by PCR¹ using AmpliTaq (Perkin Elmer) and the primers 5-CATGTCGACGGGTTCTGGTTCTGGGAGCCACGGGAAGAGAACC-3 and 5-CTAGCGGCCGCCCAGGGTCGGGTGGCTCC-3 to create a 5 SalI site, a 3 NotI site, and a 5 amino acid linker (Gly-Ser-Gly-Ser-Gly) preceeding the TIS21 coding sequence, for flexibility between the GAL4 domains and TIS21 protein. The resulting cDNA was ligated to GAL4 activation domain fusion plasmid pGAD425 (Han and Colicelli, 1995) and GAL4 DNA-binding domain fusion plasmid pGBT10 (Han and Colicelli, 1995).

Murine BTG1 cDNA was prepared by reverse transcription and PCR. Total RNA was extracted from NIH3T3 cells grown to confluence in Dulbecco's modified Eagle's medium (Life Technologies, Inc.) supplemented with 10% fetal bovine serum (Gemini Bioproducts). Poly(A)⁺ RNA was isolated using an Oligotex mRNA kit (Qiagen) and reverse transcribed using Moloney murine leukemia virus reverse transcriptase (Promega). A PCR reaction was performed using primers 5-CATGTCGACGGGTTCTGGTTCTGGGCATCCCTTCTACACCCGG-3 and 5-CTAGCGGCCGCATCCATCCAATAGACTATATC-3 to create a 5 SalI site, a 3 NotI site, and the 5 amino acid linker, as described for TIS21. BTG1 cDNA was then ligated to pGAD425 and pGBT10. The plasmids were sequenced using Sequenase 2.0, according to the manufacturer's instructions (U. S. Biochemical Corp.), to confirm the reading frame.

Yeast strain PCY2 (MAT gal4 gal80 URA3::GAL1-lacZ lys2-801^amber his3-200 tryp1 63 leu2 ade2-101^ochre) (Wang et al., 1995) was transformed simultaneously with both a pGAD and a pGBT construct, according to the method of Schiestl and Gietz (1989), to analyze homodimerization and heterodimerization of TIS21 and BTG1. Briefly, PCY2 cells were grown in YPD (Rose et al., 1990) medium until cell density reached 9-10 × 10⁶ cells/ml. After harvest, the cells were resuspended in transformation buffer (1 M sorbitol, 0.1 M lithium acetate, pH 7.5, 10 mM Tris-HCl, pH 7.5, and 0.1 mM EDTA) and incubated at 30 °C for 1 h. The cells were then mixed with plasmid DNA, carrier DNA, and PEG3350 (polyethylene glycol, molecular weight 3350), incubated at 30 °C for 30 min and heat-shocked at 42 °C for 5 min. For each transformation, 1 µg of each plasmid DNA was used with 66 µg of denatured salmon sperm DNA as carrier. Transformants containing both plasmids were selected on agar plates containing synthetic complete medium without leucine and tryptophan (SCleutrp) and analyzed for -galactosidase activity by the yeast colony filter assay (Breeden and Naysmyth, 1985), using 5-bromo-4-chloro-3-indoyl -D-galactoside as substrate.

The GAL4 activation domain vector pGAD424 (Bartel et al., 1993), which contains a leucine selective marker, was digested with SphI. The SphI cassette containing the GAL4 activation domain was removed. The SphI cassette of pBTM116, which encodes the LexA DNA binding domain (Bartel et al., 1993), was isolated by SphI digestion and gel purification, using GeneClean II (BIO-LABS 101) and ligated into the pGAD425 whose original SphI cassette had been removed. The resulting vector, containing the region encoding the LexA DNA-binding domain and the leucine selective marker, was named pLexA(L) and was used for two-hybrid interaction cloning.

TIS21 cDNA was subcloned by PCR, using pGAD425TIS21 as template and the primers 5-GCAGAATTCGGTTCTGGTTCTGGGAG-3 and 5-CGAGTCGACGGGTCGGGTGGCTCCTA-3, into pLexA(L) at an EcoRI/SalI site, to create plasmid pLexA(L)TIS21. Murine BTG1 cDNA was amplified using pGAD425BTG1 as template and the primers 5-GCAGAATTCGGTTCTGGTTCTGGGCA-3 and 5-CGTGTCGACATCCATCCAATAGACTA-3. The BTG1 cDNA was then ligated to pLexA(L) at the EcoRI/SalI site, to create plasmid pLexA(L)BTG1. The plasmids were sequenced using Sequenase 2.0 according to the manufacturer's instructions (U. S. Biochemical Corp.), or by using the DyeDeoxy Terminator Cycle sequencing kit and a 373 DNA sequencer (Applied Biosystems) to confirm the sequence and reading frame. The DNA encoding the LexA-Rin1-CT fusion was excised from pBTM116 (a gift from Limin Han and John Colicelli, UCLA) with SphI and ligated to pGAD424 whose SphI cassette had been removed, to create pLexA(L)Rin1-CT.

The pPC86 cDNA fusion library contains cDNAs fused to the GAL4 activation domain. cDNA was prepared from FAO cell poly(A)⁺ mRNA, using the SuperScript plasmid system (Life Technologies, Inc.) according to the manufacturer's instructions, and ligated to the pPC86 vector (Chevray and Nathans, 1992) at the SalI/NotI site. Plasmid pPC86 contains a tryptophan selective marker.

Escherichia coli DH5 (Life Technologies, Inc.) was the transformation recipient for all plasmid constructions, and was also used to recover expression plasmids from yeast. Yeast strain L40 (MATa his3D200 trp1-901 leu2-3, 112 ade2 LYS2::(lexAop)₄-HIS3 URA3::(lexAop)₈-lacZ gal4?? gal80??) (Hollenberg et al., 1995) was used for two-hybrid library screening. To screen the cDNA library for clones encoding proteins that interact with TIS21 protein, the yeast L40 strain was transformed to leucine prototrophy with pLexA(L)TIS21. The transformant was grown in SC-leu-ura-lys (synthetic complete without leucine, uracil, and lysine) medium overnight and used to inoculate 450 ml of YPD medium. The culture was grown until the cell density reached 9-10 × 10⁶ cells/ml, then harvested for transformation with 300 µg of the plasmid library. Cells were transfected as described in the previous section. After heat shock, the cells were incubated in YPD for 1 h at 30 °C, then washed thoroughly with TE buffer (10 mM Tris-HCl, pH 7.5, and 0.1 mM EDTA) and resuspended in SC medium lacking tryptophan, leucine, and uracil. A sample was plated on SC medium lacking uracil, leucine, and tryptophan, to calculate transformation efficiency. After overnight incubation in SC-ura-leu-trp, the transformed cells were washed twice with TE buffer and plated on SC medium lacking tryptophan, leucine, histidine, uracil, and lysine, to select for interacting clones. Three days later, viable colonies were assayed for -galactosidase activity using the yeast colony filter assay (Breeden and Naysmith, 1985). The filters were incubated at 37 °C for 2 h. Colonies positive for -galactosidase activity were grown to saturation in SC-ura-trp liquid medium, a nonselective condition for pLexA(L), then replated on YPD plates. Replica plating of these plates to SC-ura-trp and SC-ura-leu identified colonies that had lost pLexA(L)TIS21 but still retained cDNA plasmids. The colonies that continued to give a positive -galactosidase assay after loss of pLexA(L)TIS21 were eliminated.

Thirty-three yeast colonies which demonstrated pLexA(L)TIS21-dependent histidine prototrophy and -galactosidase activity were grown and treated with lyticase (3 mg/ml) and 0.1 M -mercaptoethanol in SCE (1 M sorbitol, 0.1 M sodium acetate, 60 mM EDTA, pH 7.0) at 37 °C for 1 h to release spheroplasts. The pPC86 plasmids containing cDNA inserts encoding proteins that interact with TIS21 were isolated using a Qiagen plasmid miniprep kit and subjected to PCR using a GAL4 activation domain primer (5-GGAATCACTACAGGGATG-3) and a vector primer (5-TTGATTGGAGACTTGACC-3) to amplify cDNA inserts. Cross-hybridization was performed, using randomly selected cDNA inserts from these 33 clones as probes, to identify clones with common sequences.

Clones 3G, 4A, and 5A were sequenced using the DyeDeoxy Terminator Cycle Sequencing kit and a 373 DNA sequencer (Applied Biosystems). Sequence homology searches were performed using the BLAST program, through the National Center for Biotechnology Information (Altschul et al., 1990).

The Bailey (1967) modification of the Lowry procedure was used to determine concentration of protein, after precipitation with 1 ml of 10% (w/v) trichloroacetic acid. Bovine serum albumin was used as a standard.

To construct GST plasmids expressing TIS21, BTG1, and 3G, the cDNA insert fragments were excised from plasmids pSP64XSN-TIS21, pSP64XSN-BTG1, and pPC86-3G using SalI and NotI, and ligated to plasmid pGEX-2T (Pharmacia) which had been modified to accept a SalI/NotI insert (Han and Colicelli, 1995). Expression of GST fusion proteins was induced in transformed E. coli DH5 by isopropyl-1-thio--D-galactopyranoside (1 mM) for 5 h. Bacteria were harvested by centrifugation and resuspended in extraction buffer (phosphate-buffered saline, pH 7.4, 5% glycerol, 1 mM EDTA, 1 mM EGTA, 1 mM dithiothreitol, 40 µg/ml leupeptin, 40 µg/ml aprotinin, 20 µg/ml pepstatin, 1 mM phenylmethylsulfonyl fluoride, and 0.5% Triton X-100). The cells were disrupted by sonication. Cell debris was removed by centrifugation (16,000 × g, 20 min, 4 °C), and the supernatants were collected. GST fusion proteins were isolated with glutathione-Sepharose 4B beads (Pharmacia), as described by the manufacturer.

Samples were lyophilized and hydrolyzed with 12 N HCl in vacuo (110 °C, 20 h), in a vapor-phase Waters Pico-Tag apparatus. The hydrolyzed material was resuspended in 50 µl of water and then 25 µl were analyzed on a sulfonated polystyrene column of Beckman AA-15 resin. Samples were mixed with 1 µmol each of N^G,N^G-dimethylarginine (asymmetric) and N^G-monomethylarginine (both obtained from Sigma). An equal volume of citrate sample dilution buffer (0.2 M in Na⁺, containing 2% thiodiglycol and 0.1% phenol at pH 2.2) was added, and the mixture was loaded onto the column (0.9 cm in diameter, 11 cm in length). The column was equilibrated with Na⁺ citrate buffer (0.35 M in Na⁺, pH 5.27) at 55 °C. Elution was at approximately 1 ml/min. The column was regenerated with 0.2 N NaOH for 20 min prior to the next run. One-minute fractions were collected and ³H radioactivity was determined by counting samples in a scintillation counter. Additional 100-µl samples from each fraction was analyzed for the nonradioactive methylated arginine standards, using the ninhydrin method (Gary and Clarke, 1995).

Methylation reactions were stopped by the addition of an equal volume of 2 × SDS-PAGE sample buffer (Hyrcyna et al., 1994), and subjected to SDS-PAGE analysis. The samples were heated to 100 °C for 5 min and then loaded onto slab gels using the buffer system described by Laemmli (1970). The samples were subjected to electrophoresis through either a 10 or 12.6% acrylamide, 0.28% (w/v) N,N-methylenebisacrylamide matrix by the application of a 35-mA constant current. Gels were stained with Coomassie, destained, and then treated with the fluor EN³HANCE (DuPont) as described by the manufacturer. The dried gels were subjected to fluorography at 80 °C for 7-14 days.

RAT1 cells were grown to confluence in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum. Cells were removed from the plates with trypsin, washed with cold phosphate-buffered saline, and resuspended in extraction buffer (25 mM Tris-HCl, pH 7.4, 1 mM sodium EDTA, 1 mM sodium EGTA, 40 µg/ml leupeptin, 40 µg/ml aprotinin, 20 µg/ml pepstatin, and 1 mM phenylmethylsulfonyl fluoride). Cell disruption was performed on ice, by homogenization with a glass tissue grinder. The homogenate was centrifuged at 18,000 × g for 50 min at 4 °C. Supernatants were either used immediately, or stored at 80 °C. To obtain hypomethylated cytosol, RAT1 cells were cultured with 20 µM adenosine dialdehyde (Sigma) for 2 days. After removal from the plates with trypsin, cells were washed twice with phosphate-buffered saline containing 1 mM EDTA, and disrupted as described above.

RAT1 cytosolic extract (500 µl, 233 µg of protein) or chromatographic fractions (200 µl) were incubated with either GST-TIS21 or GST immobilized to glutathione-Sepharose beads at 4 °C. After centrifugation, the supernatant was collected. The beads were washed thoroughly with buffer (50 mM Tris-HCl, pH 7.4, 100 mM KCl, 0.05% Tween 20, 1 mM dithiothreitol, and 1 mM phenylmethylsulfonyl fluoride), then incubated with 40 µl of 10 mM reduced glutathione in 50 mM Tris-HCl, pH 8.0, to elute GST-TIS21 (or GST) and the associated proteins.

Cultures (500 ml) of the odp1/rmt1 mutant strain (Gary et al., 1996) were grown to an absorbance at A_600 nm of 2.5 in YPD medium at 30 °C. Cells were harvested by centrifugation at 4,400 × g for 10 min at 4 °C. The pelleted cells were washed with buffer A (25 mM Tris-HCl, 1 mM sodium EDTA, 1 mM sodium EGTA at pH 7.5) and again harvested by centrifugation. The pellet (2.6-3.0 g) was resuspended in 2 ml of buffer A/g of cells and 40 µl of a protease inhibitor mixture was added to give final concentrations of 1 mM benzamidine, 1 mM phenylmethylsulfonyl fluoride, 2 µg/ml leupeptin, 2 µg/ml aprotinin, and 2 µg/ml pepstatin. The cell suspensions were passed twice through a French pressure cell at 20,000 psi. Homogenates were centrifuged at 23,000 × g for 40 min at 4 °C. The supernatants were stored at 20 °C.

Total RNA was isolated from RAT1 cells, using lithium chloride (Cathada et al., 1983). RNA from rat tissues was prepared using acid guanidinium thiocyanate-phenol-chloroform (Chomczymski and Sacchi, 1987). RNA was subjected to electrophoresis on formaldehyde-agarose gels and transferred to nylon membranes (Thomas, 1981). The blots were hybridized to random-primed cDNA probes (Sambrook et al., 1989) for TIS21, BTG1, 3G, and S2.

RESULTS

We prepared (i) vectors expressing both murine TIS21 and BTG1 as fusion proteins with the DNA-binding domain of the GAL4 transcription factor and (ii) vectors expressing both TIS21 and BTG1 as fusion proteins with the transcriptional activation domain of the GAL4 transcription factor (see ``Experimental Procedures''). Appropriate constructs were then transfected in pairs into the yeast PCY2 strain, to analyze by two-hybrid interaction analysis (Fields and Song, 1989; Bartel et al., 1993) the ability of TIS21 and BTG1 proteins to associate (see ``Experimental Procedures). No combination of TIS21 and BTG1 GAL4 fusion proteins could activate the expression of -galactosidase driven from a GAL4 promoter, suggesting that BTG1 and TIS21 are unable to form either heterodimers or homodimers.

To identify proteins that interact with the TIS21 protein, a yeast strain expressing a LexA-TIS21 fusion protein as ''bait`` was transfected with a plasmid library in which the GAL4 activation domain was fused to cDNAs prepared from rat FAO cells. Yeast cells in which a complex containing the LexA DNA-binding domain and the GAL4 activation domain can form are able to survive in histidine-free medium because a LexA-responsive promoter drives expression of the HIS3 protein (Fields and Sternglanz, 1994). This complex will form because of a physical association between the LexA-TIS21 protein and a library fusion protein with the GAL4 activation domain. These yeast cells also contain a reporter gene in which a LexA-responsive promoter drives expression of -galactosidase. Colonies able to survive on histidine-deficient medium were subsequently tested for -galactosidase activity. Colonies dependent on the presence of the LexA-TIS21 fusion protein for these two properties were then identified. Cross-hybridization studies indicate that, among 33 such histidine-independent clones that also express -galactosidase, three distinct cDNAs express fusion proteins (5A, 3G, and 4A) that interact with TIS21. When tested in two-hybrid interaction analysis, clones 5A and 4A do not interact with BTG1. In contrast, clone 3G does interact with both TIS21 and BTG1 (Fig. 1). Sequence analysis of clone 5A identified the encoded protein as PICK-1, a protein recently cloned from a two-hybrid screen using the catalytic region of protein kinase C as bait (Staudinger et al., 1995). Partial sequence analysis of clone 4A has not identified this open reading frame as similar to any known protein or expressed sequence tag. The sequence of the proposed open reading frame of the 3G protein is shown in Fig. 2A. The longest predicted open reading frame of the 3G protein encodes a 40.5-kDa polypeptide composed of 353 amino acid residues.

To confirm the interaction of both the TIS21 and BTG1 proteins with the 3G protein, radioactive [³⁵S]methionine-labeled in vitro translated TIS21 protein or BTG1 protein were incubated with purified glutathione S-transferase-3G (GST-3G) fusion protein or with control GST protein. GST-3G or GST protein was first immobilized on glutathione-Sepharose beads. The beads were then incubated with radioactive TIS21 or BTG1. After washing, the bound material was eluted and subjected to SDS-PAGE electrophoresis, followed by autoradiography. Both in vitro translated BTG1 and in vitro translated TIS21 are able to bind to GST-3G, but not to the GST control (Fig. 3). These in vitro binding studies confirm the two-hybrid interaction analysis, and further demonstrate the interaction of the 3G protein with both BTG1 and TIS21 proteins.

A BLAST search (Altschul et al., 1990) of the available sequence data bases for sequences similar to the 3G protein identified an open reading frame (ODP1) () from yeast chromosome II (Loubbardi et al., 1995; Feldman et al., 1994; Fig. 2A) and expressed sequence tags from the genome sequencing projects for rice, Arabidopsis, Caenorhabditis elegans, and humans. However, these proteins have no identified functions. A short region of sequence similarity between the 3G protein and the enzyme that methylates E. coli ribosomal protein L11 () was also identified (Fig. 2B). This region of sequence similarity between the 3G protein and the ribosomal protein L11 methyltransferase has previously been identified as ``methyltransferase region I,'' a consensus sequence found in a wide variety of proteins that bind S-adenosyl-L-methionine (AdoMet) (Kagan and Clarke, 1994). Crystallographic studies have demonstrated the direct interaction of region I with AdoMet, in both a small molecule methyltransferase and two DNA methyltransferases (Schluckebier et al., 1995; see also Fig. 2C). Further sequence comparisons of the 3G protein with the L11 methyltransferase, as well as other known methyltransferases, identified three additional motifs (post-I, -II, and -III) that are also highly conserved among methyltransferases and other AdoMet-binding proteins (Fig. 2A; Kagan and Clarke (1994)). The similarities in sequence for these regions of the 3G protein and members of the methyltransferase family suggested that the 3G protein and the protein encoded by the yeast ODP1 gene might also have methyltransferase activity.

Methyltransferases can transfer methyl groups from AdoMet to oxygen, nitrogen, sulfur, and carbon moieties, and can use as substrates a wide variety of both small biological molecules and macromolecules that include DNA, RNA, lipids, and proteins (Clarke, 1993). Because of the similarities between the 3G protein and the L11 protein methyltransferase, an enzyme that appears to modify the N-terminal alanine residue and/or internal L11 lysine residues (Vanet et al., 1993), we examined the possibility that the 3G protein is a protein methyltransferase. We first tested whether the 3G protein has enzymatic activity similar to that of the previously characterized types of soluble eukaryotic enzymes that catalyze methyl ester formation; e.g. protein L-isoaspartyl methyltransferase (Lowenson and Clarke, 1995) or protein phosphatase 2A C-terminal leucine methyltransferase (Xie and Clarke, 1994a). However, we found no evidence that the 3G protein has either type of activity. Recombinant GST-3G fusion protein does not methylate the isomerized aspartyl residues present in either the isoaspartyl-containing peptides KASA(isoD)LAKY or VYP(isoD)HA, nor is it able to methylate the catalytic subunit of protein phosphatase 2A from rat brain (a gift from Sandra Rossie, Purdue University).

Because the GST-3G fusion protein did not appear to be a protein carboxyl methyltransferase, we next investigated whether 3G might methylate nitrogen atoms in proteins. We began these studies by assaying the ability of GST-3G to methylate the guanidino nitrogens of arginine residues, because well defined substrates are available for these enzymes. Methylation of arginine residues is a common post-translational modification of proteins that mediate RNA processing (Rajpurohit et al., 1994a; Lischwe et al., 1985a, 1985b). Many of these proteins (e.g. fibrillarin, nucleolin, and hnRNP A1) contain N^G,N^G-dimethylarginine (asymmetric). The arginine residues subject to methylation in these RNA-binding proteins are present in a lycine and rginine ich ``GAR domain'' containing multiple repeats of the consensus arginine methylation site RGG (Najbauer et al., 1993).

The R1 peptide, GGFGGRGGFG-amide, derived from the fibrillarin consensus methylation site, has been used to characterize protein-arginine N-methyltransferase activity present in cultured cells and in mammalian tissues (Najbauer et al., 1993). When R1 peptide is incubated with GST-3G fusion protein and S-adenosyl-[methyl-³H]L-methionine ([³H]AdoMet), the peptide is readily methylated (Fig. 4, A and B). Acid hydrolysis of the methylated R1 peptide and subsequent amino acid analysis by cation exchange chromatography demonstrate that the radioactive methyl group has been incorporated to form an N^G-monomethylarginine residue (Fig. 4C). We conclude that the 3G fusion protein is a functional R1 peptide- arginine methyltransferase.

Histones, hnRNP A1, myelin basic protein, and cytochrome c have been used as substrates to identify, partially purify, and distinguish protein-arginine N-methyltransferases from several sources (Ghosh et al., 1988; Farooqui et al., 1985; Lee et al., 1977). Previous studies with partially purified mammalian enzyme preparations have suggested that one enzyme is responsible for the mono- and asymmetric dimethylation of arginine residues in both histones and hnRNP A1 (Rajpurohit et al., 1994a; Ghosh et al., 1988), while a different enzyme mono- and symmetrically dimethylates arginine 107 in myelin basic protein (Ghosh et al., 1988; Baldwin and Carnegie, 1971; Brostoff and Eylar, 1971). The protist Euglena gracilis contains a third, distinct protein-arginine N-methyltransferase that is able to methylate arginine residues in mammalian cytochrome c (Farooqui et al., 1985). The GST-3G fusion protein is able to methylate both histones and recombinant hnRNP A1 protein (Fig. 5A). Cation exchange chromatography of acid hydrolysates of these substrates demonstrates that mono- and asymmetric dimethylated arginine residues are the sole products of the enzymatic reactions (Figs. 5, B and C). In contrast, the GST-3G fusion protein cannot use either cytochrome c or myelin basic protein as substrate for methylation (Fig. 5A). We suggest that the 3G protein is the catalytic component of the previously described histone/hnRNP A1 arginine methyltransferase (Ghosh et al., 1988; Rawal et al., 1994; Liu and Dreyfuss, 1995), and have named this enzyme ``protein-arginine transferase 1,'' or PRMT1.

In parallel work, we have found (i) that the yeast ODP1/RMT1 gene also encodes a protein-arginine N-methyltransferase (Gary et al., 1996), and (ii) that soluble extract from an odp1/rmt1-deficient yeast strain contains a 55-kDa protein that is an excellent methyl-accepting substrate for the GST-PRMT1 fusion protein (Gary et al., 1996). When odp1/rmt1 mutant soluble extract is used as substrate with GST-PRMT1 and [³H]AdoMet, this 55-kDa protein is, by far, the major methylated species (Gary et al., 1996). Moreover, the 55-kDa protein present in odp1/rmt1 methyltransferase-deficient yeast cytosol can also be methylated by an endogenous methyltransferase enzyme(s) present in soluble extracts of RAT1 cells (Gary et al., 1996). We can, therefore, use the extract prepared from the odp1/rmt1 methyltransferase-deficient yeast strain as substrate to assay for the presence of protein-arginine N-methyltransferase activity in mammalian cells.

We used GST-TIS21 fusion protein immobilized on glutathione-Sepharose beads to demonstrate that TIS21 protein can sequester protein methyltransferase activity present in RAT1 cell extracts. Using the 55-kDa protein present in odp1/rmt1-deficient yeast as a methyl-accepting substrate, we demonstrated protein methyltransferase activity was bound to the GST-TIS21 fusion protein (Fig. 6A). In contrast, GST protein immobilized on glutathione-Sepharose beads cannot sequester methyltransferase activity. When extracts of RAT1 cells are analyzed by gel filtration chromatography, the rat cell enzyme capable of methylating the 55-kDa yeast substrate present in odp1/rmt1 mutant yeast extracts migrates as a single complex of approximately 180 kDa (Fig. 6, B and C). In contrast, the calculated polypeptide molecular mass of PRMT1 is approximately 40 kDa (Fig. 2). The protein methyltransferase activity present in the high molecular size fractions from the gel filtration column can be absorbed by immobilized GST-TIS21 and, when eluted, can methylate both the yeast 55-kDa substrate (not shown) and recombinant hnRNP A1 (Fig. 6D). We conclude (i) that native PRMT1 present in RAT1 cells, like the GST-PRMT1 fusion protein, is able to interact with TIS21 protein and (ii) that PRMT1 exists as a macromolecular complex in RAT1 cells.

If [³H]AdoMet is added to a soluble extract from RAT1 cells, methylated polypeptides are observed at molecular masses of 96, 60, 28, 22, and 20 kDa (Fig. 7A, lane 1), due to endogenous methyltransferase activity. If purified GST-PRMT1 fusion protein is added to the RAT1 cytosolic extract along with the [³H]AdoMet, methylation of an additional protein migrating at 55 kDa is observed (Fig. 7A, lane 2) and methylation of the 96- and 20-kDa proteins is reduced.

It is likely that many substrates for PRMT1 methyltransferase activity are already extensively methylated in cell extracts, and cannot be further methylated during in vitro reactions of the type shown in Fig. 7A. Najbauer et al. (1993) demonstrated that incubation of PC12 cells with adenosine dialdehyde (AdOx) results in the cytosolic accumulation of hypomethylated protein substrates that can subsequently be methylated in vitro. When purified GST-PRMT1 fusion protein and [³H]AdoMet are added to extracts prepared from hypomethylated RAT1 cells, we again observe the extensive methylation of a 55-kDa substrate. In addition, methylated proteins of 130, 100, 38, 34, and 16 kDa are observed (Fig. 7B, compare lanes 1 and 2), suggesting that these proteins normally exist in RAT1 cytosol in an extensively methylated form.

Purified GST-BTG1 or GST-TIS21 recombinant protein, when added to RAT1 cytosolic extracts without addition of GST-PRMT1, activate methylation of the 55-kDa protein (Fig. 7A, lanes 3 and 4). In addition, a 65-kDa protein is methylated. Neither GST-TIS21 nor GST-BTG1 fusion protein alone have methyltransferase activity; they are unable to methylate histones (data not shown). Extracts from quiescent, non-dividing RAT1 cells apparently contain latent, endogenous PRMT1 methyltransferase activity, which can be activated by addition of either GST-BTG1 or GST-TIS21 fusion proteins. The data suggest that interaction of either TIS21 or BTG1 with PRMT1 can modulate endogenous PRMT1 methyltransferase activity present in RAT1 cytosolic extracts. It will, of course, be of great interest to identify the substrates methylated both by recombinant GST-PRMT1 and by the TIS21/PRMT1 and BTG1/PRMT1 complexes.

In contrast to the results observed with extracts from untreated RAT1 cells, there does not appear to be any latent PRMT1 activity present in extracts from RAT1 cells exposed to adenosine dialdehyde. Addition of either GST-TIS21 or GST-BTG1 fusion protein, in the absence of GST-PRMT1, does not enhance methylation of proteins in the hypomethylated extract (Fig. 7B, lanes 3 and 4). These data suggest that the expression and/or stability of PRMT1 in RAT1 cells may depend on continued cellular methylation activity.

The apparent absence of endogenous PRMT1 in hypomethylated RAT1 cell extracts provided us with an opportunity to examine the effect of the GST-TIS21 fusion protein on the enzymatic activity of the GST-PRMT1 fusion protein, using endogenous rat cell substrates. When purified GST-TIS21 fusion protein is added to extracts of adenosine dialdehyde-treated RAT1 cells, along with GST-PRMT1 fusion protein, the methylation of the 34-kDa substrate is enhanced (Fig. 7B, lane 5 versus lane 2). Similar results were observed with the GST-BTG1 fusion protein (data not shown). The data from Fig. 7, A and B, suggest that the TIS21 and BTG1 proteins both qualitatively and quantitatively modulate PRMT1 activity.

We speculated that hnRNP A1 might be the 34-kDa substrate present in hypomethylated RAT1 extracts that is methylated by the GST-PRMT1 fusion protein (Fig. 7B, lanes 2 and 5). Samples of hypomethylated RAT1 extract incubated with purified GST-PRMT1 fusion protein and [³H]AdoMet were subjected to immunoprecipitation, either with an immobilized monoclonal antibody to hnRNP A1 or with an immobilized control monoclonal antibody to the HA hemagglutionation epitope. The proteins present in the supernatants and immunoadsorbed fractions were subjected to electrophoresis and fluorographic analysis. A ³H-methylated 34-kDa protein could be recovered from the hypomethylated extract following methylation by the GST-PRMT1 fusion protein, using anti-hnRNP A1 antibody (Fig. 7C), confirming that the 34-kDa substrate is hnRNP A1. Control 12CA5 monoclonal anti-hemagglutinin antibody did not immunoprecipitate any methylated proteins. These experiments suggest (i) that the bulk of hnRNP A1 present in confluent, density-arrested RAT1 cells is normally methylated at PRMT1 substrate sites (Fig. 7A), and (ii) that BTG1 protein and TIS21 protein may modulate the ability of PRMT1 to methylate hnRNP A1 in vitro (Fig. 7B). Confirmation of this hypothesis will, of course, depend on further experiments with purified PRMT1, TIS21, and BTG1, expressed without the GST moiety.

The TIS21 gene is expressed in a wide variety of tissues (Fletcher et al., 1991). BTG1 message is low in fully differentiated tissues such as brain or muscle, but is present in most other tissues (e.g. thymus, heart, lung, spleen, liver, and kidney) (Rimokh et al., 1991). Northern analysis of rat tissue extracts demonstrates that the PRMT1 gene is expressed in all rat tissues tested (Fig. 8A).

The expression of the TIS21 gene is rapidly and transiently induced by a variety of ligands, in a number of different cell types (Herschman, 1991). BTG1 gene expression has been reported to be down-regulated when cells enter S phase (Rouault et al., 1992). We compared the levels of expression of the BTG1, TIS21, and PRMT1 messages in mitogen-stimulated RAT1 cells. As expected (Fletcher et al., 1991), TIS21 message is not present in quiescent, growth-arrested cells (Fig. 8B). Stimulation with epidermal growth factor induces a rapid appearance of TIS21 message, detectable only at 30 min. Accumulation of TIS21 message is transient; within 1 h TIS21 mRNA returns to baseline, undetectable values. In contrast to results previously reported for 3T3 cells (Rouault et al., 1992), BTG1 message is also low in growth-arrested RAT1 cells (Fig. 8B). Following epidermal growth factor stimulation, BTG1 message is also elevated. The mitogen stimulation of BTG1 message accumulation peaks at a later time, 60 min, than that of TIS21 message. Even after 5-7 h, BTG1 message levels do not return to the baseline values observed in unstimulated, growth-arrested cells. Unlike the mitogen-induced elevations in BTG1 and TIS21 message levels, PRMT1 mRNA is present in growth-arrested RAT1 cells and does not change in response to epidermal growth factor stimulation. It seems likely that, if alterations in protein-arginine N-methyltransferase activity occur in response to ligand stimulation, such changes are likely to result from transient modulation of PRMT1 enzyme activity, rather than by alterations in PRMT1 gene expression.

DISCUSSION

Paik and Kim (1967) provided the first evidence for post-translational methylation of protein arginine residues. They described two novel radioactive species present in acid hydrolysates of calf thymus nuclei that had been incubated with S-adenosyl-[methyl-¹⁴C]-L-methionine. These methylated species were subsequently identified as arginine residues that had been mono- and dimethylated on their guanidino groups (Paik and Kim, 1968; Nakajima et al., 1971). Partial purification of protein-arginine transferase activity from calf brain suggested that two enzymes are present (Ghosh et al., 1988), one that methylates arginine residue 107 of myelin basic protein (Baldwin and Carnegie, 1971; Brostoff and Eylar, 1971), and a second enzyme that was initially thought to be a histone-specific methyltransferase. Subsequent studies demonstrated that this second protein-arginine transferase utilizes other substrates, e.g. hnRNP A1, much more efficiently than it does histones (Rajpurohit et al., 1994a; Liu and Dreyfuss, 1995). hnRNP A1 is methylated on an arginine residue present in the GAR domain (Rajpurohit et al., 1994a).

Although a number of purifications of histone/hnRNP A1 arginine N-methyltransferase have been reported (Ghosh et al., 1988; Rawal et al., 1994; Liu and Dreyfuss, 1995), the polypeptide composition of the methyltransferase complex has not been conclusively established. Previous chromatographic studies on partially purified preparations suggested that the histone/hnRNP A1 methyltransferase is a high molecular mass complex, estimated to be between 275 and 450 kDa, and containing multiple polypeptide components. Like GST-PRMT1, these protein-arginine transferases (i) do not methylate myelin basic protein, and (ii) mono- and asymmetrically dimethylate hnRNP A1 and histones. Ghosh et al. (1988) suggested that histone arginine methyltransferase preparations from calf brain had two major polypeptides, of molecular masses 110 and 75 kDa. In contrast, Rawal et al. (1994) report only a single 110-kDa polypeptide present in their rat liver preparation. SDS gel analysis of the preparation from HeLa cells obtained by Liu and Dreyfuss (1995) demonstrated eight distinct bands, the most prominent migrating at 110 and 45 kDa.

Our chromatographic data agree with the suggestion that native protein-arginine N-methyltransferase activity in mammalian cells and tissues exists as a macromolecular complex. Our present study describes the first cloning of a mammalian protein-arginine N-methyltransferase. The predicted molecular mass of the protein encoded by the PRMT1 cDNA is 40.5 kDa. Thus the 45-kDa band observed by Liu and Dreyfuss (1995) may represent the PRMT1 catalytic component. Co-immuoprecipitation experiments with antibody to the PRMT1 catalytic subunit should allow us to characterize the proteins present in the macromolecular complex and to investigate the effect of the TIS21 and BTG1 proteins on both the composition of the complex and the functions of its components.

The PRMT1 open reading frame has extensive sequence similarity to the open reading frame of the yeast ODP1 gene (Fig. 2). We have recently demonstrated, both by genetic and biochemical means, that the ODP1 gene also encodes a protein-arginine N-methyltransferase (Gary et al., 1996), and renamed this yeast gene RMT1, for arginine methyltransferase. Although the yeast RMT1 enzyme and the rat PRMT1 enzyme are both protein-arginine N-methyltransferases, the GST-RMT1 (yeast) fusion protein has substantially broader substrate specificity than does the GST-PRMT1 (rat) fusion protein (Gary et al., 1996). Moreover, analysis of data from the expressed sequence tag data bases suggest that more than one gene with sequence similarities to the rat PRMT1 gene may exist in humans. These data are reminiscent of other regulatory enzymatic activities that modify protein function, such as cyclin-dependent kinases. In Saccharomyces cerevisiae a single gene, CDC28, encodes a single cyclin-dependent kinase with broad substrate specificity. In contrast, mammalian cells encode a family of cyclin-dependent kinases, each with a more restricted substrate specificities.

We initiated the studies described here with the goal of identifying the role of the TIS21 immediate-early gene product in ligand-induced signal transduction. Using two-hybrid interaction analysis, we identified PRMT1 as a protein that can associate with both TIS21 protein and with BTG1 protein, a member of TIS21 gene family. In addition to identifying PRMT1 as a protein-arginine N-methyltransferase, and demonstrating its ability to interact with TIS21 and BTG1 prepared by in vitro translation, we also demonstrated that the TIS21 and BTG1 fusion proteins can modulate the methyltransferase activity of endogenous PRMT1 (Figs. 7, A and B).

It is of particular interest that extracts of RAT1 cells contain latent PRMT1 activity that cannot methylate endogenous substrates present in these extracts, unless recombinant TIS21 or BTG1 fusion protein is added. Although GST-PRMT1 can methylate the 55-kDa substrate present in RAT1 extracts, endogenous PRMT1 can only methylate this substrate if recombinant TIS21 or BTG1 GST fusion proteins are added (Fig. 7A). A 65-kDa substrate, not methylated by recombinant GST-PRMT1, can also be methylated by the endogenous PRMT1 activity only if TIS21 and/or BTG1 fusion protein is added to the extract.

The use of hypomethylated cell extracts as substrates for methylation resulted in two provocative observations. First, unlike extracts from untreated cells, extracts from cells treated with adenosine dialdehyde do not appear to have latent protein-arginine N-methyltransferase activity present. These data suggest that the presence of the catalytic component of this enzymatic activity, PRMT1, may itself be regulated by methylation. However, GST-PRMT1 does not methylate itself (Fig. 5, A and D). Thus, if PRMT1 levels are regulated by methylation, it must be by a more indirect route. Second, when GST-PRMT1 and GST-TIS21 are used together to methylate endogenous substrates in extracts from hypomethylated RAT1 cells, the methylation of the 34-kDa protein identified as hnRNP A1 is greater than that observed with GST-PRMT1 alone (Fig. 7B).

The data obtained with RAT1 cell extracts suggest that the enzymatic activity of a constitutively expressed protein methyltransferase catalytic component, PRMT1, can be modified both with respect to substrate specificity and with respect to catalytic efficiency by the interaction of ligand-induced, transiently expressed TIS21 and BTG1 regulatory subunits. These results are similar to the qualitative and quantitative modulation of enzymatic activity of constitutively expressed catalytic protein kinase molecules by transiently expressed regulatory subunits such as the G1 and G2 cyclins. We suggest that PRMT1-dependent protein arginine methylation, mediated by the TIS21/BTG1 protein family, may be an additional pathway of ligand-induced signal transduction.

Protein methylation has previously been identified as a signal transduction mechanism in several biological systems. In bacteria, the -carboxyl methylation of glutamate residues in a class of chemoreceptor/transducer proteins by the CheR methyltransferase is necessary for proper chemotactic response (Shapiro et al., 1995). In eukaryotic cells, carboxyl methylation of the COOH-terminal leucine residue in the protein phosphatase PP2A catalytic subunit has been shown to alter the subcellular localization of this molecule (Turowski et al., 1995). Moreover this methylation is reversible by the action of a methyl esterase (Xie and Clarke, 1994b). Additionally, endotoxin treatment of B cells (Law et al., 1992) and nerve growth factor treatment of PC12 pheochromocytoma cells (Kujubu et al., 1993) cause an increase in membrane-associated protein methylation that is specifically inhibited by pretreatment with the protein methylation inhibitor 5-methylthioadenosine.

hnRNP A1 plays a role in determining alternative mRNA splicing. Modulation of the level of hnRNP A1, either in vitro (Mayeda and Krainer, 1992) or in vivo (Caceres et al., 1994), can modulate 5 splice site selection during the maturation of mRNA. Moreover, methylation can alter the nucleic acid binding properties of hnRNP A1 (Rajpurohit et al., 1994b). Thus it seems likely that modulation of hnRNP A1 methylation may modify mRNA splicing in vivo. It is also possible that modulation of the methylation of hnRNP A1 may mediate other functions of this molecule, such as nuclear export (Michael et al., 1995), nuclear localization (Siomi and Dreyfuss, 1995), or other aspects of RNA biogenesis (Dreyfuss et al., 1993).

We suggest that transient expression of the TIS21 and BTG1 genes may mediate extracellular signals by modulation of splicing following ligand stimulation. Perhaps the best precedent for such a suggestion is the study of Shifrin and Neel (1993), who demonstrated that a novel form of PTP-1B, a nontransmembrane phosphotyrosine phosphatase, is expressed in human fibroblasts when they are stimulated with a variety of growth factors. The expression of the altered form of PTP-1B protein is the consequence of a growth factor-induced alteration in splicing of the PTP-1B message. Moreover, the growth factor-induced alteration in splicing of the PTP-1B message requires protein synthesis, suggesting that the product of an immediate early gene (perhaps TIS21) is necessary. Similarly, Ogimoto et al. (1993) demonstrated that either anti-IgM antibody or the combination of phorbol ester and ionomycin induces alterations in splicing of the CD45 message in murine B cells, and Sarmay et al. (1995) demonstrated that interleukin 4, anti-IgM antibody, or phorbol ester treatment can induce alterations in splicing of the Fc RII message in human B cells.

Many other protein components of the RNA processing machinery are also substrates for methylation. For example, glycine- and arginine-rich GAR domains containing multiple potential RGG substrate sites for PRMT1 methylation, are present in nucleolin and fibrillarin (Najbauer et al., 1993). These proteins contain N^G,N^G-dimethylarginine (asymmetric) in vivo (Lischwe et al., 1985a, 1985b). Nucleolin and fibrillarin participate in pre-rRNA processing, and are specifically localized to the nucleolus (Henriquez et al., 1990). It seems likely that these proteins (i) will also be substrates for PRMT1 and (ii) their methylation by PRMT1 will be subject to modification by the interaction of the TIS21 and/or BTG1 proteins with the methyltransferase enzyme.

In addition to modulating mRNA splicing, protein arginine methylation may modulate growth factor localization and function. The high molecular weight form of fibroblast growth factor has a glycine- and arginine-rich domain at its NH₂ terminus (Prats et al., 1989) that contains methylated arginine residues (Burgess et al., 1991). Arginine methylation has been suggested as a mechanism for preferential targeting of the high molecular weight form of this growth factor to the nucleus (Burgess et al., 1991).

Identification and cloning of the catalytic component of the protein-arginine transferase should now make possible previously inaccessible experiments. Elucidation of additional substrates for PRMT1, the functional consequences following methylation of PRMT1 substrates, the search for other protein-arginine transferases and their regulatory subunits, the nature of the interactions between protein-arginine transferases and their regulatory subunits, and the consequences of genetic alterations in these molecules all present important questions for future studies.