ABC50 Interacts with Eukaryotic Initiation Factor 2 and Associates with the Ribosome in an ATP-dependent Manner*

Eukaryotic initiation factor 2 (eIF2) plays a key role in the process of translation initiation and in its control. Here we demonstrate that highly purified mammalian eIF2 contains an additional polypeptide of apparent molecular mass of 110 kDa. This polypeptide co-purified with eIF2 through five different chromatography procedures. A cDNA clone encoding the polypeptide was isolated, and its sequence closely matched that of a protein previously termed ABC50, a member of the ATP-binding cassette (ABC) family of proteins. Antibodies to ABC50 co-immunoprecipitated eIF2 and vice versa, indicating that the two proteins interact. The presence of ABC50 had no effect upon the ability of eIF2 to bind GDP but markedly enhanced the association of methionyl-tRNA with the factor. Unlike the majority of ABC proteins, which are membrane-associated transporters, ABC50 associates with the ribosome and co-sediments in sucrose gradients with the 40 and 60 S ribosomal subunits. The association of ABC50 with ribosomal subunits was increased by ATP and decreased by ADP. ABC50 is related to GCN20 and eEF3, two yeast ABC proteins that are not membrane-associated transporters and are instead implicated in mRNA translation and/or its control. Thus, these data identify ABC50 as a third ABC protein with a likely function in mRNA translation, which associates with eIF2 and with ribosomes.

The initiation of translation in eukaryotes is a complex process involving approximately 15 different initiation factor proteins, the ribosome, methionyl-tRNA i , and mRNA. One of the key initiation factors is eukaryotic initiation factor 2 (eIF2), 1 a heterotrimeric protein (subunits ␣, ␤ and ␥) that binds guanine nucleotides and mediates the transfer of the initiator Met-tRNA i (as a ternary complex, eIF2⅐GTP⅐Met-tRNA i ) to the 40 S ribosomal subunit (1). This step is required for all translation initiation events, and eIF2 is known to play an important role in the regulation of translation initiation under a variety of conditions (1,2).
At a late stage in the initiation process the GTP bound to eIF2 is hydrolyzed to GDP plus P i and an inactive eIF2⅐GDP complex is released from the ribosome (reviewed in Refs. 1 and 3). GTP hydrolysis may be activated by the interaction of eIF2 with another translation factor, eIF5 (1). Because the rate of dissociation of GDP from eIF2 is very low, regeneration of active eIF2⅐GTP requires an additional protein termed eIF2B (a guanine nucleotide exchange factor; reviewed in Ref. 3). eIF2B is a heteropentameric protein (subunits ␣-⑀) and also plays an important role in the regulation of translation initiation (see below).
The ␥-subunit of eIF2 binds GDP or GTP (4), whereas eIF2␣ plays a regulatory role; phosphorylation of eIF2␣ at Ser 51 results in inhibition of the eIF2B-catalyzed recycling of eIF2 (reviewed in Refs. 2 and 3). Phosphorylation of eIF2␣ is important in the overall regulation of translation in response to numerous different types of cell stress as discussed below. The role of eIF2␤ is less clear. There is data suggesting that it is required for Met-tRNA i binding (5) and is involved in the selection of the start site for initiation of translation (6). Other studies have indicated that it interacts with eIF5 (7) and with eIF2B (8).
Phosphorylation of Ser 51 in eIF2␣ is involved in the regulation of the activity of eIF2 and the overall rate of protein synthesis under a variety of conditions in eukaryotic organisms (9). When phosphorylated on its ␣-subunit, eIF2 is a potent competitive inhibitor of eIF2B and thus causes inhibition of overall peptide chain initiation (9). A number of kinases capable of phosphorylating Ser 51 have been identified, these include the double-stranded RNA-activated eIF2␣ kinase RNAdependent protein kinase (10), the heme regulated kinase haem-regulated inhibitor (11), and most recently, the endoplasmic reticulum-associated eIF2 kinase (RNA-dependent protein kinase-like endoplasmic reticulum kinase/pancreatic eIF2␣ kinase (12,13)). In yeast, there is a further well characterized regulatory mechanism involving the phosphorylation of eIF2 by the protein kinase GCN2, which is believed to be activated by uncharged tRNA in response to amino acid starvation (14). GCN2 contains both a kinase domain and a region similar to histidinyl-tRNA synthetase that may bind tRNA. In yeast, inhibition of eIF2B through the phosphorylation of eIF2␣ leads to up-regulation of the translation of the mRNA encoding the transcriptional regulator GCN4, thus promoting transcription of genes encoding enzymes involved in amino acid biosynthesis (14). The genetic tractability of yeast has allowed the identification of several other genes involved in this regulatory mechanism. These include GCN1 and GCN20, which are both required for the activation of GCN4 synthesis, and, in particular, the activation of the kinase GCN2, in response to amino acid deprivation (15)(16)(17). GCN20 is a member of the ATP-binding cassette (ABC) family of proteins and forms a complex with GCN1 that appears to be present upon elongating ribosomes, but the precise roles played by these proteins in the regulation of GCN2 remains to be established. Recent data indicate that higher eukaryotes also possess GCN2-like enzymes (18 -20) because cDNAs for such enzymes have been cloned from fruit flies and mammals. Indeed, it has been known for some years that manipulations that lead to accumulation of uncharged tRNA (e.g. incubation of rat liver cells without essential amino acids (21), treatment of rat liver cells with an inhibitor of histidinyl-tRNA synthetase (histidinol) (22), or the use of mammalian cell lines harboring a temperature-sensitive mutant of the leucyl-tRNA synthetase (23,24)) lead to increased phosphorylation of eIF2␣, indicative of the operation of a GCN2-like system.
Here we identify a protein that co-purifies and interacts with mammalian eIF2 and show that it is the rabbit homologue of human ABC50 which had previously been identified as a protein (with no known function) whose expression is elevated in tumor necrosis factor ␣-stimulated synoviocytes (25). It contains the motifs of the ABC family of proteins and exhibits sequence similarity to the members of the eEF3 subfamily (which includes the Saccharomyces cerevisiae proteins eEF3 and GCN20) (25). We show that mammalian ABC50, like yeast eEF3 and GCN20, is associated with the translational machinery.

MATERIALS AND METHODS
Chemicals and Biochemicals-Unless otherwise indicated all chemicals and biochemicals were obtained from Sigma or Merck.
Protein Purification-eIF2 and eIF2B was purified from rabbit reticulocyte lysates or HeLa cell lysate essentially as described previously (26).
Peptide Sequence Analysis-To obtain internal peptide sequence data for p110, preparations of eIF2 rich in this polypeptide (as assessed by Coomassie Brilliant Blue staining) were subjected to SDS-PAGE, and the region of the gel containing p110 was excised. Cleavage of the protein using CNBr treatment, separation of the fragments by Tricine-SDS-PAGE, and transfer to Immobilon-P membrane were performed as described earlier (28,29). Automated Edman degradations were performed by K. Howland (Department of Biosciences, University of Kent at Canterbury) on a Procise 492 Protein Sequenator (PerkinElmer Biosystems) and phenylthiohydantoin-derivatives were identified by on-line reverse-phased high pressure liquid chromatography utilizing a phenylthiohydantoin C18 column (2.1 ϫ 220 mm, 5-mm particle size; PerkinElmer Biosystems). PerkinElmer Biosystems reagents and solvents were used where available.
Data Base Searching-DNA or protein sequences were used to search the data bases available at the National Center for Biotechnology Information web site using the Advanced Gapped Basic Local Alignment Search Tool (BLAST) program (30). DNA and protein sequences identified by these searches were recovered from the Gen-Bank TM and GenProt data bases using the National Center for Biotechnology Information Entrez program.
Alignment of DNA and Protein Sequences-DNA and protein sequences were aligned using the ClustalW Multiple Sequence Alignment program at the European Bioinformatics Institute web site or the BLAST 2 sequences program at the National Center for Biotechnology Information web site. Shading and editing of these alignments was performed using the GeneDoc Multiple Sequence Alignment Editor and Shading Utility obtained from the Pittsburg Supercomputing Center web site.
Generation of Rat Testis cDNA-Total RNA (about 2 mg) was isolated from 1 g of rat testis tissue using TRIZOL® reagent (Life Technologies, Inc.), and the mRNA (about 20 g) was isolated from this total RNA using the Oligotex mRNA mini-prep kit (Qiagen). First strand cDNA synthesis, using oligo(dT) 15 as the primer was performed using Expand Reverse Transcriptase (Roche Molecular Biochemicals).
Polymerase Chain Reaction Amplification and Cloning of the 5Ј End of the p110 cDNA-The region of the rat p110 cDNA corresponding to nucleotides 169 -1216 of GenBank TM sequence AF027302 was amplified using the polymerase chain reaction from the rat testis cDNA using the p110 FOR 5 (5Ј-CAAAGTAGTGAAGAAAGGC-3Ј) and p110 REV 2 primers and Taq DNA polymerase. The products were ligated into the pGEM-T EASY vector (Promega) and insert DNA sequenced as above. Rapid amplification of the 5Ј end of the rat p110 cDNA was performed essentially as described previously (31) using the p110 REV 4, p110 REV 3, or p110 REV 1 primer. The products were cloned into the pGEM-T EASY vector as above. Plasmid DNA was isolated, and insert DNA was sequenced.
Antibody Production-A peptide with sequence DEESQEA-PELLKRPKEC was prepared by Kevin Howland (University of Kent at Canterbury) and used to generate a rabbit anti-p110 antiserum. The anti-p110 immunoglobulins were purified from this serum using the peptide cross-linked to Affi-gel (Bio-Rad) as described (32).
Immunoprecipitation-Purified rabbit eIF2/p110 obtained from the purification procedure above (5-10 g) was added to a mixture (final volume, 75 l) consisting of IP buffer (phosphate-buffered saline (PBS), pH 7.6, 0.07% (w/v) SDS, 0.7% (w/v) sodium deoxycholate, 0.7% (v/v) Triton X-100) and a 1 ⁄10 volume of either PBS, pH 7.6, or the following antisera as indicated in the results: preimmune antiserum from rabbits prior to injection with antigen, mouse monoclonal anti-eIF2␣ antiserum, or purified rabbit anti-p110 peptide antiserum. The reactions were mixed briefly and incubated at room temperature for 1 h. PAN-SORBIN® cell suspension (20 l) in RIPA buffer (50 mM HEPES/KOH, pH 7.6, 150 mM KCl, 0.1% (w/v) SDS, 1% (w/v) sodium deoxycholate, 1% (v/v) Triton X-100) was then added to each reaction, mixed briefly, and incubated at room temperature for 20 min. The suspensions were then centrifuged at 8,000 rpm for 2 min in a benchtop microcentrifuge to pellet the antigen/antibody/PANSORBIN® cell complexes. The supernatants were removed, and the pellets were resuspended and washed in 500 l of RIPA buffer. This wash procedure was repeated a further two times and, finally, once in 500 l of PBS buffer. SDS-PAGE (2ϫ) sample buffer (25 l) was added to each of the final pellets which were then frozen at Ϫ80°C and thawed to aid resuspension of the pellet. The samples were boiled at 100°C for 5 min and centrifuged at 13,000 rpm prior to analysis by SDS-PAGE/immunoblotting. Sucrose Cushion Centrifugation-HEK 293 cells were maintained and passaged in Dulbecco's modified Eagle's medium supplemented with fetal bovine serum (10% v/v) and antibiotic/antimycotic (1% v/v; Life Technologies, Inc.). Lysis of cells at 70 -80% confluency was achieved in ice-cold SDG100 buffer (50 mM HEPES/KOH, pH 7.6, 2 mM MgCl 2 , 100 mM KCl, 1 g/ml antipain, 1 mM benzamidine, 1 g/ml leupeptin, 1 g/ml pepstatin, 1 mM phenylmethylsulphonyl fluoride, 1 M microcystin) containing 0.1% (v/v) Triton X-100. Nuclei and membranous material was removed by centrifugation (15 min at 3,500 ϫ g, 4°C). The cleared lysate was layered gently onto 0.8 M sucrose (in SDG100 buffer) and centrifuged (2 h at 290,000 ϫ g, 4°C) in a Beckman SW55Ti rotor. After centrifugation, the cleared lysate at the top of the cushion and the sucrose layer were removed separately, and after a brief rinse in 100 l of PBS, pH 7.6, the ribosomal pellet was resuspended in SDG500 buffer (as SDG100 buffer except 500 mM KCl). The salt washed ribosomes were then centrifuged as before except using a 0.8 M sucrose cushion in SDG500 buffer.
Sucrose Density Gradient Centrifugation-HEK 293 cells were maintained and lysed into SDG100 buffer (except at 7 mM MgCl 2 ) ϩ 0.1% (v/v) Triton X-100 as above. After the lysate had been cleared, it was layered onto 20 -50% (w/v) sucrose gradients (15 ml, prepared in SDG100 buffer) and centrifuged (3.5 h at 110,000 ϫ g, 4°C) in a Beckman SW28.1 rotor. Where indicated in the results, ATP or ADP was added to the cleared lysate to give a final concentration of 5 mM, and this was then layered onto a gradient that contained 5 mM ATP or ADP as appropriate. Each gradient was fractionated using a Brandel model 184 Fractionator by displacement with 60% (w/v) sucrose at a flow rate of 1 ml/min. The A 260 of the displaced gradient was analyzed in a 5-mm-path length ISCO Type 6 Optical Unit and recorded using an ISCO model UA-5 Absorbance/Fluorescence meter. Fractions (0.35 ml) were collected using a Gilson Microfraction model 203 fraction collector. Trichloroacetic acid precipitation of each of the fractions obtained was performed, and the precipitated samples were resolved by SDS-PAGE for immunoblotting analysis.
Gel Filtration-Gel filtration was performed using an Amersham Pharmacia Biotech Superose 6 HR10/30 column on a Bio-Rad Biologic FPLC System. Protein samples were dialyzed into the column running buffer (50 mM HEPES/KOH, pH 7.6, 150 mM KCl, 1 mM DTT, 5% (v/v) glycerol, unless otherwise indicated), and the dialysate (250 l) was loaded onto the column and eluted in the running buffer at a flow rate of 0.3 ml/min. Fractions (300 l) were collected, and peak fractions were analyzed by SDS-PAGE and immunoblotting using the antisera indicated in the results. The column was calibrated using the standards from the molecular mass marker kit for gel filtration chromatography (molecular mass range, 12,000 -200,000 Da; Sigma). The kit was supplemented with thyroglobulin (molecular mass, 669,000 Da) and apoferritin (443 000 Da) to increase the upper end of the calibration range. The standards were loaded and eluted under the same buffer and flow rate conditions as above.
GDP Binding and Ternary Complex Formation Assays-These assays were performed essentially as described previously (5).

RESULTS
Identification of p110 -SDS-PAGE analysis of several preparations of eIF2 that had been isolated from rabbit reticulocytes as described previously (26) indicated the consistent presence of a protein with an apparent molecular mass of approximately 110 kDa in addition to the three subunits of eIF2 (Fig. 1A). This polypeptide (henceforth termed p110) appeared to co-elute with eIF2 from the later ion exchange column steps of our standard purification procedure (Fig. 1B). This co-purification of p110 with eIF2 was analyzed further as described below.
To obtain information on the identity of this protein, the p110 protein band was excised from a SDS-polyacrylamide gel and cleaved with CNBr. After transfer to Immobilon-P membrane and Amido Black staining, five distinct bands were observed. These were excised, and the derived peptides were sequenced as described under "Materials and Methods" ( Table  I).
The data for bands 2 and 5, which gave the longest and clearest sequences, were used to search the data bases available at the National Center for Biotechnology Information web site. Note, band 3 appears to be a more complete digest of the same peptide fragment as band 2 because they both contain the same sequence in their first five residues, and band 4 appears to be a mixture of multiple fragments. At the time of the initial analysis, no sequences were identified in the nonredundant data base of sequences, suggesting that the p110 protein was novel. However, tBLASTn analysis of the data base of expressed sequence tags (dbEST) led to the identification of several mouse and human ESTs encoding for peptide sequences showing 100% identity to the band 5 sequence (when a Y residue was used at the location of the indicated ambiguity in Table I). Alignment of all the ESTs identified in this manner suggested that they all encoded the same protein (data not shown). The largest sequences were a 2,207-nt-long human EST (GenBank TM accession number U66677) and a 491-nt-long murine EST (GenBank TM accession number AA014138). No EST sequences containing the band 2 sequence were identified at this time.
Cloning of the cDNA Encoding for p110 -The polypeptide sequence predicted from the EST alignment appeared to be the C-terminal end of p110 because many of the ESTs identified contained an in-frame stop codon and a polyadenylation signal (AAUAAA) downstream. To clone and sequence the cDNA encoding for p110, primers were designed for the production of an oligonucleotide probe by the polymerase chain reaction amplification of the cDNA corresponding to nucleotides 18 -418 of the AA014138 sequence. This probe was used to screen a rat skeletal muscle cDNA library for phage containing the p110 cDNA. One clone was identified and the cDNA insert of its phagemid sequenced (GenBank TM accession number AF293383). Analysis of the insert sequence demonstrated that it was a chimeric molecule because approximately 700 nt encoded for rat parvalbumin and the remaining 2400 nt were found to show Ͼ95% identity to the alignment of the previously identified EST sequences and also 89% identity to a newly submitted human cDNA sequence for a protein of unknown function that had been termed ABC50 (with GenBank TM accession number AF027302 and GenProt accession number AAC70891). The mRNA for this protein had been identified by virtue of its increased expression in synoviocytes stimulated with tumor necrosis factor ␣ (25). The band 2/band 3 and band 5 sequences in Table I were all present in the ABC50 protein sequence. Thus, the rabbit and rat p110 proteins were identified as homologues of the human ABC50 protein. Alignment of the rat p110 cDNA sequence from the phagemid insert with the ABC50 cDNA indicated that the rat sequence lacked approximately 800 nt from its 5Ј end. To obtain these missing sequence data, the published ABC50 cDNA sequence was used in BLAST  CNBr digestion of rabbit p110 N-terminal sequence data for rabbit p110 were obtained as described under "Materials and Methods." The data for the bands resolved by Tricine-SDS-PAGE are presented in order of slowest (band 1) to fastest (band 5) migrating fragments. Ambiguities in the assignment of residues are represented within parentheses by the potential candidates at that site. Lowercase letters indicate a low degree of certainty in the assignment of that residue, and X indicates an undefined residue. p110 peptide fragment Peptide sequence obtained.
searches to identify rat EST sequences for the 5Ј end of the rat cDNA. None were found that corresponded directly to the extreme 5Ј end of the ABC50 cDNA sequence, but an EST (Gen-Bank TM accession number H34173) corresponding to nucleotides 168 -406 of AF027302 was identified. Using this EST sequence, a polymerase chain reaction primer was designed for the amplification of the rat p110 cDNA corresponding to nucleotides 169 -1216 of AF027302 as described under "Materials and Methods." Using this strategy, the sequence data were extended to within 169 nt of the 5Ј end of the published human sequence. To extend this sequence further, 5Ј rapid amplification of cDNA ends reactions were attempted but proved unsuccessful, and no further sequence data were obtained for the rat p110 cDNA. Analysis of the p110 cDNA and Polypeptide Sequences-The sequence data obtained for the rat p110 cDNA were aligned with that of the human ABC50 cDNA and found to exhibit 88% identity. The most significant difference between the two sequences was the presence of an 102-nt insert in the rat cDNA (between nucleotides 745 and 746 of AF027302). The rat p110 cDNA sequence was translated using the Translate tool at the ExPASy web site, and the sequence predicted was used to perform BLAST searches of the National Center for Biotechnology Information data bases. The sequences showing the highest similarity to the rat p110 protein were aligned using the ClustalW program at the European Bioinformatics Institute web site, and this alignment was edited using the GeneDoc program (data not shown). From this alignment, the rat p110 protein sequence was found to be 88% identical and 91% similar to ABC50, with a 34-amino acid insert (because of the nucleotide sequence insert noted above) occurring between residues 217 and 218 of the AAC70891 sequence. However, two human EST sequences (GenBank TM accession numbers AA083604 and AA587878) as well as several murine ESTs were identified, which showed high identity (Ͼ90%) to this insert sequence, suggesting that two or more isoforms of p110/ABC50 may occur within mammals. In addition, searches of the public data bases revealed four uncharacterized proteins from Caenorhabditis elegans, an uncharacterized protein from Drosoph-ila melanogaster, an uncharacterized protein from Arabadopsis thaliana, an uncharacterized protein from Schizosaccharomyces pombe, and two characterized proteins from S. cerevisiae that show Ͼ35% identity/Ͼ55% similarity to the p110/ABC50 sequences (summarized in Table II and also see below). ABC50 had been identified by Richard et al. (25) as a member of the ABC family of proteins by virtue of the presence of two ABC motifs within its sequence (Fig. 2). The proteins that show high similarity to p110/ABC50 also each contain two ABC motifs, and the arm of the ABC family that contains ABC50 is that of the eEF3 subfamily (25). This subfamily includes eEF3 and GCN20, which are yeast proteins involved in the process of translation. p110/ABC50 is homologous to the S. cerevisiae GCN20 and YER036c proteins ( Fig. 2 and Table II). GCN20 is implicated in the regulation of the eIF2␣ kinase GCN2 in response to amino acid starvation in yeast (16,17), whereas YER036c has no known function (33). The similarity between the yeast proteins and p110/ABC50 is greatest in the C-terminal two-thirds of the mammalian proteins. The N-terminal one-third of p110/ABC50 only shows approximately 20% identity and 30% similarity to the equivalent region of GCN20 and is significantly longer (between 54 and 88 residues depending on the presence of the 34-amino acid insert). The N terminus of p110/ABC50 exhibits a high degree of hydrophilicity as determined using the Kyte-Doolittle algorithm (data not shown), and the entire sequence consists of a high proportion of charged residues (approximately 36% KRED residues). No transmembrane domains could be identified in the p110 or ABC50 sequences, and this together with the other features described above suggested that p110/ABC50 was not a membrane-associated protein but was more likely to be involved in the process of translation.
p110/ABC50 Is Widely Expressed in Mammalian Cell Types- Richard et al. (25) demonstrated that the mRNA for ABC50 was expressed in a broad range of human tissues, and the source tissues for the numerous mouse, human, and rat ESTs for p110 are also very diverse. This suggested that the p110/ABC50 protein might be ubiquitously expressed. SDS-PAGE/immunoblotting analysis (using an antibody to ABC50 a The human EST sequences with GenBank™ accession numbers AA083604 and AA587878 were translated and inserted into the ABC50 protein sequence at the appropriate location to obtain the percentage of identity/percentage of similarity of a potential second isoform of the human ABC50 protein relative to the rat sequence. b The mouse p110/ABC50 protein was generated by translating the following list of mouse EST sequences and aligning them against the rat p110 and human ABC50 protein sequences: AA060014, AA404025, AA681186, AA065712, AA517413, AA543349, AA795962, AA408357, AA511948, AI156605, AA636681, AA656533, AA014138, AA144398, AI264984, AA921438, AA404181, AI156313, AI049168, AA218241, AA265043, and AA108112 (these GenBank™ accession numbers are listed in the order that they occur within the mouse p110 sequence from the 5Ј end/N terminus to the 3Ј end/C terminus). When calculating the percentage of identity/percentage of similarity between the mouse and rat sequences, the gaps in the mouse sequence because of a lack of EST sequence data were ignored. obtained from M. Richard, Center Hospitalier de L'Université Laval, Sainte-Foy, Quebec, Canada) of cell extracts from four mammalian species indicated that p110/ABC50 was present in each species (Fig. 3). The four cell types used here were also from different tissues, and therefore p110/ABC50 is expressed in various tissue types as well as in different species. The migration of the protein differed slightly in the various species, possibly reflecting the presence (in this case, in the rabbit and hamster proteins) or absence (in the rat and human proteins) of the 34-amino acid sequence identified above.
p110/ABC50 Associates with Ribosomes-Because of the similarity of p110/ABC50 to the members of the eEF3 ABC subfamily and its co-purification with eIF2, we investigated whether p110/ABC50 associated with ribosomes. Ribosomes from HEK 293 cell extracts were sedimented through a sucrose cushion and then analyzed for the presence of p110/ABC50 by SDS-PAGE/immunoblotting (Fig. 4). p110/ABC50 was found associated with the pelleted ribosomes and also in the postribosomal supernatant. This was also the case for eIF2. After the ribosomes had been resuspended in buffer containing 0.5 M KCl and resedimented, p110/ABC50 and eIF2 were no longer associated with the ribosomal material (Fig. 4). These results clearly demonstrated that p110/ABC50 can associate with the ribosome but is not an intrinsic ribosomal protein, unlike L5 and S26, which remained in the pelleted material. This behavior is typical of many translation factors including eIF2, eIF4F, and eIF3 (34 -38). To study further this association with the ribosome, sucrose density gradient analysis was performed as described under "Materials and Methods" (Fig. 5). Immunoblotting analysis of the gradient fractions revealed that p110/ ABC50 was associated with the 40 and 60 S ribosomal subunits but was not detectable in the fractions containing 80 S ribosomes or polyribosomes. Again, a substantial proportion of the p110/ABC50 was also found in nonribosomal fractions at the top of the gradient (Fig. 5B, control gradient). The distribution of eIF2 in the gradients appeared to be similar to that of p110/ABC50 as determined by immunoblotting for eIF2␤ (Fig.  5B, bottom panel). However, prolonged exposure of the Western blots to film indicated that eIF2 was also detectable on the polysomal material, whereas p110/ABC50 was not (data not shown). This might reflect greater sensitivity of the anti-eIF2␤ antiserum or may arise from a genuine difference in the actual distribution of the two proteins on the ribosomes.
Because p110/ABC50 possesses binding motifs for adenine nucleotides, we also studied the effect of the presence of ADP and ATP on the association of the protein with ribosomes. Addition of ADP to the lysate and gradient buffer caused a marked decrease in the amount of p110/ABC50 associated with the 40 and 60 S subunits (compare Control and ϩ ADP panels in Fig. 5B), whereas addition of ATP caused a marked increase in the level of association (compare Control and ϩ ATP panels in Fig. 5B). These effects were not due to any changes in the levels of 40 and 60 S subunits or in the association or distribution of eIF2, all of which were apparently identical under each of the conditions studied (data not shown).
Co-purification of p110/ABC50 with eIF2-To investigate further the co-purification of eIF2 and p110/ABC50, the distribution of the two proteins during the purification of the translation factors from rabbit reticulocyte lysate was determined by immunoblotting using the anti-ABC50 antibody in conjunction with antibodies to each of the three eIF2 subunits and to the eIF2B␣ subunit. It was found that p110/ABC50 was present in all samples containing eIF2 throughout the purification procedure, except when eIF2 was complexed with eIF2B (data not shown). However, when the purification procedure was applied to HeLa cell extract, a small fraction of the eIF2 was recovered that was deficient in p110/ABC50 (see below and Fig. 8A).
Interaction of p110/ABC50 with eIF2-The co-purification of eIF2 and p110/ABC50 was consistent with the notion that eIF2 and p110/ABC50 form a stable interaction, although it is also theoretically possible that they share identical chromatographic properties on all four column matrices used. Therefore, to investigate whether p110/ABC50 and eIF2 do interact, Mono S fractions of purified eIF2 containing p110/ABC50 were subjected to immunoprecipitation using anti-eIF2␣ and anti-p110 antisera (prepared as described under "Materials and Methods"; note that the anti-ABC50 antiserum did not efficiently immunoprecipitate the p110/ABC50 protein (data not shown)). Initial studies were hampered by high levels of nonspecific binding of both eIF2 and p110/ABC50 to the PANSORBIN® matrix in the absence of any antisera ( Fig. 6; note the appearance of eIF2 and p110/ABC50 in the No Ab (Ϫ) lane). This high degree of nonspecific binding was also observed when protein A-Sepharose was used as the matrix (data not shown). To reduce this nonspecific binding it was necessary to use high stringency buffers containing nonionic detergents in the reactions and the washes (Fig. 7, compare No Ab (Ϫ) and No Ab (ϩ) lanes). When such buffers were used, the anti-p110 or anti-eIF2␣ precipitated material was found to contain both p110/ ABC50 and eIF2, whereas the control reactions did not (Fig. 6). However, when similar immunoprecipitation reactions were performed using rabbit reticulocyte or HeLa cell lysate in place of the purified proteins, co-immunoprecipitation of the proteins could not be observed above the level of nonspecifically bound proteins (data not shown).
As an alternative strategy to analyze the interaction between the two proteins, Mono S column fractions containing the purified eIF2 and p110/ABC50 were subjected to analytical gel filtration on a Superose 6 HR 10/30 column (Fig. 7A). The two proteins co-eluted from this column at a retention time equivalent to a molecular mass of about 400 kDa. This is well in excess of the expected masses of the individual proteins (approximately 150 kDa for eIF2 and 110 kDa for p110/ABC50), which suggests that they form a complex. Repetition of the gel filtration with running buffer containing 1 M KCl did not significantly alter this retention time (Fig. 7B), whereas use of buffer containing 2.0% (v/v) Triton X-100 led to a shift in the retention times of both proteins to a position equivalent to that of the individual molecular masses of the two proteins (Fig.   7C). The apparent dissociation of eIF2 and p110/ABC50 by detergents and not by high salt suggests that the binding of the two proteins to each other involves hydrophobic interactions.
ABC50 Enhances the Formation of Ternary Complexes but Has No Effect on GDP Binding by eIF2-Purification of eIF2 from HeLa cell extract, rather than rabbit reticulocyte lysate, yielded a small amount of eIF2 that was deficient in ABC50 as determined by SDS-PAGE and by immunoblotting (Fig. 8A; the level of eIF2␣ is similar in each sample, whereas the ϪABC50 sample is evidently deficient in p110/ABC50). The availability of this material enabled us to test two of the known activities of eIF2, guanine nucleotide binding and ternary complex formation, in the presence and absence of p110/ABC50. Binding of GDP to the ABC50-deficient eIF2 was not significantly different from binding to the eIF2/ABC50 complex (compare Ϫ ABC50 and ϩ ABC50 in Fig. 8B). In contrast, binding of [ 35 S]methionyl-tRNA was reproducibly about 4-fold greater in the presence of ABC50 (compare Ϫ ABC50 and ϩ ABC50 in Fig. 8C), which indicates that p110/ABC50 plays a role in this important function of eIF2.
ABC50 Accumulates Following Stimulation of T-lymphocytes-ABC50 was first identified (25) because of the induction of its mRNA following stimulation of synoviocytes with tumor necrosis factor ␣ (which induces their proliferation). To determine whether or not stimulation of other cells to undergo proliferation leads to increased expression of the ABC50 protein, samples of stimulated T-lymphocytes were obtained. In the case of T-lymphocytes, it is known that the levels of other components of the translational machinery, such as certain translation initiation factors including eIF2, increase very markedly following stimulation (34 -36, 38). Primary human T-lymphocytes were stimulated for up to 24 h with phorbol myristate acetate and ionomycin. The cell extracts were analyzed by immunoblotting for eIF2 and ABC50 (Fig. 9). The samples exhibited increased expression of both p110/ABC50 and eIF2␣ following 8 -16 h of stimulation. Similar induction of eIF2 and ABC50 proteins was observed after stimulation of resting T-cells by the addition of anti-CD3 and anti-CD28 antibodies (data not shown).

DISCUSSION
The purification of translation factors from rabbit reticulocyte lysate by multiple ion exchange chromatographic steps consistently led to the isolation of eIF2 in association with a 110-kDa protein.
Here we identify this protein as the rabbit homologue of the human ABC50 protein. ABC50 was previously identified as a protein of no known function that was induced in tumor necrosis factor ␣-stimulated synoviocytes (25). For reasons of consistency, we suggest that the name ABC50 be used to refer to the mammalian homologues of the protein in the future. Rabbit ABC50 was shown by immunoblotting analysis to co-purify with eIF2 throughout our entire purification procedure, which suggested that the two proteins form a stable complex. The co-purification of ABC50 with eIF2 has not previously been noted. Examination of the data in previous publications indicates that eIF2 can be purified from mammalian sources in the absence of ABC50 (e.g. see figures in Refs. 39 and 40). However, eIF2 may have been purified with ABC50 in others, on the basis that a protein at the expected position for ABC50 is observed in the Coomassie-stained gels shown in Refs. [41][42][43][44][45]. The absence of ABC50 from some eIF2 samples could be due to differences in the method of purification or the absence of a particular proteinase inhibitor in the buffers used during our purification procedure. We have confirmed that our protocol also leads to co-purification of ABC50 and eIF2 from HeLa cells and HEK 293 cells, 2 which shows that it is not a peculiarity of the rabbit proteins or of rabbit reticulocyte lysates. However, a small amount of ABC50-deficient eIF2 sample was obtained from the HeLa cell extract, possibly because of degradation of the ABC50 or because of lower levels of the protein within this cell line.
An interaction between the two proteins was confirmed by immunoprecipitation analysis whereby anti-eIF2␣ and purified anti-p110 antibodies each co-immunoprecipitated rabbit ABC50 or eIF2, respectively, from the purified samples. However, similar reactions using rabbit reticulocyte or HeLa cell lysates did not lead to detectable co-immunoprecipitation of the proteins. In this case, the anti-p110 antibody did not immunoprecipitate significant amounts of ABC50 from these extracts and the anti-eIF2␣ antibody only immunoprecipitated a proportion of the eIF2 (which did not contain detectable ABC50), and a substantial amount of the eIF2 remained in the super-2 J. K. Tyzack, X. Wang, and C. G. Proud, unpublished observations. FIG. 8. p110/ABC50 enhances Met-tRNA binding to eIF2. eIF2 purified from HeLa cell extract yielded a small amount of p110/ABC50deficient eIF2 as determined by SDS-PAGE/Coomassie staining and immunoblotting (IB) using the anti-ABC50 and anti-eIF2␣ antisera as indicated (A). The ABC50-deficient eIF2 eluted from the final Mono S column at a lower concentration of KCl than the ABC50 containing fractions (data not shown). GDP and Met-tRNA binding assays were performed in duplicate on four different occasions using the p110/ ABC50-deficient eIF2 (Ϫ ABC50) and eIF2/ABC50 complex (ϩ ABC50) shown in A giving similar results on each occasion. B and C show representative results from one set of assays. Samples were diluted to give results within the linear range of the assays.

FIG. 9. p110/ABC50 and eIF2 increase in stimulated T-cells.
Resting T-lymphocytes were stimulated for the indicated times with phorbol myristate acetate and ionomycin. Cell lysates (20 g of total protein/lane) were analyzed by immunoblotting for the presence of p110/ABC50 and eIF2␣. natant (data not shown). This might be due to the necessarily high stringency of the immunoprecipitation reaction and wash buffers to prevent nonspecific binding of eIF2 and ABC50 to the immunoprecipitation matrix or due to masking of the epitope for the anti-eIF2␣ antibody when eIF2 is complexed with ABC50. The inability to demonstrate an in vivo interaction in this manner could reflect a transient interaction between the two proteins in the presence of other binding partners (e.g. the ribosomal subunits). The results of the gel filtration chromatography of the isolated eIF2/ABC50 protein samples were also consistent with an interaction between the two proteins in vitro because they co-eluted from a Sepharose 6 column at an apparent molecular mass of about 400 kDa. This may indicate a complex of two eIF2 heterotrimers to each ABC50 molecule or nonideal behavior of the complex. Our inability to isolate significant quantities of either eIF2 or ABC50 free from the other protein during purification from rabbit reticulocyte lysate prevented the determination of the retention times for the individual proteins. However, the presence of detergents in the column running buffer when used for gel filtration of our eIF2/ ABC50 samples caused a shift in the elution profile of each protein to positions expected from their individual molecular masses, suggesting that the two proteins are complexed with one another through hydrophobic interactions.
The possible functional effect of the eIF2/ABC50 interaction on the activity of eIF2 was studied by employing the ABC50deficient eIF2 sample obtained from HeLa cell lysate in GDP binding and ternary complex assays. The presence of ABC50 was found to enhance the binding of Met-tRNA to eIF2 but not the binding of GDP. This increased ternary complex formation could be due to the stabilization of a particular conformation of eIF2, which has higher affinity for Met-tRNA, as a consequence of its interaction with ABC50. This suggests a potential in vivo function for ABC50 during the initiation of translation. Initial studies indicate that phosphorylation of eIF2␣ (by recombinant RNA-dependent protein kinase) is unaffected by the presence or absence of ABC50, 3 indicating that ABC50 is unlikely to have a role in modulating the phosphorylation reaction.
The sequence of ABC50 showed it to be a member of the ABC family of proteins. These are generally membrane-associated transporters dependent upon ATP hydrolysis for their activity (46). Two known exceptions to this generalization are the S. cerevisiae proteins eEF3 and GCN20, both of which have been identified as proteins involved in the process of translation (although their precise functions remain obscure). eEF3 is thought to monitor the fidelity of the translational elongation process in an ATP-dependent fashion (47,48), whereas GCN20, together with GCN1, is involved in the modulation of the GCN2 eIF2␣ kinase activity in response to amino acid starvation (17,49). eEF3 and GCN20 each associate with the translational machinery, and the similarity exhibited by ABC50 to these members of the eEF3 subfamily of the ABC transporters suggested that it was also likely to be involved in the process of translation. Here we have shown by means of sucrose cushion sedimentation and sucrose density gradient analyses that ABC50 also associates with the ribosome. This association was with the 40 and 60 S subunits and not with the polysomal material. This strongly indicates a role for ABC50 in translation that is probably at the stage of initiation, and it is interesting to note that in activated T-cells, the expression of ABC50 increases after a period of 8 -16 h that parallels the increased expression of other initiation factors, e.g. eIF2, as demonstrated here and elsewhere (34 -36, 38, 50).
The association of ABC50 with the ribosome was markedly altered by addition of adenine nucleotides to the gradients. ADP caused a decrease in the association and ATP an increase that may indicate a role for the nucleotide binding motifs in the ABC50 sequence in regulating its interaction with ribosomes. GCN20 has also been shown by Marton et al. (17) to interact with the ribosome, and this interaction was also stimulated by ATP. However, GCN20 associates with the polysomes and not with the 40 or 60 S subunits. Initial studies performed by us have shown that expression of ABC50 in a GCN20 deficient strain of S. cerevisiae does not complement the function of GCN20. 4 However, if ABC50 is the mammalian homologue of GCN20, this inability to complement the GCN20 strain probably reflects the lack of similarity between the N-terminal sequences of the two proteins; it has been shown that it is the N-terminal one-third of GCN20 that interacts with GCN1 and that this is the only region of the protein required for complementation of a GCN20 deletion strain of S. cerevisiae (17). Comparison of the ABC50 and GCN20 sequences shows that it is the N-terminal region that is the least similar between the two proteins (in fact, this region is poorly conserved among all the identified homologues of ABC50 from the various species), indicating evolutionary and probably functional divergence. The recent identification of mammalian homologues of both GCN2 and GCN1 (17,20) suggests that the equivalent (or at least, a similar) regulatory mechanism does occur in mammalian cells, in which case it seems likely that a GCN20-like protein also exists. The results discussed above suggest that ABC50 is probably not this protein. However, it is conceivable that, in the mammalian system, two or more GCN20-like proteins have evolved to perform variations of the function of the yeast GCN20 protein at different stages in translation. ABC50 may perform this function at the stage of initiation, possibly as a regulator of eIF2 activity, whereas another as yet unidentified mammalian GCN20 homologue may perform a similar function to that of the yeast GCN20, which acts on the elongating ribosomes (17). In conclusion, ABC50 has been identified as a third member of a subfamily of the ABC proteins that are not membrane transporters but are instead associated with the translational machinery and, in this case, with eIF2.