A Rat Brain Bicistronic Gene with an Internal Ribosome Entry Site Codes for a Phencyclidine-binding Protein with Cytotoxic Activity*

The cloning and characterization of the gene for the fourth subunit of a glutamate-binding protein complex in rat brain synaptic membranes are described. The cloned rat brain cDNA contained two open reading frames (ORFs) encoding 8.9- (PRO1) and 9.5-kDa (PRO2) proteins. The cDNA sequence matched contiguous genomic DNA sequences in rat chromosome 17. Both ORFs were expressed within the structure of a single brain mRNA and antibodies against unique sequences in PRO1- and PRO2-labeled brain neurons in situ, indicative of bicistronic gene expression. Dicistronic vectors in which ORF1 and ORF2 were substituted by either two different fluorescent proteins or two luciferases indicated concurrent, yet independent translation of the two ORFs. Transfection with noncapped mRNA led to cap-independent translation of only ORF2 through an internal ribosome entry sequence preceding ORF2. In vitro or cell expression of the cloned cDNA led to the formation of multimeric protein complexes containing both PRO1 and PRO2. These complexes had low affinity (+)-5-methyl-10,11-dihydro-5H-dibenzo[a,d]cyclohepten-5,10-imine (MK-801)-sensitive phencyclidine-binding sites. Overexpression of PRO1 and PRO2 in CHO cells, but not neuroblastoma cells, caused cell death within 24–48 h. The cytotoxicity was blocked by concurrent treatment with MK-801 or by two tetrahydroisoquinolines that bind to phencyclidine sites in neuronal membranes. Co-expression of two of the other subunits of the protein complex together with PRO1/PRO2 abrogated the cytotoxic effect without altering PRO1/PRO2 protein levels. Thus, this rare mammalian bicistronic gene coded for two tightly interacting brain proteins forming a low affinity phencyclidine-binding entity in a synaptic membrane complex.

binding proteins. But the gene for the fourth subunit has not yet been cloned.
The fourth protein of the complex was identified on SDS-PAGE as an ϳ40-kDa protein. To complete the characterization of this complex of proteins, the cDNA for the fourth subunit was cloned, and a corresponding genomic sequence in rat genome was identified. The presence of two open reading frames (ORFs) in the cloned cDNA, the expression of both ORFs in a single mRNA in brain, and the translation in brain of the two proteins coded by the cDNA, led to the investigation of the mechanism of translation of both ORFs. Translation of both ORFs through an internal ribosome entry sequence (IRES) was identified, as was the need for the co-expression of the two proteins to create a functional protein, a phencyclidine-binding protein.

EXPERIMENTAL PROCEDURES
Purification of the ϳ40-kDa Protein and Raising of Polyclonal Antibodies-Synaptic membranes were prepared from two rat brains (14,15), and membrane proteins were solubilized in 25 ml of Buffer A (1% CHAPS, 0.5% n-octylglucopyranoside, 10% glycerol, 10 mM potassium phosphate buffer, pH 7.4, plus protease inhibitors) (7) and subjected to centrifugation (100,000 ϫ g for 1 h). The supernatant was then loaded onto an L-glutamate ReactiGel (Pierce) column, and the protein complex was eluted with buffer containing 5 mM NMDA (16). Following SDS-PAGE and silver staining of a single gel lane (6,14,15), the ϳ40-kDa band was identified, and the unstained portion of the gel in the location of that band was cut. The protein was then electro-eluted in 2 mM EDTA, 40 mM Tris acetate buffer, pH 8.4, dialyzed against H 2 O, concentrated, and 100 -200 g mixed with either 0.5 ml of Freund's complete or 0.5 ml of Freund's incomplete adjuvant and used to immunize rabbits (17). Serum was collected, and antibodies were characterized (15)(16)(17).
Screening of Brain cDNA Libraries with Anti-40-kDa Protein Antiserum-A cDNA expression library from brain hippocampus in Zap was plated on Escherichia coli XL-1 Blue and screened using the antiserum to the ϳ40-kDa protein (1:1,000 dilution) as described (8 -10). Approximately 10 6 plaques were screened, and phages from eight positive plaques were re-screened until a homogeneous population of immunopositive recombinant phages was obtained. Recombinant DNA was subcloned into pBS phagemid (Stratagene).
cDNA Sequencing, PCR Amplification, and Rapid Amplification of cDNA Ends (RACE)-cDNA inserts of phage and plasmid clones were sequenced on both strands (Thermo Sequenase, Amersham Biosciences; fmol Promega). DNAstar and Vector Suite software were used to analyze the sequences as follows: BLAST to search the National Center for Biotechnology Information, GenBank TM , PSORT, TMbase, and SMART to determine various protein domains, organelle-targeting sequences, and protein family relationships. RNA structure 4.3 (18) was used to determine the tertiary structure and thermodynamic stability of mRNA structures.
PCR amplification of cDNA was performed (50 l of 100 ng of DNA template, 250 M of each dNTP, 0.2 M of each primer described under "Results and Discussion," 2.5 mM MgCl 2 , and 2 l of DNA polymerase mixture, 2 l of Taq, 5 units/l, and 1 l of Pfu DNA polymerase, 2.5 units/l). PCRs were 30 cycles of 94°C (45 s), 55°C (90 s), and 72°C (2 min). The DNA was purified from agarose gels and either labeled with [ 32 P]CTP or ligated into pCR-Script (Stratagene) or pGEM-T Easy (Promega). Each PCR product was sequenced on both strands. Marathon-Ready cDNAs (Clontech) were used in 5Ј-and 3Ј-RACE of cDNA. Reactions were 30 cycles at 94°C (30 s), 94°C (5 s), and 68°C (4 min), and products were cloned into pGEM-T Easy and sequenced.
Gel Electrophoresis, Northern Blot, and Reverse Transcriptase (RT)-PCR Analyses of mRNA-Gel electrophoresis of RNA synthesized by in vitro transcription of the cloned cDNA was performed by denaturing the RNA in 50% formamide, 6% formaldehyde and subjecting it to electrophoresis on 1.2% agarose gels containing 6% formaldehyde and either 40 mM Tris acetate, 1 mM EDTA (TAE), or 0.1 M MOPS, 40 mM sodium acetate, 5 mM EDTA, pH 7.
In Situ Hybridizations-A 1.1-kb amplified fragment used for Northern blots was subcloned into PCR-script vector and used for in vitro transcription reactions to generate cRNA probes for in situ hybridization (19). The cRNA probes were synthesized in the presence of 35 S-CTP and had a specific activity of 10 8 cpm/g RNA. In situ hybridization was performed on cryostat sections of rat brain (19).
Generation of Antibodies to Peptide Sequences and Immunocytochemistry-A peptide corresponding to amino acids 2-13 from the first putative ORF of the cloned cDNA (CEQSGGDALPTE) and one to amino acids 42-54 from the second (CRGYSIQHRRLVE) were synthesized, including a C-terminal cysteine (Quality Controlled Biochemicals). Purity was established by high performance liquid chromatography and mass spectrometry (Ͼ95%). Immunization with each peptide, determination of antibody selectivity, and immunocytochemical studies on serial brain sections (20 m) were performed (19,20). Alexa 458-conjugated goat anti-rabbit secondary antibody (Jackson ImmunoResearch, 1:200 dilution) was used.
Reporter Vector Construction-The following reporter vectors were constructed: (a) fusion proteins of enhanced green fluorescent protein (EGFP) with each of the two putative ORFs of the cloned cDNA; (b) dicistronic cDNA containing red fluorescence protein (RFP) in place of ORF1, the intervening sequence between ORF1 and ORF2 (inter-ORF), and EGFP in place of ORF2; (c) dicistronic cDNA containing firefly luciferase, inter-ORF, and Renilla luciferase; and (d) dicistronic cDNA containing firefly and Gaussia luciferase in place of ORF1 and ORF2, respectively.
Dicistronic reporter vectors containing RFP and EGFP in place of ORF1 and ORF2, respectively, including the inter-ORF sequence between RFP and EGFP, were generated by digesting the ORF of RFP (BglII/NotI), cloning it into phrGFP (Stratagene), then excising the RFP insert (BamHI), and cloning it in B2 vector containing inter-ORF and EGFP.
The dicistronic vector of firefly and Renilla luciferase separated by inter-ORF was constructed using pGL3-basic and pRL-null (Promega). The oligonucleotides 5Ј-AGCTTTTC-TTGGCATTCCGGTACTGTTGGTAAA-3Ј and 5Ј-GAT-CTTTACCAACAGTACCGGAATGCCAAGAAA-3Ј were annealed, and 10 ng of this double-stranded DNA was ligated to 400 ng of pGL3 pre-digested with BglI/HindIII (removal of SV40 promoter). To this vector, the sequence 5Ј-CTTT-ACCAACAGTACCGGAATGCCAAG-3Ј, predicted to form a thermodynamically stable stem-loop hairpin (hp) in mRNA when combined with the palindromic sequence 3Ј-GAAATGGTTGTCATGGCCTTACGGTTC-5Ј in pGL3 vector, was introduced 5 bases upstream of the initiation codon of firefly luciferase. Subsequently, the luciferase and hp-Luc sequences were cut from the respective vectors and inserted in B2 (NheI site). The new vectors, pBLucMG and phpBLucMG, containing firefly luciferase, inter-ORF, and EGFP sequences, without or with the hp, were digested (EcoRI/NotI), and the EGFP ORF was excised and Renilla luciferase ORF (from pRLnull, Promega) inserted, thus forming the dicistronic vectors pBLucMRL and phpBLucMRL.
A vector with firefly and Gaussia luciferase in place of ORF1 and ORF2, respectively, was constructed by cutting (HindIII/ XbaI) the SP6 promoter sequence from pSinRep5 vector (Invitrogen) and ligating it to vector pBLucMRL that was digested with HindIII/XbaI. The resulting vector, SBMRL, contained the SP6 promoter/firefly luciferase/inter-ORF/Renilla luciferase. The inter-ORF and Renilla luciferase were removed from SBMRL (BglII/StuI) and replaced with Gaussia luciferase ORF (BamHI/XbaI of vector pNEBR-X1Glu (New England Bio-labs)). The new vector, SBMRL-EBL, was digested (HindIII/ AfeI) to remove firefly luciferase DNA and replace it with the combined firefly luciferase and inter-ORF sequence from SBMRL (HindIII/SmaI). The new vector, SMMRL, contained the SP6 promoter/firefly/inter-ORF/Gaussia luciferase.
In Vitro Transcription and Translation of the Reporter Vectors and the Cloned cDNA-The cloned cDNA was transcribed using the T7 promoter (mMessage mMachine T7 kit, Ambion). For the in vitro transcription and translation of the cloned cDNA, T 3 or SP6 RNA polymerase and reticulocyte lysate (TNT-coupled, Promega) plus phage cDNA or PCR-amplified fragments of phage cDNA were used. Reactions (reticulocyte lysate, 1 g, and RNasin, 40 units/l, final volume of 25 l) were at 30°C (90 min). Synthesized proteins were used in ligand binding assays or were labeled with [ 35 S]methionine (1000 Ci/mmol) and analyzed by SDS-PAGE.
Noncapped mRNA containing firefly luciferase/inter-ORF/ Gaussia luciferase was synthesized using vector SMMRL (20 g) cut with SspI, purified, and introduced as template (5 g) in RiboMAX TM large scale RNA production system, SP6 (Promega). Synthesis of mRNA proceeded for 2 h in the absence of 7-methylguanosine nucleotide (2 h). The DNA was digested with DNase and the RNA purified.
For transfection of CHO cells with in vitro-synthesized RNA, the RNA (2 g) was mixed with Lipofectamine 2000 (1 g/2.5 l) in serum-free Dulbecco's modified Eagle's medium, treated with proteinase K (10 g/ml, 50°C, 1 h), heated at 95°C for 30 min, cooled to 37°C, and added to cell cultures (24-well plates). Following a 4-h incubation, the medium was changed to serumcontaining Dulbecco's modified Eagle's medium, and the cells were incubated for 24 h before measuring luciferase activities.
Luciferase activities in cell extracts were measured using the dual-luciferase reporter assay (Promega). Luminescence was measured three times for 0.5 s (LumiCount, Packard Instrument Co.). Enzyme activity was expressed as relative luminescence units. Promoterless SMMRL vector was used as a transfection control in the RNA transfection experiments. Each transfection experiment was performed at least twice using triplicate or quadruplicate transfections for each experiment.
SK-N-SH cells were cultured in 24-well plates pre-coated with poly-D-lysine and laminin (0.5 ml/well, 0.5-1.0 ϫ 10 5 cells/ well), grown for 5 days in serum-containing Ham's F-12 (50 -80% confluence), and then transfected as described above. The transfection mixture was removed and 0.5 ml of Ham's F-12 serum added to each well, and 24 h from the start of transfection, the culture medium was changed to serum-free Ham's F-12 (induction of differentiation). The cells were harvested 48 h later and lysed as above.
Immunohistochemistry and Immunoprecipitation (IP) of Protein Complexes-Immune labeling of SK-N-SH cells was performed as described (20). For double labeling with SYTOX Blue (Invitrogen) and antibody, SYTOX Blue was added after all antibody labeling reactions were completed. Excess SYTOX was removed by multiple washings with Tris borate-EDTA and PBS buffers, and the cells were examined by confocal microscopy (458 nm excitation for SYTOX; 568 nm for Alexa 568 dye-secondary antibody).
Ligand Binding and Cell Viability Assays-Binding of

RESULTS AND DISCUSSION
Screening of a Brain cDNA Library and Cloning of a Dicistronic cDNA-The antiserum raised against the ϳ40-kDa protein subunit of the purified complex had a titer of 1:2,000, reacted with as low a concentration of synaptic membrane proteins as 10 ng (data not shown), and labeled a 52-kDa band in synaptic membrane proteins (supplemental Fig. S1). When the antiserum was used to immunoprecipitate proteins from synaptic membranes, it precipitated an ϳ56-kDa protein (data not shown). Thus, the molecular size of the native protein was thought to be 52-56-kDa, with the ϳ40-kDa protein possibly being a degradation product.
Screening of a rat hippocampal cDNA expression library with the antiserum identified several phage clones. The largest insert was 1.3 kb, and this was subcloned into pBS phagemid. Further library screening using the 1.3-kb cDNA insert as probe did not lead to the identification of clones with larger inserts. In addition, performance of RACE, either in the 5Ј or 3Ј direction, did not yield DNA sequence extensions.
Sequencing of the cDNA on both strands in phage and pBS yielded identical sequences (GenBank TM accession number EF495200). There were five potential initiation ATG codons in the 5Ј-terminal region, four of which were followed by in-frame termination codons leading to only short putative ORFs (Fig.  1A). The fifth initiation codon was associated with a putative ORF (ORF1) 246 bp long. Following the termination codon of ORF1, there were eight potential initiation codons, seven of which were followed by termination codons. The eighth initiation codon was associated with an ORF (ORF2) 267 bp long (Fig. 1A). ORF2 started 410 bp downstream from the termination codon of ORF1 and was offset by a Ϫ1 frameshift.
Analysis of the structure of the 410-bp sequence between the two putative ORFs (18) predicted a structure with four Y-type stem-loops. The estimated Gibbs free energy of folding of the inter-ORF mRNA was Ϫ70 kcal mol Ϫ1 , indicative of stable mRNA structures that would inhibit ribosome scanning through this region (23). Thus, it was not certain that translation of ORF2 could occur in vivo.
ORF1 encoded a protein of 81 amino acids with a predicted nonglycosylated molecular size of 8.9 kDa (PRO1) and, if ORF2 were translated, it coded for a protein containing 88 amino acids with a size of 9.5 kDa (PRO2). BLAST analyses indicated that the sequences for PRO1 and PRO2 were unique. In PRO1, there was a region of eight consecutive Ser followed by three repeats of SIYL (Fig. 1A). BLAST analyses performed using only the sequences containing the eight Ser residues revealed several proteins in plants and animals that had long tracts of Ser residues. Included among them were a rat brain and immune system phosphatase, PHLPPL, a cell wall surface anchor family protein, and a retinoblastoma-binding protein 1-like protein.
BLAST analyses also revealed several proteins in invertebrates and vertebrates with multiple SIYL repeats (frequency of 4 -6 consecutive repeats). Included among them was a mouse cerebellar granule cell marker protein. None of these proteins were homologous to the predicted whole sequence of PRO1.
Hydropathy analyses of the inferred amino acid sequence of PRO1 indicated a possible signal peptide sequence in the N-terminal region, a single hydrophobic region forming a potential transmembrane domain (Fig. 1A), and a likely localization for PRO1 in membranes with the N-terminal in the extracellular environment. The sequence of PRO1 had one consensus glycosylation site within this extracellular domain (Fig. 1A). The deduced sequence of PRO2 did not contain consensus glycosylation sites, had a signal peptide between residues 1 and 24, two putative TM regions (Fig. 1A), and a likely topography in membranes with both N and C termini in the extracellular domain.
BLAST searches using the nucleotide sequence of the cloned cDNA against that of the rat genome revealed that the cDNA sequence between bases 10 and 995, i.e. the sequence covering all of ORF1 and most of ORF2, was 100% identical to the plus strand of the genomic sequence of Sprague-Dawley rat chromosome 17 (Fig. 1B), bases 28,879 -29,864 (RGSC, NCBI). The cDNA sequence between 995 and 1291, representing the 3Ј-terminal sequence of ORF2 and the 3Ј-UTR, was 99% identical to the sequence on the same strand of rat chromosome 17, but starting 8,437 bases downstream of the first sequence, bases 38,301-38,597 (Fig. 1B). The two genomic sequences covered 98% of the total length of the cloned cDNA. The location of these two sequences on chromosome 17 appeared to represent a gene with two exons separated by a putative 8.4-kb intron. These sequences have not been identified previously as part of a specific gene. The presence of identical sequences in the cDNA and genomic DNA of chromosome 17 indicated that the mul-tiple Ser and SIYL repeats in ORF1, and the inter-ORF and ORF2 sequences, were not the result of cloning artifacts.
A short section of the nucleotide sequence in cDNA, nucleotide 1007-1291, was 88 -94% identical to a 284-bp sequence in rat, mouse, and human histone genes. In all occasions, the matches were to the inverted sequence of the sense strand of histone genes. Alignment of the sense strand of rat histone 2A and 2B cDNA with the cloned cDNA indicated only 22% overall identity and a lack of any long sequences that were identical to each other. Nor was there any homology between the protein sequences of histones and either the predicted sequence of PRO1 or PRO2. A 1.1-kb fragment of the cDNA containing both ORF sequences was used as a probe for Northern blots and in situ hybridization studies. Hybridization of this probe with commercially available rat multiple tissue Northern blot (Clontech) labeled a 2.4-kb poly(A ϩ ) mRNA in heart, liver, kidney, brain, lung, and skeletal muscle ( Fig. 2A). Brain mRNA also contained an ϳ1.8-kb band that hybridized with the probe. Liver, heart, and testis contained mRNA species around 3.5 kb and smaller than 1 kb that were labeled by the probe (Fig. 2A). A second probe representing an ϳ500-bp sequence near the 5Ј end of the cloned cDNA, i.e. avoiding the region homologous to histone genes, also labeled the 2.4-kb band in the mRNA from various tissues and the ϳ1.8 kb in brain mRNA (data not shown).
The 2.4-kb mRNA labeled by the two probes was either an mRNA that did not correspond to the cloned cDNA or an abnormally migrating mRNA that corresponded to the cDNA. CHO cells that were not transfected with the cloned cDNA vector did not express any RNA that was labeled by the 1.3-kb probe (Fig. 2B). However, Northern blot analyses of total RNA from cells transfected with the cloned cDNA revealed two prominent bands, one of 2.5 and the other of 1.3 kb (Fig. 2B), suggestive of abnormal migration of the mRNA synthesized from the 1.3-kb cDNA.
To probe this further, mRNA was synthesized by in vitro transcription of the cloned cDNA and subjected to denaturing agarose gel electrophoresis in TAE buffer. Under these conditions, the mRNA had a prominent band between 1 and 1.3-kb and two weak, diffuse bands between 2 and 2.6 kb (data not shown). Denaturing electrophoresis of the same mRNA in a MOPS buffer system resolved the mRNA into a single species with an estimated size of 1.3 kb (Fig. 2C). The reasons for the abnormal migration of RNA species in different buffers are not known but may be indicative of the presence of different structures of RNA assumed under different conditions of treatment.
Evidence That Both ORF1 and ORF2 in the Cloned cDNA Are Transcribed and Translated in Brain-Sense and antisense cRNA probes synthesized using a 1.1-kb cDNA fragment were used for in situ hybridization. Highest levels of labeling by the antisense probe were in pyramidal neurons of the hippocampus and cerebral cortex, granule cell neurons of the dentate gyrus and cerebellum, and cells of the olfactory bulb and medial habenula (Fig. 3A). The sense cRNA probe did not hybridize with mRNA in brain (Fig. 3A). The pattern of labeling in brain was similar to that observed with cRNA probes for the other three subunits of the complex (24).
To determine whether both ORFs were present in brain mRNA, RT-PCR analyses were performed with sets of primers designed to amplify either the whole sequence of the cloned cDNA or that of only ORF1 or ORF2 (Fig. 3B). Amplification of phage cDNA using these primers led to the synthesis of amplicons with the predicted sizes (Fig. 3B). Reverse transcription of rat brain poly(A ϩ ) RNA and PCR amplification yielded amplicons that had identical sizes to those obtained with phage cDNA as the template (Fig. 3B). No amplification products were detected if RT was not included in the reaction. These results indicated that mRNA with the sequence of the 1.3-kb cDNA insert, including ORF1, ORF2, and the inter-ORF sequence, was present in brain as intact mRNA. Therefore, the translation of both ORFs in brain was not likely to be due to trans-splicing of the mRNA into a monocistronic RNA.
To probe for the expression of both PRO1 and PRO2 proteins in brain, antibodies were raised to unique immunogenic domains of the predicted sequence of each protein. The anti-PRO1 and anti-PRO2 antisera recognized their cognate peptides and did not react with the noncognate peptide (supplemental Fig. S2). Of the two antibodies, the anti-PRO2 was the more reactive against the respective cognate peptide. A mixture of the two antibodies at a 1:1,000 dilution each was used in immunoblot assays against whole cell particulate proteins from non-neuronal, nontransfected cells to test their specificity. Nontransfected CHO cells did not contain RNA corresponding to the cloned cDNA (Fig. 2B) and cell extracts from nontransfected CHO cells did not react with the anti-PRO1/anti-PRO2 antibodies (supplemental Fig. S3), indicative of appropriate antibody specificity.
Both anti-PRO1 and anti-PRO2 labeled most strongly an ϳ52-kDa protein in synaptic membranes (Fig. 4A). The anti-PRO1, however, labeled additional bands at 28, 41, 65, and 82 kDa, but less strongly than the 52-kDa protein. Both antibodies labeled an ϳ56-kDa protein band in the partially purified complex of four proteins from synaptic membranes (Fig. 4B). Finally, the purified IgG from each antiserum labeled PRO1-  like and PRO2-like proteins in brain hippocampus neurons (Fig. 4, C and D). Thus, both ORFs and the inter-ORF sequence were transcribed, and the mRNA for both ORFs was translated to the respective proteins in brain cells.
Estimates of Molecular Size of Brain PRO1 and PRO2-The molecular size of the labeled proteins in synaptic membranes and in the partially purified complex was larger than predicted on the basis of ORF1 or ORF2 sequences. CHO cells transfected with the cloned cDNA also expressed a 56-kDa membraneassociated protein recognized by the antibodies (Fig. 5A). The 52-56-kDa proteins might represent stable multimeric protein structures formed through hydrophobic interactions between two relatively hydrophobic small proteins, PRO1 and PRO2. To test this possibility, synaptic plasma membrane proteins were subjected to SDS-urea gel electrophoresis (25). The presence of 8 M urea caused some of the 52-56-kDa proteins in synaptic membranes to dissociate to lower molecular size species of 46, 27, and 10-kDa that were recognized by anti-PRO2 (Fig. 5B) and anti-PRO1 (data not shown). Other known membrane-associated proteins tend to form stable aggregates. For example, a protein homologous to the GBP subunit, the 34.6-kDa RECS1 protein, is relatively hydrophobic and forms multimeric complexes in SDS-PAGE (26).
The formation of aggregates composed of PRO1 and PRO2 was further explored by the following: (a) analyzing immunoprecipitated proteins from CHO cells transfected with the 1.3-kb cDNA; and (b) conducting in vitro transcription and translation using the cloned cDNA. Protein IP from extracts of transfected CHO cells using anti-PRO1 antibodies led to the isolation of an 52-kDa protein that was recognized by anti-PRO2; IP with anti-PRO2 yielded a 52-kDa protein that was labeled by anti-PRO1 (Fig. 5C). In IP and immunoblots, purified IgGs were used. Neither of the two proteins was precipitated by preimmune IgG (Fig. 5C), and no detectable proteins were precipitated from cells transfected with vector that did not contain the 1.3-kb cDNA insert (data not shown).
The labeling by anti-PRO1 of proteins precipitated by anti-PRO2 was weaker than the labeling by anti-PRO2 of proteins precipitated by anti-PRO1 (Fig. 5C). This might have been due to either lower expression of PRO1 than PRO2, less efficient IP by anti-PRO2 compared with anti-PRO1, or less efficient labeling by anti-PRO1 versus that of anti-PRO2 (see Fig. 3C). However, because different populations of cells were transfected with the cloned cDNA and used for the IP and immunoblot studies, the absolute numbers of transfected cells and of protein expression was not expected to be identical. Thus, comparisons of the levels of expression of the two proteins in transfected CHO cells could not be made. . Evidence for the expression of PRO1 and PRO2 in brain. A and B, immune labeling of synaptic membrane proteins and of the proteins in the partially purified complex with the anti-PRO1, anti-PRO2, and anti-40-kDa protein antibodies. Synaptic membrane proteins (40 g) and purified complex (5 g) were labeled by antisera to PRO1 or PRO2 or by purified IgG from the anti-40-kDa antiserum, as indicated. Proteins in the isolated fraction were also stained with silver nitrate (lane on the right). Antisera to GBP, Gly-BP, and CPP-BP proteins labeled the other prominent bands detected by silver staining (not shown). Arrows indicate the bands labeled and their corresponding molecular sizes. Arrow on the right, major protein labeled by all antibodies. C and D, expression of PRO1 and PRO2 in rat brain hippocampus as revealed by immune labeling of the CA1 region of the hippocampus with anti-PRO1 (C) and anti-PRO2 (D) IgG. Arrowheads, neuronal cell bodies; arrows, neuronal dendrites; scale bar, 100 m.
In a separate series of IP experiments, proteins extracted from CHO cells were immunopurified through anti-PRO1 antibody-derivatized matrices and resolved by SDS-urea PAGE. The 52-59-kDa proteins expressed in CHO cells were resolved under conditions of SDS-urea into lower molecular size species of 40 -44, 18, and 9-kDa (supplemental Fig. S4). All protein bands reacted with the anti-PRO1/PRO2 antibodies. The effect of urea on the gel migration of these proteins suggested, once again, that the 52-59-kDa proteins were multimers of PRO1 and PRO2 resulting from hydrophobic interactions.
The studies described above were indicative of the formation of heteromeric complexes of PRO1 and PRO2. To explore whether in vitro synthesized proteins had a tendency to form homomeric or heteromeric complexes of PRO1 and PRO2, three different constructs based on the cloned cDNA were used for in vitro transcription and translation in the presence of [ 35 S]methionine. The DNA used in these reactions was generated by PCR amplification and contained only ORF1, only ORF2, or both ORF1 and ORF2 sequences downstream of the promoter (supplemental Table S2). SDS-PAGE of the proteins synthesized revealed an ϳ10-kDa labeled protein species for each condition. Yet, together with the 10-kDa protein, two other proteins with estimated sizes of 27 and 47 kDa were also formed under all conditions of in vitro transcription and translation (Fig. 6A). The larger molecular size proteins were less abundant than the 10-kDa protein and only became apparent after a prolonged exposure of the film to the radioactive gel (Fig.  6A). Thus, PRO1 and PRO2 exhibited a high tendency to form stable homo-or hetero-multimer complexes.
When ORF1 or ORF2 was expressed separately in CHO cells as fusion proteins with EGFP (sequence of ORF1 or ORF2 cloned upstream of and in-frame with the sequence of EGFP), they each formed proteins with the expected molecular size, and there was no evidence of multimer formation (Fig. 6B). The expected size for each fusion protein was as follows: PRO1-EGFP ϭ 35.8 kDa (8.9 PRO1 nonglycosylated ϩ26.9 EGFP); PRO2-EGFP ϭ 36.4 kDa; (9.5 PRO2 ϩ 26.9 EGFP). More than 50% of cells transfected with PRO1-EGFP or PRO2-EGFP expressed the respective fusion protein as indicated by the presence of green fluorescence in the cells.
When PRO1 was expressed as a non-fusion protein together with PRO2 as a fusion protein (PRO2-EGFP), multimers were once again present in the extracts, and the multimers contained the fusion protein as determined by labeling with anti-GFP antibodies. Two bands that were reactive with anti-GFP were observed, one at about 34 kDa, i.e. the expected size of the fusion protein, and the other at 80 kDa (data not shown). The EGFP-containing 80-kDa protein was not observed when PRO2-EGFP was expressed by itself. The 80-kDa protein was also labeled by PRO1 antibodies (data not shown), indicative of the formation of multimeric complexes between PRO1 and the PRO2-EGFP fusion protein. Thus, if one of the two proteins was expressed as a non-fusion protein, it tended to form multimeric complexes with the other, even when the second protein was expressed as a fusion protein.
Evidence for Independent Translation of the Two ORFs-To determine whether any two ORFs separated by the inter-ORF sequence of the cloned cDNA could be translated at the same time, the sequences of ORF1 and ORF2 were removed from the cloned cDNA and replaced with those of RFP and EGFP, respectively, while leaving the 410-bp inter-ORF intact. Transfection of either CHO cells (Fig. 7, A-C) or of a neuronal cell line, SK-N-SH cells (Fig. 7, D-F), with this dicistronic vector led to the expression of both red and green fluorescent proteins in the same cells. The SK-N-SH cells were transferred to a serumfree medium after 24 h to induce differentiation into neuronlike cells. Nontransfected CHO cells exhibited weak background green fluorescence and essentially undetectable red fluorescence (supplemental Fig. S5). SK-N-SH cells did not have detectable background green or red fluorescence (data not shown). Despite a higher background of green fluorescence in nontransfected CHO cells, the transfected cells were easily identified by the presence of bright green and red fluorescence (Fig. 7, A-C). Both types of fluorescence were observed in the same cells. The same was true for transfected SK-N-SH cells but expression of the two fluorescent proteins in SK-N-SH was, generally, at a lower level than in CHO cells (Fig. 7, D-F). The expression of RFP and EGFP in the same cells was an indication of the translation of both ORFs in the same cell and over the same period of time, and of the fact that any two ORFs could replace those in the cloned cDNA.  2 and 4). Protein loaded per lane was 10 g, and antibody dilution for labeling was 1:1,000. Arrow, estimated molecular size of major labeled band.
To explore whether translation of ORF2 was dependent upon translation of ORF1, a dicistronic vector was constructed in which ORF1 and ORF2 were replaced by firefly and Renilla luciferase, respectively. The two luciferase ORFs were sepa-rated by the 410-base inter-ORF sequence. In one version of this vector construct, a stem-loop hairpin with an estimated free energy of folding equal to Ϫ51 kcal mol Ϫ1 was inserted immediately preceding the initiation codon of firefly luciferase. The DNA for the vector with the hairpin loop had the identical size on gel electrophoresis as that for the vector without the hairpin loop. The presence of the stable structure of the hairpin stem-loop in the mRNA impedes ribosomal scanning and mRNA translation (23). Thus, it was expected that firefly luciferase expression would be reduced when the hairpin stem-loop was present. If translation of ORF2 were dependent on that of ORF1, Renilla luciferase expression would also be reduced. In the presence of the hairpin structure, firefly luciferase activity in transfected CHO cells was significantly reduced by 75% of that in its absence (13,898 Ϯ 1034 relative luminescence units with the hairpin versus 54,617 Ϯ 6,067 without the hairpin; units Ϯ S.E., n ϭ 3, p ϭ 0.03, two-tailed t test). On the other hand Renilla luciferase activity in the presence of the hairpin structure increased to 176% of that in its absence (1,724 Ϯ 53.9 relative luminescence units with the hairpin versus 979 Ϯ 102.7 without the hairpin). The increase in ORF2 expression was probably the result of reduced competition between ORF1 and ORF2 for translation by ribosomes. The key conclusion was that the two ORFs were translated concurrently, yet independently.
The continued expression of ORF2, even when ORF1 expression was substantially inhibited, was suggestive of a possible engagement of ribosomes by the inter-ORF RNA sequence thus leading to independent translation of ORF2, i.e. a possible IRES within the inter-ORF sequence. To exclude other potential alternatives, such as the presence of a cryptic promoter within the inter-ORF sequence or a cap-dependent continuation of translation by either frame-shifting (27,28), ribosome hopping (29), or leaky ribosome scanning (30), we synthesized in vitro noncapped mRNA that contained two luciferase ORFs, firefly and Gaussia luciferase, separated by the inter-ORF sequence. If transfection of CHO cells with in vitro transcribed mRNA led to the synthesis of Gaussia luciferase, then that would negate the possible presence of a cryptic DNA promoter. And, if following transfection with noncapped RNA, ORF2 Gaussia luciferase was predominantly or exclusively translated in comparison with firefly luciferase, then that would suggest that ORF2 translation was cap-independent and ORF1 was cap-dependent. The activity of firefly luciferase in RNA-transfected CHO cells was barely detectable above that in extracts from cells transfected with promoterless DNA vector used as control (relative luminescence units in RNA-transfected cells ϭ 212.1 Ϯ 3.4, n ϭ 18, and in promoterless DNA-transfected cells ϭ 170.8 Ϯ 18, n ϭ 12). On the other hand, the activity of Gaussia luciferase was very high in cells transfected with the noncapped mRNA and was much higher than that in lysates from cells transfected with the control promoterless DNA vector (relative luminescence units in RNA-transfected cells ϭ 5910.1 Ϯ 316.8, n ϭ 18, and in promoterless DNA-transfected cells ϭ 279.4 Ϯ 22, n ϭ 12; p Ͻ 0.0005, degrees of freedom ϭ 28). Based on these results, translation of ORF1 in the cloned cDNA appeared to be cap-dependent, whereas that of ORF2 was cap-independent and was not due to the presence of a cryptic promoter. Translation of ORF2  Table S2. Upstream primers for ORF1 and the combined ORF1/ORF2 contained the T 3 promoter; the ORF2 primer had the SP6 promoter. The proteins synthesized were subjected to SDS-PAGE (12% gel) and fluorography. Arrow, major radioactive protein bands. B, immunoblots of fusion proteins PRO1-EGFP and PRO2-EGFP expressed separately in transfected CHO cells (see text). Proteins in the particulate fraction were labeled by anti-PRO1, anti-PRO2, and anti-40-kDa protein antibodies. Arrows, major band labeled by all antibodies. had the characteristics of an IREScontrolled translation process.

PRO1-PRO2 Co-expression Needed for Formation of a TCP-binding
Protein-When either PRO1 or PRO2 was expressed singly in a cellfree transcription/translation system, there was no detectable ligand binding activity for L-  9A). The survival rate for cells transfected with vector lacking the TCP-BP insert (empty vector) was 79.8 Ϯ 2.7% (ϮS.E., n ϭ 5) that seen in cells that were never subjected to the transfection procedure, i.e. a 20% death rate among cells subjected to just the   A series of 1-aryl-1,2,3,4-tetrahydroisoquinoline derivatives compete with [ 3 H]MK-801 for binding to neuronal membrane NMDA receptor sites (33,34). These compounds were tested as possible protectants against TCP-BP transfection ( Table 1). All  compounds were introduced into the culture medium at 175 M concentration. In this study, transfection by TCP-BP reduced survival to 47.6% of that seen in cells transfected with empty vector. Only tetrahydroisoquinoline compounds that produced survival rates greater than 53% were considered as active. Two compounds were protective based on this criterion as follows: (S)-1-methyl-1-phenyl-1,2,3,4-tetrahydroisoquinolinium chloride ((S)-4) and the racemic mixture of 8-methyl-1-(2-methylphenyl)-1,2,3,4-tetrahydroisoquinolinium chloride. These two compounds have the highest affinity for binding to [ 3 H]MK-801 recognition sites in neuronal membranes (Table  1). (S)-4 was as effective as MK-801 in protecting CHO cells transfected with the TCP-BP (Fig. 9A). The half-maximal concentration of (S)-4 needed for protection was 118.7 M, i.e. very similar to that of MK-801 (134 M).
In neurons in vivo, expression of the endogenous TCP-BP does not appear to lead to cell death as indicated by the labeling of apparently intact brain neurons by anti-PRO1 and anti-PRO2 (e.g. Fig. 4). A possible explanation for differential effects of overexpression of TCP-BP in CHO cells as compared with normal expression of this protein in neurons might be that the co-expression of the other proteins of the neuronal membrane complex, e.g. GBP or Gly-BP, controls or abrogates the cytotoxic effects of excess TCP-BP expression. To test this possibility, GBP and Gly-BP were co-transfected with TCP-BP. Coexpression of either one of these two proteins completely blocked the cell death produced by the expression of TCP-BP alone (Fig. 9B). The co-expression of either GBP or Gly-BP together with TCP-BP did not affect the protein levels of TCP-BP in cells (e.g. supplemental Fig. S6). The co-expression of an unrelated protein, such as EGFP, had no effect on the cell toxicity of TCP-BP overexpression (data not shown).
To determine whether the cytotoxic effect of overexpression of TCP-BP observed in CHO cells may also be seen in neuronal cell lines, TCP-BP was expressed in the SK-N-SH neuronal cell line. SK-N-SH cells had low endogenous levels of TCP-BP and upon transfection with the TCP-BP gene, the levels were significantly increased (supplemental Fig. S7, A-D). In nontransfected SK-N-SH cells maintained in serum-free medium, the levels of proteins labeled by a mixture of anti-PRO1 and anti-PRO2 were equal to 6.9 Ϯ 0.1 pixel density units (n ϭ 137 cells), whereas in cells transfected with TCP-BP, the levels were significantly higher (12.0 Ϯ 0.2 pixel density units; n ϭ 146 cells; p ϭ 0.006). Immunoblot analyses confirmed the immunocytochemical results (data not shown).
Human neuroblastoma cells, i.e. SK-N-SH cells, were likely to have endogenous GBP or Gly-BP as the genes for these proteins (GRINA and KIAA0562) have been identified in the human genome and human neuronal tissues. Endogenous expression of GBP and Gly-BP in SK-N-SH cells was confirmed by immunoblot assays (not shown) and immunocytochemistry in both nontransfected cells (not shown) and cells transfected with only TCP-BP (supplemental Fig. S7, E and F). GBP immune reactivity was always higher than that of Gly-BP.
Overexpression of TCP-BP in SK-N-SH cells had no statistically significant effect on cell viability as TCP-BP transfected cells exhibited 75.5 Ϯ 6.0% survival compared with untreated cells, although cells transfected with empty vector or treated only with Lipofectamine had 80.0 Ϯ 9.3% survival compared with untreated cells (n ϭ 3). Thus, survival after TCP-BP transfection was at 94.4% of cells transfected with empty vector. The transfection efficiency of SK-N-SH cells was estimated by the methods described above and was 25.4% of all cells (n ϭ 6 plates). If the efficiency of transfection of SK-N-SH cells with the TCP-BP gene is taken into account, maximal cytotoxicity would be 22% of all SK-N-SH cells. This level of cell death is considerably lower than the nearly 100% cell death estimated for CHO cells after correction for the efficiency of transfection of those cells. Whether the reduced cytotoxicity in SK-N-SH cells was because of the endogenous expression of GBP and Gly-BP, or was because of other factors, has not yet been determined.
Conclusions-The cDNA for the fourth subunit of a previously characterized protein complex in brain synaptic membranes was identified as a 1.3-kb DNA that contained two ORFs. The corresponding mRNA was present in brain cells, as were the two proteins synthesized from the RNA. The proteins synthesized by translating this gene had ligand-binding sites for the ion-channel inhibitors of NMDA receptors, TCP and MK-801, hence the gene was named the TCP-BP gene. The cloned cDNA was nearly 100% identical to two sequences on the same strand of genomic DNA of rat chromosome 17, sequences that probably represent two exons of this gene separated by an 8.4-kb intron.
The presence of the complete sequence of the cDNA in rat genomic DNA indicated that the TCP-BP mRNA was a eukaryotic bicistronic mRNA. Bicistronic mRNAs in mammalian brain are rare but do exist (35); however, there is no demonstration that both ORFs of previously identified bicistronic brain mRNA are coordinately translated.
This study shows the coordinated, yet independent translation of the two ORFs. The mechanism of such translation was not because of trans-splicing of the mRNA into a monocistronic mRNA. The presence of a long intervening sequence with multiple termination codons between ORFs also negated the possibility that translation of ORF2 started from an initiation codon in close proximity to the termination codon of ORF1. Re-initiation of translation at the start of ORF2 through the continuous engagement of the ribosome was also unlikely as shown by transfecting CHO cells with in vitro synthesized, noncapped mRNA. There was no translation of ORF1 following transfection of CHO cells with in vitro synthesized, non-capped mRNA, while translation of ORF2 in the same mRNA proceeded normally. The results of transfection with synthesized mRNA also negated the possibility that there was a cryptic promoter in the inter-ORF DNA sequence. The mechanism favored to explain the translation of both ORF1 and ORF2 in TCP-BP was therefore that of re-initiation of translation by an IRES element (36,37). The 410-bp inter-ORF sequence has the characteristics of IRES elements, i.e. length of ϳ400 -600 nucleotides and low G and C content (27,28).
The cloned 1.3-kb cDNA is therefore an example of a rare mammalian bicistronic gene that allows for coordinated expression in time and space of two small proteins. The need for coordinated expression of the two proteins might be linked to the fact that both PRO1 and PRO2 were needed to form a phencyclidine and MK-801-binding protein. The two proteins, when co-expressed, formed stable oligomers composed of PRO1 and PRO2. The size of the most abundant species of the oligomeric protein, 52-59 kDa, was suggestive of a hexamer made of the 8.9 and 9.5 proteins but with an undetermined stoichiometry of the two proteins. This study also provided preliminary evidence for the fact that the oligomeric TCP-BP was part of the glutamate/glycine-binding protein complex in neuronal membranes and that the TCP-BP interactions with either the GBP or Gly-BP subunits of the complex altered its cellular function. The latter was demonstrated as the elimination of cytotoxicity by TCP-BP when either GBP or Gly-BP was coexpressed in CHO cells. The mechanism for cytotoxicity by TCP-BP in cells is currently under investigation.