γ-Glutamyltransferase and Its Isoform Mediate an Endoplasmic Reticulum Stress Response*

Although use of multiple alternative first exons generates unique noncoding 5′-ends for γ-glutamyltransferase (GGT) cDNAs in several species, we show here that alternative splicing events also alter coding exons in mouse GGT to produce at least four protein isoforms. GGTΔ1 introduces CAG four bases upstream of the primary ATG codon and encodes an active GGT heterodimeric ectoenzyme identical to constitutive GGT cDNA but translational efficiency is reduced 2-fold. GGTΔ2–5 deletes the last eight nucleotides of exon 2 through most of exon 5 in-frame, selectively eliminating residues 96–231 from the amphipathic N-terminal subunit, including four N-glycan consensus sites, while leaving the C-terminal hydrophilic subunit intact. GGTΔ7 introduces 22 bases from intron 7 causing a frameshift and a premature stop codon so a truncated polypeptide is encoded terminating with 14 novel residues but retaining the first 339 residues of the native GGT protein. GGTΔ8–9 deletes the terminal four nucleotides of exon 8 plus all of exon 9 and inserts 24 bases from intron 9 in-frame so the C-terminal subunit of the encoded polypeptide loses residues 401–444 but gains eight internal hydrophobic residues. In contrast to the product of GGTΔ1, those derived from GGTΔ2–5, Δ7, Δ8–9 all lack transferase activity and persist as single-chain glycoproteins retained largely in the endoplasmic reticulum as determined by immunofluorescence microscopy and constitutive endoglycosidase H sensitivity in metabolically labeled cells. The developmental-stage plus tissue-specific regulation of the alternative splicing events at GGTΔ7 and GGTΔ8–9 implies unique roles for these GGT protein isoforms. The ability of the GGTΔ1 and GGTΔ7 to mediate the induction of C/EBP homologous protein-10, CHOP-10, and immunoglobulin heavy chain binding protein, BiP, implicates a specific role for these two GGT protein isoforms in the endoplasmic reticulum stress response.

The entire intron/exon structure of the mouse ␥-glutamyltransferase gene (EC 2.3.2.2, GGT) 1 has been defined. This single copy gene is regulated by multiple alternative promoters that are coupled with alternative splicing mechanisms to generate several GGT cDNAs each with a unique 5Ј-noncoding region. But all encode the same protein, because coding exons appear to be spliced only in a constitutive fashion (1). The protein product is synthesized as a single-chain N-glycosylated propeptide, processed into an N-terminal amphipathic subunit and a smaller C-terminal subunit, and expressed on the cell surface where it functions as a key enzyme in glutathione metabolism (2).
Several human GGT cDNAs also exhibit unique 5Ј-ends and encode a protein that is processed in a similar fashion and shares 79% amino acid identity with that of the mouse (3). In addition, an alternatively processed human GGT cDNA has been described that contains an insertion of 22 bases within the coding domain. The extra nucleotide bases introduce a frameshift and a premature stop codon so that the predicted polypeptide would be a truncated GGT isoform. The protein product has never been characterized, but the elimination of the small subunit suggests that it would lack ␥-glutamyltransferase activity. The identification of this alternative transcript implies that human GGT gene expression may be more complex than that of other species (4).
However, while characterizing the site of a point mutation in the GGT enu1 mouse (5), we identified four previously unknown alternative splicing events involving coding exons in the normal mouse GGT gene (see Fig. 1). We studied these events to determine whether GGT protein isoforms are being generated through alternative processing of mouse GGT mRNA, to identify if these events are shared between mouse and human GGT, and finally to explore a potential role for these new mouse GGT protein isoforms in glutathione metabolism.

Source of Probes and Tissues-Tri-Reagent (Molecular Research
Center, Inc., Cincinnati, OH) was used according to the manufacturer's protocol to isolate total RNA from tissue or cells. Mouse genomic DNA was isolated from the lung. Human genomic DNA was provided by Dr. Qiang Yu from the Pulmonary Center of Boston University. Electrophoresis grade agarose was from International Biotechnologies, Inc. (New Haven, CT), and DNA standards were from Life Technologies, Inc. Materials for protein electrophoresis were from Bio-Rad (Richmond, CA), and protein standards were High-Range molecular weight markers from Amersham Pharmacia Biotech (Piscataway, NJ). X-OMAT and BioMax MR films were used for radiography and obtained from Eastman Kodak Co. (Rochester, NY). [␣-32 P]CTP, specific activity 800 Ci/mmol, was obtained from ICN (Irvine, CA) and [ 35 S]Met/Cys, specific activity ϳ1000 Ci/mmol, was from PerkinElmer Life Sciences (Wilmington, DE) as Easy Tag Express-[ 35 S]Protein Labeling Mixture. Primers were synthesized at the DNA/Protein Core Molecular Biology Unit of Boston University School of Medicine. The cDNA probes for CHOP and BiP were generated by PCR after selecting primers based on the published sequences in the GenBank. The probe for ␤-actin has been used routinely in this laboratory. The cell lines hCAR (human coxsackie and adenovirus cell surface receptor) and hMUC (human mucin 1) were available in the laboratory of Dr. Rebecca Hughey (12).
Genomic PCR, RT-PCR, Subcloning, and Sequencing-Total RNA from various tissues was used for RT-PCR as described previously (5). PCR primers were selected for primary and secondary PCR reactions, and each PCR reaction was performed for 20 cycles on a MJ Research thermal cycler. PCR products were analyzed by agarose gel electrophoresis, then eluted, cloned into an Invitrogen TA vector (San Diego, CA), and sequenced for verification at the DNA-Protein Core facility at Boston University School of Medicine. RT-PCR primers used for constitutive GGT cDNA amplification were M72, 5Ј-CCTTTCGGTTTGCC-TATGCCAAGAGGAC (upstream); 11M, 5Ј-GCGCTCCCTCTGTCCCA-CCCA (nested downstream); and 12M, 5Ј-GGCTTCCCGCAGCTTGGC-GGTGG (primary downstream). For intron 7 and intron 9 insertions, upstream hemi-nested primers were 22U, 5Ј-GCCAGCTCTGGGGTCT-CGGCAG and 24D, 5Ј-CCTGTCTCCTCCTATGGATCATAG, respectively. The primary primers used for genomic PCR and cloning of intron 7 were M71, 5Ј-AGGCACTGACGTATCACCGTATCGTG and 10.1M, 5Ј-CCTCCATCATCCTGAAGGTAGA. Primers M72 plus 10.1M were used in secondary PCR. Intron 9 was cloned using M811, 5Ј-ACCGCTCAC-CTGTCTGCGGTTTC as the upstream primer for both PCR reactions together with downstream primers 12M and 11M for primary and secondary PCR, respectively. The relative mRNA abundance of GGT⌬7 versus constitutive GGT was determining by using primers 71M plus 11M for the primary reaction, then 72M plus 810M, 5Ј-GAAACCGCA-GACAGGTGAGCGGTGCCTCC, for the secondary reaction.
Transcription/Translation in Vitro-TnT T7 Quick-Coupled Transcription/Translation System, a rabbit reticulocyte lysate system, was purchased from Promega and used according to the manufacturer's instructions. Mouse GGT constructs were generated by PCR using upstream primers with (CGGACCGGGCCCTACTGGAAGCAGACCATG-AAGA) and without (CGGACCGGGCCCTACTGGAAGACCATGAAGA-ATC) the CAG insertion 5 bases upstream of the ATG initiation codon along with a common 3Ј primer (CCGGAATTCCCGCTGAGTGGGGC-ACTGGGCACG). Template cDNA was reverse-transcribed from normal mouse kidney mRNA. The 5Ј primers contained an EcoRI and the 3Ј primer an ApaI site that were used to clone the PCR products into pCR 2.1 (Invitrogen). Plasmids were sequenced to ensure accuracy, and the T7 promoter was used for in vitro transcription. The translated products were labeled with [ 35 S]methionine, separated on a 12% polyacrylamide gel, and visualized by autoradiography. The primary translation product was predicted to have a molecular mass of 35.2 kDa with the terminal 47 amino acid residues encoded from the vector. A secondary translation product of 27 kDa was predicted from utilization of an internal translation initiation site.
Analysis of RNA-Total RNA was analyzed as previously described (5). RNA obtained from cell lines was quantitated by spectrophotometry and electrophoresed on a 1.0% agarose gel with 2.2 M formaldehyde in 1ϫ MOPS, transferred to a HyBond membrane (Amersham Pharmacia Biotech, Arlington Heights, IL) overnight, both at RT, then cross-linked with a Stratagene UV cross-linker. The membrane was prehybridized with QuickHyb (Stratagene) at 68°C for 15 min. Radiolabeled probe was then added and incubated for 2 h. The filter was washed twice at room temperature with 2ϫ SSC plus 0.1% SDS, washed twice more with 1ϫ SSC in 0.1% SDS at 60°C, and dried. An exposure of the filter was made on Kodak X-OMAT film, and the film was developed.
Stable Expression of mGGT Isoforms in CHO Cells-A full-length mouse GGT cDNA (type 3) was kindly provided by Drs. Z.-Z. Shi and M. W. Lieberman (Baylor College of Medicine, Houston, TX). This was cloned into the expression vector pCDNA3.1 (Invitrogen). To express the alternatively spliced GGT cDNAs, this wild type GGT cDNA sequence was replaced with that of GGT⌬1, ⌬2-5, ⌬7 and ⌬8 -9. All plasmids were sequenced to ensure accuracy.
Plasmids encoding the mGGT isoforms for cell transfections were prepared with the JETstar 2.0 plasmid purification kit (Genomed, PGC Scientifics, Frederick, MD). Stable clonal cell lines were obtained by transfection of Chinese hamster ovary cells (CHO) using Lipo-fectAMINE (1:3 ratio of DNA:lipid; Life Technologies Inc., Gaithersburg, MD) followed by selection in media containing G418 (0.5 mg/ml).
Transient Expression of mGGT Isoforms in CHO Cells-Transient expression of the mGGT isoforms was obtained in CHO cells using a cowpox/bacteriophage T7 (vT7CP) expression system as described previously (6,7). Confluent cultures of CHO cells were grown in 35-mm plastic dishes with a 1:1 mixture of DMEM and Ham's F-12 (1:1) supplemented with 3% FBS (normal culture media). Cells were washed with 1 ml of serum-free medium and infected with vT7CP (multiplicity of infection ϳ30) in 0.3 ml of the same media for 30 min. The media containing vT7CP was removed from the cells before transfection with a mixture of plasmid DNA (pCDNA3.1 with a T7 promoter) and Lipo-fectAMINE (Life Technologies) at a 1:3 ratio in 1 ml of the same serum-free media for 3 h. The lipid and DNA mixture was them removed, and the cells were washed with 1 ml of DMEM media lacking methionine (Met) and cysteine (Cys) (ICN, Costa Mesa, CA) and returned to culture for 15 min in the same media prior to metabolic labeling.
Metabolic Labeling of mGGT Isoforms in CHO Cells-Either stably transfected clonal cell lines expressing the mGGT isoforms or CHO cells transiently expressing the mGGT isoforms (described above) were starved for Met and Cys for 15 min in 1 ml of DMEM media lacking Met and Cys before addition of 50 -100 Ci of [ 35 S]Met/Cys for the times indicated in each experiment. Cells were chased in normal culture medium as indicated. Surface levels of the mGGT isoforms were determined by moving the dishes of cells to ice for biotinylation (8) as previously described (12). Briefly, cells were washed four times with 1 ml of PBS ϩϩ (137 mM NaCl, 2.6 mM KCl, 15.2 mM Na 2 HPO 4 , 1.47 mM KH 2 PO 4 , 0.5 mM MgCl 2 and 0.7 mM CaCl 2 ) then incubated with 0.5 mg/ml sulfo-NHS-SS-biotin (Pierce, Rockford, IL) in triethanolaminebuffered saline (10 mM triethanolamine, pH 7.6, 137 mM NaCl, and 1 mM CaCl 2 ) for 10 min. The reaction was quenched by three washes of the cells with normal culture media. Cells were solubilized as described above, and the supernatants were rotated end over end overnight at 4°C after addition of protein G immobilized on Sepharose 4B (Sigma Chemical Co., St. Louis, MO) and 1 l of goat anti-rat GGT antisera. Immunoprecipitates were recovered by brief centrifugation and washed once with 0.5 ml each of 1% Triton X-100 (Roche Molecular Biochemicals Corp., Indianapolis, IN) in HBS (10 mM HEPES-NaOH, pH 7.4, 150 mM NaCl), 0.01% SDS in HBS and HBS alone. Biotinylated mGGT was recovered by eluting the immunoprecipitates for 2 min at 90°C in 80 l of 1% SDS in HBS and further incubation with 30 l of ImmunoPure Immobilized Avidin (Pierce, Rockford, IL) after addition of 0.8 ml of HBS. After overnight rotation at 4°C, the avidin-conjugated beads were washed with 1 ml each of 1% Triton X-100 in HBS, 0.01% SDS in HBS, and HBS alone. The biotinylated mGGT isoforms were eluted by heating for 3.5 min at 90°C in 50 l of Laemmli SDS-sample buffer containing fresh 0.14 M ␤-mercaptoethanol. Samples were subjected to SDS-PAGE on 3-15% polyacrylamide gradient gels, and radioactive protein bands imaged and quantitated from the dried gel using a PhosphorImager (Bio-Rad, Richmond, CA).
Endoglycosidase H Treatment of Immunoprecipitates-Radioactive immunoprecipitates were resuspended in 80 l of 10 mM citrate buffer, pH 5.0, and 0.5% SDS by heating beads at 90°C for 2 min (9). The eluted sample was divided in half and incubated overnight at 37°C with or without 1 milliunit (mU) of endoglycosidase H (endo H, Roche Molecular Biochemicals) in the presence of protease inhibitors (Protease Inhibitor Mixture Set III, Calbiochem, La Jolla, CA).
Immunofluorescence Microscopy-COS cells were grown on glass coverslips in 35-mm dishes, transiently transfected with the cDNAs for the mGGT isoforms using LipofectAMINE, and analyzed by immunofluorescence 2 days later. All steps were carried out at room temperature. Cells were washed once with PBS ϩϩ for 5 min prior to fixation with 3% paraformaldehyde in PBS ϩϩ for 12 min. After washing with 10 mM glycine in PBS ϩϩ (PBS-Gly), cells were permeabilized by incubation for 4 min in 0.5% Triton X-100 in PBS-Gly and blocked with 5% goat serum in PBS-Gly for 5 min. Cells were incubated with rabbit anti-rat GGT antisera (diluted 1/600) for 45 min and then CY3-conjugated goat anti-rabbit IgG (diluted 1/2500; Jackson ImmunoResearch Laboratories, Inc., West Grove, PA) for 30 min, both in 1% goat serum in PBS-Gly. Cells were incubated for 5 min with 0.5 g/ml 4Ј-6-diamidino-2-phenylindole (Molecular Probes, Eugene, OR) in PBS-Gly to stain nuclei and then washed twice for 5 min with PBS-Gly before mounting on slides with 1 M n-propylgallate in glycerol and viewing with a Nikon Optiphot-2 microscope. Calnexin was stained as a positive control for localization in the endoplasmic reticulum (ER) using an antibody directed against the cytosolic tail (StressGen Biotechnologies Corp., Victoria, Canada).
Enzyme Activity-Specific GGT enzyme activity was measured at room temperature using the substrates ␥-glutamyl-para-nitroanilide and glycylglycine as previously described (10). Protein was determined by the method of Lowry.

RESULTS
Mouse GGT⌬1-The majority of our PCR-derived GGT clones generated from lung and kidney RNA of wild type and GGT enu1 mice contained the trinucleotide insert CAG five bases upstream of the ATG initiation codon (details in Fig. 1). An examination of GGT cDNA sequences in the GenBank revealed an absence of this CAG insert in all mouse GGT cDNAs, as well as those from rat and pig, but revealed its presence in most, but not all, human GGT cDNAs. Certain human lung GGT cDNAs lacked this CAG insert (11). Because this CAG insertion is located at an intron/exon junction, we compared the intron sequences from the mouse, rat, and pig to determine whether alternative splicing could account for its presence ( Fig.  2A). All three introns contain the highly conserved dinucleotides GT and AG at the 5Ј-and 3Ј-boundaries, respectively, and the AG is preceded by a cytosine residue. However, the mouse intron contains two CAGs in tandem at its 3Ј-intron boundary, whereas the rat and the pig intron each contain a single CAG. We then cloned and partially sequenced a corresponding human GGT intron sequence. We confirmed that this intron is identical in size to that of rat, mouse, and pig, ϳ0.5 kb, and contains two tandem CAG trinucleotides at the 3Ј-boundary like that of mouse.
To see if the proximity of this CAG insert to the ATG initiation site could affect the site of translation initiation, we used a transient VT7CP expression system to characterize the protein encoded by the GGT⌬1. Cow pox-infected CHO cells transfected with the GGT⌬1 plasmid were starved for Met and Cys and pulse-labeled in the same media with [ 35 S]Met/Cys, before a chase period of 0 or 2 h. SDS-PAGE analysis of GGT-specific immunoprecipitates from the cell extracts revealed a single peptide of 83 kDa at t ϭ 0, which was sensitive to treatment with endoglycosidase H and produced a product of 59 kDa (Fig.  3), consistent with expression of the full-length propeptide of 61 kDa (Table I). Because endo H removes all but one GlcNAc residue of high mannose N-glycans (M r ϳ 3000) from glycoproteins, this difference in M r (2400) is consistent with N-linked glycosylation of the GGT⌬1 at all seven consensus sites (Asn-X-Ser/Thr). After 2 h of chase, only a trace of the 83-kDa propeptide was evident, whereas the two expected subunits of the cleaved GGT propeptide were present at 50 and 23 kDa. Endo H treatment of this sample produced trace bands at 33 and 20 kDa, consistent with the presence of six N-glycans and one N-glycan on the subunits, respectively. The more diffuse gel pattern of the large subunit indicated that there is considerable microheterogeneity in the processing of these N-glycans, whereas the small subunit was more homogeneous despite its resistance to endo H treatment, indicating that the single Nglycan is minimally processed.
Similar data were obtained for the synthesis of GGT⌬1 in stable transfectants of CHO cells (Fig. 4). When these clonal CHO cells expressing GGT⌬1 were pulse labeled for 30 min and chased for 2 h prior to cell surface biotinylation with the membrane impermeant sulfo-NHS-SS-biotin, ϳ20% of the heterodimer was recovered from the immunoprecipitates with avidin-conjugated beads. This indicated that the GGT⌬1 protein did reach the plasma membrane (compare mock lanes in Fig. 4, A and B). This cell surface localization of the GGT⌬1 was confirmed by immunofluorescence analysis of COS cells 2 days after transfection with the GGT⌬1 cDNA (Fig. 5A). A similar pattern of immunofluorescence was observed in stably trans-  Table I and Figs. 2-8. fected CHO cells expressing the GGT⌬1 (data not shown). In both cases nearly all the GGT-specific immunofluorescence was found at the cell surface. An equally important result is that the clonal CHO cells expressing GGT⌬1 exhibited a much higher ␥-glutamyl transferase-specific enzymatic activity (280 FIG. 5. Immunofluorescence analysis. COS cells were transfected with plasmids encoding GGT⌬1 (A), ⌬2-5 (B), ⌬7 (C), and ⌬8 -9 (D). The steady-state localization of the expressed proteins were analyzed with rabbit anti-rat GGT antiserum followed by fluorescein isothiocyanatelabeled goat anti-rabbit IgG. COS cells (E) were stained for calnexin as a positive control for an ER protein, and primary antibody was omitted for a negative control (F). Surface staining is observed only for mGGT⌬1. Although ⌬2-5, ⌬7, and ⌬8 -9 predominantly show staining in the endoplasmic reticulum, a low level of juxtanuclear Golgi-like staining is also seen for all the proteins.   4. Only GGT⌬1 is expressed at the cell surface. Confluent cultures of clonal CHO cells stably expressing the GGT⌬1 in 21-mm wells were infected with cow pox for 30 min before transfection with 9 g of plasmid DNA encoding GGT⌬2-5, GGT⌬7, or GGT⌬8 -9 (mock received buffer and no DNA). The next day, cells were pulse-labeled for 30 min with [ 35 S]Met/Cys prior to a 2-h chase period, before biotinylation of the cell surface on ice, extraction with octyl glucoside, and immunoprecipitation of the GGT. A portion of the immunoprecipitate (75%) was further incubated with avidin-conjugated beads to obtain the biotinylated cell surface GGT. Both a portion of the total immunoprecipitate (B, 25% of total) and the biotinylated surface GGT (A) were subjected to SDS-PAGE and PhosphorImager analysis. Numbers to the right refer to the mobility of molecular mass standards in kDa. The mobility of the large (L) and small (S) subunits of the GGT⌬1 heterodimer are shown on the left, and the asterisk indicates protein isoforms. The diffuse gel pattern of the GGT⌬1 large subunit indicates that there is considerable microheterogeneity in the processing of the N-glycans. This band is less evident in B, because these lanes represent only 25% of each immunoprecipitate. mU/mg) than nontransfected CHO cells (Ͻ1 mU/mg). And after overnight accumulation of GGT⌬1 in transfected cow pox-infected CHO cells, this enzyme activity was greatly increased (200 -1000 mU/mg). Because this latter result indicated that the GGT⌬1 protein must be relatively stable, the half-life was determined for the radiolabeled GGT⌬1 during stable and transient expression by pulse-labeling cells for 30 min and immunoprecipitating GGT⌬1 after chase times of 3 and 18 h ( Table  I). Calculation of the half-life from the percent loss of radiolabeled GGT between the two time points indicated the half-life for the GGT⌬1 is similar in stable (18.7 h) and transiently transfected (17.4 h) CHO cells. Thus GGT⌬1 encodes the normal mouse GGT protein. This protein exhibited normal synthesis, cell surface expression, stability, and enzymatic activity. Because the expression of GGT⌬1 in cultured cells was indistinguishable from that described previously for rat and human GGT, we next assessed whether the CAG insert in the mouse GGT⌬1 could affect translational efficiency using an in vitro transcription/translation assay. The synthesis of the primary translation product (35.2 kDa) from the transcript containing the CAG insert was reduced ϳ2-fold when compared with synthesis from a transcript lacking the CAG sequence, whereas synthesis of an alternate protein from an internal ATG codon (27 kDa) was unchanged (Fig. 2B). Thus, the CAG appears to regulate translation of the GGT mRNA, not the protein product.
Mouse GGT⌬2-5-This alternative splicing event produces an in-frame deletion of the last eight nucleotides of exon 2, all of exons 3 and 4, and most of exon 5 (Fig. 6A). Nonconsensus 5Јand 3Ј-splice sites appear to have been utilized in exons 2 and 5, respectively. We could not identify a corresponding human GGT mRNA transcript but did find at least two alternative rat GGT cDNAs using mRNA from cultured rat alveolar type 2 cells that utilized the same 5Ј-splice site (data not shown). The encoded protein loses amino acid residues 96 through 231, including four N-glycan consensus sites from the amphipathic N-terminal subunit, while the C-terminal hydrophilic subunit remains intact (Fig. 6B).
When CHO cells transiently transfected with the GGT⌬2-5 were pulse-labeled with [ 35 S]Met/Cys for 15 min, a single labeled protein of 47 kDa was immunoprecipitated after both 0and 2-h chases (Fig. 3). At both time points the protein was sensitive to endo H treatment producing a protein of 38 kDa, consistent with the presence of three N-glycans, but which is notably smaller than the predicted sequence (46.7 kDa). Because the protein remains anchored to the membrane (data not shown), this indicates that the GGT⌬2-5 either is cleaved after the N-glycan consensus site in the small subunit or migrates anomalously on SDS gels.
The persistence of the endo H sensitivity of the GGT⌬2-5 protein after the 2-h chase also suggests that the protein is retained in the ER or early Golgi rather than moving to the cell surface. When this was tested by transient expression of GGT⌬2-5 in CHO-GGT⌬1 cells, only 1.5% of the GGT⌬2-5 was biotinylated at the cell surface after a 2-h chase, whereas ϳ20% of the GGT⌬1 was biotinylated in the same cells (Fig. 4). Localization in the ER was also evident by immunofluorescence microscopy of COS cells transiently expressing the GGT⌬2-5 (Fig. 5B). Determination of the half-life for the GGT⌬2-5 in both stable transfected (1.7 h) and transiently transfected (2.1 h) CHO cells indicates that this isoform of the mouse GGT is considerably less stable than the GGT⌬1 (Table I). Enzyme assays of these cell extracts reveal no increase in ␥-glutamyltransferase activity (Ͻ1 mU/mg) above the control levels observed in nontransfected CHO cells.
Mouse GGT⌬7-This alternative splicing event introduces a 22-base insertion within the coding domain. This insertion mimics that in humans and induces a frameshift and a premature stop codon within the open reading frame (Fig. 7A). The encoded mouse protein is a truncated GGT-like protein that retains only the first 339 amino acids of the native GGT protein and gains 14 novel residues at the C terminus, the first seven of which are identical to that in the human (Fig. 7B). Because the intron/exon structure of mouse GGT is known and the last 10 nucleotides at the 3Ј terminus of the insert agree exactly with the intron 7 sequences in the literature (1), it appears that the 22-base insertion results from an alternative splice site at intron 7. To confirm this, we cloned and partially sequenced the ϳ3 kb of mouse intron 7 (Fig. 7A). The remaining 10 nucleotides at the 5Ј-end of the insertion were identified as intron 7 sequences, and they were preceded by a CAG as a 3Ј-splice site. This 22-base mouse intron sequence is identical to the corresponding human intron sequence (Fig. 7C).
We used the 22-base insert as one of the primers in a PCR reaction to determine whether the GGT⌬7 splicing event is regulated in a developmental or tissue-specific fashion. We analyzed heart, liver, lung, kidney, and thymus RNA obtained from 1-day-old and adult mice (Fig. 7D). We detected GGT⌬7 in the RNA from the heart but not that from the liver or the lung during the neonatal period. RNA from the adult lung served as a positive PCR control here. In adult tissues, GGT⌬7 was present in RNA from the lung, the kidney, and the thymus but absent from that of the heart and the liver. It was also detectable in mRNA isolated from adult rat kidney and human peripheral blood mononuclear cells (data not shown).
To compare the relative abundance of GGT⌬7 to constitutive mGGT, we amplified the mRNAs from five different tissues using two sets of PCR primers that were common to both transcripts with a fully nested design (Fig. 7E). The 208-base pair signal amplified from constitutive mGGT was evident in all samples. The 230-base pair signal amplified from GGT⌬7 was only evident in lung, kidney, and brain and was less abundant than that of the constitutive GGT signal.
When CHO cells transiently expressing the GGT⌬7 were pulse-labeled for 15 min with [ 35 S]Met/Cys a single product of 44 kDa was obtained after chase times of both 0 and 2 h (Fig.  3). Endo H treatment produced a peptide of 32 kDa in both cases, which was slightly smaller than the peptide size calculated from the predicted sequence (38.2 kDa), but indicates that N-glycans are present at four of the five consensus sites in the GGT⌬7. When the cell surface of CHO-GGT⌬1 cells transiently expressing the GGT⌬7 were biotinylated after a 30-min pulse with [ 35 S]Met/Cys and a 2-h chase, only 0.5% of the GGT⌬7 was recovered with avidin-conjugated beads while 37% of the GGT⌬1 was recovered from the same cells (Fig. 4). Immunofluorescence microscopy of COS cells transiently expressing the GGT⌬7 (Fig. 5C) revealed staining of the ER, which is consistent with both the persistence of endo H sensitivity and the minimal cell surface expression of this isoform. Enzyme assays for both CHO cells stably and transiently ex-pressing GGT⌬7 revealed no increase in ␥-glutamyltransferase activity (Ͻ1 mU/mg) above the control levels observed in nontransfected CHO cells. The half-life for the GGT⌬7 in these same cells (6.5 and 10.2 h, respectively) was approximately half of that observed for the GGT⌬1 (Table I).
Mouse GGT⌬8 -9 -The ⌬8 -9 alternative splicing event eliminates at least the last four nucleotides of exon 8 and all of exon 9 but introduces 24 novel bases (Fig. 8A). The last 10 of these 24 nucleotides are identical to the published sequence at the 3Ј-boundary of intron 9 (1). This intron was cloned and sequenced, and the remaining 14 nucleotides were confirmed as intron 9 sequences, which were preceded by a CAG sequence. If this CAG represents the 3Ј-intron boundary, then the 5Ј-splice site lies within exon 8 but it is a nonconsensus splice site. However, other nonconsensus dinucleotide sequences could also border these boundaries as denoted in Fig. 8A, so the exact location of this splice site is unclear. This GGT cDNA has not been described in humans and we did not detect any PCR product when this 24-base insert was used as a PCR primer with human cDNA (data not shown). To confirm this, we cloned and sequenced a corresponding human intron. The 67 nucleotides at the 3Ј-boundary of our clone matched perfectly with sequences in human GGT genes 3, 6, and 11 (data not shown). However, comparison with mouse intron 9 revealed neither conserved sequences in this region nor any conserved 3Ј-splice site sequences (Fig. 8B). Usage of the 24 corresponding human nucleotides also failed to produce a PCR product. Hence, this particular GGT splicing event may be specific to mouse.  7. Characterization of GGT⌬7. A, as a result of this event, 22 bases are inserted (boldfaced) into the mouse ⌬7 cDNA, and this alters the reading frame from that of mGGT cDNA. B, a truncated mouse GGT protein (Mm) is encoded with a new C terminus that is similar to that predicted from an alternatively processed human GGT cDNA (Hs). C, this 22-base insertion in mouse GGT cDNA is identical to that described in human GGT cDNA and human lung-specific group II GGT cDNAs (Gp II) and very similar to sequences of a human GGT intronic promoter. D, PCR analysis as described under "Experimental Procedures" with primer 22U shows that this alternative splicing event is detectable in newborn mouse heart (Ht) but not liver (Li) nor lung (Lu). Adult lung (ϩ) was amplified with constitutive primers to serve as the positive control in this PCR reaction. In a separate PCR reaction for adult tissues, GGT⌬7 is undetectable in heart (Ht) and liver (Li) but evident in lung (Lu), kidney (Ki), and thymus (Thy). E, a PCR analysis was also performed as described under "Experimental Procedures" to compare the relative abundance of GGT⌬7 (230-bp product) versus constitutive mGGT (208-bp product) in heart (Ht), liver (Li), lung (Lu), kidney (Ki), and brain (Br).
To determine if ⌬8 -9 was expressed in a developmental or tissue-specific fashion, we repeated the analysis outlined for GGT⌬7 but used the 24-base insert as one PCR primer. GGT⌬8 -9 was not detected in RNA from the heart, the liver, or the lung during the neonatal period, even though it was present in the adult lung-positive control. In adult tissues, it was evident in RNA from the lung and the kidney but not that from the heart, the liver, or the thymus (Fig. 8C). Hence, GGT⌬8 -9 exhibits developmental-stage and tissue-specific regulation, but the patterns differ from that of GGT⌬7.
The elimination of the terminal four bases of exon 8 plus all of exon 9 together with the insertion of 24 bases from intron 9 removes 44 amino acids residues from the C-terminal small subunit but adds 8 novel amino acids proximal to active enzyme site residues (Fig. 8D). When CHO cells transiently transfected with the GGT⌬8 -9 were pulse-labeled with [ 35 S]Met/Cys for 15 min, a single labeled protein of 52 kDa was immunoprecipitated after both 0-and 2-h chases (Fig. 3). At both time points the band was sensitive to endo H treatment producing a protein of 37 kDa, consistent with the presence of five N-glycans on the protein. However, this is two less Nglycans than would be predicted from the GGT⌬8 -9 sequence. In addition, the protein size for the GGT⌬8 -9 calculated from the predicted sequence is considerably larger (57.6 kDa). The cumulative data would be most consistent with cleavage of the GGT⌬8 -9 into an unstable heterodimer, immediate degradation of the small subunit, and retention of the residual large subunit within the ER. When the cell surface of CHO-GGT⌬1 cells transiently expressing the GGT⌬8 -9 were biotinylated after a 30-min pulse with [ 35 S]Met/Cys and a 2-h chase, only 2.2% of the GGT⌬8 -9 was recovered with avidin-conjugated beads while 35% of the GGT⌬1 was recovered from the same cells (Fig. 4). Immunofluorescence microscopy of COS cells transiently expressing the GGT⌬8 -9 (Fig. 5D) revealed staining of the ER, which is consistent with both the persistence of endo H sensitivity and the poor cell surface expression of this isoform. The half-life of the GGT⌬8 -9 in transiently transfected cells was only 0.9 h, and there was no measurable ␥-glutamyltransferase enzymatic activity. Although stably transfected cells were not available for these studies, the similar data obtained for the GGT⌬1, GGT⌬2-5, and GGT⌬7 between transient and stably transfected cells (Table I) indicates that the half-life and the enzyme activity data for GGT⌬8 -9 in transiently transfected cells are reliable.
Stress Response Induction in the Endoplasmic Reticulum-Because GGT⌬1 is normally found at the cell surface, the localization of the GGT⌬2-5, ⌬7, and ⌬8 -9 in the ER could simply indicate that these GGT isoforms are abnormally folded products and they are being retained for subsequent degradation. Alternatively, this subcellular localization could be consistent with a new previously undefined role for these proteins within the ER. The active GGT⌬1 enzyme is essential for turnover of glutathione at the cell surface. Although the other GGT isoforms lack this enzyme activity, they could still bind substrate and act as sensors of the critical glutathione redox levels found in ER. In support of this hypothesis, we found that the phenotype of the stable CHO cells for GGT⌬1 and ⌬7 changed dramatically when the cell media was shifted from a 1:1 mixture of DMEM (cystine) and Ham's F-12 (cysteine) with 3% FBS to DMEM alone with 3% FBS. Although the GGT⌬2-5 and ⌬8 -9 cells remained unchanged, the GGT⌬1 and ⌬7 cells rounded, lifted, and detached from the culture plates within 12 h. The altered phenotype was reversed upon return of the cells to the original media containing cysteine as well as cystine. Because stress in the ER can be associated with this dramatic change in phenotype, we probed the cellular RNA from each cell line for the induction of CHOP-10 and BiP mRNAs, two markers of the ER stress response. CHOP-10 mRNA was not detected by Northern blot analysis in any of the cell lines before the change in media (data not shown) but was dramatically induced in the GGT⌬1 and ⌬7, but not the GGT⌬2-5 and ⌬8 -9 cells, during the recovery period (Fig. 9). Actin mRNA was examined as a control for RNA loading and integrity and was unchanged. To be sure this induction of CHOP-10 was not due simply to overexpression of a recombinant protein, two additional cell lines expressing high levels of FIG. 8. Characterization of GGT-⌬8 -9. A, this exact site of the splicing event for ⌬8 -9 is unclear but appears to involve alternative nonconsensus 5Ј-and 3Ј-splice sites within exon 8 and intron 9, respectively. The last four nucleotides of exon 8 are eliminated, and 24 bases from intron 9 (boldfaced) are inserted into the GGT⌬8 -9 cDNA. B, comparison of the terminal 25 nucleotides from mouse (Mm) intron 9 with those from the corresponding human (Hs) intron reveals many differences. C, PCR analysis for developmental and tissue-specific expression of ⌬8 -9 was performed as in Fig. 7, but the upstream primer in the secondary PCR reaction was 24D. D, the amino acid residues that are eliminated from the small C-terminal subunit of mGGT are boxed, and the eight novel amino acids of ⌬8 -9 are boldfaced. These hydrophobic residues are inserted proximal to the active enzyme site of mGGT (underlined). the recombinant glycoproteins hCAR or MUC1 (12) were similarly characterized and found to not induce CHOP in response to the change in media. Although hCAR is a 46-kDa glycoprotein cell surface receptor for coxsackie and adenoviruses (13), MUC1 is a very heavily O-glycosylated mucin-like transmembrane protein (M r Ͼ 220 kDa (12)). Finally, analysis of BiP mRNA levels in all these cell lines indicated that its expression paralleled that of CHOP-10.

DISCUSSION
Herein we have characterized four new GGT protein isoforms that are derived from alternative splicing events in mouse GGT cDNA. The mouse and the human not only generate GGT protein isoforms by this mechanism, but they share some of these splicing events in common. The encoded GGT protein isoforms can be expressed as transferase active heterodimeric glycoproteins on the cell surface or as transferase inactive monomeric glycoproteins in the endoplasmic reticulum. These latter findings suggest potentially novel functions for native GGT protein and its protein isoforms within the endoplasmic reticulum in addition to the known role of the native protein as a cell surface ectoenzyme.
Alternative pre-mRNA splicing of nascent eukaryotic mRNAs is a post-transcriptional process that is known to be highly regulated and widely utilized to generate multiple alternative products from a single gene (14). A prime example of the power of alternative promoters to generate such mRNA diversity can be found in the mouse GGT gene itself. Six GGT cDNAs, each with a unique 5Ј-untranslated region, are generated from this single copy gene via six alternative promoters (15). Alternative splicing events within the open reading frame can generate protein isoforms by excluding specific exon sequences or including novel intron sequences in the mature mRNA transcript. However, prior to our study, only a constitutive splicing pattern had been identified in the mouse GGT gene, which involves the invariant ligation of 12 coding exons, numbered by Arabic numerals 1 through 12, and one common noncoding exon, numbered by Roman numeral I as depicted in Fig. 1 (1).
The central difference between constitutive and alternative splicing lies in the selection and ligation of specific pairs of donor/acceptor splice sites. This is a complex process that is only partially understood, but it is clear that certain cis RNA sequences provide recognition sites for trans-splicing factors that form the spliceosome apparatus. These cis RNA sequences include the dinucleotides GT and AG located at the 5Ј-and the 3Ј-boundaries of an intron, respectively (16). These conserved dinucleotide sequences are found at all of the intron boundaries in the constitutively spliced mouse GGT gene (1). The factors involved in the selection of these specific GT and AG residues from a much large number of potential choices are not yet fully known for any gene. However, additional consensus sequences surrounding these dinucleotides as well as internal consensus sequences at the branch point participate in the selection process. Selection of the AG residue at the 3Ј-splice site appears to involve a scanning process initiated from the branch point and a competition among various intervening AG dinucleotides with the nucleotide preceding the AG having a profound influence on its selection in the order CAGϳTAG Ͼ AAG Ͼ GAG (16). In GGT⌬1 and GGT⌬7 the alternative splicing event was limited to the 3Ј-intron boundary and utilized an alternative AG dinucleotide that was preceded by a cytosine residue. In GGT⌬2-5 and GGT⌬8 -9 the alternative splicing events involved 5Ј-and 3Ј-splice sites as well as non-GT/AG dinucleotides.
In GGT⌬1 two CAGs in tandem are present at the 3Ј-boundaries of the corresponding introns in mouse and human GGT but absent from those of the rat and the pig. Our results show that alternative utilization of these sites can explain the presence or absence of a CAG trinucleotide insert upstream of the ATG initiation codon specifically in mouse and human GGT cDNAs. The scanning model predicts that the first CAG should identify the 3Ј-boundary of the intron while the second should be present in the cDNA. Our data agrees with this model, because 80% of the clones contained the CAG insertion. Despite the proximity of these bases to the translation initiation site, our protein expression studies clearly show that the heterodimer derived from GGT⌬1 is a stable protein and identical to that described previously in the rat and in humans (9,17). Hence, the mouse and humans express two GGT cDNAs that encode a GGT propeptide, which is heavily glycosylated, especially on the large subunit, and processed into the heterodimer before or after delivery to the cell surface. Our study now represents the first detailed characterization of the GGT glycoprotein in the mouse. The activity associated with this enzyme initiates the hydrolysis of reduced or oxidized glutathione at the cell surface, a process that is essential for recovery and uptake of cysteine/cystine. However, this metabolism also appears to generate the pro-oxidant hydrogen peroxide in the presence of iron (18). Hence, it is likely that the level of GGT activity is regulated. Our in vitro transcription/translation data shows that mouse and human GGT can use the insertion of this CAG sequence to down-regulate GGT protein production at the level of translation. The utilization of this same alternative splicing strategy to alter mRNA translational efficiency has already been described in the gene for human surfactantassociated protein A2, so it is not unique to GGT (19,20).
GGT⌬7 also utilizes a nearby AG dinucleotide located only 24 bases pairs upstream in intron 7 as the alternative 3Ј-intron boundary, and this is preceded by a cytosine residue. Two factors could have selected against this CAG as the constitutive 3Ј-splice site according to the scanning hypothesis. The first is a distance of less than 12 nucleotides from the branch point; the second is a location within a region of secondary mRNA structure such as stem loop. Our partial sequence analysis of mouse intron 7 supports the first mechanism. We found that there is a potential branch point consensus sequence that would place this alternative CAG within this distance restriction. There is also a second potential branch point sequence even further upstream that would also place it as the first downstream CAG trinucleotide (16). The factors that determine how these potential branch points are utilized in the FIG. 9. Expression of GGT⌬1 and ⌬7 mediate the ER stress response in CHO cells. Control nontransfected CHO cells, and CHO expressing GGT⌬1, ⌬2-5, ⌬7 or ⌬8 -9, or the recombinant control proteins hCAR or hMUC1 were maintained in DMEM/Ham's F-12 with 3% FBS. The media was changed to DMEM alone with 3% FBS for 12 h, and RNA was extracted from cells after 7 days recovery in DMEM/ Ham's F-12 with 3% FBS. RNA was analyzed by filter hybridization as described under "Experimental Procedures" using CHOP-1, BiP, and ␤-actin probes. No signal for CHOP-10 was evident in cells at baseline. mouse GGT gene are not yet known. The presence of these intron sequences in a human GGT gene together with their regulation in a developmental-stage and tissue-specific fashion in the mouse GGT gene suggests that they serve a common role in GGT gene expression in these two species. Determination of the degree of conservation will require further examination of the GGT genes of additional species such as the rat, the dog, the pig, and the cow. But these sequences are not found in the GGT gene(s) of bacteria, yeast, flies, or worms, suggesting that they are a relatively late addition to the genomes of metazoans. Further studies will be required to fully understand the complete role of these intron sequences for GGT gene expression. However, we note that the identical intron sequences are also found in the human lung-specific group II GGT cDNAs where they reside not in the coding domain but in the 5Ј-untranslated region (11). Similar, but not identical, sequences are also present in a human intronic GGT promoter (21).
Our PCR data on the relative expression of GGT⌬7 indicate that this alternative splicing event is minor compared with the constitutive event. However, we also found that the GGT⌬7 splicing variant was detectable in tissues where constitutive GGT mRNA was highly expressed, like the kidney, but not in tissues where GGT mRNA abundance is very low, like the liver. Hence, the expression of constitutive GGT and GGT⌬7 appears to be linked. An alternative splicing event like GGT⌬7 has previously been described in a human GGT mRNA, but three potential protein products were predicted based on the presence of different open reading frames (4). Our study in the mouse is the first to characterize the encoded protein product. Our characterization of GGT⌬7 expression in stable and transiently transfected CHO cells reveals that a single truncated form of GGT protein is synthesized. It is a glycoprotein of 44 kDa and has 14 novel residues at the C terminus. It is a relatively stable protein with a half-life of ϳ8 h and is localized to the ER rather than the cell surface. Because it lacks the C-terminal small subunit residues of the active site, it predictably has no transferase activity. However, it does retain the critical arginine 107 for substrate binding (23). Because recent studies have established an essential role for the GGT substrate glutathione in regulating the redox state of the ER (24), it is possible that GGT⌬7 can bind glutathione and acts as a sensor for glutathione levels within this compartment. Alternatively, the GGT⌬7 isoform may act as a chaperone for the synthesis of the native enzymatically active GGT, blocking its activity within the ER where degradation of glutathione would be unwanted. This would be supported by finding that GGT⌬7 predominantly exists in tissues where high levels of GGT are synthesized, and the low level of GGT⌬7 would reflect that amount needed to balance the low transient levels of newly synthesized native GGT in the ER.
The dinucleotides delimiting the possible alternative 5Ј-and 3Ј-splice sites in GGT⌬2-5 and GGT⌬8 -9 differ from the consensus GT and AG sequences. Hence, it is unclear exactly how these alternative events were processed but there are certainly many examples in other genes of splice sites that differ from the GT-AG consensus (25), including the human GGT genes (22). The presence of such sites led to the recent search and identification of an alternative intron subclass that is bounded by AT-AC dinucleotides. Therefore, other intron subclasses may also occur (26). Until further information becomes available in this area, the mechanism for these alternative splicing events in mouse GGT will remain uncertain. Nonetheless, the similarities between the developmental-stage and tissue-specific expression of GGT⌬8 -9 and GGT⌬7 suggests that these GGT mRNA splicing events are highly regulated, whereas the differences in tissue-specific expression suggest they can be regulated independently. We were not able to find a human correlate for GGT⌬8 -9, so this event may be specific to the mouse. The protein isoforms derived from GGT⌬2-5 and GGT⌬8 -9 lack ␥-glutamyltransferase activity, like GGT⌬7. In each case, a residue required for GGT activity, glutamic acid 108 (23) and aspartic acid 423 (27), respectively, is eliminated along with several other amino acids. GGT⌬2-5 and ⌬8 -9 were also localized to the ER. Coexpression studies failed to show any effect of these isoforms, including GGT⌬7, on total endogenous GGT activity or delivery of active transferase to the cell surface (data not shown). Hence, their exact function remains obscure. Certainly the more rapid turnover rates for ⌬2-5 and ⌬8 -9 could indicate that these glycoproteins are sensed as abnormal GGT products within the ER and targeted for degradation.
Nonetheless, the expression of GGT⌬1 and GGT⌬7 protein isoforms appears to be able to impact the environment of the ER as suggested by their ability to trigger an ER stress response with a change in cell culture conditions. This stress was demonstrated by the parallel induction of the mRNAs for CHOP, a nuclear protein that is regulated by ER stress, and BiP, a chaperone whose expression during ER stress is coordinately regulated with CHOP (28). The message for CHOP is not normally expressed in cells but is strongly induced by ER stress. CHOP protein then forms stable heterodimers with C/EBP family members and binds to novel DNA target sequences to alter the pattern of cellular gene expression in response to ER stress. When the ER stress is severe, CHOP can activate a programmed cell death pathway. The ability to trigger this response appears to be limited to the full-length mouse GGT protein and the truncated GGT⌬7 protein isoform under the conditions tested here. But the change in cellular phenotype as well as the level of induction of CHOP mRNA was dramatic in these two cell lines. CHOP can be strongly induced by deprivation of nutrients such as glucose as well as the amino acids arginine, leucine, lysine, methionine, phenylalanine, and threonine, but none of these molecules were limiting in our conditions (29). However, reduced Cys is found in Ham's F-12 and not in DMEM, which has only oxidized Cys, and this could directly impact the redox state of the ER. The lack of CHOP induction in two other stably transfected CHO cells lines overexpressing human glycoproteins hCAR or hMUC1 suggests that this triggering effect may be specific for these GGT isoforms. Further study will be required to understand the basis for the ER stress and the outcome of CHOP induction in these cell lines, but this will provide new insight into a role for GGT and its isoforms in this intracellular compartment. In addition, the GGT enu1 mouse, an animal model of oxidant stress due to GGT deficiency, allows one to study how the loss of GGT enzyme activity affects the pattern of GGT mRNA splicing. In preliminary experiments, we found that the GGT⌬7 splicing event is increased in the lung of these animals, further supporting a connection between GGT activity and the ⌬7 splicing event. The common expression of this event in human GGT suggests that the results of further studies on mouse GGT will be directly relevant to our understanding of human GGT gene expression and ultimately the role of GGT in mammalian cellular physiology and glutathione metabolism.