GNIP, a novel protein that binds and activates glycogenin, the self-glucosylating initiator of glycogen biosynthesis.

Glycogenin is a self-glucosylating protein involved in the initiation of glycogen biosynthesis. Self-glucosylation leads to the formation of an oligosaccharide chain, which, when long enough, supports the action of glycogen synthase to elongate it and form a mature glycogen molecule. To identify possible regulators of glycogenin, the yeast two-hybrid strategy was employed. By using rabbit skeletal muscle glycogenin as a bait, cDNAs encoding three different proteins were isolated from the human skeletal muscle cDNA library. Two of the cDNAs encoded glycogenin and glycogen synthase, respectively, proteins known to be interactors. The third cDNA encoded a polypeptide of unknown function and was designated GNIP (glycogenin interacting protein). Northern blot analysis revealed that GNIP mRNA is highly expressed in skeletal muscle. The gene for GNIP generates at least four isoforms by alternative splicing. The largest isoform GNIP1 contains, from NH(2)- to COOH-terminal, a RING finger, a B box, a putative coiled-coil region, and a B30.2-like motif. The previously identified protein TRIM7 (tripartite motif containing protein 7) is also derived from the GNIP gene and is composed of the RING finger, B box, and coiled-coil regions. The GNIP2 and GNIP3 isoforms consist of the coiled-coil region and B30.2-like domain. Physical interaction between GNIP2 and glycogenin was confirmed by co-immunoprecipitation, and in addition GNIP2 was shown to stimulate glycogenin self-glucosylation 3-4-fold. GNIPs may represent a novel participant in the initiation of glycogen synthesis.

The biosynthesis of glycogen involves two distinguishable stages. The initiation step involves the formation of a glycoprotein primer, by self-glucosylation of glycogenin to form a covalently linked oligosaccharide. Elongation involves the bulk synthesis of glycogen through the reactions catalyzed by glycogen synthase and branching enzyme (for reviews see Refs. [1][2][3]. Humans express two isoforms of glycogenin, one widely distributed (4,5) and the other, glycogenin-2, predominantly expressed in liver (6). One of the most important properties of glycogenin is its ability to self-glucosylate, using UDP-glucose as a glucose donor (6,7). In rabbit glycogenin-1, self-glucosylation results in the formation of a C-1-O-tyrosyl linkage between glucose and Tyr 194 (8,9). Self-glucosylation continues with the formation of ␣-1,4-glycosidic linkages, until a chain of 8 -12 residues has formed. This form of glycogenin serves as a substrate for glycogen synthase (1)(2)(3).
Several lines of evidence suggest an important role of protein-protein interactions for the function of glycogenin. First, glycogenin is capable of forming dimers (10 -12), resulting in self-glucosylation via an inter-subunit mechanism (12,13). It was proposed that interaction between subunits of glycogenin is relatively weak (12). Second, there are protein-protein interactions between glycogenin and glycogen synthase. For example, glycogenin co-purified with glycogen synthase from rabbit skeletal muscle in a stoichiometric 1:1 complex, indicating a relatively strong interaction (7). Moreover, two-hybrid analysis revealed the interaction between yeast glycogenin, Glg2p, and yeast glycogen synthase, Gsy2p (14). Third, protein-protein interactions may be involved in the subcellular distribution of glycogen. For example, glycogenin expressed in mammalian cells as a green fluorescent fusion protein was partly co-localized with actin (15). Binding to actin is mediated by the carboxyl-terminal part of glycogenin. Additionally, glycogenin was found in the cytoplasm and the cell nucleus (15)(16)(17).
In order to identify other proteins that interact with glycogenin, we used the yeast two-hybrid system with rabbit glycogenin-1 as bait. We identified an as yet undescribed protein, which we named glycogenin-interacting protein (GNIP), 1 that activates glycogenin in vitro.

EXPERIMENTAL PROCEDURES
Plasmids-The cDNA for rabbit skeletal muscle glycogenin was generated by cutting pET15b-GN (11) with NdeI, blunting with Klenow fragment, and subsequent digestion with SalI. The cDNA was ligated into pGBDU-C2 (18) that was cut with EcoRI, blunted, and digested with SalI. The resulting plasmid, which contains the glycogenin coding sequence in-frame with the DNA binding domain of Gal4p, was designated pGBDU-GN. The plasmid containing a cDNA fragment for GNIP, GNIPt-h, flanked by EcoRI-NotI-SalI adapter sequences (pGAD-GNIPt-h) was isolated from the two-hybrid library and used to create several GNIPt-h expression vectors. To construct pET28-GNIPt-h, pGAD-GNIPt-h was digested with NotI and blunt-ended, and the 1.6-kb fragment encoding GNIPt-h was ligated into pET28a (Novagen), which had been previously cut with NheI and blunt-ended. To construct pFLAG-GNIPt-h, pGAD-GNIPt-h was digested with NotI, and the 1.6-kb fragment was inserted into pFLAG-CMV-2 (Sigma), which was digested with NotI. To construct the vectors expressing GNIP1, the 0.4-kb fragment encoding the 5Ј end of GNIP1 was generated by BglII and NotI digestion of DNA from EST clone AI 492496 (Fig. 1). The middle fragment of GNIP1 was amplified from a human skeletal muscle library by PCR using GNIP1-and library adapter-specific primers, as described below. This PCR product was digested with NotI and EcoRI to generate a 0.5-kb fragment. Both the 0.4-and 0.5-kb fragments were ligated with BglII-and EcoRI-cut pET32a to construct pET32-GNIP-N. The 1.2-kb fragment encoding the 3Ј-terminal part of GNIP1 was obtained by digestion of pGAD-GNIPt-h with EcoRI and SalI. The 1.2-kb fragment was ligated into EcoRI-and SalI-cut pET32-GNIP-N to produce pET32-GNIP1. To construct pGADT7-GNIP1, pET32-GNIP1 was digested with BglII and XhoI, and the 2.1 fragment was inserted into BamHI-and XhoI-cut pGADT7 (CLONTECH). To generate GNIP2, the 0.3-kb fragment of GNIP2 encoding amino acid residues from Met 1 to Glu 98 and carrying an NdeI site at the 5Ј end was generated by PCR. This product was digested with NdeI and EcoRI to produce the 0.2-kb fragment. The 1.2-kb fragment encoding the 3Ј-terminal part of GNIP2 was obtained by digestion of pGAD-GNIPt-h with EcoRI and XhoI. Both the 0.2-kb and the 1.2-kb fragments were ligated with NdeI-and XhoIcut pGADT7 to construct pGADT7-GNIP2. To construct pET28-GNIP2, pGADT7-GNIP2 was digested with NdeI and EcoRI, and the 0.2-kb fragment was used to substitute the 0.3-kb NdeI/EcoRI fragment in pET28-GNIPt-h.
Yeast Two-hybrid Screen-The yeast strain PJ69-4A (18) was sequentially transformed with pGBDU-GN and the human skeletal muscle matchmaker cDNA library in pGAD10 plasmid (CLONTECH) using the lithium-acetate method. The transformants were plated on synthetic medium lacking uracil, leucine, and adenine. Colonies were picked from 4 to 12 days and re-plated on the second generation plates with synthetic medium deficient in uracil, leucine, and histidine. Library plasmids from the second generation colonies were rescued into Escherichia coli RRI cells plated on M9 medium lacking leucine and were analyzed by yeast two-hybrid tests and DNA sequencing. The purified pGAD plasmids were retransformed into Saccharomyces cerevisiae carrying pGBDU-GN, and quantitative ␤-galactosidase solution assays (19) were performed.
Northern Blot Analysis-A human multiple tissue Northern blot (CLONTECH) was probed with a 32 P-labeled 1.6-kb GNIP fragment excised from pGAD-GNIPt-h by NotI digestion. The procedure was performed according to the manufacturer's protocol.
5Ј-Rapid Amplification of cDNA Ends (RACE)-Marathon-Ready cDNA (CLONTECH) prepared from human skeletal muscle was used as template for 5Ј-RACE. Two sets of nested primers were designed to extend from exon six: SAS14, CTTAAGATCCAGAGAGAGGATGAGG; SAS12, CTCAGAAGAGACTGTGGTTGGCTTG; and from exon eight: SAS19, ACATGTGACCTCAGGAAGGGAACACC; SAS20, CCTCAAG-GCCAGATTCGCAAGTAGG. Two approaches were used to amplify and identify the 5Ј end of GNIP1. In the first approach, the primary amplification was performed on 5 l of cDNA template with 0.2 mM dNTPs, 10 pmol of adapter primer (AP1, CLONTECH), 10 pmol of the genespecific primer (SAS14), 10ϫ reaction buffer containing 15 mM MgCl 2 (Roche Molecular Biochemicals), and 1 unit of Expand High Fidelity Taq polymerase (Roche Molecular Biochemicals) in a total volume of 50 l. Hot start cycling conditions were used with an initial denaturation step at 94°C for 2 min followed by addition of the enzyme at 85°C. This was followed by 30 cycles of 94°C for 30 s and 65°C for 45 s and a final extension step 68°C for 5 min. The last cycle was at 68°C for 8 min. For secondary amplification, components and conditions were identical except that nested AP2 (CLONTECH) and gene-specific (SAS12) primers were used; 1 l of the primary product was used as template and 35 cycles of 94°C for 15 s, 68°C for 30 s, and 68°C for 5 min were performed. Cloning of amplified products was carried out using TOPO TA Cloning kit (Invitrogen). Plasmid DNA was purified using Plasmid Miniprep Kit (Bio-Rad), and PCR products were identified by automated DNA sequencing.
By using this protocol, the 5Ј end of original two-hybrid clone (GNIPt-h) was extended (PCR 403, Fig. 1). Search of the human EST data base in GenBank TM using the sequence of clone PCR 403 as a query identified one EST clone (GenBank TM accession number AI 492496) with sequence that partially overlaps the sequence of clone PCR 5-61. This clone was obtained from the ATCC collection and analyzed by automated DNA sequencing. Combining the sequences from EST clone AI 492496, PCR 403 and GNIPt-h would generate a cDNA sequence for an isoform designated GNIP1 (GenBank TM accession number AF396651). The original two-hybrid clone, GNIPt-h, represents the 3Ј fragment of GNIP1 (Fig. 1). The cDNA for GNIP1 was assembled in pET32a and pCMV-Tag-3B vectors using the BglII/NotI fragment of EST clone AI 492496, the NotI/EcoRI fragment of PCR 403, and the EcoRI/SalI fragment GNIPt-h (see above). Another clone identified in this screen, PCR GP9 ( Fig. 1), has a 5Ј end sequence that does not match the sequence of GNIP1. In this clone, two codons for methionine precede and are in-frame with the initiating ATG of GNIPt-h. At the 5Ј end, clone PCR-GP9 contains a stop codon, which precedes and is in-frame with new ATG codons. This clone predicts the isoform designated GNIP3 (GenBank TM accession number AF396655).
To confirm that the entire coding sequence for GNIP1 is present in human skeletal muscle, a second PCR approach was applied. In this method, the same template and nested primers SAS19/SAS20 were used with Advantage-GC 2 PCR Kit (CLONTECH) according to the manufacturer's protocol for amplification of DNA with target size Ͻ5 kb. The amplified products were inserted into the pCR4 Blunt-TOPO cloning vector (Invitrogen) followed by transformation of One Shot TOP10 cells (Invitrogen). The resulting clones were analyzed by colony hybridization using a 32 P-labeled 0.4-kb BglII/NotI fragment from EST clone AI 492496 as a probe. Plasmid DNAs from positive clones were isolated, and the PCR fragments were analyzed by automated DNA sequencing. One of the clones (PCR 18, Fig. 1) contained the entire open reading frame for GNIP1. This new clone extends the 5Ј end of cDNA for GNIP1 by 43 bp, as compared with the EST clone AI 492496, and contains a stop codon, in-frame with and preceding the initiating ATG codon.
In another effort to extend the 5Ј end of GNIP, 5Ј-RACE was performed on a human skeletal muscle 5Ј-STRETCH PLUS cDNA library (CLONTECH) using GNIP-specific and TriplEx-specific nested primers. The largest PCR fragments were selected for sequence analysis. Three clones, PCR38D, PCR38E, and PCR39H, have an identical 3Јterminal region that is indistinguishable from the corresponding sequence in GNIP1. However, sequences in the 5Ј ends of these clones are different. All three clones contain in-frame stop codons preceding the initiating ATG codon. Despite differences in 5Ј-UTR sequences, these clones encode the same protein product, which was designated GNIP2. Later the sequences from the 5Ј regions of three clones were found in the GNIP gene corresponding to the alternatively spliced gene transcripts GNIP2a (GenBank TM accession number AF396653), GNIP2b (GenBank TM accession number AF396652), and GNIP2c (GenBank TM accession number AF396654). These products include the sequences from clones PCR38D (GNIP2a), PCR38E (GNIP2b), and PCR39H (GNIP2c). The existence of three alternatively spliced mRNA products for GNIP2 was confirmed by sequencing of several clones obtained from human skeletal muscle Marathon-Ready cDNA as a result of 5Ј-RACE analysis.
Transient Transfection-COS-M9 cells were transiently transfected by using LipofectAMINE (Invitrogen protocol). Briefly, 0.5-1 g of plasmid DNA per 6 l of LipofectAMINE was used to transfect cells in a 35-mm dish. Cells were grown for 2 days and subsequently harvested.
Immunoprecipitations-COS cells expressing the protein(s) of interest were homogenized in 0.2 ml of buffer B, containing 50 mM Tris-HCl, pH 7.8, 10 mM EDTA, 2 mM EGTA, 100 mM NaF, 1 mM dithiothreitol, and protease inhibitors. The supernatants were isolated after centrifugation at 14,000 ϫ g for 10 min at 4°C. The pellets were resuspended in 0.2 ml of buffer B. The supernatants were subsequently incubated with antibodies and protein A-agarose in the presence of 0.1% Triton X-100. The precipitates were washed three times with phosphate-buffered saline containing 0.1 M NaCl and 0.1% Triton X-100, and proteins were eluted with SDS-PAGE loading buffer. The supernatants, the pellet fractions, and the immunoprecipitates were separated by SDS-PAGE and analyzed by Western blot.
Expression and Purification of Recombinant Proteins-Expression of His-tagged GNIP2 was performed in E. coli BL21/DE3 from the plasmid pET28-GNIP2. Transformed cells were cultured at 37°C until the A 600 reached 0.4 and then incubated at 18°C for ϳ20 h. Cells were harvested by centrifugation at 4500 ϫ g for 10 min. The cell pellet was resuspended with lysis buffer (10 ml per 1 g of cell pellet) containing 50 mM Tris-HCl, pH 7.9, 300 mM NaCl, 0.1% Triton X-100, 15% glycerol, 1 mM phenylmethylsulfonyl fluoride, 0.1 mM 1-chloro-3-tosylamido-7-amino-2-heptanone, 2 mM benzamidine, 0.5 mM 2-mercaptoethanol, 1 M leupeptin, 0.3 M aprotinin, 20 mM imidazole. The cells were broken by passage through a French pressure cell at 1,000 pounds/inch 2 . After centrifugation of the cell homogenate at 10,000 ϫ g for 30 min, the supernatant was collected and mixed with Ni 2ϩ -nitrilotriacetic acidagarose at 4°C for 1 h. After incubation, the resin was collected by centrifugation, resuspended in lysis buffer, washed with lysis buffer containing 65 mM imidazole, and eluted with lysis buffer containing 200 mM imidazole. Fractions were collected and dialyzed against the buffer containing 50 mM Tris-HCl, pH 7.9, 100 mM NaCl, 2 mM dithiothreitol, and 20% glycerol. NH 2 -terminally His-tagged rabbit skeletal muscle glycogenin was expressed and purified as described previously (12).
Glucosylation Assays-The assay measures the amount of [ 14 C]glucose incorporated into glycogenin essentially as described previously (20). The reaction was typically carried out in 50 mM Hepes, pH 7.5, 2 mM dithiothreitol, 5 mM MnCl 2 , 20 mM NaCl, 4.5% glycerol, and 38.5 M UDP-[ 14 C]glucose at 30°C for 5 min or as indicated. Two methods were applied to quantitate glycogenin glucosylation. In one, aliquots of the reaction mixture were spotted onto P81 chromatography paper which was washed three times with 5% phosphoric acid for 20 min each. The paper was dried and subjected to scintillation counting. In the second method, an aliquot was added to 0.25 volume of 5ϫ SDS-PAGE loading buffer and subjected to SDS-PAGE and autoradiography.

RESULTS
Screening for Glycogenin-1-binding Proteins Using the Twohybrid System-To search for cDNA clones encoding proteins that interact with glycogenin-1, we fused the coding sequence for rabbit skeletal muscle glycogenin with the Gal4p DNA binding domain and screened a human skeletal muscle library expressed from the pGAD-10 vector. From 3.1 ϫ 10 6 transformants harboring bait and library cDNA, five clones were isolated representing three different cDNAs as judged by cDNA sequencing. One clone contained a 3.2-kb cDNA for glycogen synthase including 157 bp of the 5Ј-untranslated region (clone GN-11). Two clones contained a 1.5-kb cDNA for glycogenin-1 with 82 (clone GN-6) or 91 bp (clone GN-10) of 5Ј-untranslated region. The sizes of the isolated cDNAs exceed the lengths of coding regions for glycogen synthase and glycogenin indicating that both polypeptides are expressed in yeast as full-length proteins, with a segment of the fusion protein coded by the 5Ј-UTR. Two other clones contained identical 1.6-kb fragments encoding a 340-residue polypeptide of an unknown protein (clones GN-1 and GN-7). This protein was designated glycogenin-interacting protein (GNIP), and the entire protein sequence encoded by two-hybrid library plasmid was designated GNIPt-h (see Fig. 1). To confirm the interactions, the plasmids isolated from two-hybrid clones were introduced into yeast that had been pre-transformed with pGBDU-GN. The resulting transformants were able to express both ADE2 and HIS3 genes, which are under control of the GAL4 upstream-activating sequence in PJ69-4A yeast strain (data not shown). Based on quantitative ␤-galactosidase assays the interaction between GNIPt-h and glycogenin-1 was relatively strong compared with the interaction between glycogen synthase and glycogenin-1 (Fig. 2). The glycogenin-1 inter-subunit interaction was weak, in agreement with previous data (12). GNIPt-h interacts with human glycogenin-1 as judged by using ADE2, HIS3, and ␤-galactosidase expression as reporter genes (data not shown).
GNIP Interacts with Glycogenin in Mammalian Cells-To determine whether GNIP interacts with glycogenin in mammalian cells, we expressed NH 2 -terminally FLAG-tagged protein together with glycogenin in COSM9 cells. GNIPt-h was detected as a polypeptide of ϳ39 kDa in both the soluble and the pellet fractions of the transfected cells (Fig. 3A). The soluble fractions were immunoprecipitated with anti-FLAG antibody. As shown in Fig. 3B, FLAG-GNIPt-h protein was detected in the anti-FLAG immunoprecipitates from cells transfected with FLAG-GNIPt-h or cells co-transfected with both FLAG-GNIPt-h and glycogenin. With a specific anti-glycogenin-1 antibody, glycogenin was only detected in immunoprecipitates from cells co-transfected with glycogenin and FLAG-GNIPt-h (Fig. 3C). These data demonstrate that the interaction between glycogenin and GNIP occurs in mammalian cells.
GNIP Contains a B30.2-like Domain and a Predicted Coiledcoil Domain-The nucleotide sequence of the 1579-bp GNIPt-h insert predicted an open reading frame (ORF) encompassing nucleotides 1-1021, in-frame with the GAL-4 activation domain of pGAD-10. The ORF was followed by a TGA stop codon at nucleotides 1022-1024. Near the 5Ј end, there was a putative ATG start codon at nucleotide 113. The deduced amino acid sequence of the ORF (bp 113-1021) encodes a protein of 303 amino acids with theoretical molecular mass of 34213 Da. GenBank TM searches of the protein data base using the BLASTP algorithm indicated several homologous proteins including butyrophylin, RET finger protein (RFP), pyrin/marenostrin (the Mediterranean fever protein), 52-kDa Ro/SSA autoantigen, and several RING finger and zinc finger proteins of unknown function. The homologous region was at the COOH terminus of these proteins and contained a B30.2-like domain (21). Analysis of the GNIPt-h protein sequence with the Simple Modular Research Tool (SMART) (22) indicated that 114 residues of COOH terminus contain a SPRY domain, which is a subdomain of the B30.2 region (23,24). Toward the NH 2 terminus of GNIPt-h, the SMART program identified a stretch of 28 residues with the potential of forming an ␣-helical coiled-coil The numbers indicate the scale and the length of the cDNA for GNIP1 expressed in base pairs. Restriction sites for BglII, NotI, and EcoRI are indicated by arrows. Putative initiating ATG codons are indicated by dashed vertical ticks. The terminating TGA codon is indicated as solid vertical tick.

FIG. 2. Quantitative two-hybrid assay for glycogenin interactions.
The plasmid expressing rabbit muscle glycogenin (pGBDU-GN) was tested in combination with plasmids isolated from clones GN-1, -6, and -11 or control pGAD-10 plasmid. Plasmids from these clones express GNIP (GNIPt-h, clone GN-1), glycogen synthase (clone GN-11), or glycogenin itself (clone GN-6). The two-hybrid assay was quantitated using ␤-galactosidase measurements (see "Experimental Procedures"). Values represent an average of two independent experiments. structure. The B30.2-like domain and coiled-coil region occur in some members of the protein family known as RBCC. Members of this family contain a tripartite motif of a RING finger, one or two B boxes, and a predicted coiled-coil domain (25,26) (Fig. 4). Because we had not unequivocally identified the start codon, it was possible that the cloned GNIPt-h was not full length and that it lacked NH 2 -terminal sequence corresponding to the RING finger and B box domains. This hypothesis was initially supported by searching the human EST data base in Gen-Bank TM where we found several EST clones (GenBank TM accession numbers AI139356, AI290319, and AI347404) containing sequences identical to 105 bp in the 5Ј-terminal region of the GNIPt-h cDNA, allowing us to extend the 5Ј region of GNIPt-h. However, the extended region did not contain an in-frame ATG. From extensive 5Ј-RACE analyses, described in detail under "Experimental Procedures," we were able to establish the existence of multiple GNIP transcripts that correspond to at least three different isoforms, which we designate GNIP1, GNIP2, and GNIP3.
The cDNA for GNIP1, the longest isoform, encodes a 511amino acid protein with a predicted molecular mass of 56,630 Da that contains, according to SMART analysis, an NH 2 -terminal RING finger domain followed by a B box domain in addition to the coiled-coil region and B30.2-like domain of the initial GNIPt-h clone (Fig. 4). GNIP2 is a truncated version of GNIP1, which lacks the NH 2 -terminal RING finger and B box domains. GNIP3 would have a unique NH 2 -terminal sequence, not present in GNIP1, and the COOH terminus that is identical in all three isoforms (Fig. 4). Note that we are predicting the GNIP3 transcript based only on amplification of the unique 5Ј end and not the entire sequence (see "Experimental Procedures"). Also, we are assuming that the first ATG is the initiating codon, rather then one of two downstream ATGs. If the third ATG were the initiating codon, the product of translation would be the same as GNIP2. Searching GenBank TM with the GNIP1 protein sequence as query, we retrieved the sequence of the tripartite motif protein TRIM7 (GenBank TM accession number AF220032). The NH 2 -terminal 206 residues of TRIM7 are identical to the NH 2 terminus of GNIP1 and contain RING finger and B box domains (Fig. 4). The COOH terminus of TRIM7 is short and contains 15 residues unique to TRIM7.
Mouse Homologues of GNIP-From the mouse EST data bank, we identified three homologous mouse clones. Two of these EST clones were obtained from ATCC (GenBank TM accession numbers AA517788 and AW012184) and sequenced.
We found several sequencing errors in the published sequence of clone AA517788 and resubmitted the corrected 1,454-bp sequence (GenBank TM accession number AF396656). This sequence contains an ORF encoding a polypeptide of 303 residues, corresponding to GNIP2. The initiating ATG codon for the ORF is preceded by an in-frame stop codon suggesting that it corresponds to a full-length protein. The mouse GNIP2 is 93% identical to its human counterpart. The sequence of the second EST clone, AW012184, is identical to AF396656 only in the 3Ј part that encodes GNIP2. The 5Ј end of this clone is different and does not contain an in-frame terminating codon, suggesting that a larger isoform of mouse GNIP might exist.
Human GNIP Gene-We identified a clone of 183 kb in the unfinished high throughput genomic sequences that contained matches to GNIP. The clone (GenBank TM accession number AC008620) is derived from chromosome 5. Later, we found the GNIP gene in contig NT 006854.3, a working draft sequence segment of human chromosome 5. Exon-intron boundaries were established by comparison of the genomic sequences with GNIP and TRIM7 cDNA sequences. All the exon-intron boundaries conformed to the expected GT and AG sequences ( Table  I). The gene for GNIP contains eight exons and seven introns (Fig. 5). The transcripts that we have identified can be explained by alternative splicing, by skipping exons, or by using alternate splice acceptor sites. Thus, splicing can generate smaller versions of exons 2, 3, and 4 (designated 2a, 3a, 3b, and 4a in Fig. 5). The cDNA for GNIP1 utilizes all exons, except exon 2, and exons 3 and 4 in this cDNA are represented by the exons 3a and 4a. We identified three different GNIP2 cDNAs that are distinguished by their 5Ј-UTR. These variants all contain exons 4a to 8 but have variant-specific regions from exons 2 and 3 (Fig. 5). GNIP3 cDNA is composed of five exons, from 4 to 8. The shortest isoform TRIM7 is derived from exons 1 and 3. The size of the exons varies from 23 bp (exon 5) to more than 1000 bp (exons 4 and 8). The first exon encompasses the 5Ј-UTR for GNIP1 and TRIM7 and the region encoding the RING finger and B box. The second exon encodes the 5Ј-UTR for GNIP2. Exon 3a is part of the 5Ј-UTR for GNIP2 but encodes part of the coiled-coil region in GNIP1. Exon 3b provides the 5Ј-UTR for GNIP2a. The entire exon 3 is present only in the cDNA for TRIM7 and encodes the coiled-coil region and the 3Ј-UTR. The larger version of exon 4 was found in the cDNA for GNIP3. This exon encodes 5Ј-UTR and part of coiled-coil motif. Exon 4a contains the initiating ATG codon for GNIP2 and encodes part of coiled-coil region in both GNIP1 and Interestingly, another sequence, which is almost identical to the fragment of cDNA for GNIP1 (1-1516 bp, more than 99% identity), was reported to derive from chromosome 13 (contig NT009891.7). It is not yet clear if this represents a second GNIP gene or an annotation problem.
Expression of the GNIP Gene-To examine the tissue distribution of GNIP expression, we probed a human multiple tissue Northern blot (CLONTECH) using the entire cDNA fragment from the two-hybrid clone. This probe detected major bands of ϳ3.4 kb in skeletal muscle and of 1.3 kb in placenta (Fig. 6). The 3.4-kb band of significantly lower intensity was detected in heart, brain, and pancreas. Additionally skeletal muscle contained species of 1.4 and 5.8 kb. Heart and brain contained species of 1.7 and 5.8 kb. The multiplicity of mRNA species is consistent with the multiplicity of transcripts indicated by the cDNA cloning.
GNIP2 Activates Glycogenin Self-glucosylation-In order to examine the possible biological relevance of the interaction between GNIP and glycogenin, we analyzed the effect of GNIP on the self-glucosylation reaction catalyzed by glycogenin. GNIP2 and glycogenin were produced as NH 2 -terminally His 6tagged proteins expressed in E. coli. Expression of GNIP1 resulted in the accumulation of insoluble protein in bacteria, and no soluble protein was obtained. Incubation of glycogenin with UDP-[U-14 C]glucose followed by SDS-PAGE and fluorography revealed a single labeled polypeptide with ϳ42 kDa, corresponding to His 6 -glycogenin (Fig. 7). In the presence of GNIP2, incorporation of [ 14 C]glucose into glycogenin was markedly increased indicating that GNIP2 activates the reaction. No labeling was associated with other proteins suggesting that His 6 -GNIP2 is not an acceptor of [ 14 C]glucose. Maximal activation occurred at equal concentrations of both proteins, suggesting a likely stoichiometry for the binding of glycogenin to GNIP2 of 1:1. Glycogenin also transfers glucose from UDPglucose to low molecular weight acceptors, like n-dodecyl ␤-Dmaltoside. However, GNIP2 had no effect on glycogenin-mediated incorporation of glucose into n-dodecyl ␤-D-maltoside suggesting that the effect of GNIP2 was restricted to the selfglucosylation reaction (data not shown). A more detailed characterization of glycogenin activation by GNIP2 revealed that the GNIP2 caused an almost 4-fold increase in V max with little or no change in K m for UDP-glucose (Fig. 8). DISCUSSION We have identified a novel family of proteins, GNIPs, that interact with glycogenin, a critical component of the glycogen biosynthetic pathway. The search for interacting proteins also selected glycogen synthase and glycogenin, whose ability to bind glycogenin was already known. GNIP1 belongs to the RBCC subgroup of RING finger proteins, which contain a second zinc finger known as the B box, followed by a coiled-coil domain (26,27). COOH-terminal to the RBCC motif, GNIP1 contains a domain similar to the B30.2-like domain, which was initially identified as a product of a coding sequence in the chromosomal region containing the human major histocompatibility complex class I (21). Thus, GNIP1 is a member of the more specific RBCC-B30.2 protein family (28). Proteins from this family are involved in various processes including cell growth and differentiation. This category includes the following: the RET finger protein, RFP, which becomes oncogenic when it recombines with the tyrosine kinase domain of the RET protooncogene (29); putative transcription factors, the estro-  (39,40). Partial loss of the B30.2 domain in the midline 1 (MID1) protein caused by mutations is responsible for the Opitz G/BBB syndrome, characterized by developmental midline defects (41). Multiple alignment of protein sequences (us-ing the ClustalW algorithm) indicated considerable homology of GNIP1 with the RBCC-B30.2 proteins, i.e. 36% identity with RFP, 35% identity with testis-abundant finger protein/Ring finger protein 23 (42), 32% identity with TERF, 32% identity with SSA/Ro, and 30% identity with Xnf7.
In attempts to amplify the 5Ј end of GNIP cDNA obtained in two-hybrid screen, we found five distinct extensions. One would predict the cDNA for largest isoform of GNIP, GNIP1. Three distinct cDNA extensions predict three different 5Ј-noncoding regions for another protein isoform of GNIP, GNIP2 (Fig. 5). GNIP2 represents the COOH-terminal region of GNIP1, lacking the N-terminal RING finger and the B box domains (Fig. 4). The fifth cDNA extension generates the third isoform of GNIP, GNIP3. This isoform would contain a unique 26-residue sequence attached to the NH 2 terminus of GNIP2. By searching GenBank TM with the GNIP1 protein sequence, we identified a fourth isoform of GNIP, TRIM7, which is identical in sequence to the NH 2 -terminal 206-residues of GNIP1 and contains a unique COOH-terminal tail of 15 residues. All variants are derived from one region of human chromosome 5, corresponding to the GNIP gene. Characterization of the human GNIP gene demonstrated a complex gene structure, with multiple differentially spliced transcripts. Some of the variation results from splicing occurring inside certain exons to produce smaller sub-exons such as 2a, 3a, 3b, and 4a (Fig. 5). The significance of this observation is not clear and could represent the existence of allelic variants.
The existence of multiple products derived from the GNIP gene is confirmed by other experiments. Northern blot analysis demonstrated transcripts of different sizes in skeletal muscle and one distinct species in placenta (Fig. 6). Significantly lower levels of GNIP expression were found in heart, brain, and pancreas. Consistent with our data, other studies demonstrated that one of the products of the GNIP gene, TRIM7, is selectively expressed in skeletal muscle in adult mice (43). However, TRIM7 is ubiquitously distributed in embryonic mouse tissues. Independent evidence for expression of GNIP2 in skeletal muscle was obtained from sequencing of a mouse EST clone (GenBank TM accession number AA517788) from the Barstead myotubes library. This clone contains the entire coding sequence for GNIP2 flanked by 5Ј-UTR and 3Ј-UTR, the latter including the poly(A) tail. Another EST clone (Gen- Bank TM accession number AW012184) from mouse kidney might represent a partial sequence of the GNIP1 isoform. Interestingly, both full-length and truncated versions were found for other members of the RBCC-B30.2 protein family (37,43). For example, the gene for RNF21 generates at least three isoforms, due to alternative splicing (37). Expression of one specific form in HeLa cells was dramatically up-regulated by interferon. It is possible that expression of individual forms of GNIP in skeletal muscle is selectively regulated by different stimuli and may be important for specific cell functions.
Interaction between isoforms of GNIP and glycogenin might have important physiological consequences. It is intriguing to speculate that GNIP may act to target or sequester glycogenin to specific cellular locations where glycogen is in demand. The systematic study of cellular localization of several members of RBCC protein family demonstrated that some of these proteins might target unique cellular compartments (43). To identify other proteins interacting with GNIP, we performed a twohybrid screen of a human skeletal muscle library using GNIPt-h as bait. In this screen, we found interaction between GNIPt-h and desmin. 2 We hypothesize that GNIP might target glycogenin to intermediate filaments of the muscle cytoskeleton. However, the biological significance of this targeting is not clear.
Synthesis of biopolymers is often regulated at the initiation stage, and so glycogenin is a candidate to control glycogen accumulation. However, our understanding of the control of glycogenin is still quite limited. In previous work (44,45), we have shown that the ability of purified glycogenin to serve as substrate for glycogen synthase depends on its glucosylation state. Identification of protein(s) interacting with glycogenin leads us to seek regulatory functions. We found that GNIP2 activates the ability of glycogenin to self-glucosylate, increasing the V max 3-4 times with little or no change in K m for UDP-glucose (Fig. 8). However, GNIP2 does not change the rate of glucosylation of low molecular weight acceptors indicating that GNIP2 does not act as regulator of enzymatic activity of glycogenin. Therefore, the stimulatory effect of GNIP2 on self-glucosylation presumably occurs via a change in the conformation of the glycogenin dimer that creates a better condition for attachment of glucose to the existing polysaccharide chain. However, we cannot exclude the possibility that the activation is secondary to a more important targeting role.
Our finding of GNIP in skeletal muscle might help to explain the results obtained by Smythe and colleagues (10). These authors demonstrated that electrical stimulation of or epinephrine administration to skeletal muscle caused degradation of 50% of the glycogen molecules in the muscle resulting in glycogen-free glycogenin. The subsequent reassociation of glycogenin and glycogen synthase was slow and could be rate-limiting for glycogen re-synthesis in the period of recovery. The authors suggested that muscle contains factors that might be responsible for controlling the rate of reassociation of glycogenin and glycogen synthase (10), and we propose that GNIP might be such a factor.