Glycogenin-2, a Novel Self-glucosylating Protein Involved in Liver Glycogen Biosynthesis*

Glycogenin is a self-glucosylating protein involved in the initiation phase of glycogen biosynthesis. A single mammalian gene had been reported to account for glycogen biogenesis in liver and muscle, the two major repositories of glycogen. We describe the characterization of novel forms of glycogenin, designated glycogenin-2 (GN-2), encoded by a second gene that is expressed preferentially in certain tissues, including liver, heart, and pancreas. Cloning of cDNAs encoding glycogenin-2 indicated the existence of multiple species, including three liver forms (GN-2α, GN-2β, and GN-2γ) generated in part by alternative splicing. Overall, GN-2 has 40–45% identity to muscle glycogenin but is 72% identical over a 200-residue segment thought to contain the catalytic domain. GN-2 expressed in Escherichia coli or COS cells is active in self-glucosylation assays, and self-glucosylated GN-2 can be elongated by skeletal muscle glycogen synthase. Antibodies raised against GN-2 produced in E. coli recognized proteins of M r ∼66,000 present in extracts of rat liver and in cultured H4IIEC3 hepatoma cells. In H4IIEC3 cells, most of the GN-2 was present as a free protein but some was covalently associated with glycogen fractions and was only released by treatment with α-amylase. H4IIEC3 cells also expressed the muscle form of glycogenin (glycogenin-1), which was attached to a chromatographically separable glycogen fraction.

Glycogen, a branched polymer of glucose, is a metabolic energy reserve accumulated in many cell types (1). In mammals, the major glycogen deposits in absolute amount are those of skeletal muscle and liver. These glycogen reserves have somewhat different functions, but both pools contribute to blood glucose homeostasis. In eukaryotes, the pathway of glycogen biogenesis consists of an initiation step and a subsequent phase of bulk polysaccharide synthesis mediated by glycogen synthase and the branching enzyme (see, for example, Ref. 2). The existence of a specific initiator protein, postulated from the work of Krisman and colleagues (3), was substantiated by characterization of a protein, glycogenin, that was covalently linked to glycogen and that acted as a primer for the action of glycogen synthase (4,5). Glycogenin was also found to be enzymatically active (4,5). A model for muscle glycogen biosynthesis evolved in which glycogenin would undergo a self-glucosylation reaction to generate the oligosaccharide primer for glycogen synthase action (reviewed in Refs. 6 -8). Glycogenin is an oligomer, probably a dimer, and it has been proposed that the self-glucosylation, while intramolecular, is intersubunit (9).
Both the enzymic machinery and the regulatory mechanisms for glycogen metabolism differ, at least in part, between liver and muscle. Muscle and liver isoforms of glycogen synthase (10) and phosphorylase (11) have been defined by biochemical analysis and cDNA cloning, whereas to date there appear to be single genes encoding the branching and debranching enzymes (12,13). The first glycogenin to be characterized at the molecular level was from mammalian muscle. Campbell and Cohen (14) purified and sequenced rabbit skeletal muscle glycogenin, and a cDNA encoding this protein was subsequently cloned (15). Glycogenin was also purified from liver as a protein associated with glycogen, and sequencing of several derived peptides indicated sequence identity with muscle glycogenin (16). It was therefore concluded that the same form of glycogenin was responsible for glycogen biogenesis in both liver and muscle (16). Ercan et al. (17) have also described a self-glucosylating protein of similar size present in rat liver, whereas several other reports have noted mammalian proteins, of different sizes, that had some of the properties of glycogenin (reviewed in Refs. 7 and 18). Whether or not these species are related by amino acid sequence to the known glycogenin was not established. In other organisms, however, there is clear evidence for multiple glycogenin-like proteins (see Ref. 6). Two self-glucosylating proteins, Glg1p and Glg2p, from the yeast Saccharomyces cerevisiae have been characterized and shown to be required for glycogen accumulation in that organism (19). In addition, searching GenBank revealed that Caenorhabditis elegans, Caenorhabditis briggsae, and Arabidopsis thaliana all express multiple glycogenin-like messages (6). We report here the existence of a second mammalian glycogenin gene preferentially expressed in liver, heart, and, to a lesser extent, pancreas. The protein, which we designate glycogenin-2, has all the properties expected of glycogenin; it can self-glucosylate, can act as a substrate for glycogen synthase, and is released from glycogen by ␣-amylase treatment. This finding makes it necessary to reconsider the enzymic components implicated in the pathway for glycogen biogenesis in these tissues.
in the original sequences in the data base, and complete corrected sequences for the two cDNA clones have been deposited in GenBank with accession numbers U94357 (brain) and U94358 (breast). Sequences from the other ends of the breast clone (accession number R71875) and brain clone (accession number H04789) were already present in the data base. A small overlapping region (80 base pairs) of identical sequence was found in the brain and breast cDNA clones. These cDNAs were used as probes to screen human brain and human liver cDNA libraries (5Ј-stretch plus cDNA libraries, in gt10, from CLONTECH). A 320-base pair HindIII-EcoNI fragment (probe 1) of the brain clone was used to screen the brain library, and a 0.3-kb EcoRI-NcoI fragment from the 5Ј-end of brain clone 9 -1 (see Fig. 1 and "Results") was used for liver library screening. Totals of 4 ϫ 10 5 plaques were screened for each library. Probes were randomly labeled by [␣-32 P]dCTP using a New England Biolabs NEBlot™ kit. Approximately 2 ϫ 10 6 cpm/ml probe was used for hybridization which was carried out in 5 ϫ SSPE (1 ϫ SSPE: 0.15 M NaCl, 0.01 M NaH 2 PO 4 ⅐H 2 O, 1 mM EDTA, pH 7.4), 10 ϫ Denhardt's solution (1 ϫ Denhardt's solution: 0.2 g/liter Ficoll, 0.2 g/liter polyvinylpyrrolidone, 0.2 g/liter bovine serum albumin), 100 g/ml single-stranded DNA, 0.2% SDS, 50% formamide, 0.05% pyrophosphate at 42°C for 18 h. Filters were first washed twice in 2 ϫ SSC (1 ϫ SSC: 0.15 M NaCl, 0.015 M sodium citrate, pH 7.0) and 0.5% SDS at room temperature for 20 min each and then washed twice for 20 min in 1 ϫ SSC and 0.1% SDS at 65°C. Positive clones were picked, replated, and taken to plaque purity. phage DNA was purified using polyethylene glycol/NaCl precipitation of the plate lysate, followed by phenol:chloroform (1:1) DNA extraction. Purified DNA preparations were digested with EcoRI and subcloned into the EcoRI site of the pGEM7 vector (Promega) for sequencing. DNA was sequenced by the dideoxy termination method of Sanger (20), either manually or with an Applied Biosystems automated sequencer. All DNA sequences reported in this work result from sequencing the corresponding cDNAs on both strands. For cDNA clones with internal EcoRI sites, the order of the EcoRI fragments was confirmed by polymerase chain reaction (PCR) using uncut DNA as templates and primers that straddle the EcoRI sites.
Northern Blot Analysis-A human multiple tissue Northern blot was obtained from CLONTECH. The blot was hybridized with probes from the brain clone (H04789/H04891) using either probe 1 or a 1.2-kb PstI-NotI fragment (probe 2). The filter was also probed with a 0.9-kb NotI-SalI fragment (probe 3) from the breast clone (R71874/R71875). Probes were randomly labeled by [␣-32 P]dCTP, as described above. Normally 2 ϫ 10 6 cpm probe/ml hybridization solution was used. Hybridization was performed in 5 ϫ SSPE, 10 ϫ Denhardt's solution, 100 g/ml single-stranded DNA, 0.1% SDS, 50% formamide, 0.05% pyrophosphate at 42°C for 18 h. The filter was washed for 20 min twice in 2 ϫ SSC and 0.1% SDS at room temperature and then washed for 20 min twice in 0.1 ϫ SSC and 0.1% SDS at 60°C. The filter was exposed to x-ray film at Ϫ80°C. Before each re-use, the filter was stripped in boiling 0.5% SDS solution for 10 min, followed by a wash with 2 ϫ SSC for 10 min according to the manufacturer's recommendations, and then exposed for several days at Ϫ80°C to confirm that no signal remained.
Expression, Purification, and Assay of Recombinant Glycogenin-2-The coding sequences from the liver cDNAs encoding glycogenin-2␣ and glycogenin-2␥ (clones 7Ј and 1Ј, respectively) were cloned into the E. coli expression vector pET-28a so as to express NH 2 -terminally His 6 -tagged forms of the proteins. Thus, EcoRI-ClaI fragments (0.4 kb for clone 1Ј and 0.6 kb for clone 7Ј) containing the ATG sites were excised and ligated into the corresponding sites in the pGEM7 vector. Site-directed mutagenesis was then used to create NdeI sites at the ATG start codons by changing the sequence from ACCATG to CATATG using the Quick-Change™ site-directed mutagenesis kit (Stratagene). The mutagenic oligonucleotides for clone 1Ј were: CCCGCGGCGCCACT-GCATATGTCGGCCACCATTG(ϩ) and CATTGGTGGCCGACA-TATGCAGTGGCGCCGCGGG(Ϫ). The oligonucleotides for clone 7Ј were: CCCGCGGCGCCACTGCATATGTCGGAGACAGAG(ϩ) and CT-CTGTCTCCGACATATGCAGTGGCGCCGCGGG(Ϫ). All regions that were subjected to mutagenesis or to PCR amplification were sequenced. Finally, NdeI-ClaI fragments encoding the NH 2 -terminal segment of the protein and ClaI-EcoRI fragments encoding the COOH-terminal portion were ligated into pET-28a via a three-piece ligation. The resulting plasmids (pET-28a-7Ј and pET-28a-1Ј) were transformed into E. coli BL21/DE3 cells for expression. Conditions for cell culture and purification of the His 6 -tagged proteins were essentially as described previously (21). For analysis of self-glucosylation, recombinant GN-2␣ protein, at 120 g/ml, was routinely incubated for 30 min with 77 M UDP-[U-14 C]glucose (specific activity 265 mCi/mmol) in buffer containing 50 mM HEPES, pH 7.5, 5 mM MnCl 2 , 2 mM dithiothreitol, at 30°C. To test the ability of the proteins to serve as substrates for elongation by glycogen synthase, after 30 min of self-glucosylation, purified glycogen synthase (from rabbit skeletal muscle) and unlabeled UDP-glucose were added, to final concentrations of 100 g/ml and 27 mM, respectively, and incubated for another 120 min. To test the inhibitor sensitivity, the self-glucosylation was done as described above in the presence of 10 mM effectors but was measured using a filter assay similar to that described previously (21). Briefly, an aliquot (5 l) removed from the reaction was spotted onto P81 chromatography paper followed by three washes (30 min each) in 0.5% phosphoric acid and once in ethanol. The dried paper was counted by a liquid scintillation counter.
Expression of Glycogenin-2 in COS-1 Cells-All three full-length forms of liver glycogenin-2 (from clones 1Ј, 2Ј, and 7Ј) were expressed in COS-1 cells. The coding regions were excised from the corresponding pGEM7 vectors with EcoRI and cloned into the EcoRI site of the mammalian expression vector pcDNA3 (Invitrogen) for transfection in COS-1 cells. Transient transfection was achieved using LipofectAMINE from Life Technologies, Inc. Normally, 5 g of plasmid was used for transfection of each 35-mm dish at a stage when the cells were ϳ80% confluent. After transfection, the cells were maintained in the Dulbecco's modified Eagle's medium (Sigma) containing 10% fetal bovine serum for 2 days before they were harvested in buffer containing 50 mM HEPES, 0.5 mM phenylmethylsulfonyl fluoride, 0.1 mM N ␣ -p-tosyl-Llysine chloromethyl ketone, 2 mM benzamidine, 0.5 mM ␤-mercaptoethanol, and 0.5% Triton X-100, pH 7.4. Cells were broken by freezing in liquid nitrogen and thawing. The plates were then scraped with a rubber policeman and the material further homogenized by pipetting. Total cell lysate (1 mg/ml protein) was used to assay self-glucosylation (as described above) and Western analysis (see below).
Antibody Production and Western Blot Analysis-Recombinant His 6tagged glycogenin-2␣ purified by Ni 2ϩ -agarose chromatography was used for antibody production in guinea pigs (Cocalico Biologicals, Inc). The specificity of the antisera was tested by Western hybridization using extracts from COS-1 cells expressing glycogenin-2 (see "Results"). The Western blot procedure was essentially as described previously (21). Signals were visualized by using either the Enhanced Chemiluminescence (ECL) system (Amersham) or 125 I-protein A (5 Ci/10 ml hybridization). Normally, the antiserum was diluted 1:2000 for Western analysis. Rat hepatoma H4IIEC3 cells were grown in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum. Confluent monolayers were harvested in 1 ml of buffer A (50 mM Tris-HCl, pH 7.8, 10 mM EDTA, 2 mM EGTA, 100 mM NaF, 0.1 mM N ␣ -p-tosyl-L-lysine chloromethyl ketone, 1 mM phenylmethylsulfonyl fluoride, 1 mM dithiothreiotol) and frozen in liquid nitrogen. After thawing, cells were homogenized by repeated pipetting. The supernatant obtained by centrifuging the cell lysate at 17,000 ϫ g for 10 min at 4°C was used for Western analysis. Some samples were treated with 60 g/ml ␣-amylase (Worthington) for 30 min at 30°C before Western blot. Previously, we had produced guinea pig polyclonal antibodies to rabbit skeletal muscle glycogen synthase (22) that are effective for immunoprecipitation but not for Western analysis. These were available as a purified IgG fraction. Serum containing guinea pig polyclonal antibodies to rabbit muscle glycogenin (23), also produced in this laboratory, are effective for both immunoprecipitation and Western hybridization. Serum containing polyclonal antibodies to rabbit skeletal muscle glycogen synthase produced in chicken was kindly provided by John C. Lawrence Jr., University of Virginia. It was used for Western analysis of glycogen synthase at 1:4000 dilution. For the Western analysis of rat liver samples, liver from fed rats was homogenized in cold buffer (0.25 M sucrose, adjusted to pH 7.4 with Tris base) and centrifuged at 700 ϫ g for 10 min. The supernatant from this low speed spin was centrifuged at 8500 ϫ g for 10 min. Protein taken from the pellet fraction was used for Western analysis.
Immunoprecipitation-Supernatants from extracts of H4IIEC3 cells were obtained as described above and incubated with 60 g/ml ␣amylase for 30 min at 30°C. Antisera to glycogenin-2 was added to 1 ml of supernatant (1.2 mg of protein/ml) at 1:200 dilution; for the antiglycogen synthase IgG, 2 g of antibody was added. Cell extracts were incubated on ice for 1 h before addition of 20 l of protein A-agarose (Life Technologies, Inc.). The samples were then incubated for another 1 h on a nutator at 4°C, followed by two washes, first with 1 ml of buffer containing 1 ϫ phosphate-buffered saline, 0.1% Triton X-100 and 0.3 M NaCl and then with 1 ml of 1 ϫ phosphate-buffered saline, 0.1% Triton X-100 and 0.1 M NaCl. After the second wash, the pellet was resuspended in 25 l of electrophoresis sample buffer (62 mM Tris-HCl, pH 6.8, 2% (w/v) SDS, 37.9 mM dithiothreitol, 10% (v/v) glycerol, and 0.01% (w/v) bromphenol blue) and subjected to SDS-PAGE, followed by Western blot analysis.

Fractionation of H4IIEC3 Extracts by Concanavalin A-Sepharose
Column-Confluent monolayers of H4IIEC3 cells, grown in forty 100-mm plates, were rinsed twice in 6 ml of buffer B (containing 50 mM Tris HCl, pH 7.8, 100 mM NaCl, 1 mM EDTA, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, 0.1 mM N ␣ -p-tosyl-L-lysine chloromethyl ketone) and were frozen in liquid nitrogen. After thawing, lysed cells were homogenized and centrifuged at 16,000 ϫ g for 15 min, and the supernatant (containing a total of 210 mg protein) was adjusted with MnCl 2 and CaCl 2 to final concentrations of 3 mM. The protein was loaded onto a column (5 ml) of concanavalin A-Sepharose 4B equilibrated with buffer C (containing 50 mM Tris HCl, pH 7.8, 100 mM NaCl, 1 mM dithiothreiotol, 1 mM phenylmethylsulfonyl fluoride, 3 mM MnCl 2 , and 3 mM CaCl 2 ). The column was washed with buffer D (buffer C containing 0.5 M NaCl) and was then eluted with a linear gradient of glucose formed of 50 ml of buffer D and 50 ml of buffer D plus 1 M glucose. Fractions of 2 ml were collected.
Other Methods and Materials-Protein was quantitated by the method of Bradford (24), using bovine serum albumin as a standard. Glycogen was measured as described by Skurat et al. (25). SDS-PAGE followed the method of Laemmli (26) as described previously (21). Recombinant glycogenin-1, with an NH 2 -terminal hexahistidine tag, was expressed in E. coli and purified as described previously (9).

Identification of Novel
Glycogenin-like cDNAs-Searches of expressed sequence tag (EST) data bases with the muscle glycogenin sequence revealed numerous entries with identical or almost identical sequences. In addition, there were two matching EST sequences that were similar, but not identical, to that of muscle glycogenin. These cDNAs, which were derived from human brain (accession numbers H04789/H04891) and breast (accession numbers R71874/R71875) libraries, were sequenced in their entirety (see Fig. 1). The two clones shared a region of identity but contained 3Ј regions that differed completely from one another. Both were partial clones. Using the brain clone as a probe, we screened a human brain cDNA library and identified four related cDNAs, none of which was full-length ( Fig. 1). Two clones were identical. We also screened a human liver cDNA library and isolated four apparently full-length cDNAs. Two of the four liver cDNAs were identical, leaving three independent clones that differed at their 5Ј-ends (Fig. 1).
Despite their multiplicity, all of the cDNAs appear to be derived from a single gene since there were common segments of absolute sequence identity. We propose to name the corresponding protein glycogenin-2 because, as described below, it has properties consistent with its serving a similar function to the known glycogenin. Muscle glycogenin can be termed glycogenin-1 where needed for clarity. After skeletal muscle, liver is the major repository of glycogen in the body, and so most effort was applied to analyzing clones from this source. The three independent cDNAs analyzed each had long open reading frames and differed in regions immediately following the start codon; they are named glycogenin-2␣ (accession number U94362), glycogenin-2␤, (accession number U94363), and glycogenin-2␥ (accession number U94364). Glycogenin-2␣ (501 amino acids) has an insertion of 40 residues compared with glycogenin-2␥, whereas glycogenin-2␤ has an insert of 9 residues that aligns with the extreme NH 2 terminus of glycogenin-1. This 9-amino acid insert in glycogenin-2␤ is also present in glycogenin-2␣. It seems most likely that the three different forms are generated by alternative splicing. Where they overlap, the three liver cDNAs have identical sequences except in three locations. First, the cDNA encoding glycogenin-2␥ has a region of 5Ј-sequence that does not match glycogenin-2␣ or glycogenin-2␤. The region of sequence dissimilarity ends with an EcoRI site and is probably due to an artifact during library construction. Second, glycogenin-2␣ and ␥ share a codon insertion with respect to glycogenin-2␤. A possible explanation for this difference is the existence of different alleles. Finally, glycogenin-2␣ has a G 3 T substitution in the 3Ј noncoding sequence. The liver glycogenin-2 isoforms ␣, ␤, and ␥ have predicted M r of 55,150, 51,841, and 50,995 with predicted pI values of 4.75, 4.72, and 4.76, respectively. Overall, the proteins would have 42-45% amino acid identity to muscle glycogenin (Fig. 2). However, there is 72% identity over an NH 2terminal region of about 200 amino acids thought to be involved in catalysis (6).
The other partial cDNAs potentially define as many as four more isoforms of glycogenin-2. H04789/H04891 and brain clone 9 -1 (accession number U94360) have identical sequences where they overlap and would predict a protein with an internal deletion of 71 COOH-terminal residues with respect to liver glycogenin-2␣. We designate this isoform glycogenin-2␦. Brain clones 4 -1 and 9 -2 are identical (accession number U94361) and similarly code for a protein with a deletion in the COOH terminus with respect to glycogenin-2␣, but in this case the deletion is only 36 residues. This form is termed glycogenin-2. Brain clone 2-1 (accession number U94359) has two internal EcoRI sites, and the two EcoRI fragments at the 5Ј-end are most likely due to artifacts during library construction. The 3Ј EcoRI fragment has a region of sequence matching the 3Ј regions of the liver glycogenin-2 messages except for a 13-base insertion in the noncoding sequence. However, preceding this segment of identity is a sequence not present in any of the other clones and which does not have an open reading frame. We are therefore doubtful as to the legitimacy of this clone. Finally, the breast clone would encode a partial protein sequence consistent with that, for example, predicted by the liver clones except that the sequence is truncated by a stop codon introduced by a single base change. This form is designated glycogenin-2⑀. The full significance of the various partial glycogenin-2 cDNAs will require further study.
The tissue distribution of glycogenin-2 expression was analyzed by Northern blotting of poly(A) ϩ RNA isolated from several different tissues. Using three different probes derived from the original brain (H04789/H04891) and breast (R71874/ R71875) clones (see "Experimental Procedures"), we obtained essentially the same results. A single hybridizing signal corresponding to a size of 3.6 kb was detected, with the strongest signal for liver RNA (Fig. 3). The band was broad enough, however, that we would not have distinguished the small differences in size predicted by the cDNAs for glycogenin-2␣, ␤, and ␥. A species of similar size was clearly visible with heart and, more weakly, with pancreas. The longest liver cDNA was 3.2 kb, suggesting that we lack little of the mRNA sequence. Indeed, if the 3Ј of the brain H04789/H04891 clone, which does extend to a poly(A) tail, is shared by the glycogenin-2␣ clone, we can account for 3.4 kb of the message.
Biochemical Properties of Glycogenin-2-To test whether glycogenin-2 has properties expected of a self-glucosylating initiator protein, we expressed it in bacterial and mammalian cells. The glycogenin-2␣ coding sequence was inserted into a pET vector so that it would be produced in E. coli as a fusion protein with an NH 2 -terminal polyhistidine tag. The protein was expressed in E. coli and purification by Ni 2ϩ -agarose chromatography yielded a major species of apparent M r 66,000, compared with a predicted M r of 57,329 for the His 6 -tagged protein (Fig.  4A). The reason for this aberrant electrophoretic mobility is not known, but it is probably not due to glucosylation since it was not significantly affected by ␣-amylase. Some smaller protein species were present and these probably represent proteolytic degradation products. After incubation with UDP-[U-14 C]glucose, glycogenin-2␣ incorporated radioactive label indicating that it was capable of self-glucosylation. Note that several lower M r labeled species accumulated suggesting that any proteolysis did not eliminate self-glucosylation. For comparison, muscle glycogenin-1 was similarly analyzed (Fig. 4B) using 7-fold less protein. After the self-glucosylation reaction, puri-fied glycogen synthase was added to the glycogenin together with an excess of unlabeled UDP-glucose; any high molecular weight radioactivity would then track the glycogenin protein attached to high M r carbohydrate. For both glycogenin-1 and -2, a significant amount of label was moved to higher M r , including material too large to enter the gel, indicating that glycogenin-2␣ was able to serve as a substrate for elongation by glycogen synthase. Similar efforts to analyze glycogenin-2␥ resulted in expression of protein that was not active under the FIG. 3. Northern analysis of glycogenin-2 distribution. A human multiple tissue Northern blot was hybridized with probes from both brain and breast EST clones (see "Experimental Procedures"). The same pattern was obtained with either probe. The numbers to the right indicate the molecular weights (in kb) of standards.

FIG. 2. Alignment of glycogenin-2␣
with other glycogenins or glycogenin-like proteins. The alignment is of all current full-length sequences for glycogenins or glycogenin-like proteins. Shown, however, are only the regions of similarity: the first 281 residues of glycogenin-2␣ and the extreme COOH terminus. The C. elegans sequence CeGlg-a has accession number Z72514, and Ce-Glg-c is CE04664 from the St. Louis C. elegans sequencing project. The alignment was made with the MACAW program, followed by manual adjustments. Uppercase letters indicate the domains selected by the computer. Several conserved domains, as defined by Roach and Skurat (6) are indicated. Note that here the alignment of domain VII includes mammalian and C. elegans sequences. Residues conserved in five or more of the six sequences have an asterisk beneath. The bullets mark the Tyr residues involved in selfglucosylation in glycogenin-1, Glg1p, and Glg2p.
conditions of the self-glucosylation assay (data not shown). The three liver glycogenin-2 coding regions were also inserted into mammalian expression vectors, as described under "Experimental Procedures," and transiently expressed in COS-1 cells (Fig. 5). In all three cases, antibodies raised to glycogenin-2 recognized species of M r ϳ66,000. No signal was observed from control COS cell extracts, suggesting that endogenous COS cell glycogenin either was not recognized by the antibodies or else was present at very low levels. Lysates from the expressing cells were subjected to self-glucosylation assays. Both glycogenin-2␣ and ␤ were active, whereas little or no label became associated with glycogenin-2␥ consistent with the results of expression in E. coli.
Several enzyme kinetic properties of glycogenin-1 and glycogenin-2 were compared. Self-glucosylation of glycogenin-2, like glycogenin-1, was dependent on the presence of divalent metal ions of which Mn 2ϩ was the most effective (data not shown). The reaction was not significantly dependent on pH from pH 6 to pH 9 (data not shown). In absolute terms, self-glucosylation by glycogenin-2␣ was slower than that by recombinant glycogenin-1, ϳ1 nmol/min/mg compared with ϳ11 nmol/min/mg, while the apparent K m for UDP-glucose was significantly higher than that of glycogenin-1 (Table I). Like glycogenin-1, glycogenin-2 was also capable of transferring glucose to small molecule acceptors such as n-dodecyl ␤-D-maltoside. Under standard conditions (21), glycogenin-1 catalyzed transglucosylation at a rate more than 50 times greater than self-gluco-sylation, whereas, for glycogenin-2␣, self-glucosylation and transglucosylation occurred at similar rates (data not shown). A number of potential effectors of glycogenin-2 activity were also tested for their effect on self-glucosylation (Table II). In general, the most effective inhibitors of glycogenin-1 were similarly potent for glycogenin-2␣. For example, CDP and UDPxylose were strong inhibitors of both glycogenins, as reported previously for glycogenin-1 (27,28). The same was true of some polyphosphate compounds. The most striking difference was the effect of transglucosylation acceptors, like maltose and n-dodecyl ␤-D-maltoside, which presumably inhibit self-glucosylation by competing as substrates. Consistent with the finding that glycogenin-2 was less active in transglucosylation, neither of these compounds was an effective inhibitor of glycogenin-2 self-glucosylation. Finally, ATP was reproducibly a stronger inhibitor of glycogenin-2 than glycogenin-1.
Analysis of Glycogenin-2 in Cells-Extracts from H4IIEC3 hepatoma cells or from the liver of fed rats were analyzed by Western hybridization using anti-glycogenin-2 antibodies. A single immunoreactive species of M r ϳ66,000 was detected in rat liver. The fraction giving the strongest signal was an 8500 ϫ g pellet (Fig. 6C), whereas only a weak signal was seen in the the 8500 ϫ g supernatant (data not shown). We also analyzed extracts from H4IIEC3 cells and likewise observed a single immunoreactive species of similar size (Fig. 6A). When the H4IIEC3 cell extract was first treated with ␣-amylase to degrade glycogen and release free glycogenin, there was only a small change in the signal strength for glycogenin-2, suggesting that most of this protein was not attached to carbohydrate

TABLE I Enzyme kinetic properties of glycogenin-1 and glycogenin-2
The self-glucosylation of glycogenin-1 (55 g/ml) and glycogenin-2 (120 g/ml) was measured as described under "Experimental Procedures" with UDP-glucose concentrations over the range of 1.5-300 M. Assays were for 20 min. The kinetic parameters were estimated by non-linear regression using the GraFit package (Erithacus Software, Staines, UK). Note that values of K m and V max for the hexahistidinetagged glycogenin-1 used in this study differ a little from those observed with wild-type glycogenin-1. The dried gel was exposed to x-ray film for 5 days. The signal for Western blotting was visualized by interaction with 125 I-protein A, followed by autoradiography. The numbers to the left indicate the molecular mass values (in kDa) of standards. (Fig. 6A). From densitometric analysis of four different experiments, we observed on average a 17% increase in glycogenin-2 signal after ␣-amylase treatment (range 5-30%). Immunoblotting using anti-glycogenin-1 antibodies indicated that the H4IIEC3 cells contained an immunoreactive species of M r ϳ41,000, consistent with the presence of glycogenin-1; this protein was only detected after treatment of the extracts with ␣-amylase, suggesting that it was covalently bound to glycogen (Fig. 6B). An immunoreactive species of M r ϳ45,000 was present whether or not the extracts were treated with ␣-amylase, and the significance of this species is not known. Based on densitometric analyses using the corresponding recombinant glycogenin as the standard, we estimate that glycogenin-2 is at least 10 times more abundant than glycogenin-1 in H4IIEC3 cells. 2 The H4IIEC3 cell extract was also analyzed by chromatography on concanavalin A-Sepharose which binds carbohydrates such as glycogen (Fig. 7). The glycogen content of H4IIEC3 cells grown under these conditions is 19 g of glycogen/mg of protein. Most glycogenin-2 did not bind to the column and was present in the flow-through fraction as a species of M r 66,000, as judged by SDS-PAGE (data not shown). The column was eluted with a glucose gradient and the fractions were subjected to Western analysis using antibodies against glycogenin-1 or glycogenin-2. No signal was seen for fractions not treated with ␣-amylase (data not shown). However, after treatment with ␣-amylase, both glycogenin-1 (Fig. 7C) and glycoge- 2 These values are based on comparison of the signal intensities with purified recombinant standards of human glycogenin-2␣ and rabbit glycogenin-1. Over the 160 residues for which the sequences are known, rat glycogenin-1 is 92% identical to rabbit glycogenin-1 at the amino acid level. There are no non-human sequences for glycogenin-2 to make a similar comparison. The antibodies are polyclonal and were raised against whole recombinant proteins. The standards are not perfect, however, and there may be some error in this quantitation.

TABLE II Effectors of glycogenin self-glucosylation
The self-glucosylation of glycogenin-1 (55 g/ml) and glycogenin-2 (120 g/ml) was measured as described under "Experimental Procedures" with the indicated additions. The UDP-glucose concentrations were 19 M and 77 M for glycogenin-1 and glycogenin-2␣, respectively. Assays were for 5 min or 30 min for glycogenin-1 or glycogenin-2␣, respectively. nin-2 ( Fig. 7B) were detected. Glycogenin-1 and glycogenin-2 eluted in overlapping fractions, but their elution was not coincident and the peak fractions were clearly separated. The peak of glycogenin-2 elution was fraction 15 or 16, whereas the glycogenin-1 was maximal in fraction 19. This result suggests that glycogenin-1 and -2 are attached to populations of glycogen molecules that are chromatographically separable. The fractions from concanavalin A chromatography were also analyzed by Western blotting with anti-glycogen synthase antibodies. An immunoreactive species of M r 82,000 was also detected in a wide range of fractions. Previously, we had observed glycogen synthase in H4IIEC3 cells as species with M r ϳ86,000 or ϳ90,000 (29). The difference may be due to the use of different molecular weight standards or due to proteolysis, since the liver isoform of glycogen synthase is known to be particularly susceptible to degradation (30). The strongest signal was in coincidence with glycogenin-1. However, with longer exposures, signal was also clearly present in earlier fractions where glycogenin-2 was dominant. We found that these chicken antibodies, although raised against muscle glycogen synthase, cross-reacted weakly with purified liver glycogen synthase (data not shown). Therefore, we cannot be certain which glycogen synthase isoform is being detected, although it seems probable that the liver isoform is present. Further evidence for an association between glycogen synthase and glycogenin-2 came from immunoprecipitation experiments (Fig. 8). Anti-glycogenin-2 immunoprecipitates contained glycogen synthase, as judged by Western blotting. Likewise, anti-glycogen synthase immunoprecipitates contained material cross-reacting with anti-glycogenin-2 antibodies and of appropriate size to be glycogenin-2. Shown in Fig. 8 are immunoprecipiates from cell extracts after treatment with ␣-amylase so that any interactions detected could not be mediated by glycogen itself. However, we obtained essentially the same results if the ␣-amylase digestion was omitted, presumably because most of the glycogenin-2 was free anyway. This result provides evidence for a physical association between glycogen synthase and glycogenin-2 in H4IIEC3 cells. DISCUSSION The most significant outcome of the present study is to recognize the existence of a novel glycogenin gene expressed in tissues that are important for glycogen metabolism. Liver is a major site of glycogen synthesis, and, in the heart, glycogen may have a special significance for the function of cardiac muscle. Expression of glycogenin-2 in the pancreas, although weaker, is also intriguing because of the connection between glycogen metabolism and blood glucose homeostasis. Although not all glycogenin-like species previously reported matched glycogenin-1 in every respect, there has been relatively little discussion of the presence of a separate gene. In liver, glycogenin-1 was found to be covalently linked to liver glycogen and its low abundance compared with skeletal muscle was rationalized by the fact that liver glycogen molecules are much larger (16). Ercan et al. (17) had likewise identified a self-glucosylating species of the correct size to be glycogenin-1 but had questioned whether glycogenin-1 was involved in the major pathway of liver glycogen synthesis. Interestingly, Calder and Geddes (31) did describe a protein of 60,000 daltons that was released from rat liver glycogen after digestion. If the amino acid composition analysis of this protein was correct, however, it is unlikely to be glycogenin-2 (or glycogenin-1) since it was extremely rich in Ser and Gly, which accounted for 43% of the residues. Glycogenin-2␣ has a Ser ϩ Gly content of 16%. Thus, the significance of the Calder and Geddes protein is unknown. A substantial fraction of glycogenin-2 is not covalently attached to glycogen and would not have been purified by methods aimed at isolating covalently linked proteins. On the other hand, one might have expected to detect glycogenin-2 as a self-glucosylating protein in experiments such as those of Ercan et al. (17). The assay of Ercan et al. (17) used 5 M UDP-glucose, commensurate with the low K m of glycogenin-1 for the nucleoside diphosphate sugar, so that conditions would not have been optimal for the higher K m glycogenin-2. However, we have no certain explanation of why Ercan et al. (17) failed to detect glycogenin-2. Krisman and colleagues (32,33) have described multiple glycosylated heart proteins, including one of M r ϳ60,000 whose formation is inhibited by Mn 2ϩ and which they associate with an activity termed "glycogen initiator 2." It is not clear, however, whether this species was considered capable of supporting heart glycogen synthesis independent of other self-glucosylated species of 38,000 and 42,000 daltons, of similar size to glycogenin-1. Detection of a second glycogenin gene has been slow to emerge in the sequence data bases since the glycogenin-2 message is simply not well represented in the libraries that are the subject of mass sequencing projects. For example, the number of human and mouse glycogenin-1 ESTs exceeds 100 and is growing steadily, whereas we are aware of only about 20 ESTs (representing 12 clones) of glycogenin-2, most of which contain only the 3Ј-noncoding sequence.
Glycogenin-2 has strong sequence similarity to glycogenin-1. Recombinant protein expressed in heterologous systems is active as a self-glucosylating protein, and it can serve as a substrate for purified glycogen synthase. The protein can be detected in rat hepatoma cells and in rat liver. Furthermore, a portion of the glycogenin-2 can only be detected after treatment with ␣-amylase. Glycogenin-2 and glycogen synthase co-immunoprecipitate from H4IIEC3 cells. Therefore, glycogenin-2 has all of the properties that would be expected for it to function as an initiator of glycogen synthesis, and we propose that it functions in this capacity in vivo. This proposal would force consideration of two new concepts regarding glycogen synthesis. First, at least in hepatoma cells, both glycogenin-1 and glycogenin-2 appear to be simultaneously involved in glycogen synthesis since both are released from glycogen by ␣-amylase. Furthermore, the corresponding glycogen molecules were separable by chromatography on concanavalin A-Sepharose. One interpretation would be the the existence of two populations of glycogen associated with the different glycogenin isoforms. Further work is needed to in-FIG. 8. Co-immunoprecipitation of glycogenin-2 and glycogen synthase from H4IIEC3 cell extracts. Immunoprecipitation from cell extracts was performed as described under "Experimental Procedures" using guinea pig preimmune serum (C), anti-glycogenin-2 (GN-2), or anti-glycogen synthase (GS) antibodies. Cell extract (30 g), not subjected to immunoprecipitation, was also run as a positive control. After SDS-PAGE and transfer to nitrocellulose filters, the samples were probed with anti-glycogen synthase (A) or anti-glycogenin-2 (B) antibodies.
vestigate this suggestion. In muscle, glycogen synthase and glycogenin-1 are found in approximately equimolar amounts (34), whereas, in liver, this ratio is very much in favor of glycogen synthase. Our work suggests that a major proportion of the glycogenin bound to liver glycogen synthase is likely to be glycogenin-2. Second, unlike glycogenin-1 in muscle extracts (35), most glycogenin-2 was present as a low molecular weight species even when glycogen was present. Free glycogenin-1 in fasted rat liver had been reported previously by Ercan et al. (17). The presence of significant amounts of free glycogenin-2 could be explained if only a small proportion of the molecules at any given time are recruited to glycogen biosynthesis. Future work should address whether the amount of free glycogenin varies under different physiological conditions. Alternatively, there could be "recycling" whereby the glycogenin is removed from the glycogen at some stage in the biosynthesis (discussed in Refs. 6 and 18) and the glycogenin bound to glycogen would represent the small fraction not yet released from the polysaccharide. This hypothesis would imply that most of the glycogen molecules have no attached glycogenin-2. Interestingly, Stetten and Katzen (36) many years ago reported the presence of reducing groups in liver glycogen, and it is likely that, unlike in muscle, most liver glycogen molecules are not covalently linked to protein (see discussion by Gannon and Nuttall (18)). An interesting consequence of such a model would be to predict the occurrence of a cleavage reaction to release the glycogenin. It will be of interest to test whether such an enzymic activity exists in liver.
The glycogenin-2 sequence is 72% identical to glycogenin-1 over the NH 2 -terminal 200 residues. Like glycogenin-1 and the yeast Glg proteins, it is acidic. Conserved domains I-IV, as defined by Roach and Skurat (6), are easily discerned. There is also a short region of strong similarity at the COOH terminus with glycogenin-1 and the CeGlg-c sequence, containing the motif WEXXXDY(M/L) (Fig. 2). In Fig. 2, we have shown this motif in alignment with the WEXXDYL sequence of Glg1p and Glg2p (domain VII of Ref. 6). In the Glg proteins, the extreme COOH terminus may have a function in interactions with glycogen synthase (19,21). Glycogenin-2 has a Tyr in correspondence to rabbit muscle Tyr-194, the site of self-glucosylation, even though overall this is not a region of particularly high sequence conservation in different glycogenin and glycogenin-like proteins (6).
The multiplicity of the glycogenin-2 cDNAs analyzed so far suggests that expression of this gene could be more complex than that of the glycogenin-1 gene. From the latter, a single 1.8-kb transcript appears to account for glycogenin-1 in most tissues of rabbit (2.4 kb in human tissues). A second transcript, cloned from keratinocoytes (accession number X79537), has a 220-base pair deletion, with respect to the muscle cDNA, that changes the reading frame after residue 202 and leads to a shorter protein. A cDNA with a small insert has also been identified. 3 For glycogenin-2, the cDNAs analyzed define as many as six different isoforms, and the potential exists for substantial diversity generated by alternate splicing. It is premature, however, to conclude that all forms are physiological and some might be artifactual or allelic forms. Several cDNAs came from a brain library, but, if these forms are expressed in brain, the level must be relatively low based on Northern analysis. We are also not certain as to whether all three liver glycogenin-2 isoforms are expressed in that tissue. The three differ by relatively small deletions close to the 5Ј of the cDNAs, and so the fact that the Northern analysis gave no evidence for multiple mRNAs is not informative since such small size differences would not be resolved. Likewise, Western analysis of the proteins is not definitive. The available antibodies are not isoform-specific, and the size differences among isoforms are quite small, with glycogenin-2␣ no more than 4 kDa larger than the other forms. For the most part, the different potential splice variants would provide alternative NH 2 and COOH termini attached to a common 335-residue segment containing the region with homology to other self-glucosylating proteins and implicated in catalysis. Whether or not all would self-glucosylate is difficult to predict. We have shown that recombinant glycogenin-2␥ was inactive, but it is difficult to interpret such a failure since minor changes in sequence can sometimes affect whether a functional protein is produced in bacteria. Although a role in some form of glycosyl transfer reaction is the most logical expectation for the various glycogenin-2 isoforms, it is possible that not all isoforms are involved in glycogen metabolism.
With the original identification of muscle glycogenin, it was attractive to propose regulation at the initiation step of glycogen biosynthesis, as happens for the synthesis of many other biopolymers. To date, there is relatively little evidence for such regulation. In particular, the idea that nutritional cues might set the glycogen storage capacity of cells by regulating the glycogenin level has not been substantiated. Now, the discovery of glycogenin-2 requires that some aspects of glycogen biosynthesis and its control be reevaluated, especially in liver and heart. Of specific interest would be any hormonal controls such as by insulin or glucagon. Nutritional control of gene expression is common in liver, and we can ask again whether longer term controls of glycogen storage could be exerted through the levels of glycogenin-2 expression. Finally, the metabolism of glycogen is so closely linked to blood glucose homeostasis that any gene acting positively in the pathway of glycogen deposition is a candidate to be impaired in non-insulindependent diabetes mellitus. Thus, glycogenin-2 will be examined in this context and cloning of the glycogenin-2 gene is in progress.