Identification of Novel Glycogen Synthase Kinase-3β Substrate-interacting Residues Suggests a Common Mechanism for Substrate Recognition*

Substrate recognition and specificity are essential for the reliability and fidelity of protein kinase function. GSK-3 has a unique substrate specificity that requires prior phosphorylation of its substrates. However, how the enzyme selects its phosphorylated substrates is unknown. Here, we combined in silico modeling with mutagenesis and biological studies to identify GSK-3-substrate interaction sites located within its binding cleft. Protein-protein docking of GSK-3β and the phosphorylated cAMP responsive element binding protein (pCREB) (using the available experimentally determined structures), identified Phe67, Gln89, and Asn95 of GSK-3β as putative binding sites interacting with the CREB phosphorylation motif. Mutations of these residues to alanine impaired GSK-3β phosphorylation of several substrates, without abrogating its autocatalytic activity. Subsequently, expression of the GSK-3β mutants in cells resulted in decreased phosphorylation of substrates CREB, IRS-1, and β-catenin, and prevented their suppression of glycogen synthase activity as compared with cells expressing the wild-type GSK-3β. Our studies provide important additional understanding of how GSK-3β recognizes its substrates: In addition to prior phosphorylation typically required in GSK-3 substrates, substrate recognition involves interactions with GSK-3β residues: Phe67, Gln89, and Asn95, which confer a common basis for substrate binding and selectivity, yet allow for substrate diversity.

Glycogen synthase kinase 3 (GSK-3) 4 is a ubiquitous serine/ threonine kinase expressed as two isoforms (␣ and ␤) (1), and has been implicated in many biological processes, including glucose metabolism, cell apoptosis, and embryonic develop-ment (reviewed in Refs. [2][3][4]. The cellular activity of GSK-3 is stringently controlled in response to growth factors and hormones. However, unlike most protein kinases, GSK-3 is constitutively active in resting cells and becomes inhibited upon stimulation of the cells. This inhibition is achieved through direct phosphorylation of N-terminal serine residues (Ser 21 or Ser 9 in ␣, ␤, respectively) by several protein kinases, such as PKB, p90RSK, PKA, and PKC (2)(3)(4). GSK-3 also may be phosphorylated on Tyr 216 located in the activation loop (5). This phosphorylation is an autophosphorylation event as demonstrated by in vitro and in vivo cell systems (5)(6)(7).
Elevated activity of GSK-3 is associated with several diseases, including type 2 diabetes, neurodegenerative diseases, and affective disorders (8 -10). Hence, selective inhibitors of GSK-3 may be of therapeutic value and are currently under extensive development (11)(12)(13)(14). Thus, understanding of how GSK-3 interacts with its substrates may pave the way for design and development of new specific substrate competitive GSK-3 inhibitors.
Substrate specificity of protein kinases is a fundamental determinant for the integrity and fidelity of signaling pathways. Previous studies formulated consensus sequences for optimal phosphorylation motifs of protein kinases using oriented peptide libraries, bioinformatics, and computational molecular modeling (15)(16)(17)(18)(19). The increasing number of three-dimensional structures of protein kinases complexed with substrates had provided an important basis for understanding the mechanism of molecular recognition (20 -23). Still, our knowledge of the mechanisms by which protein kinases recognize their substrates is rather limited. GSK-3 has a unique substrate specificity that requires prior phosphorylation of its substrates in the context motif SXXXS(p), where S(p) is the phosphorylated "priming" site (24 -26).
The three-dimensional structure of GSK-3␤ showed that three basic residues within the catalytic core, Arg 96 , Arg 180 , and Lys 205 , form a positive pocket that most likely serves as the docking site for the phosphorylated moiety of GSK-3 substrates (27)(28)(29). GSK-3␤-binding of Axin and APC was localized to a hydrophobic site in the C-terminal helical domain (29,30). This interaction site, however, is downstream from the catalyticsubstrate binding cleft and is not directly involved in the phosphorylation process (29). Therefore, detailed knowledge of the interactions between the substrate amino acids near the priming phosphorylation site and GSK-3␤ is still desirable.
In this study, we sought to determine the important substrate recognition sites in GSK-3␤ by combining in silico proteinprotein docking of GSK-3␤ and the phosphorylated cAMP responsive element-binding protein (pCREB), based on the crystal structures of GSK-3␤ (28,31) and the NMR structure of pCREB (32), with biological tools. We present a model structure of the ternary complex of GSK-3␤, ATP, and the pCREB peptide. The docking model identified specific electrostatic and hydrophobic interactions between pCREB and three amino acids in GSK-3␤. Mutagenesis of these sites impaired GSK-3 ability to phosphorylate CREB, confirming their importance for substrate recognition. Importantly, additional GSK-3 substrates were affected by the mutations as well. Hence, our studies identified novel GSK-3␤ sites involved in recognition of diverse substrates, and provide important data for rational drug design of compounds targeting GSK-3.
Molecular Modeling-The available x-ray structures of GSK-3␤ (28,31), one with phosphorylated Tyr 216 (PDB (Ref. 34) code 1o9u) and another with bound non-hydrolysable analog of ATP, ANP, (PDB code, 1pyx), were used to model the structure of a phosphorylated GSK-3⅐ATP complex (using the homology module of InsightII, Accelrys, San-Diego, CA). A model of the ternary complex GSK-3⅐ATP⅐CREB was obtained by protein-protein docking, using the program MolFit. The p9CREB fragments (residues 127-135) from the 17 NMR models of pCREB in complex with the co-activator CBP (32) (PDB code, 1kdx) present some backbone variation in the turn region. We selected 3 variants of the p9CREB fragment, with different backbone and side chain conformations. In the first proteinprotein docking step, we docked each fragment to the GSK-3⅐ATP model structure, employing the geometric (35), weighted-geometric (36), geometric-electrostatic (37), and geometric-hydrophobic (38) options in MolFit. Standard translation and rotation grid intervals were used (1.05 Å and 12°, respectively). The surface grid points that belong to the side chains of residues Arg 96 , Arg 180 , and Lys 205 of GSK-3 (the primed phosphate binding site) were up-weighted in the weighted geometric docking search, thereby biasing the docking results to include more models in which these residues interact with any residue of p9CREB. The lists of solutions from the four individual docking scans were intersected. Thus, the final list of models included only models that appeared in all 4 docking searches, and each model was evaluated by a weighted-geometric-electrostatichydrophobic complementarity score, which is the sum of the (weighted-geometric) ϩ (geometric-electrostatic Ϫ geometric) ϩ (geometric-hydrophobic Ϫ geometric) scores.
The first docking step clearly preferred one of the 3 conformers of p9CREB (see "Results" and "Discussion"). In a second docking step, the preferred conformer was extended by including also the N-terminal helix (N-pCREB; residues 119 -135) and docked to GSK-3␤⅐ATP. The best GSK-3␤⅐ATP⅐N-pCREB model, which was very similar to the preferred GSK-3␤⅐ATP⅐p9CREB model, was refined by testing small local relative rotations of the molecules (Ϯ2°, Ϯ4°, and Ϯ6°) and searching for the best shape and chemical complementarity. Next, we superposed each of the 17 NMR structures of pCREB onto the refined GSK-3␤⅐ATP⅐N-pCREB model, using the common C␣ atoms. This showed that although there is a large variation in the positions of the C-terminal helices of pCREB, only a few of them interact with GSK-3␤. We selected a model with only few clashes with GSK-3␤ for the final modeling step, which consisted of 20 iterations of intermittent energy minimizations of the GSK-3␤⅐ATP⅐pCREB ternary complex and dynamics simulations (10,000 steps of 1fs in each iteration). These computations also included a layer of water molecules around the complex (10-Å thick). We used the Discover-3 module in the InsightII package (Accelrys Inc., San Diego, CA) for these simulations, employing the CVFF force field.
Plasmids and Mutants-We previously described Histagged GSK-3␤ construct and GSK-3␤ in pCMV4 plasmid (6). GSK-3␤ fused to an N-terminal FLAG-tag was initially cloned into pCMV-Tag 2B (Stratagene) in ECoRV and BamHI1. These 3 expression vectors were used as templates for mutagenesis of GSK-3 by the QuikChange site-directed mutagenesis kit (Stratagene) to replace Phe 67 , Gln 89 , Asn 95 , Glu 97 , and Ser 66 to alanine. All constructs were sequenced to confirm the presence of mutations. The sequences of mutagenic oligonucleotides are available from the authors upon request. pCREB-EGFP plasmid was purchased from BD Biosciences Clontech (Palo Alto, CA). pCMV4IRS-1 plasmid was described (33). GFP-␤-catenin was kindly provided by Dr. Rina Abersfeld from Tel Aviv University.
In Vitro Kinase Assays-The GSK-3␤ mutant proteins purified from bacteria or prepared from HEK-293 cells were incubated with indicated substrates (200 M) in a reaction mixture (50 mM Tris-HCl, pH 7.3, 10 mM Mg-Acetate, and 0.01% ␤-mercaptoethanol) together with [␥-32 P]ATP (100 M, 0.5 Ci/assay) for 20 min or as indicated in figure legends. Reactions were stopped, spotted on p81 paper (Whatman) washed with phosphoric acid, and counted for radioactivity, as described (6). For CREB phosphorylation, CREB-EGFP was immunoprecipitated with anti-GFP antibody MBL (Woburn, MA) in complex with protein A-Sepharose. GSK-3 proteins were added to the immunoprecipitates under conditions similar to those described above. In "hot" assays, the reactions were boiled with SDS sample buffer and subjected to gel electrophoresis. In cold assays, the reactions were boiled with SDS sample buffer and subjected to immunoblot analysis with ␣pCREB 129/133 . Similar experiments were performed in autophosphorylation assays except that the substrate was omitted, and ATP concentration was raised to 300 M.
In Silico Molecular Modeling of the Ternary Complex GSK-3⅐ATP⅐pCREB-The structure of the ternary complex GSK-3␤⅐ATP⅐CREB was modeled as described under "Experimental Procedures." We used the structures of the non-ATP-bound GSK-3␤ (28) and the ANP-bound GSK-3␤ (31) to model the complex between phosphorylated GSK-3␤ and ATP. The two structures differ only slightly in the position of the P-loop, but not in the substrate binding cavity. Hence, the phosphorylated GSK-3␤⅐ATP model was constructed by combining these two structures; the conformation of the P-loop backbone and side chains was as in the ANP-bound GSK-3␤, and the conformation of Tyr 216 (p) was as in the phosphorylated GSK-3␤.
Currently, there is only one available structure of pCREB (32) bound to the KIX domain of CBP. pCREB consists of two helices (N-and C-terminal) joined by a loop, and the 17 NMR models of pCREB present considerable variability in the relative position of the N-and C-terminal helices. Our protein-protein docking program MolFit treats the docked molecules as rigid bodies and cannot cope with such structural flexibility. Therefore, we docked pCREB to GSK-3␤ in steps. First, we docked three representing conformers of p9CREB (residues 127-135) to GSK-3⅐ATP employing a weighting scheme that emphasized contacts involving the primed binding site of GSK-3␤, but did not specify the binding partner (see "Experimental Procedures"). Thus, it was rewarding to find that the first docking step clearly preferred one of the three conformers of p9CREB, and ranked near the top a model in which the primed binding site was occupied by CREB-Ser 133 (p) (rank 31 out of 9,415 models obtained after intersection; see "Experimental Procedures"). The same model was obtained in the second docking step in which a larger fragment of pCREB (N-pCREB; see "Experimental Procedures") was docked to GSK-3⅐ATP.
In the GSK-3␤⅐ATP⅐pCREB model the substrate phosphorylation site, CREB-Ser 129 , was found at a distance of 5 Å from the ␥-phosphate of ATP, adequate for phosphorylation. This was a plausible starting structure for the third modeling step, which consisted of energy minimization of the GSK-3⅐ATP⅐pCREB (residues 119 -146) model that included the C-terminal helix of pCREB as well, all soaked in water. The list of interactions in the final energy-minimized model is given in Table 1. In addition to the interactions between the Glycogen Synthase Kinase-3␤ Substrate Recognition phosphorylation site and the primed site of pCREB and GSK-3⅐ATP mentioned above, we found that CREB-Arg 135 (within the p9CREB fragment) forms hydrogen bonds with Asn 95 and Gln 89 of GSK-3␤ and CREB-Tyr 134 interacts with Phe 67 of the kinase and also makes a hydrogen bond to the ATP ␥-phosphate. The interactions of p9CREB or pCREB with GSK3␤ are shown in Fig. 1B.
The amino acids Gln 89 and Asn 95 are good candidates for substrate binding. They are conserved preferentially in GSK-3␤, but not in other protein kinases, including GSK-3 paralogs (Fig. 1C). In addition, these amino acids are positioned away either from the conserved ATP binding pocket or the activation segment, and they are surface-exposed. The amino acid Phe 67 , on the other hand, is more generally conserved and located in the conserved P-loop, which binds ATP, (41). Phe 67 is not directly involved in ATP binding (according to Ref. 31), but it may be important for stabilizing the conformational change in the P-loop brought about by ATP binding. In addition, however, Phe 67 is surface exposed and in our model it points toward  , 1o9u, 1jst, 2erk, and 2cpk, respectively). The numbering corresponds to that of GSK-3. Potential residues for substrate recognition are denoted by yellow or green (identical residues) background. It shows that Gln 89 and Asn 95 are preferentially conserved in GSK-3. Conserved residues in all sequences are denoted by red letters; residues conserved in two or three sequences are denoted by blue letters. The residues mutated in this study are marked with an asterisk. D, ability of GSK-3 to phosphorylate p9CREB variants, was examined in assay conditions similar to those described in A (and under "Experimental Procedures"). Results are presented as the percentage of the phosphorylation obtained with p9CREB as a substrate that was set to 100%, and are mean of three independent experiments.  OCTOBER 13, 2006 • VOLUME 281 • NUMBER 41 the substrate binding cavity, making it a potential hydrophobic contact with substrates.

Glycogen Synthase Kinase-3␤ Substrate Recognition
In Vitro-To test the molecular modeling results, we replaced Arg 9 in p9CREB (corresponding to Arg 135 ) by lysine or alanine, and Tyr 8 (corresponding to Tyr 134 ) by phenylalanine. In vitro phosphorylation assays showed reduced ability of GSK-3␤ to phosphorylate R9K and Y8F. Replacement of R 9 by alanine completely abolished phosphorylation (Fig. 1D). These results supported the model structure of GSK3⅐ATP⅐CREB described above, indicating that the hydrogen bonding interactions between Arg 9 in p9CREB and GSK-3␤ are essential for substrate binding. Similarly, the hydrogen bonding interactions of Tyr 8 have a role in substrate recognition.
Expression of GSK-3 Mutants in Escherichia coli and Their Autocatalytic Activity-To explore the role of residues Phe 67 , Gln 89 , and Asn 95 in GSK-3-substrate recognition, we generated GSK-3␤ mutants and investigated their ability to phosphorylate substrates using in vitro and cellular systems. GSK-3 mutants in which residues Phe 67 , Gln 89 , or Asn 95 were replaced by alanine were generated and expressed as His-tagged proteins (termed here Q89A, N95A, and F67A). Three additional mutants were used as controls. A kinase-dead mutant KK85,86MA in which ATP binding was prevented (6), termed here KI, E97A mutant in which the conserved glutamic acid located in the ␣C-helix participating in phosphoryl transfer (42) was replaced by alanine, and S66A mutant in which Ser 66 was replaced by alanine. Ser 66 is located near the enzyme-substrate binding site; however its hydrogen bond interaction with the substrate, as predicted by the model, is solvent-exposed and is not likely to contribute significantly to the binding (Fig. 1B and Table 1).
Expression of purified GSK-3␤ proteins was determined by Coomassie-stained gels or by immunoblot analysis using the GSK-3␤ antibody (Fig. 2, A and B). Both analyses confirmed that all proteins were expressed at similar levels. Phosphorylation of GSK-3␤ at Tyr 216 reflects its autocatalytic activity (7). Thus, Tyr 216 phosphorylation levels were determined by immunoblot analysis using the anti-phospho-GSK-3-Tyr 216 antibody. Results show that, similar to wildtype, Q89A, N95A, F67A, and S66A mutants were tyrosinephosphorylated. KI and E97A were not phosphorylated on Tyr 216 (Fig. 2A). These results indicated that mutation at Gln 89 , Asn 95 , Phe 67 , and Ser 66 did not impair the intrinsic catalytic activity of the enzyme, whereas mutation at Glu 97 resulted in an inactive mutant. The ability of F67A to autophosphorylate suggested that Phe 67 is not critical for ATP binding as indeed previously indicated (31).

Phosphorylation of CREB Peptide by GSK-3 Mutant Proteins-
The following experiments examined whether F67A, Q89A, and N95A mutant proteins can phosphorylate p9CREB. GSK-3␤ proteins were subjected to in vitro kinase assays using p9CREB as the substrate, and the amount of incorporated phosphorylation into the peptide was determined. As shown in Fig.  2B, Q89A displayed reduced phosphorylation ability toward p9CREB, as compared with WT GSK-3 (about 75% reduction),

FIGURE 2. Expression of GSK-3 mutants in E. coli.
A, His-tagged wild-type GSK-3 and GSK3 mutant proteins (F67A, S66A, Q89A, N95A, KK85, 86MA, E97A) were purified by affinity chromatography on Talon resin, as described under "Experimental Procedures." Total protein from each purification was subjected to gel electrophoresis and either stained with Coomassie Blue (upper), or subjected to immunoblot analysis with ␣GSK-3␤ antibody or ␣pY 216 GSK-3 as indicated. B, phosphorylation of p9CREB with GSK-3 proteins was performed as described in the legend to Fig. 1A under "Experimental Procedures." The phosphorylation of p9CREB by mutant GSK3 proteins was expressed as a percentage of that obtained with wild-type which was set to 100%. The results are the mean of six independent experiments Ϯ S.E. Results were statistically significant at p Ͻ 0.01.

TABLE 1 The interactions between GSK-3⅐ATP and pCREB as seen in our final model structure
The interactions of p9CREB are emphasized in bold font.

In Vivo Expression of GSK-3 Mutants in Cells and
Phosphorylation of Substrates-It was important to demonstrate that the defected phosphorylation ability of GSK-3 mutants observed in vitro also occurs in vivo. For that, wild-type and Q89A, N95A, F67A, KI, and E97A cDNA expression vectors were transiently transfected in HEK-293 cells. FLAG-tagged or non-tagged GSK-3 constructs were used initially. All mutant proteins were expressed at similar levels and FLAG-tagged proteins were readily distinguished from endogenous GSK-3␤ (Fig.   3A). Immunoblot analysis of the same samples with the antiphospho-Tyr 216 antibody revealed that both tagged and nontagged GSK-3 mutant proteins Q89A, N95A, and F67A were tyrosine-phosphorylated, whereas E97A and KI displayed (as expected) a very weak signal (Fig. 3A). These results indicated that mutations at Gln 89 , Asn 95 , and Phe 67 did not abrogate the catalytic activity, because tyrosine phosphorylation of GSK-3 is mainly an intramolecular autophosphorylation process (6,7). This conclusion is further supported by the observation that the inactive mutants were not tyrosine-phosphorylated (Fig. 3A).
Because overexpression of GSK-3 proteins was considerably above endogenous GSK-3 (Fig. 3), we chose to perform our experiments with the non-tagged GSK-3 constructs. The ability of GSK-3 mutants to autophosphorylate was examined in in vitro kinase assays. GSK-3 proteins were partially purified from cell extracts by ion exchange chromatography. The enzymes

. Expression of GSK-3 mutants in cells.
A, cells were transiently transfected with DNA constructs expressing wild-type GSK-3, Q89A, N95A, F67A, KI, and E97A mutant proteins, as described under "Experimental Procedures." Cell extracts were subjected to Western blot analysis with either ␣GSK-3␤ antibody or ␣pTyr 216 GSK-3 antibody, as indicated. Control (C) represents extracts from cells expressing the empty vector. Left panel shows results obtained with non-tagged GSK-3␤, and right panel presents results obtained with FLAG-tagged GSK-3␤. B, GSK-3␤ mutants were partially purified from expressing cells, as described under "Experimental Procedures," and were incubated for 20 min at 30°C in the presence of [␥-32 P]ATP. Reactions were subjected to gel electrophoresis, and autoradiographed. Indicated is phosphorylated GSK-3␤. C, phosphorylation of peptide substrates by GSK-3 mutant proteins. Equal amounts of partial purified GSK-3 proteins prepared as described under "Experimental Procedures" were subjected to in vitro kinase assays with p9CREB, PGS1, and pIRS-1 peptide substrates as described in the legend to Fig. 1A under "Experimental Procedures." 32 P incorporation into substrates was determined, and the activity obtained from control non-transfected cells was subtracted from each sample. Results present the percentage of substrate phosphorylation obtained with wild-type GSK-3 which was set to 100%, and are mean of five independent experiments each performed in duplicates. Expression levels of GSK-3 used in the assays are shown. Results were statistically significant (p Ͻ 0.01 mutant versus WT). For A and B shown representative gel of three independent experiments.
were incubated with [␥-32 P]ATP, and phosphorylation was detected by gel electrophoresis (Fig. 3B). All three mutants were able to autophosphorylate supporting the view that their catalytic activity is not impaired. The ability of GSK-3 mutants to phosphorylate p9CREB substrate was tested next. As shown in Fig. 3C, Q89A and N95A displayed reduced phosphorylation toward p9CREB as compared with the wild-type, and F67A completely failed to phosphorylate the substrate. To further examine whether these findings are general, two additional substrates were used: PGS-1, a phosphorylated peptide derived from a GSK-3-substrate, glycogen synthase (24,26), and pIRS-1, a phosphorylated peptide based on GSK-3-phosphorylation sequence in insulin receptor substrates-1 (IRS-1) (33). Similar results were obtained: Q89A and N95A displayed reduced phosphorylation toward PGS-1 and pIRS-1 substrates (about 50%), as compared with the wild-type, and F67A did not phosphorylate any of these substrates. These results suggested that Gln 89 and Asn 95 are indeed important determinants for substrate recognition by GSK-3. We noted that the phosphorylation ability of cellular-expressed N95A differs from the corresponding bacterially expressed mutant that was unable to phosphorylate p9CREB (Fig. 2B). We cannot provide a full explanation for this observation, although it is quite possible that post-translational modifications that do not occur in prokaryotes improved the ability of N95A to interact with the substrate. In any event, the results obtained from both expression systems (i.e. bacteria and mammalian cells) indicated that Gln 89 and Asn 95 play important roles in the substrate recognition mechanism. Phe 67 , which is conserved among protein kinases, may also have a general role in kinase function, such as stabilization of the correct conformation for ATP binding. We cannot distinguish between these two roles, namely substrate binding (as predicted by the model) and conformational stabilization of the P-loop.

Phosphorylation of Cellular CREB by GSK-3 Mutant Proteins-
The phosphorylation of cellular CREB by GSK-3 mutants was tested next. GSK-3 mutants were co-transfected with EGFP-CREB in HEK-293 cells. Notably, GSK-3 requires pre-phosphorylation of CREB at Ser 133 (40). Therefore, cells were first treated with forskolin to activate PKA phosphorylation of CREB. The phosphorylation of CREB at the GSK-3 phosphorylation site Ser 129 was determined by immunoblot analysis using a specific anti-phospho-CREB antibody that recognizes CREB phosphorylated at both Ser 129/133 . Expression of wildtype GSK-3 increased phosphorylation of CREB Ser 129/133 significantly. However, expression of Q89A or N95A mutants resulted only in very weak CREB phosphorylation at these sites (Fig. 4A); expression of F67A did not increase CREB phosphorylation, which was comparable to that observed in "control" cells expressing EGFP-CREB alone. Notably, phosphorylation of CREB at Ser 133 , as determined by a specific anti-phospho antibody, was unchanged in all samples, thus verifying that the changes observed with the "double" anti-phospho antibody reflected the changes in the GSK-3 phosphorylation site, Ser 129 . Hence, mutations at Phe 67 , Gln 89 , and Asn 95 impaired GSK-3 ability to phosphorylate cellular CREB.
To further investigate the above conclusion, we performed in vitro analyses with the "whole" protein CREB. In this experiment, EGFP-CREB was immunoprecipitated with anti-GFP antibody from overexpressing cells. The immunoprecipitate was incubated with GSK-3 wild-type and mutant proteins in hot or cold conditions. The hot reactions included [␥-32 P]ATP and were subjected to gel electrophoresis; 32 P incorporation into CREB protein was observed (Fig. 4B). The cold reactions were subjected to immunoblot analysis with the anti-phospho-CREB 129/133 antibody (Fig. 4C). Results showed that CREB protein was barely phosphorylated by Q89A, N95A, and F67A.

The Effect of GSK-3 Mutant Proteins on Glycogen Synthase Activity and Phosphorylation of IRS-1 and ␤-Catenin Substrates-
The following experiments examined how GSK-3 mutants affect additional cellular substrates including glycogen synthase, insulin receptor substrate-1 (IRS-1), and ␤-catenin. Glycogen synthase (GS) is inhibited by GSK-3 via phosphorylation on a cluster of serine sites (43). We showed previously that overexpression of GSK-3 suppressed GS activity (6). Hence, it was possible to determine whether GSK-3 mutants suppress GS activity. We found that expression of wild-type GSK-3 suppressed GS activity by 50% Ϯ10 (Fig. 5A), the mutant Q89A suppressed the enzyme by only 27% Ϯ 8, and mutants N95A and F67A had no effect on GS activity (Fig. 5A). Thus F97A and N95A showed severely impaired ability toward suppression of GS activity.
In a different set of experiments, GSK-3 constructs were coexpressed with wild-type IRS-1 plasmid in HEK 293 cells. The GSK-3 phosphorylation site on IRS-1 is Ser 332 and its phosphorylation can be detected by a specific antibody (33). IRS-1 phosphorylation increases significantly when co-expressed with GSK-3. Co-expression with Q89A increased IRS-1 phosphorylation to a much lesser extent and co-expression with N95A did not lead to detected increased phosphorylation of IRS-1 (Fig.  5B). These results are in line with the results obtained in the GS phosphorylation experiments, namely that Asn 95 and Gln 89 do not play equivalent roles in affecting IRS-1 or GS phosphorylation.
Collectively, it appears that Asn 95 is more important than Gln 89 for GS and IRS-1 recognition. Interestingly, both GS and IRS-1 have aspartic acid residues in the position corresponding to Arg 9 of p9CREB (position ϩ2 from the primed site). Possibly the shorter side chain of aspartic acid, compared with arginine, interacts with Asn 95 and less with Gln 89 . The difference between the results obtained for substrate peptides and substrate proteins can be attributed to the greater flexibility of peptides as compared with the corresponding fragments within proteins, which enables the peptides to adjust and form better contacts with the GSK-3 mutants than the corresponding proteins.
␤-Catenin is a key downstream target in Wnt signaling pathway and a substrate of GSK-3 phosphorylated at Ser 33,37 and Thr 41 (44). Phosphorylation of Ser 45 in ␤-catenin by casein kinase-1 (CKI), serves as a priming site for subsequent phosphorylation by GSK3 (44,45). GSK-3 constructs were co-expressed together with GFP-␤-catenin plasmid in HEK-293 cells. Phosphorylation of ␤-catenin at GSK-3␤ phosphorylation sites was detected with specific anti-phospho ␤-catenin antibody as shown in Fig. 5C. Q89A and N95A were able to phosphorylate ␤-catenin albeit to a significantly lower extent as compared with wide-type GSK-3. Consistently, F67A was unable to phosphorylate ␤-catenin. Notably, position ϩ2 in ␤-catenin is occupied by a small polar side chain (serine), which is likely to make fewer interactions with GSK-3.
The similar results obtained for the various GSK-3 substrates suggested that a common mechanism controls substrates recognition. This mechanism involves interactions with Gln 89 and Asn 95 and possibly Phe 67 in addition to the interactions with the primed phosphate binding pocket. The ability to interact with diverse substrate motives is explained by the polar nature of Gln 89 and Asn 95 , which enables them to form hydrogen bonds with various polar/charged residues. As discussed above, the exact role of Phe 67 is not clear.
In conclusion, our studies gained important additional understanding of how GSK-3␤ recognizes its substrates. On one level, GSK-3␤ binds a phosphorylated substrate by a welldefined, positively charged pocket (27,28,46), which filters away non-phosphorylated substrates. On the second level, additional interactions are necessary to facilitate precise positioning of the substrate within the substrate binding pocket, with the target serine located next to the ATP ␥-phosphate. FIGURE 5. The effect of GSK-3 mutants on glycogen Synthase, IRS-1, and ␤-catenin. A, GS activity was measured in extracts prepared from cells expressing GSK-3 mutants. Results are expressed as activity ratio of reactions performed with low G6P or high G6P and are mean of three independent experiments Ϯ S.E. Control (C) represents GS activity ratio measured in cells expressing pCMV4 empty vector only. *, p Ͻ 0.01 WT or mutant versus control. B, cells were transiently transfected with GSK-3 and IRS-1 constructs. Equal amounts of protein aliquots prepared from cells were subjected to gel electrophoresis, followed by immunoblot analysis with ␣pIRS-1 Ser 332 , ␣IRS-1 antibody, or ␣GSK-3␤. Control (C) cells expressing IRS-1, and the empty vector (pCMV4) C, cells were co-expressed with GSK-3 and GFP-␤-catnein constructs, and cell extracts were prepared as described under "Experimental Procedures." Equal amounts of protein aliquots were subjected to gel electrophoresis, followed by immunoblot analysis with ␣p ␤-catenin antibody (Ser 33,37 and Thr 41 ) or ␣,␤-catenin antibodies as indicated. The expression of GSK-3 mutants in each sample is shown. Control (C) cells expressing GFP-␤-catenin and empty vector (pCMV4). For B and C, a representative gel of three independent experiments is shown. This is achieved by interaction with Gln 89 , Asn 95 , and possibly Phe 67 . We highlight the roles of Gln 89 and Asn 95 , which are preferentially conserved in GSK-3␤, and are polar residues that can participate in hydrogen bonding with various polar/ charged residues, allowing both tight binding and the ability to interact with a broad selection of substrates. Recognition, thus, combines the highly specific primed phosphorylation recognition, and moderately specific hydrogen bond interactions that together determine the substrate specificity toward the kinase.