Inhibitory Phosphorylation of Glycogen Synthase Kinase-3 (GSK-3) in Response to Lithium

Glycogen synthase kinase-3 (GSK-3) is a critical, negative regulator of diverse signaling pathways. Lithium is a direct inhibitor of GSK-3 and has been widely used to test the putative role of GSK-3 in multiple settings. However, lithium also inhibits other targets, including inositol monophosphatase and structurally related phosphomonoesterases, and thus additional approaches are needed to attribute a given biological effect of lithium to a specific target. For example, lithium is known to increase the inhibitory N-terminal phosphorylation of GSK-3, but the target of lithium responsible for this indirect regulation has not been identified. We have characterized a short peptide derived from the GSK-3 interaction domain of Axin that potently inhibits GSK-3 activity in vitro and in mammalian cells and robustly activates Wnt-dependent transcription, mimicking lithium action. We show here, using the GSK-3 interaction domain peptide, as well as small molecule inhibitors of GSK-3, that lithium induces GSK-3 N-terminal phosphorylation through direct inhibition of GSK-3 itself. Reduction of GSK-3 protein levels, either by RNA interference or by disruption of the mouse GSK-3β gene, causes increased N-terminal phosphorylation of GSK-3, confirming that GSK-3 regulates its own phosphorylation status. Finally, evidence is presented that N-terminal phosphorylation of GSK-3 can be regulated by the GSK-3-dependent protein phosphatase-1·inhibitor-2 complex.

Lithium salts have provided effective pharmacotherapy for bipolar disorder for decades (1,2); however, the underlying mechanisms of lithium action in psychiatric disorders are unknown (3)(4)(5). Lithium also has diverse effects in other settings, including glycogen synthesis, embryonic development, and synaptogenesis (4 -6). Identification of the molecular target of lithium action in these settings has been a focus of research over the past twenty years, and several potential targets have been identified. For example, lithium inhibits glycogen synthase kinase-3 (GSK-3), 1 and this has been proposed to explain the effects of lithium on embryonic development and glycogen synthesis (7). However, lithium also inhibits other targets, including inositol monophosphatase (8), other phosphomonoesterases structurally related to inositol monophosphatase (5,9), phosphoglucomutase (10), and possibly other enzymes. For many of the known effects of lithium, especially in the treatment of bipolar disorder, the relevant target(s) of lithium has not been defined. Thus, to establish a prospective candidate as the target of lithium in a given setting, new approaches that specifically interfere with the function of that candidate, including development of more specific inhibitors and loss-offunction mutations in the genes of interest, will be required, in addition to demonstrating in vivo inhibition of a putative target by lithium.
Inhibition of GSK-3 by lithium in vivo has been demonstrated in a number of systems, including cultured cells, Xenopus oocytes and embryos, and adult rat brain (11)(12)(13)(14)(15)(16)(17). In many of these studies, lithium activates pathways known to be inhibited by GSK-3. For example, GSK-3 inhibits glycogen synthase, and this is reversed by lithium (mimicking insulin action), most likely through direct inhibition of GSK-3 (18). Similarly, GSK-3 antagonizes the canonical Wnt signaling pathway, and lithium activates this pathway in cell culture and in intact embryos (11,12,19,20). The argument that these effects of lithium are caused by inhibition of GSK-3, rather than another target of lithium, is supported by loss-of-function mutations in GSK-3, which are phenocopied by lithium in diverse organisms, including Dictyostelium and mouse (21,22).
In addition, numerous alternative inhibitors of GSK-3 have been described recently, and where tested, these agents also mimic lithium effects on the Wnt pathway. Synthetic small molecule inhibitors include hymenialdisine, paullones, indirubins, and maleimide compounds, most of which are ATP competitors (23)(24)(25)(26)(27). However, many of these compounds inhibit protein kinases other than GSK-3. Polypeptide inhibitors of GSK-3 have also been described, including the protein frequently rearranged in advanced T-cell lymphoma/GSK-3-binding protein (FRAT/GBP) and peptides derived from the GSK-3␤-interacting domain (GID) of Axin (28,29). FRAT/GBP inhibits GSK-3␤ activity in vivo and is required for dorsal development in Xenopus embryos (28). Axin is a scaffold protein that binds multiple components of the canonical Wnt pathway, including GSK-3 and ␤-catenin, to facilitate GSK-3-mediated phosphorylation and degradation of ␤-catenin (30,31). In contrast to the function of full-length Axin, fragments of Axin that bind to GSK-3 (GID) inhibit GSK-3 activity in Xenopus oocytes (29). Although the mechanism of GSK-3 inhibition by FRAT/GBP or GID peptides is not known, these molecules interact with GSK-3␤ at a site remote from the ATP binding pocket, suggesting that the mode of inhibition is distinct from the small molecule inhibitors described above (32,33).
One intriguing consequence of lithium treatment is the increased phosphorylation of GSK-3 itself (see Refs. 25 and 34 -38 and data presented here). Several signaling pathways, including insulin, platelet-derived growth factor, fibroblast growth factor, epidermal growth factor, and reelin cause increased inhibitory phosphorylation of GSK-3 on either serine 21 (in GSK-3␣) or serine 9 (in GSK-3␤) (39). The ability of lithium to mimic this effect in cerebellar granule cells was reported to be caused by transient activation of Akt, but the direct target of lithium responsible for this regulation was not identified (34). Lithium-induced inhibitory phosphorylation of GSK-3 is especially interesting, as it provides additional, indirect inhibition of GSK-3 that could reinforce or amplify the well established direct inhibition of GSK-3 by lithium.
We have addressed whether lithium-induced serine 21/9 phosphorylation arises through direct inhibition of GSK-3 or through another lithium-sensitive target using GID peptides, small molecule GSK-3 inhibitors, and GSK-3 loss-of-function approaches. We have further characterized GID peptides to show that they inhibit GSK-3 in multiple mammalian cell types and that they are direct inhibitors of GSK-3 in vitro. We then show that increased serine 21/9 phosphorylation can be caused by GSK-3 inhibitors that function through several, distinct mechanisms and that reduction of GSK-3 expression by either RNA interference or GSK-3␤ knock-out also results in increased serine 21/9 phosphorylation, providing strong support for the autoregulation of GSK-3 activity. Finally, we propose that this autoregulation of GSK-3 is mediated in part through GSK-3 regulation of the protein phosphatase-1⅐inhibitor-2 complex.
Cell Culture and Transfection-293T cells, Neuro2A, and NIH3T3 cells were obtained from American Type Culture Collection and cultured as described elsewhere (43). Wild-type and GSK-3␤ knock-out mouse embryonic fibroblasts were gifts from Dr. James Woodgett (22). For transfections, cells were plated in 6-well dishes at a density of 2.5 ϫ 10 5 cells per well. Each plasmid was transfected as follows: 1 g of Axin, ⌬RGS, GID (or as dose-curve as indicated in the figures), or empty vector control (either pCS2ϩ or pCS2ϩMT (myc tag), indicated as pCS2 control in the figures), 0.25 g of tau T40/pSG5, 0.2 g of Lef-OT or Lef-OF, 10 ng of pRL-SV40 (Renilla luciferase), 0.5 g of dnTCF. All transfections included 0.3 g of pEGFP to assess transfection efficiency (typically 50 -80%). FuGENE 6 (Roche Applied Science) was used for transfecting 293T or Neuro2A cells. Firefly and Renilla luciferase activities were measured on a Monolight 3010 luminometer (Turner Designs) using the dual luciferase assay kit (Promega). For drug treatment experiments, cells were serum-starved overnight before addition of drugs at concentrations indicated in the figures and harvested after indicated period of time.
Immunofluorescent Staining-NIH3T3 cells grown on coverslips were co-transfected with 2 g of GID 380 -404 , GID(Leu 3 Pro) 380 -404 , full-length Axin, or pCS2ϩMT, together with 0.3 g of pEGFP, using LipofectAMINE Plus. 24 h after transfection, cells were fixed with chilled 4% paraformaldehyde in PBS for 15 min, rinsed with PBS, permeabilized in 1% Triton X-100/10% goat serum in PBS for 10 min, blocked in 0.1% Tween 20/10% goat serum in PBS for 30 min, incubated with ␤-catenin antibody diluted 1:200 in blocking buffer for 60 min at room temperature, washed with PBS, incubated with rhodamine-conjugated anti-mouse antibody (Jackson ImmunoResearch) diluted 1:200 in blocking buffer for 45 min at room temperature, washed with PBS, and then mounted onto slides with VECTASHELD mounting medium with DAPI (Vector Laboratories). Cells were viewed at 40-fold magnification with a Leica DMR fluorescence microscope.
In Vitro Kinase Assays-Recombinant GSK-3␤ (500 units/l) was purchased from New England Biolabs. GSK-3␤ assay was performed as described previously (7), except that MgCl 2 was 0.5 mM. GID 320 -429 -His was affinity-purified by nickel-nitrilotriacetic acid (Qiagen). GID 380 -404 and GID(Leu 3 Pro) 380 -404 peptides were chemically synthesized by the HHMI Structure Laboratory at the University of California, San Francisco, CA and were purified by high pressure liquid chromatography. FRATide was purchased from Upstate Biotechnology. Recombinant ␤-catenin protein purified from Sf9 cells was a generous gift from Barry Gumbiner. Recombinant inhibitor-2 was purchased from New England Biolabs, and tau was purified from bacteria as described (29). Prior to performing in vitro GSK-3 assays, we tested each substrate at multiple concentrations and chose a concentration within the linear range for the respective substrates. The final concentrations added to the reaction were as follows: ␤-catenin at 25 g/ml, tau at 12.5 g/ml, and I-2 at 25 g/ml. Tau protein was pre-phosphorylated with MAP kinase as described previously (45). Reactions were stopped by adding Laemmli sample buffer and boiling for 5 min. Samples were separated in 10% SDS-PAGE, followed by Coomassie Blue staining. The gel was dried and scanned with a STORM 840 PhosphorImager (Amersham Biosciences). The intensity of the bands was quantified with Image-Quant, version 1.0 (Amersham Biosciences). The stained gel was scanned to confirm equal loading of GSK-3 and substrates. PKA, p34 cdc2 /Cdk-1, p42 MAP kinase/Erk2, and casein kinase II assays were performed as described previously (46, 90 -92). 20 M H-89 was used for control as a PKA inhibitor, and 1 M staurosporine was used for control as a p34 cdc2 inhibitor.
PP-1 Assay-PP-1 assay was performed as described previously (47). Recombinant GSK-3␤ phosphorylated on serine 9 was used as substrate. I-2 (New England Biolabs) was added to a final concentration of 1 M. ATP was added to a final concentration of 0.2 mM. LiCl and GID 380 -404 were added at the concentrations indicated in the figures. Serine 9 phosphorylation of GSK-3␤ was detected by WB with phospho-GSK-3␤ (serine 9) antibody. GSK-3-mediated phosphorylation of I-2 was detected by autoradiography by adding [␥-32 P]ATP.

A Short GSK-3 Binding Peptide Mimics Lithium Action in
Mammalian Cells-We have shown previously that a construct encoding 268 amino acids derived from the GID of Axin (GID 277-545 ) mimics lithium action in Xenopus embryos and that a sequence as short as 25 amino acids (GID 380 -404 ) inhibits overexpressed GSK-3␤ in Xenopus oocytes (29). To test whether GID peptides inhibit endogenous GSK-3 activity in mammalian cells, we have transfected myc-tagged GID 380 -404 into mouse Neuro2A cells and assayed phosphorylation of the microtubule-binding protein tau, a well characterized substrate for GSK-3␤ (44). Overexpression of GID 380 -404 reduced tau phosphorylation, as determined by Western blotting with a phospho-tau-specific antibody (PHF-1) and by the increased electrophoretic mobility of tau, mimicking the effects of lithium (Fig. 1A). The dominant-negative form of Axin, ⌬RGS, also inhibited tau phosphorylation in Neuro2A cells. Mutation of leucine 390 to proline blocks interaction of Axin-GID with GSK-3 (40); GID 380 -404 containing this Leu 3 Pro mutation (GID(Leu 3 Pro) 380 -404 ) failed to inhibit tau phosphorylation, indicating that GID must interact with GSK-3␤ to inhibit tau phosphorylation. In contrast, full-length Axin did not inhibit GSK-3␤-mediated tau phosphorylation under these conditions (see also Ref. 29).
Inhibition of GSK-3␤ by Wnts or by lithium results in reduced phosphorylation and subsequent accumulation of ␤-catenin protein, and this has been widely employed as an indirect assay of GSK-3␤ activity (12,28,48,49). We therefore tested whether GID causes accumulation of ␤-catenin in Neuro2A cells. In contrast to full-length Axin, overexpression of GID 320 -429 or GID 380 -404 caused accumulation of ␤-catenin in a dose-dependent manner, mimicking the effect of lithium ( Fig.  1, B and C), and similar to observations in Xenopus oocytes (29). GID(Leu 3 Pro) 320 -429 or GID(Leu 3 Pro) 380 -404 did not cause ␤-catenin accumulation. GID-induced ␤-catenin accumulation was evident in both the cytoplasm and the nucleus, as visualized by immunofluorescence for ␤-catenin in NIH3T3 cells (Fig. 2, A and B). Thus GID mimics lithium in causing ␤-catenin accumulation in mammalian cells.
As ␤-catenin binds to TCF/Lef transcription factors to activate Wnt-responsive genes, we evaluated the ability of GID to activate Wnt-dependent transcription using a TCF/Lef luciferase reporter (Lef-OT) (41). GID 380 -404 activated Lef-OT in a dose-dependent manner, mimicking lithium, with a maximum activation over 200-fold (Fig. 2, C and D). Lithium-or GIDinduced Lef-OT activation was abolished by co-expression of a dnTCF that lacks the ␤-catenin binding site (42). Neither GID nor lithium activated the Lef-OF luciferase reporter in which the TCF binding sites are mutated (41). This suggests that activation of Lef-OT by lithium and GID is mediated specifically by ␤-catenin/TCF. These findings are similar to observations showing activation of TOPFLASH by a larger fragment encoding GID derived from mouse Axin (40). We have also found that a 23-residue GID peptide (amino acids 382-404 of Xenopus Axin) retains GSK-3 binding and potently activates Lef-OT activity, whereas removal of two additional residues from either end markedly reduces activation of Lef-OT (data not shown). Therefore short GID peptides mimic lithium in activating the canonical Wnt/␤-catenin signaling pathway in mammalian cells.
Thus inhibition of GSK-3␤ by GID peptides appears to depend on the substrate, but from the data presented so far it is not clear whether this is related to substrate sequence or the presence of a priming phosphate. Therefore tau protein prephosphorylated with MAP kinase was compared with nonprimed tau in the in vitro GSK-3␤ assay. As reported previously (45,55), pre-phosphorylation of tau protein by MAP kinase facilitated GSK-3␤-mediated phosphorylation of tau (Fig. 3B, compare lane 8 with lane 3). Inhibition of GSK-3␤ by GID was significantly less efficient when pre-phosphorylated tau was used as substrate (Fig. 3B, compare lane 11 with lane 6). However, a 5-fold higher concentration of GID inhibited phosphorylation of both non-primed and primed tau. This suggests that GID preferentially inhibits GSK-3␤ activity toward non-primed substrates but can also inhibit phosphorylation of primed substrates at higher concentrations.
To determine whether GID inhibits other serine/threonine protein kinases, we assayed PKA, p34 cdc2 /Cdk-1, casein kinase II, and MAP kinase in vitro in the presence of GID 320 -429 -His or GID 380 -404 peptide. PKA-mediated phosphorylation of kemptide was not inhibited by GID even at high concentrations but was effectively inhibited by the PKA inhibitor H-89 at 20 M (Fig. 3C) (56,57). Similarly, p34 cdc2 phosphorylation of histone-H1 was not inhibited by GID but was inhibited by staurosporine at 1 M (24). GID also did not inhibit the in vitro activity of casein kinase II or MAP kinase (data not shown). Therefore GID is not a general inhibitor of serine/threonine protein kinases.
GSK-3 Inhibitors Cause Increased Phosphorylation of GSK-3-Although lithium is a direct inhibitor of GSK-3, in vivo it also causes increased phosphorylation of GSK-3␤ on serine 9 (25, 34 -37), which inhibits GSK-3␤ activity (39, 59 -61). Various mechanisms have been proposed to explain this observation, including activation of Akt/protein kinase B (34,62). We have also observed robust and rapid (within 5 min) serine 9 phosphorylation after treatment with lithium in 293T, Neuro2A, and NIH3T3 cells (see Figs. 5 and 8). This effect of lithium could be mediated through inhibition of GSK-3 or another target of lithium. Therefore we tested whether alternative inhibitors of GSK-3 also cause increased phosphorylation of GSK-3␤ Ser9 . We reasoned that if GSK-3␤ Ser9 phosphorylation is increased by treatment with diverse agents that inhibit GSK-3 through distinct mechanisms, it is likely that this response is mediated through inhibition of GSK-3 per se. We found that, similar to lithium treatment, GID expression caused increased GSK-3␤ Ser9 phosphorylation in a dose-dependent manner in both 293T cells (Fig. 5C) and Neuro2A cells (data not shown). This is in contrast to full-length Axin, which reduced GSK-3␤ Ser9 phosphorylation (Fig. 5B) (63). GBP expression also caused increased phosphorylation of GSK-3␤ Ser9 (data not shown).
Reduced Expression of GSK-3 Leads to Increased N-terminal Phosphorylation-Because a number of GSK-3 inhibitors act-ing through at least three distinct mechanisms similarly increased GSK-3␣/␤ phosphorylation, the simplest explanation is that this phosphorylation is regulated by GSK-3 itself. However, the formal possibility remains that each of these inhibitors affects target(s) other than GSK-3 that all regulate GSK-3 phosphorylation in some manner. Therefore we tested the hypothesis that GSK-3 negatively regulates its own phosphorylation by reducing GSK-3 expression using RNA interference (RNAi) (69,70) or disruption of the mouse GSK-3␤ locus (22). Using siRNAs targeted specifically against GSK-3␣ or GSK-3␤, we reduced the protein level of GSK-3␣ or ␤, respectively (Fig.  7A). When GSK-3␣ was depleted, phosphorylation of GSK-3␤ Ser9 increased without a change in GSK-3␤ protein level (Fig.  7A, lane 3). In addition, the amount of phosphorylated GSK-3␣ did not change despite the reduced level of GSK-3␣ protein, indicating a relative increase in GSK-3␣ phosphorylation. Conversely, when GSK-3␤ was depleted (Fig. 7A, lane 4), phosphorylation of GSK-3␣ Ser21 increased (the relative phosphorylation

FIG. 4. Inhibition of GSK-3 activity by GID and GBP/FRAT. A,
293T cells were co-transfected with pCS2 control (C), GID 380 -404 (GID), or myc-tagged GBP at the concentrations indicated, together with tau, Lef-OT, and Renilla luciferase (pRL-SV40). The expression levels of GID 380 -404 and GBP were detected by WB with myc antibody (bottom panel), tau phosphorylation was detected by WB with PHF-1 antibody (p-tau; top panel), and total tau was detected by WB with T14/46 antibodies (tau; second panel). Endogenous GSK-3␤ protein levels detected by WB are shown as loading controls. B, lysates from samples shown in A were assessed for luciferase activity. -Fold activation represents normalized luciferase activity compared with pCS2 control. C, in vitro GSK-3␤ assay was performed using non-primed tau (12.5 g/ ml) as substrate. Phosphorylation of tau was detected by autoradiography and quantified as described under "Experimental Procedures." Synthetic GID 380 -404 or FRATide peptides were added at the concentrations indicated. The IC 50 for GID 380 -404 peptide is ϳ50 nM, and IC 50 for FRATide is ϳ160 nM. Data shown are representative of at least three independent experiments with similar results. of GSK-3␤ Ser9 also increased, when normalized to overall protein levels). Therefore, depletion of either isoform of GSK-3 increases N-terminal phosphorylation. Alternatively, GSK-3␤ expression is completely absent in mouse embryonic fibroblasts (MEFs) derived from mice in which the GSK-3␤ gene has been disrupted, but these cells still express GSK-3␣ (Fig. 7B) (22). In these GSK-3␤ Ϫ/Ϫ MEFs, serine 21 phosphorylation of GSK-3␣ is increased compared with GSK-3␣ in wild-type MEFs (Fig.  7B, GSK-3␤ ϩ/ϩ (lane 1) versus GSK-3␤ Ϫ/Ϫ (lane 4)). Therefore, reduction of GSK-3 activity either by RNAi or by gene disruption of a specific GSK-3 isoform leads to increased N-terminal phosphorylation of the complementary isoform. Furthermore, overexpression of the constitutively active GSK-3␤ S9A mutant causes decreased serine 9 phosphorylation of endogenous GSK-3␤ (71). Taken together with the effects of GSK-3 inhibitors such as GID and lithium, these data strongly support a role for GSK-3 in the autoregulation of serine 21/9 phosphorylation.
Consistent with previously reported data in cerebellar granule cells (34), lithium caused increased phosphorylation of Akt on serine 473 (associated with activation of Akt) in Neuro2A cells (Fig. 8B). However, this lithium-induced Akt phosphorylation/ activation appears to be cell type-specific. In NIH3T3 (Fig. 8C), 293T (Fig. 8D), and mouse embryonic fibroblasts (data not shown) lithium treatment does not affect Akt Ser473 phosphorylation at anytime between 5 min and 24 h, whereas the increase in GSK-3 phosphorylation occurs within 5 min of lithium treatment (Fig. 8, C and D). This suggests that lithium-induced serine 9 phosphorylation does not require activation of Akt. Furthermore, pretreatment with LY 294002, an inhibitor of phosphatidylinositol 3-kinase (74), did not block GSK-3 phosphorylation caused by lithium in Neuro2A cells (Fig. 8B, lane 7) or in NIH3T3 cells (Fig. 8C, lanes 4, 7, and 9), supporting that the increase in GSK-3 phosphorylation caused by lithium is largely independent of phosphatidylinositol 3-kinase and Akt. To test whether GSK-3 inhibitors cause increased GSK-3 Ser21/9 phosphorylation by activating p90 rsk2 , we compared the time course of p90 rsk2 activation with GSK-3␤ Ser9 phosphorylation after lithium or Rottlerin treatment. Although both lithium and Rottlerin activated p90 rsk , as detected by an antibody recognizing the phospho-p90 rsk1/2 , the time course was significantly delayed relative to phosphorylation of GSK-3␤ Ser9 (Fig.  8E). This suggests that although p90 rsk2 may be involved in phosphorylation of serine 9 in vivo, activation of p90 rsk2 is not sufficient to explain the rapid increase in phospho-GSK-3␤ Ser9 caused by GSK-3␤ inhibitors. Pretreatment with a PKA inhibitor (Rp-cAMPS) (75) or a p70 S6K inhibitor (rapamycin) (76) has no effect on either endogenous or lithium-induced GSK-3␤ Ser phosphorylation (Fig. 8F, lane 4) (data not shown), suggesting that lithium-induced GSK-3 phosphorylation is not mediated through activation of PKA or p70 S6K .
Supporting these in vitro observations, overexpression of mouse I-2 in 293T cells (Fig. 9D) or Neuro2A cells (data not shown) causes increased phosphorylation of GSK-3␣ Ser21 and ␤ Ser9 , suggesting that PP-1 can regulate GSK-3 phosphorylation in vivo. Furthermore, GSK-3␤ activity has also been proposed to be regulated by PP-1, based on the effects of phosphatase inhibitors in cultured cells and brain slices (83,84). Taken together, these results suggest a mechanism by which GSK-3, through phosphorylation and inhibition of I-2, activates PP-1 to reduce serine 21/9 phosphorylation and that increased serine 21/9 phosphorylation by lithium or GID is a consequence of inhibition of GSK-3.

DISCUSSION
Lithium inhibits GSK-3 in vitro and in vivo, and this inhibition may explain many of the effects of lithium; however, lithium also inhibits, at similar or lower concentrations, phosphoglucomutase, a family of phosphomonoesterases that includes inositol monophosphatase, and possibly other targets yet to be identified. Because lithium salts are widely used therapeutically, as well as experimentally in diverse settings, it is important to develop approaches to distinguish which, if any, of these direct targets is responsible for a given effect of lithium. Several groups have reported recently that lithium causes increased phosphorylation of GSK-3␤ on serine 9, but the direct target of lithium responsible for this effect has not been determined. We have therefore used pharmacological and genetic approaches to test whether the target of lithium in this setting is GSK-3. We find that inhibition of GSK-3 by lithium, by peptide or small molecule inhibitors, or through reduction in GSK-3 expression leads to increased N-terminal phosphoryla-tion of GSK-3␣ and ␤, demonstrating autoregulation of GSK-3 N-terminal serine phosphorylation.
One approach to validating a proposed target of lithium is to test whether alternative inhibitors of the putative target mimic the effect of lithium. To this end, we have developed a short peptide inhibitor of GSK-3 derived from the GSK-3 interaction domain of Axin, which we have termed GID. We showed previously that GID peptides as short as 25 amino acids inhibit overexpressed GSK-3␤ in Xenopus oocytes. Further characterization here shows that GID potently inhibits endogenous GSK-3 activity in mammalian cells, as measured by reduced phosphorylation of tau protein, accumulation of ␤-catenin protein, and robust (over 200-fold) activation of ␤-catenin/TCF-dependent transcription, mimicking lithium-mediated inhibition of GSK-3. Activation of a Wnt-dependent transcription reporter has also been reported by others (40) using a larger fragment of Axin containing the GID; although this effect was interpreted as a consequence of disruption of the Axin complex, our data show that GID peptides directly inhibit GSK-3 activity in vivo and in vitro (these distinct mechanisms are not mutually exclusive). In this respect, GID peptides are similar to FRAT/ GBP, which also inhibits GSK-3 activity in Xenopus oocytes and mammalian cells and in in vitro assays (28,45,58). We had reported previously (29) that a larger fragment of Axin containing the GID did not inhibit in vitro GSK-3␤ phosphorylation of a prephosphorylated peptide derived from glycogen synthase (GS-2). However, new insights into the structure of GSK-3 have suggested that the mechanism of phosphorylation may differ between prephosphorylated and unphosphorylated substrates, and prior observations with FRAT showed that, in vitro, a FRAT peptide inhibits GSK-3␤ phosphorylation of unphosphorylated, but not prephosphorylated, tau protein (45). We therefore re-examined GID inhibition of GSK-3␤ in vitro and found that GID more potently inhibits GSK-3 activity toward unphosphorylated substrates, indicating an unexpected substrate preference for inhibition. In contrast to our findings that fulllength Axin does not inhibit GSK-3 in vivo when expressed at levels similar to GID or ⌬RGS (see this work and Ref. 29), a recent paper (85) reports that overexpression of full-length Axin can inhibit GSK-3-mediated phosphorylation of tau. Although the level of Axin overexpression was not indicated in that work, we also find that, at considerably higher levels of expression, full-length Axin can inhibit GSK-3 overexpressed in Xenopus oocytes (data not shown), consistent with those reported findings. The mechanism of this inhibition is currently being investigated.
We have also found that several of the small molecule inhibitors of GSK-3, like lithium, cause rapid (within 5 min) increased N-terminal serine phosphorylation of GSK-3␣ and ␤. Most of these inhibitors appear to function by occluding the ATP binding pocket, and in many cases these inhibitors have been shown to inhibit other protein kinases. GID and FRAT/ GBP interact with residues on the GSK-3 dimer interface, far from the ATP binding pocket, suggesting that these polypeptides inhibit GSK-3 through a distinct mechanism (32,33). A crystal structure of GSK-3␤ bound to a 19-residue peptide from the Axin-GID has been described recently (86) that confirms this site of interaction with GSK-3␤; this 19-mer did not cause a significant change in the conformation of the GSK-3␤ active site; however, in our assays, a similar peptide also did not inhibit GSK-3␤ activity (data not shown). Furthermore, TDZD, a novel GSK-3 inhibitor that reportedly does not compete with ATP, also causes increased serine 21/9 phosphorylation. Thus, increased serine 21/9 phosphorylation can be caused by agents that inhibit GSK-3 through several, distinct mechanisms (lithium, GID, TDZD, and ATP site competitors). A few of (1 M). ATP was added at 0.2 mM to activate GSK-3␤. LiCl (20 mM) or GID 320 -429 -His peptide (105 nM) was added as indicated. GSK-3␤ and phospho-GSK-3␤ were detected by WB as in previous figures. C, LiCl (20 mM) or GID 320 -429 -His (105 nM) was added in the absence of I-2 to verify that these inhibitors do not inhibit PP-1 directly. PP-1 protein levels are visualized by Coomassie staining. D, 293T cells were transfected with pCS2 control (C) or myc-tagged mouse I-2 and were harvested after 24 h. GSK-3␤ and phospho-GSK-3␤(S9) were detected as above, and I-2 expression was detected with myc antibody. Data shown are representative of at least four independent experiments with similar results. the ATP competitors, including Ro31-8220, SB-216763, and SB-415286, inhibit GSK-3 but do not cause increase serine 21/9 phosphorylation (Fig. 6A) (23). However, these compounds also inhibit p90 rsk2 , and we find that inhibition of p90 rsk2 , even with inhibitors that do not inhibit GSK-3 (e.g. BIS-III; see Fig. 8F), reduces basal and lithium-stimulated serine 21/9 phosphorylation.
Although the most likely explanation for the parallel effect of diverse GSK-3 inhibitors on GSK-3 phosphorylation is that they act through inhibition of GSK-3, the possibility remains that each of these agents interacts with a cryptic target that regulates GSK-3 phosphorylation independently. To test further whether GSK-3 regulates its own phosphorylation, we have used two loss-of-function approaches: RNA interference in Chinese hamster ovary cells and disruption of the GSK-3␤ gene, using an MEF cell line derived from the GSK-3␤ knockout mouse (22). When GSK-3␣ levels are reduced by RNAi, phosphorylation of GSK-3␤ Ser9 increases markedly without a change in overall level of GSK-3␤ (the relative phosphorylation of GSK-3␣ Ser21 also increased). Similarly, when GSK-3␤ levels are reduced, either by RNAi or in GSK-3␤-deficient MEFs, phosphorylation of GSK-3␣ Ser21 is increased. Furthermore, overexpression of GSK-3␤ S9A , which raises GSK-3 activity without changing the level of GSK-3␤ Ser9 , decreases phosphorylation of endogenous GSK-3␤ on serine 9 (71). Taken together with the inhibitor data presented here, these observations demonstrate autoregulation of GSK-3␣ and ␤ and strongly support the argument that the effect of lithium on N-terminal phosphorylation is mediated by inhibition of GSK-3.
GSK-3 autoregulation could involve inhibition of a kinase or activation of a phosphatase. Protein kinases known to phosphorylate GSK-3␣/␤ at serine 21/9 include Akt, p90 rsk2 , PKA, and p70 S6K . Modest activation of Akt by lithium in cerebellar granule cells has been reported previously, and we find similar modestly increased phosphorylation of Akt, indicative of activation, in Neuro2A cells, supporting a role for GSK-3 in the regulation of Akt activation (87). However, we also observe lithium-induced serine 21/9 phosphorylation that is independent of Akt activation. In fibroblasts, where basal Akt phosphorylation and activity are low, or in 293T cells, where basal Akt activity and phosphorylation are high, lithium increases serine 21/9 phosphorylation without changing the level of Akt phosphorylation, and this increased GSK-3 phosphorylation is not blocked by inhibition of phosphatidylinositol 3-kinase, indicating that the effect does not require activation of Akt. Furthermore, activation of p90 rsk2 is also not sufficient to explain lithium-induced phosphorylation of GSK-3␣/␤, as p90 rsk activation is significantly slower than the phosphorylation of GSK-3␣/␤ following addition of lithium (although p90 rsk may contribute to basal GSK-3 phosphorylation). Inhibition of PKA (with Rp-cAMPS) or p70 S6K (with rapamycin) does not block endogenous or lithium-induced GSK-3 phosphorylation, suggesting that lithium-induced phosphorylation of GSK-3␣/␤ is not mediated through activation of PKA or p70 S6K . Although these findings cannot rule out regulation of another serine 21/9 kinase (such as protein kinase C or integrin-linked kinase; see Ref. 88), none of the four protein kinases that we tested appears to explain the autoregulation of GSK-3␣/␤, and we turned to GSK-3-mediated activation of a phosphatase as an alternative mode of regulation.
It is well documented that GSK-3 activates PP-1 by phosphorylating I-2, an inhibitory subunit of the PP-1 complex (77)(78)(79)(80)(81)(82). Furthermore, PP-1 and GSK-3, as well as PKA, interact with the A-kinase anchoring protein 220, and, based on the effect of phosphatase inhibitors in cultured cells or rat brain slices, PP-1 has been proposed to activate GSK-3 (83, 84), although PP-1 regulation of GSK-3 N-terminal phosphoryla-tion was not examined in those studies. If PP-1 dephosphorylates serine 21/9, then this would be inhibited by I-2. Increased GSK-3 activity would then enhance phosphorylation and inhibition of I-2, leading to activation of PP-1 and thus diminished serine 21/9 phosphorylation, whereas inhibition of GSK-3 would allow inhibition of PP-1 by I-2, with consequent increased serine 21/9 phosphorylation. We demonstrate this by showing that GSK-3␤ Ser9-p can be dephosphorylated by PP-1 and that, in the absence of ATP, this is inhibited by I-2. Importantly, addition of ATP activates GSK-3, allowing phosphorylation of I-2 and reactivation of PP-1, resulting in GSK-3␤ Ser9-p dephosphorylation. GSK-3 inhibitors (lithium or GID) block phosphorylation of I-2, leading to suppression of PP-1, and thus serine 9 phosphorylation is maintained. Therefore, GSK-3 can regulate its own phosphorylation status through regulation of the PP-1⅐I-2 complex.
Protein phosphatase 2A (PP-2A) is also able to dephosphorylate GSK-3␤ in vitro (60,66) and has been proposed to regulate GSK-3␤ activity and N-terminal phosphorylation in vivo (37,89). Furthermore, lithium can block ceramideinduced activation of a protein phosphatase with characteristics similar to PP-2A (89). This finding could imply that GSK-3 regulates PP-2A in a manner similar to the regulation of PP-1. Although GSK-3 regulation of PP-2A has not been described, our data do not exclude this possibility. Nevertheless, our findings, taken together with previous work (83,84), strongly support a role for PP-1 in the autoregulation of GSK-3 N-terminal phosphorylation/dephosphorylation.
In most settings, GSK-3 is a constitutively active protein kinase that is inhibited in response to extracellular signals. Regulation of GSK-3 N-terminal phosphorylation by GSK-3 itself confers an unusual feedback system that could help to maintain GSK-3 in an active state but could also amplify an inhibitory signal and accelerate the conversion of GSK-3 to an inactive state. In the response to lithium, this autoregulation implies two levels of inhibition: rapid, direct inhibition of GSK-3 by lithium, followed by inactivation of PP-1 and consequent increased inhibitory phosphorylation of GSK-3. Because GSK-3 function in the Wnt pathway appears to be insulated from the effects of inhibitory, N-terminal kinases such as Akt (63), we predict that this increased inhibitory phosphorylation of GSK-3 would primarily affect Wnt-independent GSK-3-regulated pathways. This suggests an unexpected level of signaling pathway selectivity in lithium action that warrants further investigation.