Molecular Recognition of the Protein Phosphatase 1 Glycogen Targeting Subunit by Glycogen Phosphorylase*

Disrupting the interaction between glycogen phosphorylase and the glycogen targeting subunit (GL) of protein phosphatase 1 is emerging as a novel target for the treatment of type 2 diabetes. To elucidate the molecular basis of binding, we have determined the crystal structure of liver phosphorylase bound to a GL-derived peptide. The structure reveals the C terminus of GL binding in a hydrophobically collapsed conformation to the allosteric regulator-binding site at the phosphorylase dimer interface. GL mimics interactions that are otherwise employed by the activator AMP. Functional studies show that GL binds tighter than AMP and confirm that the C-terminal Tyr-Tyr motif is the major determinant for GL binding potency. Our study validates the GL-phosphorylase interface as a novel target for small molecule interaction.

Disrupting the interaction between glycogen phosphorylase and the glycogen targeting subunit (G L ) of protein phosphatase 1 is emerging as a novel target for the treatment of type 2 diabetes. To elucidate the molecular basis of binding, we have determined the crystal structure of liver phosphorylase bound to a G L -derived peptide. The structure reveals the C terminus of G L binding in a hydrophobically collapsed conformation to the allosteric regulator-binding site at the phosphorylase dimer interface. G L mimics interactions that are otherwise employed by the activator AMP. Functional studies show that G L binds tighter than AMP and confirm that the C-terminal Tyr-Tyr motif is the major determinant for G L binding potency. Our study validates the G L -phosphorylase interface as a novel target for small molecule interaction.
Diabetes is one of the major public health problems. Approximately 194 million people worldwide, or 5.1%, in the age group 20 -79 were estimated to have diabetes in 2003. This estimate is expected to increase to some 333 million, or 6.3% of the adult population, by 2025 (1). Type 2 diabetes, the most common form of diabetes is characterized by defects in insulin secretion, insulin resistance, and elevated hepatic glucose production. Both increased gluconeogenesis and increased glycogenolysis contribute to excessive hepatic glucose output despite hyperglycemia (2). Several novel pharmacological strategies are aiming to treat hyperglycemia by normalizing or increasing depleted glycogen stores (3). For example, drug discovery has focused on competitive as well as allosteric inhibition of glycogen phosphorylase activity (4).
Glycogen phosphorylase (GP) 2 is an important allosteric enzyme in carbohydrate metabolism that catalyzes phosphorolysis of an ␣-1,4-glycosidic bond of glycogen to glucose-1phosphate. In humans there are three GP isoforms (liver, muscle, and brain GP), which are named after the tissues where they are predominantly expressed. Glycogen phosphorylase is a homodimer that cycles between two conformations: active (R) and inactive (T) state. Phosphorylation of Ser 14 by phosphorylase kinase and active site as well as allosteric binders modulate the equilibrium between both states (5), but the isozymes differ in their responsiveness to regulatory mechanisms. In the liver, phosphorylation is the major regulator of GP activation. Conversion of unphosphorylated liver GPb to phosphorylated GPa fully activates the enzyme. AMP stimulates liver GPb by 10 -20%, whereas it does not further activate GPa (6). In contrast, AMP activates the unphosphorylated muscle isoform to 80% of the maximal activity and increases the activity of phosphorylated muscle GPa by a further 10%. Crystallographic studies have shown endogenous and synthetic modulators bound to four major sites (see Fig. 1a): active site (7), purine site (8), central cavity (9), and allosteric AMP site (10,11).
Important for glycogen metabolism is the strong reciprocal control between GPa and glycogen synthase activity. Activation of glycogen synthase via its phosphatase (protein phosphatase 1 (PP1)) can be allosterically inhibited by binding of GPa (12) to G L (13), a glycogen targeting subunit of PP1. PP1 in turn suppresses GP and phosphorylase kinase activities through dephosphorylation. Glycogen targeting subunits bind to PP1, modulate its activity toward substrates, localize it to specific cellular sites, and are proposed to function as a scaffold for the assembly and regulation of glycogen metabolizing enzymes. There is an increasing number of glycogen-targeting subunits. So far, seven family members of glycogen-targeting subunits are described in humans: G M (PPP1R3A) (14), G L (PPP1R3B) (15), R5/PTG (PPP1R3C) (16,17), R6 (PPP1R3D) (18), PPP1R3E (19), and PPP1R3F and PPP1R3G (20). Mutational analysis of the rat liver targeting subunit G L has identified three separate regions that are responsible for binding to PP1 (residues 59 -94), glycogen (residues 94 -257), and phosphorylase a (residues 269 -284 at the G L C terminus) (21). The G L C-terminal region is unique in G L and absent in other glycogen targeting subunits but is conserved between rodent and human. Pharmacological inhibition of the interaction of phosphorylase a with G L could provide a novel mechanism to lower blood glucose levels by inducing the dephosphorylation and activation of glycogen synthase (3).
Here we show that the G L C-terminal region structurally and functionally mimics AMP binding to human liver glycogen phosphorylase (hlGP). Using x-ray crystallography we identify the C terminus of G L bound in the allosteric regulator site of hlGP. Using functional assays we map binding * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. contributions of the G L peptide and show that it activates phosphorylase b, in vitro.

EXPERIMENTAL PROCEDURES
Protein Production-Human liver glycogen phosphorylase was prepared as described (7). Briefly, full-length hlGPa was expressed in insect cells and purified using copper-chelating, anion exchange, and size exclusion chromatography. hlGPb was expressed in Escherichia coli and purified analogously.
Crystallization and Structure Determination-Crystals were obtained at 20°C in 1 ϩ 1 l hanging drops from hlGPa concentrated to 8 mg/ml in 20 mM BES, pH 6.7, 1 mM EDTA, 0.5 mM dithiothreitol, 50 mM glucose, 0.5 mM AMP over a reservoir of 0.1 M Tris, pH 8.5, 7-8% (w/v) polyethylene glycol 8000. Macroseeding improved crystal size and reproducibility. The G L complex was obtained by soaking crystals for 24 h in reservoir supplemented with 1 mM G L -Cterm peptide (see below). For cryoprotection the mother liquor was incrementally exchanged to 0.1 M Tris, pH 8.5, 20% glycerol, 20% polyethylene glycol 8000, 1 mM G L -Cterm peptide. The crystals were then flash frozen in a 100 K nitrogen stream. The diffraction data were collected on the PX-1 beamline at the SLS (Villigen, CH) and processed with XDS (22) (see Table 1). The complex structure was solved using difference Fourier methods with the coordinates of hlGPa-AMP (Protein Data Bank accession code 1FA9) as a template. Following rigid body refinement the (F o Ϫ F c )␣ calc maps included easily interpretable electron density for the bound ligand. Restrained refinement was performed with REFMAC (23) and BUSTER (Global Phasing Ltd.) iterated with model building in COOT (24). The final model has been completed to residues 4 -838 of hlGPa, residues 281-284 of G L , and 154 water molecules. 98.5% of residues are in the most favored and additionally allowed regions of the Ramachandran plot, and 0.3% are in disallowed regions. Several regions were not well defined in the final electron density and are omitted from the final model (residues 162-165, 251-262, 282-285, and 317-322). The final model statistics are listed in Table 1. Surface calculations were made with the EBI PISA server. The figures were prepared using PyMOL (DeLano Scientific LLC).
Coordinates-The coordinates and structure factors have been deposited in the Protein Data Bank (Protein Data Bank accession code 2QLL).
SPA Assay for G L /hlGPa Interaction-A SPA was used to measure the interaction between 33 P-radiolabeled hlGPa and a biotin-labeled G L -peptide (G L probe). A one-step phosphorylation reaction by phosphorylase kinase (Sigma) transformed hlGPb into 33 P-radiolabeled hlGPa. The incubation reaction contained test compound, 5 g of 33 P-hlGPa, 50 pmol of G L probe, 0.4 mg SPA-Beads (streptavidin-SPA beads; Amersham Biosciences, RPNQ 0007) in test buffer (50 mM Tris, 5 mM EDTA, pH 7.5)/well in a 384-well format. Scintillation was measured after overnight incubation at room temperature. Triplicate determinations were made in all of the binding experiments.
rmGPb Activation Assay-Activation of enzymatic activity of rabbit muscle glycogen phosphorylase b (rmGPb from Sigma) was measured in direction of glycogen degradation by coupling glucose-1-phosphate production to NADP consumption (25). Activation experiments were performed as triplicate determinations. All of the reagents and enzymes were purchased from Sigma.

RESULTS
Structure of G L Bound to hlGPa-To identify the molecular basis of G L binding, we diffused a peptide comprising the last 16 residues of G L (GL-Cterm, 269 GL -284 GL ; numbering refers to the sequence of rat G L ) into crystals of hlGPa and determined the crystal structure of the resulting hlGPa-G L complex ( Fig. 1 and Table 1). Only the terminal four amino acids (residues 281 GL -284 GL , Gly-Pro-Tyr-Tyr) were visible in difference electron density maps ( Fig. 2A), whereas the remainder of the peptide was disordered. The G L -binding site is located at the subunit interface and overlaps with the binding site for the allosteric regulator AMP (Fig. 2C). The binding site is ϳ14 Å from the Ser 14 phosphorylation site, 32 Å from the catalytic site, and 25 Å from the central cavity where the allosteric inhibitor CP-403700 binds (9). The allosteric site is lined by helices ␣2 and ␣8, and a short strand, ␤7, and is closed by the capЈ region (residues 36Ј-47Ј; the prime symbol refers to residues from the second chain of the homodimer) from the other subunit. G L binds in an U-shaped, hydrophobically collapsed conformation where it exploits numerous polar and hydrophobic contacts to hlGPa (Fig. 2, A and B). The G L peptide protrudes deep into the pocket, and 78% (542 Å 2 ) of its total solvent-accessible surface becomes buried upon binding.
The terminal carboxylate group mimics the AMP phosphate by addressing an arginine-rich region that also comprises the binding site for the allosteric effector phosphate. The Tyr 284 GL side chain points into the ribose binding region, where it forms hydrophobic contacts to Trp 67 , Gln 71 , Tyr 75 , and Val 45 Ј and a strong hydrogen bond to Asp 42 Ј. Tyr 283 GL protrudes into a region that is not contacted by AMP. Its phenol group is incorporated in a hydrogen bonding network with Asp 306 and Arg 242 . We further observe an edge-to-face interaction with Phe 196 and stacking of the phenol ring against the Arg 309 guanidinium group. The pyrrolidine of Pro 282 GL is packed through hydrophobic stacking interactions between the two subunits: the phenolic side chain of Tyr 75 from subunit A and two residues from the subunit B cap region (carbonyl oxygen of Asn 44 Ј and CG2 of Val 45 Ј). In this it partially mimics the stacking interactions of the AMP adenine between Tyr 75 and Asn 44 Ј (Fig. 2C). With Gly 281 GL , the G L peptide leaves the pocket and protrudes into the solvent. The last notable interaction is a van der Waals' contact of its carbonyl oxygen to C␤ of Ala 313 .
Conformational Changes-Prior to soaking with G L , hlGPa was crystallized in the active, AMP-bound conformation (7), and on the subunit level there are no large conformational changes upon G L binding. The backbone atoms of hlGP-G L and hlGP-AMP align with an root mean square deviation of 0.48 Å. However, the side chains of three residues undergo local reorganization to accommodate the Tyr 283 GL phenol (Phe 196 and Arg 309 ) or adapt to the terminal carboxylic acid (Tyr 155 ). Moreover, there is a subtle change in quaternary structure as the two subunits rotate ϳ2°toward each other (not shown). Within the allosteric pocket this translates to a 0.7-Å shift of the capЈ loop (Fig. 2C) to improve contacts to Pro 282 GL . Phosphorylase a is a potent allosteric inhibitor of the PP1-G L complex (12,21), whereas the inactive phosphorylase b is not (12). To address the molecular basis of this finding we compared hlGPa-G L to a complex with the inhibitor N-acetyl-␤-Dglucopyranosylamine (hlGPa-GlcNac), which defines the inac-tive (T state) conformation (7). Upon inhibitor binding the subunits rotate "outward" by ϳ7°, and the dimer interface (including the allosteric site) is remodeled (Fig. 2D). Consequently, several interactions that stabilize the G L complex are lost (for example: capЈ interactions, stacking with Tyr 75 , hydrogen bond to Tyr 155 ) or would lead to steric conflicts (for example: Arg 309 , Phe 196 , and Asp 306 ).
Functional Analysis of G L Recognition by Glycogen Phosphorylase-After solving the complex structure, we sought to elucidate the determinants that are crucial for conferring binding potency. We developed a SPA to measure the competitive binding of a biotinylated G L peptide to 33 P-labeled hlGPa. The dissociation constant (K D ) for the interaction of the G L peptide with hlGP was determined from a saturation binding curve (Fig. 3A) to be 145 nM. Next, we measured the competitive displacement of the G L probe through several peptide variants as well as AMP. Representative binding curves for the displacement of the G L probe by GL-Cterm and AMP are shown in Fig. 3B. The apparent IC 50 values of the different peptides ( Table 2) allow an assessment of the relative importance of individual amino acids for binding. We find that the hexa-to tetrapeptides inhibit the hlGPa-G L interaction in a similar range as the complete GL-Cterm. This is in agreement with the structure, because residues 269 GL -280 GL are not ordered and therefore are not supposed to contribute strongly to binding. An approximately 8-fold drop in affinity is observed once truncation includes Gly 281 GL , and no binding could be detected when only the two terminal tyrosines were probed. Likewise, mutation of either tyrosine to alanine or truncation of Tyr 284 GL was deleterious for binding. Again, this is in good agreement to the numerous contacts these residues involve in the structure. AMP is not a good competitor and inhibits the hlGPa-G L interaction only with an IC 50 of 21 M, roughly 10 times weaker than GL-Cterm.
We then tested the ability of G L derivatives to activate the inactive rabbit muscle phosphorylase b (rmGPb; Table 2) and compared it with AMP-induced activation (6). Interestingly, G L peptides are able to activate rmGPb. Activation through GL-Cterm matches that of AMP at 30 M, which represents a maximally active concentration. Also the three shorter, tightly binding peptides still activate rmGPb between 20 and 30%.

DISCUSSION
G L as an AMP Mimic-In vitro binding studies have shown that AMP is able to inhibit the specific interaction between phosphorylase a and recombinant G L protein (15). The structural analysis of the GPa-G L complex revealed that the very C terminus of G L binds in an AMP-competitive fashion, mimicking many of the nucleotides interactions. Additional contacts (mostly provided by Tyr 283 GL ) result in a 10-fold tighter binding compared with AMP. The improved potency might be necessary to counterbalance the high cellular AMP levels. Phosphorylase itself can accommodate G L smoothly, without undergoing any larger conformational changes. A structural comparison between active and inactive states of hlGP explains why the latter cannot directly accommodate G L and consequently is not effective as an inhibitor of PP1 (12). Nevertheless, G L peptides are able to activate muscle GPb. Because G L cannot directly bind to GPb in its inactive (T state) conformation, we propose that the peptide acts analogous to AMP (5) and shifts the equilibrium between active and inactive conformations toward the R state.
The hlGP-G L Interface as a Drug Target-Inhibiting the interaction of phosphorylase a and G L has the potential to block the allosteric inhibition of the PP1/G L activity on glycogen synthase by phosphorylase a. It has been hypothesized that a stimulation of the glycogen synthase pathway via pharmacological dissociation of phosphorylase a from G L could help to normalize hyperglycemia (3). We identified the AMP site as a high affinity binding site for G L . Our data indicate that it might be feasible to antagonize binding of G L protein to GPa with small molecules.
Because in humans G L is expressed in liver and muscle cells, glycogen synthesis might be activated in both tissues. Increasing G L activity by overexpression was found in cultured primary human myotubes (26) as well as in primary hepatocytes (27) to stimulate glycogen synthase activity and to exhibit a high glycogenic effect. The effects of increasing activity of different glycogen targeting subunits by hepatic overexpression have also been studied in ani- GL and Val 45 Ј are not labeled for clarity. B, schematic representation of the interactions observed between G L and hlGP. hlGP residues in contact to G L are color-coded according to physicochemical property: green, nonpolar; blue, polar; red circle, acidic; blue circle, basic. Hydrogen bonds and salt bridges are shown as green arrows, and solvent-accessible G L atoms are surrounded by a blue halo. C, superposition of hlGPa-G L (color coding is as described for A) with the hlGPa-AMP complex (rosy, AMP carbon atoms in magenta). The protein backbones are shown as ribbons. Tyr 283 GL is not labeled for clarity. D, superposition of hlGPa-G L (color coding is as described for A) with the inactive hlGPa-GlcNac complex (carbon atoms in blue). The protein backbones are shown as a ribbon. Note that not all residues of hlGPa-GlcNac are labeled for clarity. mal models (28,29). Despite the profound effects of hepatic G L overexpression on glycogen stores in the livers of these animal models, no or only modest and transient effects on hyperglycemia are reported. This is in contrast to the improved glycemic situation following the expression of G M ⌬C, a truncated version of the muscle-targeting subunit in diabetic rats. The difference was explained by a larger increment in hepatic glycogen storage under oral glucose tolerance test with G M ⌬C overex-pression than with G L overexpression. Higher glycogen stores even in the fasted state following G L overexpression were discussed as indicative for reduced glycogenolytic sensitivity.
Therefore, the potential in vivo profile of an inhibitor of G L binding to GP with respect to effects on glycogenic and glycogenolytic pathways is hard to predict. It is clear that our small peptides might not exhibit the ideal profile, because they behave like AMP with respect to rmGPb activation. Increasing GP activity might counterbalance the stimulation of glycogen synthesis, thus neutralizing potential beneficial effects. Therefore, drug development strategies should aim for compounds lacking GP activating properties.
Acknowledgments-We thank Katja Mück for initial support in protein purification; Petra Wieland and Rene Schiller for assistance with the biochemical assays; Stefan Hörer, Herbert Nar, and Stefan Kauschke for discussions; and Clemens Schulze-Briese and the beamline staff of SLS-PX1 for support during data collection.
Addendum-While this manuscript was under revision, an independent study was published (30) that employed calorimetric measurements on G L -derived peptides to locate the interaction site to the last five residues. The authors also show that an indole-2-carboxamide drug (which binds to the central cavity and stabilizes the T state) can block the interaction. The study clearly supports our conclusions.

TABLE 2 Functional analysis of ligand binding
Displacement of a biotinylated G L -probe from 33 P-radiolabeled hlGPa was measured in a scintillation proximity assay. rmGPb activation was measured in the direction of glycogen degradation by coupling glucose-1-phosphate production to NADP consumption. The relative increase in rmGPb activity in the presence of 30 M AMP is used as reference and set to 100%.

Ligand
a The values are the means of three independent measurements unless noted otherwise. Standard deviations are shown as ϮS.D. Activation of rmGPb through G L -Cterm was performed as a single experiment only. b The values are the means of two independent measurements.