Research ArticleBIOCHEMISTRY

Identification of the substrate recruitment mechanism of the muscle glycogen protein phosphatase 1 holoenzyme

See allHide authors and affiliations

Science Advances  14 Nov 2018:
Vol. 4, no. 11, eaau6044
DOI: 10.1126/sciadv.aau6044


Glycogen is the primary storage form of glucose. Glycogen synthesis and breakdown are tightly controlled by glycogen synthase (GYS) and phosphorylase, respectively. The enzyme responsible for dephosphorylating GYS and phosphorylase, which results in their activation (GYS) or inactivation (phosphorylase) to robustly stimulate glycogen synthesis, is protein phosphatase 1 (PP1). However, our understanding of how PP1 recruits these substrates is limited. Here, we show how PP1, together with its muscle glycogen–targeting (GM) regulatory subunit, recruits and selectively dephosphorylates its substrates. Our molecular data reveal that the GM carbohydrate binding module (GMCBM21), which is amino-terminal to the GM PP1 binding domain, has a dual function in directing PP1 substrate specificity: It either directly recruits substrates (i.e., GYS) or recruits them indirectly by localization (via glycogen for phosphorylase). Our data provide the molecular basis for PP1 regulation by GM and reveal how PP1-mediated dephosphorylation is driven by scaffolding-based substrate recruitment.


The Ser/Thr protein phosphatase 1 (PP1) regulates diverse cellular processes, including neuronal plasticity, cell division, and protein synthesis (1). However, PP1 was originally discovered for its ability to direct glycogen metabolism in skeletal muscle. Specifically, PP1 dephosphorylates glycogen synthase, phosphorylase kinase, and glycogen phosphorylase, essential enzymes that are regulated by insulin and together control glycogen synthesis and breakdown (2). The dephosphorylation of glycogen synthase and phosphorylase a (glycogen phosphorylase phosphorylated on Ser14 is commonly referred to as phosphorylase a) by PP1 has opposing effects, namely, the activation of glycogen synthase and the inactivation of phosphorylase a. As a consequence, glycogen synthesis is robustly stimulated (3, 4).

During the last 30 years, it has become apparent that PP1 associates with scores of different PP1 regulatory subunits to form distinct, heterodimeric holoenzymes (5, 6). These PP1 regulatory subunits function to both target PP1 to its cellular point of action and selectively recruit specific substrates for PP1-mediated dephosphorylation. This latter function, substrate recruitment, is typically achieved by protein domains outside the primary PP1-anchoring domain. The ability of PP1 to regulate glycogen synthesis also requires its association with a specific regulatory subunit, the glycogen-targeting subunit in muscle, GM. Although GM was the first PP1-specific regulatory protein discovered, it is now known that it is one of seven genes in mammalian genomes that constitute the glycogen-targeting PP1 family (G-subunits): GM (RGL) is expressed in skeletal and cardiac muscle (4, 7); GL (FLJ14005) is most abundantly expressed in the liver (8); and GC (PTG, R5) and GD (R6) are ubiquitously expressed (9). A comprehensive molecular understanding of how these G-subunits direct both PP1 targeting and dephosphorylation of glycogen synthase and phosphorylase a is currently missing.

All G-subunits include a highly conserved RVxF motif, which is essential for PP1 binding, and a family 21 carbohydrate binding module [CBM21; also known as starch binding domains (SBDs)], which is responsible for binding glycogen (1012). Typically, SBDs have two sugar binding sites, known as sites 1 and 2 (13). It is currently unknown whether only one or both of these sites in GMCBM21 are necessary for glycogen targeting. Further, it was suggested that the GMCBM21 domain may also facilitate glycogen synthase substrate recruitment (14). However, whether or how this occurs is unknown. The first structure of PP1 bound to any of its regulators was that of PP1 bound to a short RVxF peptide from GM (15). This structure identified the RVxF binding pocket in PP1, which showed that it is more than 20 Å away from the PP1 active site. This structure also revealed that RVxF binding does not alter the conformation of the PP1 active site, explaining why it does not affect PP1 catalytic activity. However, it did not provide any insights into whether and how GM binds to PP1 beyond the RVxF motif, how GM binds glycogen, or how these interactions facilitate glycogen-specific substrate recruitment.

Here, we used nuclear magnetic resonance (NMR) spectroscopy, x-ray crystallography, and enzymatic studies to determine how GM recruits and targets PP1 to phosphorylase a and glycogen synthase. Unexpectedly, we found that PP1 interacts with GM outside its RVxF sequence via an extended ΦΦ motif and the GMCBM21 domain; this results in extremely tight binding. After determining the structure of the GMCBM21 domain, we then used NMR chemical shift perturbation (CSP) mapping to show that only one of its two carbohydrate sites binds directly to glycogen. This led to the discovery that the second site has a different function, namely, binding and recruiting the PP1-specific substrate glycogen synthase. We then showed that while holoenzyme formation with GM does not enhance phosphorylase a dephosphorylation, the simultaneous recruitment of both GM:PP1 and phosphorylase a to glycogen does. Together, these structural and enzymatic data reveal how, at a molecular level, GM targets PP1 to its glycogen-specific substrates phosphorylase a and glycogen synthase to specifically and robustly dephosphorylate both enzymes. Thus, this study provides the most comprehensive molecular understanding of how a specific PP1 holoenzyme, GM:PP1, mediates the rapid and selective dephosphorylation of its specific substrates.


GM has two PP1 interaction sites

The skeletal muscle glycogen–binding subunit [PP1 regulatory subunit 3A (PPP1R3A); hereafter referred to as GM] is a 1109–amino acid protein (124 kDa) (Fig. 1A). It was previously shown that residues 1 to 240 are required for PP1 binding and regulation [if not otherwise noted, PP1α residues 7 to 330 (PP1α7–330) are used in all experiments]. Bioinformatics predicts that residues 1 to 100 are intrinsically disordered (IDR; typical for PP1 regulatory/binding regions), while residues 102 to 237 form a well-folded domain that is a member of the CBM21 family. To define the GM residues that bind directly to PP1, we used NMR spectroscopy, isothermal titration calorimetry (ITC), and surface plasmon resonance (SPR). ITC showed that GM residues 2 to 64 do not bind PP1, as both GM2–237 and GM64–237 bind PP1 with the same affinities [statistically identical dissociation constant (KD) values of 27 and 21 nM, respectively; Fig. 1, B and C, and Table 1]. Next, we used NMR spectroscopy to investigate the solution behavior of GM64–237. A 2D [1H,15N] HSQC spectrum confirmed that GM64–93 is unstructured in solution, based on the limited chemical shift dispersion in the 1HN dimension (Fig. 1D). In contrast, GM102–237 (hereafter referred to as GMCBM21) has a 2D [1H,15N] HSQC spectrum that is typical for a folded protein (Fig. 1E). Furthermore, overlaying the NMR spectra of GM64–93 and GMCBM21 with that of GM64–237 shows that both domains are independent in solution, as their individual spectra overlap nearly perfectly with that from GM64–237 (Fig. 1F). Last, using SPR, we showed that GM64–105 binds PP1α7–330 with a KD of 114 ± 4 nM (fig. S1, A to D), revealing that GMCBM21 directly contributes to PP1 binding (~4-fold increase in PP1 binding with GM64–237 versus GM64–105) and thus PP1 and GM bind one another via two distinct interaction sites.

Fig. 1 GM PP1 interaction.

(A) GM domain structure, highlighting the PP1 binding short linear motifs (SLiMs) RVxF (cyan) and ΦΦ (salmon), which are followed by a well-folded CBM21 family domain (purple). The constructs examined in this study are shown below. ITC curves for (B) GM2–237 and (C) GM64–237 show similar binding affinities for PP1α. Two-dimensional (2D) [1H,15N] heteronuclear single-quantum coherence (HSQC) spectrum of (D) GM64–93 (inset, Trp NHε1) and (E) GMCBM21. GM64–93 is an IDR; GMCBM21 is well folded. (F) Overlay of the 2D [1H,15N] HSQC spectrum of GM64–237 (black) with that of GMCBM21 (red, left) or GM64–93 (magenta, right); unassigned GM94–102 residues are highlighted by blue asterisks (*).

Table 1 Thermodynamic and dissociation constants for the GM interaction with PP1 and cyclodextrins derived from ITC experiments at 25°C.

n.d., not determined.

View this table:

We then performed ITC measurements to test whether GM64–237 interacts with PP1 in an isoform-specific manner. ITC showed that GM64–237 binds PP1β6–327 (17.3 ± 4.0 nM) and PP1γ7–323 (21.4 ± 11.6 nM; PP1γ1) with statistically identical KD values, demonstrating that the interaction of GM is not PP1 isoform specific (fig. S1, E to H).

The GM PP1–anchoring domain

To determine how GM64–237 binds PP1 at atomic resolution, we used x-ray crystallography. Despite extensive efforts, crystallization of the GM64–237:PP1α7–300/330 holoenzyme was unsuccessful. However, we were able to determine the 3D structure of the GM64–93:PP1α7–300 holoenzyme (hereafter referred to as GM:PP1) to a resolution of 1.45 Å (table S1). The structure shows that GM residues 64 to 85 become ordered when bound to PP1 (Fig. 2A). They bind in a largely extended manner at multiple sites across the top of PP1, including the RVxF motif binding site and the ΦΦ motif binding site (Fig. 2B), both of which are used by a large number of PP1 interactors. The interaction between PP1 and GM is extensive, with the complex burying 2194 Å2 of a solvent-accessible surface area. As expected for a targeting protein, the PP1 catalytic site is accessible, and PP1 is catalytically active in the GM64–237:PP1α7–300 holoenzyme, as it is capable of dephosphorylating model substrates, such as p-nitrophenyl phosphate (fig. S2, A and B).

Fig. 2 GM:PP1 holoenzyme structure.

(A) Crystal structure of GM64–93:PP1α7–300. GM64–93 shown in pink with a 2FoFc electron density map contoured at 1σ (2 Å) and PP1 shown as surface (gray). RVxF binding pocket (cyan), ΦΦ binding pocket (salmon), and the active site (orange) of PP1 are highlighted. PP1 substrate binding grooves (A, acidic; H, hydrophobic; C, C-terminal) are marked. (B) Close-up of GM bound to PP1. GM residues 65RVSF68 bind the PP1 RVxF binding pocket (cyan), and GM residues 79VK80 bind the PP1 ΦΦ binding pocket (salmon). (C) GM residues 78 to 81 form a β strand (green) to cap β strand 14 of PP1 via hydrogen bonds (dotted blue lines). (D) Greek key turn residues (raspberry), D70GM:R261PP1 salt bridge (yellow dotted lines), L76GM (lid), and F82GM are shown as sticks; ΦR pocket is highlighted.

GM residues 65RVSF68 form the RVxF motif, which binds the PP1 RVxF binding pocket (Val66GM and Phe68GM are the anchoring hydrophobic residues that bind deeply in this pocket). This interaction is highly similar to those observed in other PP1 holoenzyme complexes, as well as the previously determined PP1:GMpeptide structure (GMpeptide includes residues 63 to 75 with density visible only for residues 63 to 68; fig. S2, C and D) (15). As has been shown for other RVxF containing PP1 regulatory subunits, phosphorylation by protein kinase A (PKA) of the “x” residue in the GM RVxF motif, Ser67GM, inhibits PP1 binding (16). Val79GM and Lys80GM form the GM ΦΦ motif, which binds the PP1 ΦΦ binding pocket. Like the RVxF interaction, the ΦΦ interaction is highly similar to those observed in other PP1 holoenzyme complexes. Further, these residues are part of a short β strand (GM residues 78 to 81) that hydrogen bonds with PP1 β strand β14 to extend one of PP1’s two central β sheets (Fig. 2C).

Notably, a distinctive feature of the GM:PP1 holoenzyme structure is the “extended kink” between the RVxF and ΦΦ motifs (Fig. 2, B and C). In GM, these two motifs are separated by 12 residues, the longest insert observed to date for any PP1 regulator. It is long because it forms an ~8-residue “Greek key” turn. This structural feature is stabilized by two anchoring interactions: (i) a strong intermolecular bidentate salt bridge between PP1 residue Arg261PP1 and Asp70GM and (ii) hydrophobic interactions made by the GM “lid” residue (Leu76GM), with the hydrophobic ΦR pocket adjacent to the RVxF binding pocket (Fig. 2D). As a consequence, the side chains of GM residues Phe72GM and Phe74GM point up away from the holoenzyme surface, where they bury the side chain of PP1 residue Met290PP1.

A second distinctive feature of the GM:PP1 complex is that the GM:PP1 interaction extends beyond the ΦΦ motif, with Phe82GM binding in a deep pocket immediately adjacent to Pro298PP1 (Fig. 2D). Comparison with other PP1 holoenzyme structures reveals that, similar to the RVxF and ΦΦ binding pockets, this pocket is also frequently used by regulators to bind PP1. Namely, spinophilin (Thr461spino), PNUTS (Phe413pnuts), and RepoMan/Ki67 (Phe404RM) bind this same pocket (1719). Binding in this pocket requires a conformational change in PP1 to make it accessible. Namely, in apo-PP1, this pocket is occupied by Arg74PP1; regulator binding causes the Arg74PP1 side chain to rotate about its Cβ carbon by 90° to make this pocket accessible.

Last, an overlay of the GM:PP1 holoenzyme with all other PP1 holoenzyme structures shows that the GM backbone (N to C terminus) follows a “hybrid” binding path of RepoMan (optimal overlay for the RVxF motif), PNUTS (optimal overlay from the lid motif to the ΦΦ binding pocket), and again RepoMan (optimal overlay residues from the ΦΦ motif to the C terminus; fig. S2E). Together, this mode of interaction is typical for a scaffolding function of regulatory proteins but likely does not affect substrate selectivity toward PP1.

The GM carbohydrate binding domain also interacts with PP1

As shown by ITC, GM64–237 binds PP1 ~4-fold more tightly than the primary PP1-anchoring domain (GM64–105). To understand this in more detail, we determined the 3D structure of GMCBM21 (16 kDa) using solution NMR spectroscopy. The NMR data of GMCBM21 are of outstanding quality, allowing for the sequence-specific backbone assignment of 128 out of an expected 132 residues and a 98% completeness for the side-chain assignment (fig. S3A). A total of 2709 nuclear Overhauser effect (NOE)–based unambiguous distance restraints (~20 restraints per residue) and 110 dihedral angle restraints were used for the final structure refinement of 200 structures (table S2). The 20 lowest-energy conformers show excellent stereochemistry, with a backbone root mean square deviation of 0.56 Å (secondary structure). GMCBM21, like all known SBD structures, adopts an immunoglobulin-like β-sandwich fold (fig. S3B) (13, 20, 21). The β-sandwich consists of nine antiparallel β strands: β0 (GM residues 108 to 110), β1 (130 to 137), β2 (144 to 151), β3 (158 to 164), β4 (174 to 178), β5 (187 to 193), β6 (208 to 215), β7 (218 to 221), and β8 (229 to 235) (fig. S3C). At its N terminus, GMCBM21 has an additional α helix, α1 (118 to 127). SBDs adopt one of two topologies, type I or type II, which differ in the order and orientation of a single β strand (fig. S3D) (22). The GMCBM21 structure adopts a type II topology; this is consistent with the observation that the majority of N-terminal SBDs adopt a type II topology, whereas most C-terminal SBDs adopt a type I topology (23).

We then used NMR spectroscopy and 2H,15N-labeled GM64–237 (here, only 1H attached to carbons are exchanged to 2H; nitrogen-bound 1H are rapidly back exchanged) to test the interaction with PP1. Two results were readily detected. First, peaks corresponding to GM64–85 are missing as they move to their folded PP1-bound state. This demonstrates that the residues observed in the GM:PP1 holoenzyme structure also bind PP1 in solution (Fig. 3A). Second, several peaks corresponding to residues in GMCBM21 have CSPs or intensity changes, including Leu165GM, Trp168GM, Gln169GM, Thr170GM, Leu196GM, Val197GM, Asn225GM, and Thr230GM. Mapping these residues onto the GMCBM21 structure shows that the largest changes correspond to residues within the GMCBM21 β3-β4 loop (Fig. 3, B and C). 15N[1H]-NOE data of GMCBM21 showed that this loop is very rigid, showing no differences in dynamics when compared to the rest of GMCBM21 (fig. S3E). Unexpectedly, variants of these loops fail to express solubly, making ITC measurements to further confirm the interaction impossible. ITC data testing a direct interaction of GMCBM21 with PP1 showed no binding isotherm, confirming that GM residues 64 to 85 are strictly required for the interaction (fig. S1I).

Fig. 3 The GMCBM21 domain binds glycogen and recruits substrates.

(A) Overlay of the 2D [1H,15N] TROSY spectrum of [2H,15N]-labeled GM64–237 in the presence (red) and absence (black) of PP1α7–330. GMCBM21 residues that interact with PP1 are labeled. Residues that belong to GM64–93 are highlighted (*). (B) Interacting residues mapped onto the surface of GMCBM21 show that they form a contiguous surface around the β3-β4 loop (pink). (C) Model of the interaction of GMCBM21 and PP1. Residues of GMCBM21 shown to interact with PP1 are depicted as blue sticks. (D) CSP mapping of GMCBM21 bound to α-CD (blue) or β-CD (black) versus residue number (right, GMCBM21 secondary structure). Site 1 and site 2 are labeled and marked with dotted lines. (E) CSPs shown in (D) mapped onto the structure of GMCBM21 and colored by magnitude of the chemical shift difference (Δδ). One, two, and three SDs away from the mean are colored cyan, light blue, and red, respectively. Maltoheptose (G7) has been docked onto the GMCBM21 structure by superimposing GMCBM21 with the crystal structure of the RoGACBM21-G7 complex, highlighting the two sugar binding sites: site 1 (bottom) and site 2 (top).

Structural comparison of GMCBM21 with the CBM21 family

Unlike the majority of SBDs, the N terminus of GMCBM21 contains an additional β strand, β0, and an α helix, α1. The 15N[1H]-NOE data did not show any additional or different flexibility of β0 and α1, when compared with the core β-sandwich structure, showing that these additional regions are not flexible and therefore compose an integral part of the structured domain (fig. S3E). The only other SBD that contains these additional N-terminal secondary structure elements is GLCBM21 [Protein Data Bank (PDB) ID 2EEF; fig. S3F]. A structural homology search conducted using the Dali server shows that GLCBM21 and the Ro glucoamylase CBM21 [RoGACBM21; (23)] are the two most similar structures to GMCBM21 with Z scores of 12 and 10, respectively. The GM, GL, and RoGA SBDs are currently the only proteins for which structural information is available within the CBM21 family (fig. S3F).

Despite belonging to the same CBM21 family, each protein has a distinct surface, both in shape and in charge distribution (fig. S3G). Moreover, the two sugar binding sites in CBM21 family members (site 1 and site 2), which are essential for recognition of intracellular glycogen [α-cyclodextrin (α-CD) and β-cyclodextrin (β-CD), six- and seven-membered sugar ring, respectively, produced from starch], are very different among the three proteins. In RoGACBM21, sugar binding site 1 adopts a rigid and curved surface, which imposes a strict curvature on the sugar molecule (fig. S3F) (20). Site 1 in GMCBM21 and GLCBM21 is flatter (fig. S3G), which may affect the binding affinity of these proteins for glycogen/β-CD. In RoGACBM21, sugar binding site 2 forms a small narrow groove, with two protruding aromatic residues that clamp down over the sugar molecule (fig. S3G) (20). GMCBM21 has a similar “closed clamp” architecture at site 2, whereas GLCBM21 adopts an “open clamp” conformation. This may simply reflect the structural plasticity of site 2 (21), a site known to undergo large conformational rearrangement upon β-CD binding. These data, in addition to the presence of β0 and α1 in only GM and GL CBM21, suggest that these domains may differ in function, despite being in the same CBM21 family.

GM contains a single glycogen binding site

In GMCBM21, the two sugar binding sites are formed by Trp168GM, His171GM, Cys210GM, Trp221GM, Asn223GM, and Asn228GM (site 1), and by Asn152GM, Phe155GM, Glu156GM, Lys157GM, Tyr178GM, and Asp188GM (site 2; fig. S4A). Site 1 is a shallow binding pocket whose curvature is defined by two to three aromatic rings. Site 2 is generally formed by the β2-β3 and β4-β5 loops and is typically dominated by two aromatic residues that undergo large conformational rearrangements upon β-CD binding, acting as a clamp (fig. S4B). Unlike site 1, site 2 is highly conserved in sequence among GMCBM21, GLCBM21, and RoGACBM21 (fig. S4A).

To examine the sugar binding properties of GMCBM21, we performed NMR CSP experiments titrating 15N-labeled GMCBM21 with α-CD or β-CD at 1:0.5, 1:1, 1:2, 1:5, 1:20, and 1:40 (protein:ligand) molar ratios, achieving saturation of binding at a ratio of 1:20 (fig. S4, C and D). In general, both α-CD and β-CD caused similar CSPs, with the largest CSPs observed for Phe155GM, Glu156GM, and Glu186GM in site 2, and for Thr215GM and Ser216GM in the β6-β7 loop immediately adjacent to site 2 (Fig. 3, D and E). Unexpectedly, no CSPs were observed for the residues that form site 1, with an average Δδ of 0.057 parts per million (ppm) for site 1 and 0.333 ppm for site 2 when bound to β-CD. Consistent with this observation, ITC of GMCBM21 with both α-CD and β-CD showed a binding stoichiometry of 1 and KD values of 27.6 ± 5.8 μM for α-CD (fig. S1J) and 8.2 ± 0.1 μM for β-CD (fig. S1K). Together, our data show that soluble cyclodextrins bind GMCBM21 at a single binding site, centered on Phe155GM.

In the presence of either α-CD or β-CD, most 2D [1H,15N] HSQC cross-peaks shift “linearly,” such that the beginning, middle, and end titration points can all be connected with a straight line in the overlaid spectra (fig. S4, C and D). However, a subset of perturbed cross-peaks deviate from linearity and are curved (fig. S4E). Curved cross-peak patterns generally represent a convolution of two or more processes that occur as a result of ligand binding. In GMCBM21, these residues map to the central β-sandwich, not to the sugar binding sites (fig. S4F), suggesting that cyclodextrin binding to site 2 is accompanied by a conformational change that is translated across the center of the protein, but does not allosterically perturb/activate site 1; i.e., these sites are functionally independent.

To confirm that the same CSPs are observed between the GMCBM21 domain and β-CD when GM is bound to PP1, we repeated the CSP NMR titration experiments of β-CD with the GM64–237:PP1 holoenzyme (fig. S5, A and B). Site 2 showed the same CSPs as identified for GMCBM21 alone. These data show that GM is able to bind glycogen and PP1 simultaneously, further supporting its function as a scaffolding protein.

Glycogen enhances dephosphorylation of phosphorylase a by scaffolding

Substrate recruitment is poorly understood for most PP1 holoenzymes, mainly because very few substrates have been identified. This is different for the GM:PP1 holoenzyme, whose primary substrates are phosphorylase a and glycogen synthase. This enables a detailed study of the mechanism(s) by which these substrates are recruited to PP1. Phosphorylase a is phosphorylated on Ser14, which is the specific substrate of PP1. To understand the recruitment of the substrate phosphorylase a by the GM64–237:PP1 holoenzyme, we measured the dephosphorylation of phosphorylase a by PP1 alone or by the GM49–86:PP1, GM64–105:PP1, GM64–237:PP1, NIPP1158–216:PP1, and spinophilin417–602:PP1 holoenzymes (fig. S6A). Different GM constructs were used to determine how GM-interacting domains influence substrate recruitment, while the PP1-specific regulators nuclear inhibitor of protein phosphatase 1 (NIPP1) (24) and spinophilin (25) were used as negative controls [phosphorylase a is not an endogenous substrate of either the NIPP1:PP1 or the spinophilin:PP1 holoenzyme; further, previous studies showed that both holoenzymes inhibit phosphorylase a dephosphorylation (17, 26)]. The data show that PP1 alone and all GM:PP1 holoenzymes were equally effective at dephosphorylating phosphorylase a (Fig. 4A and fig. S6B). This demonstrates that GMCBM21 does not alter PP1’s activity toward phosphorylase a. In contrast, and consistent with previous data, the PP1 binding domains from both NIPP1 and spinophilin inhibit phosphorylase a dephosphorylation.

Fig. 4 Dephosphorylation of phosphorylase a and GYS1 by GM:PP1.

(A) Relative phosphorylation level of phosphorylase a following incubation alone (brown circle) or with PP1 (coral square), various GM:PP1 holoenzymes (green triangle, cyan triangle, blue diamond), NIPP1158–216:PP1 (orange circle), spinophilin417–602:PP1 (pink circle), NG1:PP1 (green circle), or NG2:PP1 (light blue circle). Only NIPP1158–216, spinophilin417–602, and NG1 inhibit the PP1-mediated dephosphorylation of phosphorylase a [error bar corresponds to the SD with an n between 4 and 12; statistical significance was determined by one-way analysis of variance (ANOVA): ****P < 0.0001, Tukey’s multiple comparisons test]; n.s., not significant. (B) Glycogen enhances the GM64–237:PP1-mediated dephosphorylation of phosphorylase a. Time course of the relative phosphorylation of phosphorylase a in the presence of PP1 (black circle), GM64–237:PP1 (red square) or GM64–237:PP1, and glycogen [glycogen (4 mg/ml); purple triangle; two-way ANOVA, P < 0.0001]. (C) Phosphorylase a and the GM64–237:PP1 holoenzyme bind directly to glycogen. Glycogen pelleting assay using Con A beads incubated with the biomolecules indicated (n = 3). (D) Gel shift analysis shows that GM64–237:PP1 causes the largest shift in the migration of the GYS1 band (error bar corresponds to the SD, n = 5; one-way ANOVA with post hoc Tukey test); n.s., not significant. (E) Time course of dephosphorylation measured by gel shift. GM64–237 (red square) enhances the dephosphorylation of GYS1 by PP1 compared to PP1 alone (black circle; two-way ANOVA, P = 0.0017). The single-point mutation GMCBM21 N228A (blue triangle) eliminates this enhancement.

Because phosphorylase a, like GM, binds directly to glycogen (via phosphorylase residues 397 to 437), we hypothesized that the dephosphorylation of phosphorylase a by the GM64–237:PP1 holoenzyme would be enhanced in the presence of glycogen. To test this, we measured phosphorylase a dephosphorylation at multiple time points in the presence and absence of glycogen. The data show that glycogen greatly enhances phosphorylase a dephosphorylation (Fig. 4B and fig. S6C). Thus, this leads to a model where the recruitment of both phosphorylase a and the GM64–237:PP1 holoenzyme to glycogen is necessary to achieve the highest dephosphorylation. To test this, we performed a glycogen pelleting assay using glycogen-bound concanavalin A (Con A)–Sepharose beads (27). The data showed that phosphorylase a, GM64–237, and the GM64–237:PP1 holoenzyme (both alone and in combination with phosphorylase a and/or GM64–237) robustly bind glycogen. In contrast, PP1 alone does not (Fig. 4C). This demonstrates that glycogen serves as the scaffold that recruits both phosphorylase a and the GM64–237:PP1 holoenzyme to facilitate the GM64–237:PP1 holoenzyme–mediated dephosphorylation of phosphorylase a.

PP1 regulator mediated inhibition of phosphorylase a dephosphorylation

PP1 has ~200 confirmed distinct regulators, including GM. However, while the GM64–237:PP1 holoenzyme is fully capable of dephosphorylating its endogenous substrate, phosphorylase a, other PP1 regulators can inhibit PP1-mediated dephosphorylation. Previously, we found the molecular mechanism by which spinophilin prevents PP1 from dephosphorylating phosphorylase a (17). Namely, our crystal structure of the spinophilin:PP1 holoenzyme showed that spinophilin binds the PP1 C-terminal binding groove, blocking access to Asp71PP1 (28). This residue is essential for phosphorylase a dephosphorylation. Thus, spinophilin and other PP1 regulators that bind the same substrate binding groove on PP1 (i.e., PNUTS) inhibit phosphorylase a dephosphorylation by steric exclusion. That is, the regulators block phosphorylase a from binding PP1 in the C-terminal groove and, as a consequence, it is not dephosphorylated. While it is well documented that NIPP1 also inhibits the dephosphorylation of phosphorylase a, the molecular mechanism by which this is achieved is still unknown (our crystal structure of the NIPP1:PP1 holoenzyme revealed that, unlike spinophilin, NIPP1 does not bind the C-terminal groove and thus Asp71PP1 is accessible) (26).

To further understand the molecular basis of NIPP1-mediated inhibition, we generated two chimeric PP1 regulators: NG1:NIPP1158–198GM64–100 and NG2:NIPP1175–198GM64–100. NG1 includes the NIPP1helix (residues 158 to 174), the NIPP1connector (residues 175 to 198; not visible in the NIPP1:PP1 complex crystal structure), and the full primary PP1-anchoring domain from GM (GM64–85). NG2 is identical to NG1, with the exception that it does not include the NIPP1helix (fig. S6A). The data show that the NG1:PP1 complex inhibits phosphorylase a dephosphorylation to the same extent as NIPP1:PP1, showing that the inhibition is due to either the NIPP1helix or the NIPP1linker, or both, but not the RVxF or ΦΦ interaction (i.e., GM or NIPP1 is exchangeable) (Fig. 4A and fig. S6D). By comparison, the NG2:PP1 complex inhibits phosphorylase a dephosphorylation poorly and allows for rapid dephosphorylation similar to GM64–237:PP1. We have previously shown that mutating positively charged residues within the NIPP1linker to a poly-A stretch (NIPP1 residues 193KRKRK197 mutated to 193AAAAA197) in the NIPP1linker does not change inhibition significantly (26). Thus, these data demonstrate that the NIPP1helix is likely critical for the ability of NIPP1-mediated inhibition of PP1’s ability to dephosphorylate phosphorylase a. These data correlate with those obtained for the PP1 regulator MYPT1, which has an MYPT1helix that binds PP1 in an area similar to that of the NIPP1helix (29). Together these data provide insight into how the substrate phosphorylase a engages PP1 and highlights that GM does not alter PP1’s activity toward phosphorylase a itself; rather, it forms a glycogen recruitment platform that allows phosphorylase a recruitment via glycogen scaffolding.

GM enhances the dephosphorylation of glycogen synthase

A second confirmed substrate of GM64–237:PP1 is GYS1 (muscle specific glycogen synthase; 84 kDa) (14). GYS1 catalyzes the conversion of glucose to glycogen and is activated by two mechanisms: (i) allostery, as it is positively allosterically regulated by glucose-6-phosphate, and (ii) dephosphorylation, as the dephosphorylation of specific Ser/Thr residues increases GYS1 activity (GYS1 is phosphorylated by glycogen synthase kinase 3, AMP-activated protein kinase (AMPK), PKA, and casein kinase 2, which inhibit GYS1 activity). Dephosphorylation, which activates GYS1, is mainly controlled by PP1 via the recruitment by the GM:PP1 holoenzyme. To understand how GM:PP1 dephosphorylates GYS1, we produced functional human muscle GYS1 by coexpressing it in complex with human glycogenin-1 (GYG1; muscle specific; 66 kDa; glycogenin-1 binds directly to glycogen). Production of GYS1 in this manner results in an enzyme that is phosphorylated at both well-characterized and uncharacterized Ser/Thr residues (30). Incubation of GYS1 with PP1 shows a significant shift in the migration of the GYS1 band in SDS-PAGE (polyacrylamide gel electrophoresis), as often seen when an extensively phosphorylated protein is dephosphorylated (fig. S6, E and F). The same shift is also detected when GYS1 is incubated with GM49–86:PP1 and GM64–105:PP1, showing that the GM anchoring PP1 binding domain does not specify or influence the PP1 activity toward GYS1. However, GM64–237:PP1, which includes GMCBM21, showed a significantly larger shift, indicating that the CBM21 domain facilitates the recruitment and dephosphorylation of substrate GYS1 (Fig. 4D).

To identify the residues selectively dephosphorylated by the GM64–237:PP1 holoenzyme, we used liquid chromatography–tandem mass spectrometry (LC-MS/MS). It is well known that the phosphorylation state of multiple residues of GYS1 is directly correlated with its activity, including residues 8 and 11 (sites 2 and 2a) and residues 641, 645, and 649 (sites 3a/b/c) (31). We focused our analysis on Ser641GYS1 (3a) and Ser645GYS1 (3b), as the dephosphorylation of both residues is essential for GYS1 activity (sites 2 and 2a were not identified by LC-MS/MS and thus were not analyzed) (fig. S7, A and B). The MS data showed that only singly (3a) or doubly (3a/b) phosphorylated peptides were detected for GYS1; no unphosphorylated or singly phosphorylated (3b) peptide was observed. However, incubation with PP1 or any of its holoenzymes [GM49–86:PP1, GM64–105:PP1, GM64–237:PP1, and spinophilin:PP1 (used as control)] resulted in the complete disappearance of the single phosphorylated 3a peptide (fig. S7C). This demonstrates that this residue is rapidly dephosphorylated by all versions of PP1. In contrast, the doubly phosphorylated 3a/b peptide was most effectively dephosphorylated by the GM64–237:PP1 holoenzyme compared to either PP1 alone or any other PP1 holoenzymes (fig. S7D). This enhanced dephosphorylation by GM64–237:PP1 was also observed for the singly phosphorylated 3b peptide (note that this peptide is only present after incubation with PP1 or one of its holoenzymes due to the dephosphorylation of 3a) (fig. S7E). Further, the amount of unphosphorylated 3a/b peptide present after incubation is highest for the GM64–237:PP1 holoenzyme (fig. S7F). Last, the data also show that GYS1 phosphorylated residues that are not correlated with GYS1 activity, such as Thr278 and Ser412, were not dephosphorylated by either PP1 or any of the PP1 holoenzymes (fig. S7, G and H). Together, these data show that recruitment of PP1 to the GMCBM21 via GM64–85 creates a specific enzyme that activates GYS1 and that GMCBM21 functions as the specifier, but not any GM residue that is directly interacting with PP1.

To confirm these results, we performed two additional experiments. First, we repeated the experiment with glycogen (fig. S8). In contrast to phosphorylase a, no dephosphorylation enhancement is detected upon the addition of glycogen, showing that the enhancement comes solely from GMCBM21. Second, we generated a GMCBM21 N228A variant, a residue that is part of a highly conserved patch of residues and was previously speculated to be involved in GYS1 recruitment (14). Repeating the dephosphorylation assay with GM64–237:PP1 N228A showed a significant reduction in dephosphorylation, confirming that N228A is essential for the recruitment of GYS1 to PP1 (Fig. 4E).


The balance between glucose storage in the form of glycogen and its subsequent breakdown is controlled by phosphorylation. A single phosphatase, the GM:PP1 holoenzyme, specifically dephosphorylates three of the key enzymes that control glycogen synthesis and breakdown: phosphorylase kinase, phosphorylase a, and glycogen synthase. While many insights into PP1 activity and function of PP1 have been obtained by studying GM:PP1, a molecular understanding of how PP1 generally and GM:PP1 in particular recruits and selectively dephosphorylates its specific substrates is still largely missing. GM:PP1 is uniquely positioned to answer these questions, as its substrates are well described. Phosphorylase a is the canonical substrate for measuring PP1 activity, and thus a detailed molecular understanding of substrate recruitment will have a profound impact on understanding PP1 regulation.

Our NMR spectroscopy, crystallography, enzymatic, and molecular binding data show that GM binds PP1 via a much longer domain than previously thought, including the canonical RVxF motif and an unusually extended ΦΦ motif that is connected by a highly structured, kinked linker. As is typical for PP1 holoenzymes, these interactions do not alter the conformation of PP1, nor do they block its active site. Thus, GM binding does not alter the catalytic activity of PP1 toward its substrates. Yet, the GM:PP1 holoenzyme is highly selective for its endogenous substrates. Previous data suggested this specificity resides in GM residues 102 to 240, which include the GMCBM21 domain. Thus, we determined the 3D structure of GMCBM21, which showed that it belongs to the family of SBDs. Unexpectedly, using CSP mapping, we found that GMCBM21 has only one starch binding site, unlike many of its closest family members, which commonly have two (20). This site is necessary and, as we have now shown, sufficient for the effective recruitment of glycogen. In GM, however, we show that the second starch binding site has evolved to (i) interact with PP1 and (ii) bind glycogen synthase (Fig. 5A). These data, coupled with multiple enzymatic assays, led to the discovery that GMCBM21 achieves GM-mediated substrate recruitment via two distinct mechanisms. In the first mechanism, GM functions as a scaffold to localize PP1 near its substrates; i.e., the GM PP1 binding domain binds PP1, while the GMCBM21 domain binds glycogen. Glycogen binding, in turn, targets GM:PP1 to one of its glycogen-specific substrates, phosphorylase a (Fig. 5B). This is because ~70% phosphorylase a is always glycogen bound (32). The phosphorylation of Ser67GM by PKA releases PP1 from GM (16), leading to a model in which GM stays localized near its substrate via glycogen binding, while PP1 (via the phosphorylation state of Ser67GM) is recruited only when needed. The second mechanism is direct binding between GM and a GM:PP1-specific substrate. Namely, the GMCBM21 domain binds directly to GYS1 to recruit this substrate to the GM:PP1 holoenzyme for the PP1-mediated dephosphorylation (Fig. 5C).

Fig. 5 Substrate recruitment by GM:PP1.

(A) Model showing the interaction of GM64–237 with PP1; GYS1 (site 1) and glycogen binding sites (site 2) within the GMCBM21 are highlighted. (B) Glycogen-mediated recruitment of PP1 and phosphorylase a leads to phosphorylase a dephosphorylation by PP1. (C) Dephosphorylation of GYS1 by PP1 is mediated by direct substrate recruitment to site 1 within GMCBM21.

As previously highlighted, phosphorylase a is also the canonical substrate used to measure PP1 activity. This has led to the discovery that many PP1-specific regulatory proteins potently inhibit PP1-mediated dephosphorylation of phosphorylase a. One of these is the PP1-specific regulator spinophilin. Using x-ray crystallography and enzymatic assays, we previously showed that spinophilin inhibits the dephosphorylation of phosphorylase a by blocking its access to the PP1 C-terminal substrate binding groove (17). This demonstrates that regulators can achieve substrate selectivity, in part, by sterically excluding binding site for subsets of substrates. A second regulator that inhibits dephosphorylation of phosphorylase a is NIPP1. However, its structure showed that the C-terminal substrate binding groove is fully accessible, and thus, that it inhibits phosphorylase a dephosphorylation by a distinct mechanism (26). Here, we used NIPP1-GM chimeras to show that the NIPP1helix, which binds at the entrance of a second substrate binding groove in PP1, the hydrophobic substrate binding groove, is important for NIPP1’s ability to inhibit the dephosphorylation of phosphorylase a. Thus, the NIPP1helix likely sterically blocks the access of phosphorylase a to the PP1 hydrophobic groove. Together, these data reveal that robust dephosphorylation of phosphorylase a requires its ability to bind both the hydrophobic and the C-terminal substrate binding grooves.

This study provides the most comprehensive molecular understanding of how a specific PP1 holoenzyme, GM:PP1, mediates the rapid and selective dephosphorylation of its specific substrates. In particular, it highlights the essential role of distinct recruitment domains present in PP1-specific regulatory proteins for directing the specificity of PP1. These domains (GMCBM21 in GM) recruit PP1 to their cellular points of action (glycogen) that, in turn, convert PP1 into an exquisitely specific enzyme, either indirectly via localizing PP1 to its specific substrates (i.e., phosphorylase a, which, like GM:PP1, also binds directly to glycogen) or directly by binding to the substrate itself (i.e., GYS1, which binds directly to GMCBM21). Similar mechanisms for substrate recruitment have also been observed in other PP1-specific regulators. For example, the PSD95/Discs large/ZO-1 (PDZ) domain of the PP1-specific regulator spinophilin binds to the C termini of GluR2/3 subunits of the AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid) receptor, and thus functions to localize PP1 to its specific substrate, Ser845 of the GluR1 (33). Likewise, the forkhead associated (FHA) domain in the regulator NIPP1 binds directly to specific substrates of the NIPP1:PP1 holoenzymes (34). Thus, these data show that to fully understand the function of a particular PP1 holoenzyme, and especially its specificity toward distinct substrates, it is essential to understand the functions of all the domains in cognate PP1 regulatory protein. As we and others have shown, substrate recruitment sites on PPP holoenzymes are functional drug binding sites (e.g., as seen for PP2B/PP3/calcineurin and FK-506) (35). Thus, the identification and characterization of these specific and unique interactions, especially in PP1 holoenzymes, will lead to the design of potent, effective PP1-selective drugs.


Protein expression

The coding sequences of rabbit GM64–93, GM64–237, and GM2–237 were subcloned into a pET-M30-MBP vector containing an N-terminal His6-tag followed by maltose binding protein (MBP) and a tobacco etch virus (TEV) protease cleavage site. GMCBM21 (residues 102 to 237) was subcloned into pRP1B containing an N-terminal His6-tag followed by a TEV protease cleavage site. Escherichia coli strain BL21-Codon-Plus (DE3)-RIL (Agilent) cells were transformed with the GM expression vectors. Freshly transformed cells were grown at 37°C in LB medium containing selective antibiotics until they reached an OD600 (optical density at 600 nm) of 0.8 to 1.0. Protein expression was induced by addition of 1 mM β-d-thiogalactopyranoside to the culture medium, and cultures were allowed to grow overnight (18 to 20 hours) at 18°C. Cells were harvested by centrifugation (6000g, 15 min, 4°C) and stored at −80°C until purification.

Expression of uniformly 13C- and/or 15N-labeled protein was carried out by growing freshly transformed cells in M9 minimal media containing [13C]-d-glucose (4 g/liter) and/or 15NH4Cl (1 g/liter) (Cambridge Isotopes Laboratories) as the sole carbon and nitrogen sources, respectively. Expression of uniformly [2H,15N]-labeled GM64–237 was achieved by growing cells in D2O-based M9 minimal media containing 15NH4Cl (1 g/liter) as the sole nitrogen source. Multiple rounds (0, 30, 50, 70, and 100%) of D2O adaptation were necessary for high-yield expression. Cloning, expression, and purification of PP1α7–300, PP1α7–330, PP1β6–327, PP1γ7–308, and PP1γ7–323 were performed as previously described (18); human and rabbit PP1 are 100% identical. GM64–93, GM49–86, and GM64–105 peptides were purchased from Bio-Synthesis Inc. GM64–237 N228A was generated by site-directed mutagenesis and expressed using the same methods as those used for wt-GM64–237.

The plasmid of human glycogen synthase (pFastBacDual GST-GYG1+GYS1) was a gift from E. Zeqiraj, University of Leeds. Expression and purification were carried out according to previously published methods (30). Recombinant bacmid was generated in DH10Bac cells (Thermo Fisher Scientific) and purified using the PureLink HiPure Plasmid Maxiprep Kit (Thermo Fisher Scientific). Sf9 cells were cultured as suspension in Sf-900 III SFM medium (Thermo Fisher Scientific; 110 rpm at 26°C). P1 baculorvirus was produced in monolayer cultures by transfecting the recombinant bacmid using Cellfectin II Reagent (Thermo Fisher Scientific). P2 virus was generated by infecting Sf9 cultures (1.6× 106 cells/ml) with P1 virus at 400 μl per 100 ml of cells. The supernatant from P2 was harvested 3 days after infection, and 3 ml was used to infect 600 ml of Sf9 suspension culture to produce the P3 culture. P3 was used at a 1:10 ratio to infect 3 liters of Sf9 cell culture at 2.0 × 106 cells/ml for protein production. Cells were grown in suspension for 3 days, and the cell pellets were washed in phosphate-buffered saline before they were frozen and stored at −80°C until used.

Protein purification

GMCBM21, GM64–237, GM2–237, and GM64–237 N228A cell pellets were resuspended in ice-cold lysis buffer [50 mM tris (pH 8.0), 500 mM NaCl, 5 mM imidazole, 0.1% Triton X-100, and an EDTA-free protease inhibitor tablet (Roche)] and lysed by high-pressure homogenization (Avestin EmulsiFlex C3). Lysate was clarified by centrifugation (45,000g, 45 min, 4°C), and the supernatant was loaded onto a HisTrap column (GE Healthcare) pre-equilibrated with 50 mM tris (pH 8.0), 500 mM NaCl, and 5 mM imidazole. Protein was eluted using a gradient of 5 to 500 mM imidazole. Fractions containing GMCBM21 were pooled and dialyzed overnight at 4°C against 50 mM tris (pH 8.0), 500 mM NaCl, and 0.5 mM tris(2-carboxyethyl)phosphine (TCEP) to cleave the His6-tag. The cleaved protein was incubated with the Ni+2-NTA (nitrilotriacetic acid) beads (GE Healthcare) to remove the TEV protease and cleaved His-tag. The flow-through was collected, concentrated, and further purified using size exclusion chromatography [SEC; Superdex 75 26/60 (GE Healthcare)] equilibrated in NMR buffer [20 mM sodium phosphate (pH 6.5), 50 mM NaCl, and 10 mM dithiothreitol (DTT) or 20 mM bis-tris (pH 6.8), 150 mM NaCl, and 0.5 mM TCEP] or ITC buffer [20 mM tris (pH 8.0), 0.5 M NaCl, 0.5 mM TCEP, and 1 mM MnCl2]. Fractions were pooled, concentrated, and stored at −20°C. 15N-labeled GM64–93 was purified identically except that the protein was heat purified at 95°C (15 min), and the supernatant was collected and concentrated before SEC. PP1 was purified as previously described (18).

The GST-GYG1:GYS1 complex was purified using glutathione agarose beads (Pierce). The Sf9 cell pellet (8 g) was lysed in ice-cold lysis buffer [50 mM tris (pH 8.0), 150 mM NaCl, 5% glycerol, 1 mM EDTA, and 0.1% Triton X-100] with cOmplete mini protease inhibitor cocktail tablets (Roche). Lysate was clarified by centrifugation (40,000g, 45 min) and filtered through a 0.22-μm syringe filter. The supernatant was incubated on a rolling platform for 1 hour at 4°C with 1-ml bed volume of glutathione agarose resin pre-equilibrated in low-salt buffer [50 mM tris (pH 8.0), 150 mM NaCl, 5% glycerol, and 1 mM EDTA]. The beads were then washed with 10 column volumes (CVs) of low-salt buffer, followed by 50 CVs of high-salt buffer [50 mM tris (pH 8.0), 500 mM NaCl, 5% glycerol, and 1 mM EDTA], and followed again by 10 CVs of low-salt buffer. The complex was eluted with 10 mM fresh reduced glutathione, concentrated to 1 mg/ml, flash frozen in liquid nitrogen, and stored at −80°C until needed.

GM64–237:PP1α7–300/330 complex formation

To purify the GM64–237:PP1α7–330 complex for NMR spectroscopy analysis, PP1α7–330 was lysed in PP1 Lysis Buffer [25 mM tris (pH 8.0), 700 mM NaCl, 5 mM imidazole, 1 mM MnCl2, and 0.1% Triton X-100], clarified by ultracentrifugation, and immobilized on Ni2+-NTA resin. Bound His6-PP1 was washed with PP1 Buffer A [25 mM tris (pH 8.0), 700 mM NaCl, 5 mM imidazole, and 1 mM MnCl2], followed by a stringent wash containing 6% PP1 Buffer B [25 mM tris (pH 8.0), 700 mM NaCl, 250 mM imidazole, and 1 mM MnCl2] at 4°C. The protein was eluted using PP1 Buffer B and purified using SEC [Superdex 200 26/60 (GE Healthcare)] pre-equilibrated in ITC Buffer [20 mM tris (pH 8), 500 mM NaCl, 0.5 mM TCEP, and 1 mM MnCl2]. Peak fractions were incubated overnight with TEV protease at 4°C. The cleaved protein was incubated with Ni2+-NTA beads (GE Healthcare), and the flow-through was collected. The flow-through was combined with excess 2H/15N-labeled GM64–237 and concentrated, and the complex was purified using SEC [pre-equilibrated in 20 mM bis-tris (pH 6.8), 150 mM NaCl, and 0.5 mM TCEP]. Fractions containing the holoenzyme complex were concentrated to 0.1 mM for NMR studies.

To generate the GM64–93:PP1α7–300 complex for crystallization, purified PP1 was incubated with microcystin-LR (MC-LR), and GM64–93 was added to a final ratio of 1:1:5 in crystallization buffer [20 mM tris (pH 8.0), 50 mM NaCl, 0.5 mM TCEP, and 1 mM MnCl2]. The GM64–93 peptide was prepared by dissolving 2 mg of peptide in 1% NH4OH before diluting with buffer [20 mM bicine (pH 9.0) and 5 mM DTT]. Final complex concentration was ~8 mg/ml for crystallization trials using vapor diffusion (sitting drop).

NMR spectroscopy

All NMR experiments were acquired at 298 K on Bruker Avance 500 or 800 MHz spectrometers, both equipped with a TCI HCN-z cryoprobe. The following spectra were used to complete the sequence-specific backbone assignment (recorded at 500-MHz 1H Larmor frequency): 2D [1H,15N] HSQC, 3D HNCACB, 3D CBCA(CO)NH, 3D HNCA, 3D (H)CC(CO)NH, and 3D HBHA(CO)NH. Together with these spectra, 3D HC(C)H–total correlation spectroscopy (TOCSY) (Tm = 11.3 ms) was used for the assignment of aliphatic side-chain 1H and 13C resonances. Aromatic side chains were assigned using 2D [1H,1H] NOE spectroscopy (NOESY) (Tm = 70 ms), 2D [1H,1H] TOCSY (Tm = 60 ms), and 2D [1H,1H] correlation spectroscopy (COSY) spectra of GMCBM21 in 20 mM sodium phosphate (pH 6.5), 50 mM NaCl, 10 mM DTT, and 100% D2O. A 2D 15N[1H]-NOE (heteronuclear NOE) experiment was recorded at 500 MHz 1H Larmor frequency with a saturation delay of 5 s and evaluated using the Dynamics Center 2.0 software (Bruker).

All spectra were processed using Topspin 2.1/3.0/3.1 (Bruker, Billerica, MA), and chemical shift assignments were achieved using Cara ( NMR spectra of GMCBM21 were acquired using either 15N- or 15N,13C-labeled protein at a final concentration of 0.8 mM in 20 mM sodium phosphate (pH 6.5), 50 mM NaCl, 10 mM DTT, and 90% H2O/10% D2O. The interaction of GMCBM21 with carbohydrates was tested by NMR titration experiments using α-CD (Thermo Fisher Scientific) and β-CD (Acros Organics).

For GMCBM21, only Asn224, Asn228, and two cloning artifacts (His2 and Met3) have no sequence-specific backbone assignment. The high-quality spectral data also enabled a 98% completeness of the side-chain assignment, except Pro110 Cδ/Qδ resonances and all of the side-chain resonances for Gly1, His2, Phe192, Phe220, and Tyr106. Aromatic side-chain assignment was more challenging owing to the large number of aromatic residues in GMCBM21—a total of 16—a characteristic feature of SBDs.

The interaction between GM64–237 and PP1α7–330 was studied by direct comparison of 2D [1H,15N] TROSY spectra of free and PP1α7–330-bound (2H,15N)-labeled GM64–237. The final concentration used was 0.1 mM GM64–237:PP1α7–330 complex in 20 mM bis-tris (pH 6.8), 150 mM NaCl, 0.5 mM TCEP, and 90% H2O/10% D2O. The spectra were processed using Topspin 4.0.3 (Bruker, Billerica, MA) and analyzed using Sparky. The NMR spectra were acquired on a Bruker Avance NEO 800 MHz 1H Larmor frequency NMR spectrometer equipped with a TCI-active HCN-cooled z-gradient cryoprobe at 298 K.

Structure calculation of GMCBM21

The following spectra were used for structure calculation: 3D 15N-resolved [1H,1H] NOESY (Tm = 70 ms, 800 MHz 1H Larmor frequency), 3D 13C-resolved [1H,1H] NOESY (Tm = 70 ms, 800 MHz 1H Larmor frequency), and 2D [1H,1H] NOESY (Tm = 70 ms, 100% D2O solution, 500 MHz 1H Larmor frequency). Automated NOESY peak picking and NOE assignment were carried out using ATNOS/CANDID (automated NOESY peak picking/combined automated NOE assignment and structure determination module) (36). A total of 2709 unambiguous NOESY-derived distance restraints along with 110 dihedral angle restraints derived from 13C-chemical shifts were used in the initial structure calculations performed using CYANA (combined assignment and dynamics algorithm for NMR applications). Final energy minimization and structure refinement were performed in explicit solvent using CNS 1.3 (Crystallography and NMR system), along with the RECOORD (Recalculated Coordinates Database) script package. A total of 200 structures were generated, and the 20 conformers with the lowest restraint violation energies were selected as the final representative model. The quality of the structures was assessed by the programs WHATCHECK, AQUA (Analyzing the Quality), NMR-PROCHECK, and MOLMOL. Ramachandran analysis showed that the final bundle of 20 lowest-energy conformers of GMCBM21 has excellent stereochemistry, with 97.8% of residues in the most favored and allowed region, 2.0% in the generously allowed region, and 0.2% in the disallowed region.

CSP experiments with α-CD and β-CD

The interaction of GMCBM21 with carbohydrates was tested by NMR titration experiments using α-CD (Thermo Fisher Scientific) and β-CD (Acros Organics). To this end, 2D [1H,15N] HSQC spectra at 298 K were recorded at 500 MHz 1H Larmor frequency. Experiments were performed in 20 mM phosphate (pH 6.5), 50 mM NaCl, 10 mM DTT, and 90% H2O/10% D2O, with α-CD and β-CD titrated into 15N-labeled GMCBM21 at 1:1, 1:2, 1:5, 1:10, 1:20, and 1:40 (protein:sugar) molar ratios. 15N-GMCBM21 was used at concentrations of 300 μM for the 1:0.5, 1:1, and 1:2 titrations, 190 μM for the 1:20 titration, 100 μM for the 1:40 titration, and 40 μM for the 1:5 titration. In the 1:20 and 1:40 titration points, the concentration of GMCBM21 was limited by the maximum solubility of β-CD in this buffer (less than 10 mM). Saturation of sugar binding by GMCBM21 was achieved at a ratio of 1:20 for both α-CD and β-CD. Chemical shift differences (Δδ) between free GMCBM21 (no α-CD or β-CD) and sugar-bound GMCBM21 (1:20 molar ratio) spectra were calculated usingEmbedded Image

To test the interaction of carbohydrates with GM64–237:PP1α7–330, a 2D [1H,15N] TROSY spectrum was recorded with β-CD (1:20 molar ratio; for GMCBM21 alone, saturation was observed at this ratio). β-CD was chosen because of its stronger binding affinity to GMCBM21 when compared to α-CD. Chemical shift differences between GM64–237:PP1α7–330 and GM64–237:PP1α7–330:β-CD were calculated as described above. The data were recorded on a Bruker Avance NEO 800MHz 1H Larmor frequency equipped with a cryoprobe at 298 K.

Crystallization and structure determination of the GM64–93:PP1α7–300:MC-LR complex

GM64–93:PP1α7–300:MC-LR holoenzyme crystallized as clusters or single rod-shaped crystals in 0.4 M magnesium formate dihydrate and 0.1 M sodium acetate trihydrate (pH 4.6) at 4°C. For x-ray diffraction, crystals were cryoprotected in 30% glycerol and immediately flash frozen in liquid N2. Diffraction data were collected at the Stanford Synchrotron Radiation Lightsource beamline 12-2 at 100 K using a Dectris PILATUS 6M detector. The structure of GM64–93:PP1α7–300:MC-LR was determined by molecular replacement using Phaser as implemented in Python-based Hierarchical Environment for Integrated Xtallography (PHENIX) (37). The PDB of PP1 (4MOV) was used as the search model. A solution was obtained in space group C2221. The model was completed using iterative rounds of refinement in PHENIX and manual building using Coot (Ramachandran statistics: 95.8% favored and 4.2% allowed).

Isothermal titration calorimetry

ITC experiments testing the interaction between GMCBM21 and α-CD and β-CD were performed at 25°C using a VP-ITC microcalorimeter (Malvern). Both α-CD and β-CD were dissolved in 20 mM sodium phosphate (pH 6.5), 50 mM NaCl, and 10 mM DTT. Concentrations of GMCBM21 between 14.5 and 16 μM were used in the sample cell. Ligand was titrated in 10-μl increments over 20 s at concentrations of 410, 440, and 1030 μM for α-CD and 450 μM for β-CD (performed in duplicate). Twenty-eight injections were delivered during each experiment, with a 250-s interval between titrations to allow for complete equilibration and baseline recovery, and the solution in the sample cell was stirred at 307 rpm to ensure rapid mixing. To determine the thermodynamic parameters (ΔH, ΔS, and ΔG) and binding constant (Ka), data were analyzed with a one-site binding model assuming a binding stoichiometry of 1:1 using the Origin 7.0 software.

His6-tagged PP1s (PP1α7–300, PP1α7–330, PP1β6–327, PP1γ7–308, and PP1γ7–323) were purified as described for ITC analysis (18, 19). GM (30 or 40 μM) was titrated into PP1 (3 or 4 μM) using a VP-ITC microcalorimeter (Malvern) or an Affinity SV ITC (TA Instruments) at 25°C. Data were analyzed using NITPIC, SEDPHAT, and GUSSI for a one-site binding model.

Surface plasmon resonance

Measurements were conducted using a BI-4500A five-channel SPR with autosampler and degasser pump (Biosensing Instrument Inc.) and a Ni-NTA chip. His6-tagged PP1α7–330 (62.5 nM) in 20 mM tris (pH 8.0), 500 mM NaCl, 0.5 mM TCEP, 1 mM MnCl2, and 0.005% Tween-20 was loaded onto a Ni2+-NTA chip (Biosensing Instrument Inc.) using different loading times (20, 40, 60, and 80 s) to achieve different PP1 densities on the His6-sensor chip in four different channels (channel 1 was the reference channel). GM64–105 was prepared in the same buffer as PP1α7–330 using 1:3 serial dilutions (31.25 to 500 nM). Kinetic parameters were determined by curve fitting using Scrubber (BioLogic Software).

Dephosphorylation assay

Rabbit liver glycogen (Sigma-Aldrich, G8876) and rabbit phosphorylase a (Sigma-Aldrich, P1261) were purchased. All assays were carried out at 30°C for 30 min if not otherwise stated. Dephosphorylation of phosphorylase a was performed at a final concentration of 2 μM in reaction buffer [50 mM tris (pH 7.8), 150 mM NaCl, and 1 mM TCEP]. To investigate the effects of the PP1 interactors GM49–86, GM64–105, NIPP1158–216, spinophilin417–602, NG1, or NG2 toward phosphorylase a dephosphorylation, the interactors were added at a 10 M excess to PP1 and incubated for 30 min at room temperature before the initiation of the assay. The final concentration of PP1 for the steady-state experiments was 0.2 μM, except for the time point experiment for which 0.04 μM PP1 was used. To test the effect of glycogen on the dephosphorylation of phosphorylase a, glycogen (4 mg/ml) was added to phosphorylase a before the addition of the GM64–237:PP1 holoenzyme. The reactions were terminated by the addition of 5× SDS loading buffer, and the samples were boiled (95°C) for 5 min. The samples were analyzed using SDS-PAGE, fixed, and stained with Pro-Q Diamond and Sypro Ruby (Thermo Fisher Scientific) to quantify the phosphor-protein (phosphorylase a) and total proteins, respectively. Gel images were captured using a ChemiDoc MP Imaging system or a Pharos FX Imager (Bio-Rad), and the densitometry of protein bands was analyzed using Image Lab 6.0 (Bio-Rad).

The GST-GYG1:GYS1 complex (final concentration of 0.1 mg/ml) in assay buffer [50 mM tris (pH 7.8), 150 mM NaCl, 1 mM TCEP, and 5% glycerol] was used in all assays. A 10 M excess of GM49–86, GM64–105, GM64–237, or spinophilin417–602 was incubated with PP1 (0.2 μM) for 30 min before the initiation of the assay. For the time point study, reactions were initiated by the addition of 0.2 μM His6-PP1α7–330 with or without GM64–237 or GM64–237 N228A. To test the effect of glycogen, glycogen (4 mg/ml) was added to GYS1 before the addition of the GM64–237:PP1 holoenzyme. The reactions were carried out at 30°C and were terminated at different time points (3, 6, 9, 12, 15, and 30 min) by the addition of 5× SDS loading buffer and boiling at 95°C for 5 min. Samples were analyzed using SDS-PAGE and stained with Sypro Ruby. Dephosphorylation of GYS1 can be readily visualized as a band shift in SDS-PAGE (30). The relative distance between the GYS1 bands relative to GST-GYG1 was used to assay the dephosphorylation of GYS1. Last, Pro-Q Diamond (Invitrogen), which stains for total phosphorylation, and mass spectrometer analysis were used to confirm the dephosphorylation states of GYS1.

Con A–Sepharose glycogen binding assay

Con A–Sepharose (GE Healthcare) beads were used to assay glycogen binding. Fresh Con A beads were washed in Con A buffer [67 mM Hepes (pH 6.8), 0.2 mM CaCl2, 10 mM MgCl2, 500 mM NaCl, and 4 mM DTT] and incubated with rabbit liver glycogen (50 mg/ml; Sigma-Aldrich, G-8876, 1:1 volume) at 4°C for 1 hour. Glycogen-conjugated Con A beads were washed three times with Con A buffer and resuspended as 50% slurry. Con A–glycogen Sepharose beads (30 μl) were incubated with 15 μg of phosphorylase a, GM64–237, and PP1α7–330 alone and in various combinations thereof in a total volume of 250 μl for 1 hour at 4°C under gentle mixing. The beads were washed three times with Con A buffer (750 μl) and recovered by centrifugation (2000g, 1 min). The supernatant was removed using gel-loading tips, leaving behind the beads that were then incubated with 25 μl of 1 M α-d-methylglucoside (in Con A buffer) to elute all proteins bound to glycogen. Eluate (10 μl) was carefully transferred to new tubes to avoid contamination with residual beads and analyzed by SDS-PAGE. SDS-PAGE was stained using Sypro Ruby (Invitrogen) for visualization of total proteins.

MS analysis

Samples were separated by SDS-PAGE, and bands were excised and digested with trypsin in 50 mM ammonium bicarbonate overnight at 37°C. Peptides were extracted using 5% formic acid/50% acetonitrile and dried. Peptides were analyzed on a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific) equipped with an Easy-nLC 1000 (Thermo Fisher Scientific). Raw data were searched using COMET in high-resolution mode (38), with a precursor mass tolerance of 1 Da, trypsin enzyme specificity with up to three missed cleavages, and carbamidomethylcysteine as fixed modification. Oxidized methionine and phosphorylated serine, threonine, and tyrosine were searched as variable modifications. Probability of phosphorylation site localization was determined by PhosphoRS (39). Quantification of LC-MS/MS spectra was performed using MassChroQ with retention time alignment for smart quantification (40).


Supplementary material for this article is available at

Fig. S1. Interaction of GM with PP1.

Fig. S2. The GM:PP1 holoenzyme.

Fig. S3. Solution structure of GMCBM21 shows that the PP1 G subunits, GM and GL, have structurally similar CBM21 domains.

Fig. S4. Site 2, but not site 1, of GMCBM21 is involved in sugar binding and is structurally conserved with RoGACBM21.

Fig. S5. PP1 binding does not affect β-CD binding to GMCBM21.

Fig. S6. Dephosphorylation of phosphorylase a and human GYS1 by GM:PP1.

Fig. S7. GYS1 dephosphorylation analysis by MS.

Fig. S8. GYS1 dephosphorylation assays.

Table S1. Data collection and refinement statistics for GM64–93:PP1α7–300:MC-LR.

Table S2. NMR statistics for the structure determination of GMCBM21.

Data file S1. GSY1 MS data.

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: The GM2–240 plasmid was provided by D. Brautigan (University of Virginia), while the GYS1/GYG1 plasmids were provided by E. Zeqiraj (Beatson Institute for Cancer Research) and F. Sicheri (Lunenfeld-Tanenbaum Research Institute). We thank M. Gentry (University of Kentucky) for discussions and protocols regarding the Con A assays and glycogen types and quality. We thank T. B. Jayasundera and M. Ragusa for help at early stages of the project. We also thank N. Ly and A. Ueki (Biosensing Instrument Inc.) for help with SPR. Funding: This work was supported by the American Diabetes Association Pathway to Stop Diabetes Grant 1-14-ACN-31 to W.P., R35GM119455 to A.N.K., and R01GM098482 to R.P. Author contributions: G.S.K., M.S.C., D.M.K., and M.K.L. expressed and purified proteins. G.S.K. and D.M.K. performed the NMR experiments. G.S.K., D.M.K., and M.K.L. performed ITC experiments. M.S.C. performed the dephosphorylation assays and x-ray crystallography. S.P.L. and A.N.K. performed and analyzed the MS data. R.P. and W.P. conceived the experiments. G.S.K., M.S.C., D.M.K., R.P., and W.P. analyzed and interpreted the data and wrote the manuscript. Competing interests: The authors declare that they have no competing interests. Data materials and availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. Chemical shifts were deposited in the BioMagResBank ( under accession number 19225. Atomic coordinates were deposited in the Protein Data Bank under PDB codes 2M83 and 6DNO.

Stay Connected to Science Advances

Navigate This Article