Biochem. J. (2015) 468, 245–257

245

doi:10.1042/BJ20150270

*Department of Biochemistry and Molecular Biology, University of Melbourne, Parkville, Victoria 3010, Australia †Bio21 Molecular Science and Biotechnology Institute, University of Melbourne, Parkville, Victoria 3010, Australia ‡Eurogentec Biologics Division, Rue Bois Saint-Jean, 14, 4102 Seraing, Belgium §CSL Behring AG, Wankdorfstrasse 10, CH-3000 Bern 22, Switzerland St. Vincent’s Institute of Medical Research, Fitzroy, Victoria 3065, Australia ¶The Florey Institute of Neuroscience and Mental Health, The University of Melbourne, Victoria 3010, Australia

AMP-activated protein kinase (AMPK) is an αβγ heterotrimer that is important in regulating energy metabolism in all eukaryotes. The β-subunit exists in two isoforms (β1 and β2) and contains a carbohydrate-binding module (CBM) that interacts with glycogen. The two CBM isoforms (β1- and β2CBM) are near identical in sequence and structure, yet show differences in carbohydrate-binding affinity. β2-CBM binds linear carbohydrates with 4-fold greater affinity than β1-CBM and binds single α1,6-branched carbohydrates up to 30-fold tighter. To understand these affinity differences, especially for branched carbohydrates, we determined the NMR solution structure of β2CBM in complex with the single α1,6-branched carbohydrate glucosyl-β-cyclodextrin (gBCD) which supported the dynamic nature of the binding site, but resonance broadening prevented defining where the α1,6 branch bound. We therefore solved the X-ray crystal structures of β1- and β2-CBM, in complex with gBCD, to 1.7 and 2.0 Å (1 Å = 0.1 nm) respectively. The

additional threonine (Thr101 ) of β2-CBM expands the size of the surrounding loop, creating a pocket that accommodates the α1,6 branch. Hydrogen bonds are formed between the α1,6 branch and the backbone of Trp99 and Lys102 side chain of β2CBM. In contrast, the α1,6 branch could not be observed in the β1-CBM structure, suggesting that it does not form a specific interaction. The orientation of gBCD bound to β1- and β2-CBM is supported by thermodynamic and kinetic data obtained through isothermal titration calorimetry (ITC) and NMR. These results suggest that AMPK containing the muscle-specific β2-isoform may have greater affinity for partially degraded glycogen.

INTRODUCTION

(γ 1, γ 2 and γ 3) [15]. These subunits can form up to 12 different heterotrimer complexes with different tissue distributions. For example the dominant AMPK isoform in skeletal muscle is α2β2γ 3 [16]. The α-subunit contains the catalytic serine/threonine kinase domain (KD) that is phosphorylated at the conserved Thr172 for activity. The γ -subunit contains four nucleotide-binding sites, two of which are important for regulating AMPK activity [17]. The β-subunit contains a carbohydrate-binding module (CBM) which is thought to be involved in regulating AMPK activity via glycogen binding [7] or localizing AMPK to glycogen [18], although the details of this mechanism are not clearly understood. Recently crystal structures have been obtained for AMPK in the presence of activating compounds (A-769662 and 991). In these crystal structures, the CBM is phosphorylated at Ser108 and forms interactions with the KD, whereas the activating compounds bind to the KD–CBM interface, stabilizing this active conformation [19]. Recent data suggest that binding of carbohydrate to the CBM reduces its ability to form this stable interaction with the KD [20]. However, inhibition of AMPK in the presence of glycogen was

AMP-activated protein kinase (AMPK) is a key enzyme involved in energy regulation in nearly all eukaryotes. AMPK acts as a balance, responding to levels of AMP and ATP to keep sufficient amounts of energy available to the cell [1]. The enzyme is phosphorylated by upstream kinases for activation [2,3] and can be further regulated by nucleotides (AMP, ADP and ATP [4]), hormones [5], drugs [6] and possibly glycogen [7]. When activated AMPK up-regulates catabolic pathways, such as glucose uptake [8], glycolysis [9], fatty acid uptake [10] and fatty acid oxidation [11], in addition to down-regulating anabolic pathways, including glycogen synthesis [12], fatty acid synthesis [13] and protein synthesis [14]. Due to these wide roles in metabolism AMPK is believed to be an important drug target for treating disorders such as Type 2 diabetes [15]. AMPK is a trimeric complex consisting of three different subunits; in mammals these subunits are known as α, β and γ . Each subunit has multiple isoforms: there are two α isoforms (α1 and α2), two β isoforms (β1 and β2) and three γ isoforms

Key words: AMP-activated protein kinase (AMPK), carbohydrate-binding module (CBM), glycogen, isothermal titration calorimetry, nuclear magnetic resonance spectroscopy, X-ray crystallography.

Abbreviations: AMPK, AMP-activated protein kinase; BCD, β-cyclodextrin; CBM, carbohydrate-binding module; CPMG, Carr–Purcell–Meiboom– Gill; gBCD, glucosyl-β-cyclodextrin; ITC, isothermal titration calorimetry; KD, kinase domain; NCS, non-crystallographic symmetry; TLS, TranslationLibration-Screw-rotation. 1 Correspondence may be addressed to either of these authors (email [email protected] or [email protected]).  c The Authors Journal compilation  c 2015 Biochemical Society

Biochemical Journal

Jesse I. Mobbs*†, Ann Koay*†, Alex Di Paolo*†‡, Michael Bieri*†§, Emma J. Petrie*†, Michael A. Gorman, Larissa Doughty, Michael W. Parker*†, David I. Stapleton¶, Michael D.W. Griffin*†1 and Paul R. Gooley*†1

www.biochemj.org

Determinants of oligosaccharide specificity of the carbohydrate-binding modules of AMP-activated protein kinase

246

J.I. Mobbs and others

not detected in that work, as was previously seen by McBride et al. [7]. The AMPK β-subunit CBM has a β-sandwich fold with the conserved residues Trp100 , Lys126 and Trp133 (residue numbers according to β1-CBM), classifying it under the CBM48 family [21]. The X-ray crystal structure of β1-CBM in complex with β-cyclodextrin (BCD) (PDB code 1Z0M) shows that the two tryptophan residues form a cradle making hydrophobic contacts with two sequential glucosyl units, whereas the side chain NH3 + of the lysine residue forms hydrogen bonds with the glucosyl unit near Trp133 . Along with these conserved residues, Leu146 forms additional hydrophobic contacts and the side chains of Thr148 and Asn150 form hydrogen bonds with the glucosyl unit in contact with Trp100 . The CBMs have a preference for carbohydrates with seven glucose units connected via α1,4 linkages [21]. The two isoforms of the CBM (β1- and β2-CBM) are more than 80 % identical and their carbohydrate contacting residues are 100 % identical. Yet previous research by us has shown that β2-CBM always binds unbranched (linear maltoheptaose or cyclic BCD) carbohydrates four times more tightly [22]. The β2-CBM also appears to have a stronger preference for carbohydrates with a single α1,6-glucose branch [linear glucosyl-maltoheptaose or cyclic glucosyl-β-cyclodextrin (gBCD)], binding single-branched carbohydrates 30 times more tightly than the β1-CBM [22,23]. Whereas site-directed mutagenesis and protein dynamic NMR studies suggested the presence of an additional threonine (Thr101 ) may be indirectly related to this difference, the molecular details accounting for these affinity differences have not been elucidated [23]. In the present study, we aimed to understand why β2CBM binds carbohydrate with higher affinity than β1-CBM and specifically understand the preference of single α1,6-branched carbohydrates. We solved the NMR solution structure of β2CBM in the presence of the branched carbohydrate 6-O-α-DgBCD which supported the dynamic nature of Trp133 observed previously [23]. We also solved the X-ray crystal structures of β1- and β2-CBM in complex with gBCD, providing atomic resolution detail of the site in which the branched glucosyl unit contacts β2-CBM and the ligand–protein interactions formed. We further characterized the carbohydrate-binding properties of β1and β2-CBM by studying the energetics via isothermal titration calorimetry (ITC) and the kinetics via a combination of ITC and NMR experiments. Our results suggest that the additional threonine residue (Thr101 ) of the β2-CBM confers the ability to bind single α1,6-branched carbohydrates better than β1-CBM by increasing the length of the loop surrounding Thr101 and creating a pocket that accommodates the α1,6 branch. AMPK complexes containing the muscle-specific β2-sububunit may therefore have enhanced affinity for partially degraded glycogen. EXPERIMENTAL Protein expression and purification

β1- and β2-CBMs were produced in a pGEX-6P-3 vector as GST-fusion proteins. Samples for ITC and crystallography were expressed in Escherichia coli BL21(DE3) bacteria in 2YT medium [1.6 % (w/v) tryptone, 1 % (w/v) yeast extract and 0.5 % NaCl] at 37 ◦ C. Expression was induced at an OD600 of 0.8 by the addition of 1 mM IPTG for 4 h. NMR samples were uniformly 15 N- or 13 C,15 N-labelled according to the protocol of Cai et al. [24]. For all samples, bacterial pellets were resuspended at pH 7.4 in 10 mM Na2 HPO4 , 2 mM KH2 PO4 , 137 mM NaCl and 3 mM KCl, 1 mM EDTA, 1 mM DTT and a protease inhibitor cocktail tablet (Roche). Cells were lysed using an Avestin EmulsiFlex c The Authors Journal compilation  c 2015 Biochemical Society

C3 homogenizer before being applied to glutathione-Sepharose resin, washed and then eluted with 10 mM glutathione in the same buffer, pH-adjusted to 7.4. Protein samples were further purified by size-exclusion chromatography using a Superdex 75 16/60 column in 20 mM Tris/HCl, pH 8.0, 150 mM NaCl and 1 mM EDTA. For a final step samples were desalted before being applied to a Mono Q anion-exchange column and eluted in a gradient manner from 0 to 1 M NaCl in 20 mM Tris/HCl and 1 mM EDTA, pH 8.0. Samples were buffer-exchanged and concentrated to 500 μM using an Amicon Ultra 3 kDa centrifugation filter (Millipore) before flash freezing in liquid nitrogen and storage at − 80 ◦ C. NMR and ITC samples were exchanged to 100 mM Na2 HPO4 /NaH2 PO4 , pH 6.8, and crystallography samples to 20 mM HEPES, pH 7.0. NMR assignment and solution structure elucidation

All NMR data were collected at 25 ◦ C using either a Bruker 600 MHz AVANCE III or a Bruker 800 MHz AVANCE II spectrometer. All experiments utilized the cryogenically cooled triple resonance probes of the spectrometers to maximize sensitivity and the standard solution conditions were 0.5– 1.5 mM β2-CBM, dissolved in 50 mM phosphate buffer at pH 6.9. For assignment of free and gBCD-bound β2-CBM, the following standard spectra were collected: 3D CBCA(CO)NH, HNCACB, HN(CA)CO, HNCO, HBHA(CO)NH, H(CCO)NH, (H)C(CO)NH, HCCH-TOCSY and 2D TOCSY. For distance restraints 2D or 3D 13 C-edited and 15 N-edited NOESY and 2D NOESY were collected using 100 ms mixing times. To guide interpretation of stereo assignments and torsion angle restraints 3D HNHA, HNHB HACACB-COSY were collected and a 2D 13 C-HSQC was collected on 10 % 13 C-labelled samples to determine the stereo assignments of the methyls of valine and leucine residues. Additional spectra collected on the gBCD–β2CBM complex were 3D 13 C-filtered, edited NOESY, 2D 13 C,15 Nfiltered NOESY and TOCSY. All data were processed in NMRPipe [25] and inspected in Sparky (Goddard, T.D. and Kneller, D.G., University of California, San Francisco). All data were essentially multiplied by a Lorentz–Gaussian function in the detected dimension and cosine bells function in the indirect dimensions. Data were zerofilled once prior to Fourier transformation. Initial assignments were obtained by submitting data to the PINE server [27] which were then manually verified and supplemented. NOE data were auto-assigned using the CANDID algorithm [28,29] within CYANA (version 2.1). Structures were then iteratively calculated in CYANA using 100 trial structures and selecting the best 20 for inspection and validation of the experimental data. Crystallization and X-ray diffraction data collection

Crystals of β1- and β2-CBM were grown at 8 ◦ C using the sitting-drop vapour-diffusion method. All crystals were grown by mixing equal amounts (1 μl) of protein solution (13 mg/ml) with corresponding reservoir solution prior to equilibration with the reservoir solution to 8 ◦ C. In order to crystallize the CBM– gBCD complexes, 13 mg/ml protein samples (1.3 mM) were incubated with 6 mM gBCD before crystallization drops were produced. β2-CBM in the unliganded state was crystallized in 0.17 M ammonium sulfate, 15 % (v/v) glycerol and 25.5 % (w/v) PEG4000. β2-CBM was crystallized in complex with gBCD in 0.2 M lithium chloride, 20 % (w/v) PEG 6000 and 0.1 M sodium HEPES, pH 7. β1-CBM was crystallized in complex with gBCD in 0.2 M lithium sulfate, 25 % (w/v) PEG 8000 and 0.1 M sodium acetate, pH 4.5. All crystals grew within 1–2 weeks. β2-CBM crystals were flash-cooled in liquid nitrogen directly

Specificity of carbohydrate binding to AMPK

from the crystallization drop before data collection. Crystals of β1- and β2-CBM in complex with gBCD were cryoprotected by passing through their respective reservoir solutions supplemented with 7.5 % glycerol and 7.5 % ethylene glycol before flash cooling. X-ray diffraction data were collected at 100 K at the MX1 beamline (β2-CBM and gBCD–β2-CBM complex) and MX2 beamline (gBCD–β1-CBM complex) of the Australian Synchrotron. Data collection was controlled using Blue-Ice software [30]. Crystal structure phasing and model refinement

Diffraction data were indexed and integrated using XDS [31] before scaling and merging with AIMLESS [32] from the CCP4 suite [33]. Initial phases were solved by molecular replacement using the respective CBM structure (β1-CBM PDB: 1Z0N and β2-CBM PDB: 2F15) and the program PHASER [34]. Restrained refinement was carried out with REFMAC5 [35] and model building, corrections and the addition of solvent using COOT [36]. Early addition of solvent was performed using ARP/wARP [37]. During model building, side chains with insufficient electron density were not modelled. All molecular replacement solutions were subjected to a single cycle of simulated annealing using PHENIX [38] in order to minimize model bias. TranslationLibration-Screw-rotation (TLS) refinement was performed in the final rounds of refinement using REFMAC5, with each protein molecule defined as an individual TLS group. Crystals of β2-CBM belonged to space-group C2 and the structure was solved to a resolution of 1.6 Å. The structure comprises two molecules in the asymmetric unit. Due to the plasmid cloning strategy, both constructs (β1- and β2-CBM) have an additional five N-terminal residues (GSPNS) that remain after purification. All β2-CBM residues (Gln75 to Lys156 ) are visible for both molecules and two additional N-terminal residues (Asn73 and Ser74 ) from the plasmid cloning strategy are observed in monomer B. Crystals of β2-CBM in complex with gBCD belonged to space-group C2 and the structure was refined to a resolution of 2.0 Å. Chemical restraints for the branched gBCD were constructed using the existing restraints for BCD (monomer code BCD) by adding a single α1,6-branched glucose unit using SKETCHER and LIBCHECK from the CCP4 suite [39]. The structure comprises 18 protein molecules in the asymmetrical unit. All β2-CBM residues (Gln75 to Lys156 ) are visible in majority of the molecules, with some containing up to three additional N-terminal residues visible (Pro72 , Asn73 and Ser74 ) that are a product of cloning. One molecule comprises residues Pro78 to Lys156 . In nine of the 18 monomers the α1,6 glucosyl unit formed crystal contacts with Lys88 and Glu89 of neighbouring protein monomers. In all other molecules the sugar did not form direct crystal contacts with other monomers of the asymmetric unit. The structure of β1-CBM was solved in space-group P1 to a resolution of 1.72 Å, with eight monomers in the asymmetric unit. The number of observed residues in the eight structures differs with respect to one another. The largest molecule contains all β1-CBM residues (Gln76 to Lys156 ) and has five residues that are a product of cloning (Gly71 , Ser72 , Pro73 , Asn74 and Ser75 ), whereas the shortest contains residues Pro79 to Val155 . The crystals of β1-CBM in complex with gBCD exhibited a rare crystal lattice pathology which has previously been described as ‘partial rotational lattice order–disorder’ [40,41]. To refine this crystal structure we followed the protocol described in detail for the crystal structure of stefin B [40]. Two of the eight monomers were refined at 50 % occupancy, as these molecules occupy

247

the same space in the asymmetric unit and represent a single monomer in alternate orientations (refer to Supplementary Figure S1 for more information). The two monomers at 50 % occupancy (G and H) and their corresponding gBCD ligands were refined with non-crystallographic symmetry (NCS) restraints to the six well-defined monomers (A–F) and gBCD ligands. Molecules G and H and their ligands, were not used for structural analysis. For monomers A–F, electron density was not available to guide building of the α1,6 glucosyl-branch unit of the ligand and, therefore, this was not modelled. Solvent molecules surrounding gBCD were added very late in refinement and some density surrounding gBCD was not modelled due to ambiguity. Omit maps were calculated by removing all gBCD molecules from the liganded structures, followed by simulated annealing in PHENIX [38] and restrained refinement using REFMAC5 [35]. Refinement statistics for all crystal structures are summarized in Table 1. All images were prepared using PyMOL (http://www.pymol.org). Isothermal titration calorimetry

β1- and β2-CBM protein samples were dialysed overnight in 100 mM sodium phosphate buffer (pH 6.8) before experiments and the dialysis buffer was used to prepare solutions of gBCD and BCD. To obtain affinity measurements at 25 ◦ C titrations were performed using a MicroCal iTC200 (GE Healthcare) in triplicate. Titrations were performed with 16 injections of 2.5 μl of 300 μM carbohydrate into 30 μM protein samples. Injections were performed for 5.0 s with 180 s between injections. A blank sample of injecting carbohydrate into buffer was subtracted from each result so that at saturation the binding curves approached 0 kcal·mol − 1 . Titration data were analysed using the Origin MicroCal package and fitted to a single-site binding model. Values of stoichiometry (N), binding affinity (K a ) and enthalpy (H) were obtained from the fit. Dissociation constant (K d ) could be obtained by the inverse of association constant (K a ), whereas Gibbs free energy (G) and entropy (TS) were obtained by the following equations: G = −RT ln (K a )

(1)

G = H − T S

(2)

where R is the gas constant (1.987 cal K − 1 ·mol − 1 ) and T is the absolute temperature in K. Temperature dependence of binding

To study the temperature dependence of binding for both β1- and β2-CBM to both carbohydrates (BCD and gBCD), ITC titrations were carried out at 10, 15, 20, 25 and 30 ◦ C. Titrations were also analysed by a one site-binding model using Origin MicroCal software. The temperature dependence of binding is governed by the heat capacity (Cp ) of binding and could be obtained by the following equations: H = H 0 + C p (T − T 0 )   T S = S0 + C p ln T0   −H + T S K d = exp RT

(3) (4) (5)

H 0 and S0 are the enthalpy and entropy of dissociation at a reference temperature T 0 .  c The Authors Journal compilation  c 2015 Biochemical Society

248 Table 1

J.I. Mobbs and others X-ray data collection and refinement statistics.

ASU, asymmetric unit.

PDB accession code Data collection Space-group Wavelength (A˚) Number of images Oscillation range per image (◦ ) Detector Cell dimensions a, b, c (A˚) α, β, γ (◦ ) Resolution range (A˚)

R sym R means R Pim I /σ I Total observations Unique reflections Completeness (%) Multiplicity Wilson B -factor (A˚2 ) Matthews coefficient, V M (A˚3 ·Da − 1 ) Solvent (%) Refinement Resolution range (A˚) R work R free Protein molecules in ASU Total number of residues Number of atoms Protein Ligands Water Average B factor (A˚2 ) Protein Ligands Water RMSD Bond lengths (A˚) Bond Angles (◦ ) Ramachandran statistics Favoured/allowed/outliers (%)#

β2-CBM APO

β2-CBM + gBCD

β1-CBM + gBCD

4Y0G

4YEE

4YEF

C2 0.953700 360 1 ADSC Quantum 210r

C2 0.953700 720 0.5 ADSC Quantum 210r

P1 0.953700 360 1 ADSC Quantum 315r

113.0, 27.6, 78.9 90, 133.9, 90 40.72–1.60 (1.62–1.60) 0.080 (0.669) 0.093 (0.814) 0.047 (0.415) 17.8 (2.8) 175003 (7801) 23221 (1088) 97.6 % (90.7) 7.5 (7.2) 20.68 2.22 44.6

204.5, 96.4, 118.7 90, 125.5, 90 48.38–2.00 (2.03–2.00) 0.084 (0.668) 0.098 (0.778) 0.050 (0.397) 16.9 (2.8) 961536 (45123) 127116 (6168) 99.9 (97.8) 7.6 (7.3) 32.78 2.65 53.6

10.7, 68.4, 92.2 111.7, 95.6, 90.1 43.71–1.72 (1.75–1.72) 0.043 (0.628) 0.060 (0.888) 0.043 (0.628) 18.6 (1.9) 368361 (17354) 95126 (4503) 97.1 (92.9) 3.9 (3.9) 28.19 3.39 63.7

40.72–1.60 (1.62–1.60) 0.150 0.183 2 164 1693 1338 54 301 27.4 26.7 38.4 31.9

48.38–2.00 (2.03–2.00) 0.169 0.224 18 1505 15355 11980 1698 1677 34.13 32.22 39.45 42.38

43.71–1.72 (1.75–1.72) 0.187 0.222 8* 633* 6310 4968 658 684 30.89 24.84 30.89 37.11

0.017 1.9

0.012 1.7

0.014 1.8

98.2/1.8/0

98.3/1.7/0

97.7/2.0/0.3

*These values include the two overlapping molecules, each included at 50 % occupancy; refer to Supplementary Figure S1 for more information. #Obtained from MolProbity [53]

These equations could be fitted simultaneously using a leastsquares method using the software MATLAB. ZZ-exchange NMR spectroscopy

ZZ-exchange NMR spectroscopy experiments [43] were performed for β2-CBM at multiple temperatures (5, 10, 15 and 20 ◦ C). β2-CBM samples at 500 μm were prepared with the addition of either gBCD or BCD (∼250 μM) such that free peaks had a similar intensity to bound peaks. ZZ-experiments were performed on a Bruker Avance III HD 700 MHz spectrometer with various mixing times (0.03, 0.05, 0.09, 0.13, 0.15, 0.35, 0.55, 0.75 and 0.95 s). The experiments were performed with 2048 t2 and 128 t1 points with 64 scans per t1 point. Spectra were processed using NMRPipe [25] and peak volumes were measured with Sparky (Goddard, T.D. and Kneller, D.G., University of California, San Francisco). Peak volumes could then be analysed with MATLAB scripts generously donated by Dr Demers and Dr Mittermaier  c The Authors Journal compilation  c 2015 Biochemical Society

[44]. K off values were obtained by a global fit of all the data and K on values were determined with the use of ITC-derived K d values. Residues used in the analysis were Gly94 , Ser95 , Asn97 , Asn98 , Trp99 , Thr101 , Lys126 , Phe128 , Val134 , Ser144 and Thr148 . To estimate dissociation rates at 25 ◦ C, the K off values were fitted to the following Eyring equation:   −G ‡ BT exp (6) Kof f = h RT G ‡ = H ‡ − T S‡ (7) H ‡ = H ‡0 + C p (T − T 0 )   T S‡ = S‡0 + C p ln T0

(8) (9)

Where G‡ , H ‡ and S‡ are the change in Gibbs free energy, enthalpy and entropy between a transition state and bound state.

Specificity of carbohydrate binding to AMPK

Figure 1

249

Solution structures of β2-CBM

(A and C) Stereo rainbow views of the backbone atoms (Cα, C , N) for an ensemble of 20 structures of the (A) apo and (C) gBCD-bound β2-CBM. Cartoon representations of the immunoglobulin fold of β2-CBM in the (B) apo and (D) ligand-bound states. β-Strands and the N- and C-termini of the domain are labelled. NOEs to the α1,4 glucosyl of gBCD from residues of the β-hairpin (β7 and β8) and the top of the β-sheet show that gBCD binds similarly to cyclodextrin binds to β1-CBM [18], but no unambiguous NOEs were observed to locate the α1,6 branch on the protein.

B is the Boltzmann constant (cal·K − 1 ), h is Planck’s constant (cal·s − 1 ) and R is the gas constant (cal·K − 1 ·mol − 1 ). We set Cp to those obtained by temperature-dependence experiments of ITC and used a Monte Carlo simulation to solve for K off at 25 ◦ C and obtain errors. Relaxation dispersion experiments 15 N Carr–Purcell–Meiboom–Gill (CPMG) relaxation dispersion experiments [45] were performed at 10 ◦ C and 25 ◦ C for β1CBM. β1-CBM samples at 500 μM were prepared with substoichiometric amounts of each of the carbohydrates BCD and gBCD (20 μM). Experiments were performed on Bruker Avance III 600 MHz and Bruker Avance II 800 MHz spectrometers with various CPMG pulse frequencies (0, 50, 2 × 75, 100, 2 × 150, 200, 300, 500, 2 × 700, 800, 900, 1000, 1500 Hz) and a constant delay (T cpmg ) of 80 ms. Experiments were performed with 2048 t2 and 256 t1 points and 24 scans for each t1 point. Spectra were processed using NMRPipe [25] and peak intensities were measured with Sparky (Goddard, T.D. and Kneller, D.G., University of California, San Francisco). The automatic relaxation dispersion analysis software NESSY [46] was first used to analyse data on a per residue basis and residues with no exchange were removed from further analysis. R2eff dispersion profiles were then created using the online server ShereKhan [47]. The dispersion profiles could then be further analysed using software generously donated by Dr Korzhnev [48]. This software allowed for global analysis of all residues at both 10 ◦ C and 25 ◦ C. Residues used in the analysis were Ser94 , Gly95 , Ser101 , Lys126 , Phe127 , Trp133 , Thr134 , Thr143 , Ser144 , Thr148 , Thr149 , Asn150 and Asn151 .

RESULTS Structure of apo and glucosyl-β-cyclodextrin-bound β2-CBM by NMR spectroscopy

The solution structures of β2-CBM (Figure 1) were determined from 1793 and 1612 inter-residue distance and dihedral angle restraints (Table 2) corresponding to an average of 21 and 19 restraints per residue in the apo and gBCD-bound states respectively. The distribution of short-, medium- and long-range restraints is represented in histograms in the Supplementary Figures S2(C) and S2(D). Regions with NOEs close to or lower than the per residue average are mostly found to lie within loops and turns of the β2-CBM. Additionally, a lack of restraints between β-strands 3 and 4 is also observed in both the free and the gBCD-bound states and is reflected by increases in backbone RMSD within the final ensemble of 20 structures in both the freeand the ligand-bound states (Supplementary Figure S2E). The overall number of NOEs observed in the bound state was fewer compared with the free-state. This can be attributed to the effect of modest line-broadening caused by gBCD binding to the CBM resulting in fewer NOE peaks. In each β2-CBM state, 20 calculated structures with the lowest target function were used to represent the β2-CBM solution structures (Figure 1). The final structures are welldefined and converge to a backbone RMSD of 0.27 Å (apo) and 0.32 Å (gBCD-bound) for residues 15–98 (pairwise comparison). In these structures, no distance and dihedral angle violation was >0.3 Å and >2◦ respectively. Further structural statistics of these calculated structures are summarized in Table 2.  c The Authors Journal compilation  c 2015 Biochemical Society

250

Figure 2

J.I. Mobbs and others

Proton resonances from the gBCD that interact with the respective protons on a sample of 13 C,15 N-labelled β2-CBM

NOEs are observed in the 13 C,15 N-filtered-edited NOESY spectra in the 3.5–4.0 ppm region of the indirect 1 H dimension, with successful elimination of intramolecular β2-CBM peaks. Assignments for 1 H groups of β2-CBM are indicated on the x -axis and those for the α1,4 glucosyl units of the gBCD on the y -axis.

Table 2 Structural statistics for the ensemble of 20 structures of apo β2CBM and β2-CBM bound to gBCD in solution

Protein structural restraints (residues 15–98) Total number of NOE restraints used*: Short-range |i − j| = 1 Medium-range 1 < |i − j| < 5 Long-range |i − j|  5 Dihedral angle restraints TALOS + † Structural statistics Average maximum upper distance violation Average maximum angle violation Average CYANA target function (A˚2 ) Ramachandran statistics‡ (residues 18–98) Most favoured Additionally allowed Generously allowed Disallowed RMSDs to mean structure (residues 15–98) Average backbone (Cα, C , N) All heavy (non-hydrogen) atoms

Apo β2-CBM

Bound β2-CBM

complex revealed that the intensity of several intraprotein NOEs differ indicating that several side chains have reoriented. In particular, the intensity of NOEs between the side chain protons of Trp99 and Trp133 , both located in the carbohydrate-binding site, show significant differences that result in a reorientation of Trp133 (Supplementary Figure S3).

519 267 877

467 239 774

Structure of glucosyl-β-cyclodextrin bound to β1- and β2-CBM by X-ray crystallography

130

132

0.2 1.51◦ 0.87

0.12 0.59◦ 1.01

84.1% 15.7% 0.1% 0.0%

82.2% 16.6% 1.2% 0.0%

0.27 + − 0.10 0.74 + − 0.09

0.32 + − 0.11 0.79 + − 0.12

* Number of peaks from the cycle7.noa file of an automated calculation in CYANA. †Dihedral angles ϕ and ψ predicted with TALOS + using (HN, HA, CA, CB, CO, N) chemical shift assignments [54] matched against a high-resolution structural database. ‡Statistics derived from PROCHECK-NMR [55]

To analyse the gBCD-bound complex, we conducted 13 Cfiltered, edited NOESY and 13 C,15 N-filtered NOESY and TOCSY experiments. To our surprise, whereas the majority of carbohydrate resonances were observed, the anomeric protons are completely broadened under the experimental conditions of 1:3 protein/gBCD and therefore no intermolecular NOEs could be assigned to this group. In general no NOEs could be unambiguously assigned to the α1,6 glucosyl-branch. This linebroadening, however, is consistent with the broadening of some protein resonances suggesting the presence of conformational heterogeneity [23]. Nevertheless, intermolecular NOEs are observed from all expected carbohydrate-binding residues, except Asn150 , to the remaining carbohydrate signals (Figure 2). Furthermore, comparing the apo β2-CBM to the gBCD-bound  c The Authors Journal compilation  c 2015 Biochemical Society

Although the solution NMR structures were well-defined, the frustration of lack of assignable NOEs between the protein and the α1,6 glucosyl-branch prompted us to crystallize the protein. For a complete comparison, we attempted crystallizations of β1and β2-CBM in both the apo- and the gBCD-bound states. In the course of the present work, we were able to solve the Xray crystal structure of apo (unbound) β2-CBM to a higher resolution (1.6 Å) than previously (2.0 Å, PDB code: 2F15; Walker, J.R., Wybenga-Groot, L., Finerty, Jr, P.J., Newman, E., MacKenzie, F.M., Weigelt, J., Sundstrom, M., Arrowsmith, C., Edwards, A., Bochkarev, A. and Dhe-Paganon, S. Structural Genomics Consortium, unpublished work). The construct used for the new apo β2-CBM structure was shorter (Gln75 to Lys156 ) than the construct used for the previous β2-CBM structure (Gln75 to Phe161 ). The new unbound structure of β2-CBM was solved in the space-group C2 with two molecules in the asymmetric unit. The new structure has nearly identical conformation to the previous structure with an overall RMSD of 0.4 Å. The largest structural change is apparent in the binding loop with an RMSD of 1.1 Å (residues 138–151), reflecting the dynamic nature of this region that we have previously characterized [23]. The crystal structure of β2-CBM in complex with gBCD was solved to a resolution of 2.0 Å and has 18 protein molecules in the asymmetric unit. Strong and well-defined electron density for gBCD was apparent, allowing accurate placement and orientation of the ligand to each protein molecule. In addition, electron density provided unambiguous location and orientation of the α1,6-branched glucosyl unit and this was equivalent for all molecules of the asymmetrical unit (Figure 3A; Supplementary Figure S4). The α1,4-linked glucosyl units of the cyclic core of gBCD form similar contacts to those observed in the crystal structure of β1-CBM in complex with BCD. Two glucose units

Specificity of carbohydrate binding to AMPK

of the cyclic core form hydrophobic contacts with the two critical tryptophan residues (Trp99 and Trp133 ) and the side chain of Leu146 which pierces the cyclic ring. Hydrogen bonds are formed to the cyclic ring via the side chains of Lys126 , Ser144 , Thr148 and Asn150 as well as the backbone carbonyls of Glu145 and Leu146 . The α1,6 glucosyl-branch of gBCD sits in a pocket between the two tryptophan residues and forms hydrogen bonds with the backbone carbonyl of Trp99 and the side chain of Lys102 and hydrophobic contacts with the ring of Trp133 (Figure 3C). No other protein residues appear to be involved in forming the complex. In nine of the 18 β2-CBM–gBCD protein monomers of the asymmetrical unit, the α1,6 glucosyl-branch forms crystal contacts to Lys88 or Glu89 of the neighbouring protein subunit. However, in the other nine complexes the α1,6 glucosyl-branch is instead contacted by solvent molecules and does not form direct interactions with other monomers in the asymmetric unit. A single crystal was also obtained which was solved by molecular replacement in the hexagonal space-group P63 22 with a single β2-CBM–gBCD complex in the asymmetric unit. This structure was not able to be reasonably refined; however, gBCD was in the same location and orientation as observed in the C2 crystal form described in the present study (Supplementary Figure S5). These results suggest that the observed orientation of gBCD in the crystal structure is representative of the complex in solution and is not constrained significantly by crystal contacts. We also solved the crystal structure of β1-CBM–gBCD complex to a resolution of 1.7 Å. Due to poor density and necessary NCS restraints associated with molecules G and H, ligand-binding analysis was carried out for subunits A–F only (Supplementary Figure S1). For molecules A–F, electron density for the gBCD was clear for all α1,4 glucosyl units of the cyclic core and the same protein–ligand interactions were observed for the cyclic core as in the gBCD–β2-CBM complex. In contrast with the β2-CBM–gBCD complex, no continuous electron density corresponding to the α1,6 glucosyl-branch was apparent in the maps in close proximity to the cyclic core of the carbohydrate and as a result the α1,6 glucosyl-branch could not be modelled. Whereas some density was observed surrounding the glucose units of the gBCD cyclic core, attempts to model the α1,6 glucosyl-branch into this density failed or resulted in very poor fits (Figure 3B; Supplementary Figure S4). Overall these results suggest that the α1,6 glucosyl-branch of gBCD forms specific contacts with the β2-CBM but does not form these interactions with the β1-CBM. The lack of electron density apparent for the α1,6 glucosyl-branch of gBCD in the case of β1-CBM suggests that gBCD could bind to β1-CBM in multiple orientations of the gBCD ring. Thermodynamics of carbohydrate binding to β1- and β2-CBM

ITC was used to determine the affinity of the two carbohydrates gBCD and the non-branched BCD to both modules. β1CBM bound to both carbohydrates with similar affinities of 4.39 + − 0.19 μM (BCD) and 4.44 + − 1.09 μM (gBCD) whereas β2CBM binds more tightly to both carbohydrates with affinities of 0.98 + − 0.07 μM (BCD) and 0.32 + − 0.07 μM (gBCD) (Figure 4) in agreement with previous fluorescence experiments on fulllength trimeric AMPK [23]. ITC has the added ability of being able to directly measure the change in enthalpy (H) of binding and from this the entropy (TS) and free energy (G) can be calculated (Table 3). Overall, the binding of the two CBMs to carbohydrates is enthalpically driven with small contributions from entropy. β1-CBM has less contribution from enthalpy −1 [ − 5.7 + − 0.3 and − 6.5 + − 0.4 kcal·mol (1 cal ≡ 4.184 J) for BCD and gBCD respectively] and a more favourable contribution

251

−1 from entropy (1.6 + BCD and − 0.4 and 0.8 + − 0.5 kcal·mol gBCD respectively), with similar results for both carbohydrates. On the other hand β2-CBM has clear differences between the two carbohydrates. β2-CBM binding to gBCD has larger −1 but enthalpy contributions ( − 10.0 + − 0.7 kcal·mol ) and a small 0.8 kcal·mol − 1 ) as unfavourable entropy contribution ( − 1.1 + − opposed to BCD, which has smaller and favourable contributions −1 from enthalpy ( − 7.6 + − 0.4 − 0.5 kcal·mol ) and entropy (0.6 + −1 kcal·mol ). In the present study, we have also fully characterized the binding affinity and thermodynamics of the previously identified Thr101 deletion, which was shown to significantly modify the affinity to the branched carbohydrate glucosyl-maltoheptaose but not the unbranched maltoheptaose [22]. The trend in binding affinity for the mutations is similar for gBCD and BCD (Table 3). Deletion of Thr101 from β2-CBM (β2-Thr101 ) caused a 2-fold reduction in the binding affinity to gBCD (0.66 + − 0.06 μM) and only a minor change to BCD (1.15 + − 0.08 μM). Insertion of a threonine into β1-CBM at a corresponding position to that of β2-CBM (β1-Thr101ins ) resulted in approximately 5fold increase in binding affinity for gBCD (0.81 + − 0.12 μM) and 2-fold increase for BCD (2.29 + 0.27 μM). The largest − change in thermodynamics occurred for β1-Thr101ins binding to −1 gBCD, where enthalpy increased to − 8.4 + − 0.4 kcal·mol −and 1 entropy became mildly unfavourable ( − 0.1 + − 0.3 kcal·mol ). There were also smaller changes in enthalpy and entropy for β1-Thr101ins binding to BCD, − 6.7 + − 0.4 and 1.0 + − 0.4 kcal·mol − 1 respectively. For β2-Thr101 binding to gBCD there was a decrease in both enthalpy and entropy ( − 9.1 + − 0.5 and −1 respectively) and for BCD enthalpy − 0.7 + 0.6 kcal·mol − −1 entropy became increased ( − 8.4 + − 0.7 kcal·mol ) whereas −1 slightly unfavourable ( − 0.3 + − 0.6 kcal·mol ). Another advantage of ITC is that enthalpy is determined directly allowing one to perform titrations at multiple temperatures to obtain a slope of the heat capacity Cp , as opposed to van’t Hoff methods where one must assume a Cp = 0 to determine approximate enthalpy. The heat capacity of binding is typically associated with the hydrophobic effect and can be a measure of the amount of water returned to the bulk solvent [49]. Heat capacity for CBM binding to carbohydrates could be obtained from simultaneously fitting eqns 3–5 and resulted in plots of H and TS compared with temperature with very similar slopes for all carbohydrates (Supplementary Figure S6). Heat capacity changes for both CBMs to both carbohydrates are all similar ranging from − 92 to − 106 cal·mol − 1 ·K − 1 (Table 3). These small negative changes are consistent with previously seen protein–carbohydrate interactions and represent a balance of polar and hydrophobic interactions [50]. This agrees well with the structure of CBM in complex with carbohydrate as there is evidence of hydrophobic interactions from the two tryptophan residues (Trp99 , Trp133 ) as well as hydrogen bonding from polar residues (Lys126 , Thr148 and Asn150 ).

Binding kinetics of carbohydrate to β1- and β2-CBM

For β2-CBM binding to gBCD or BCD we were able to use ZZexchange NMR spectroscopy to determine the dissociation rates (K off ) and using ITC-derived dissociation constants (K d ) we could determine the association rates (K on ). ZZ-exchange experiments performed at 25 ◦ C for both carbohydrates gave poor profiles, probably due to their high off-rates. Performing the experiments at lower temperatures allowed us to fit the off-rates (K off ) to an Eyring plot and estimate the off-rate for gBCD at 25 ◦ C (Supplementary Figure S7). At 10 ◦ C, fits could be obtained for both gBCD and BCD to β2-CBM and examples of the plots are  c The Authors Journal compilation  c 2015 Biochemical Society

252

Figure 3

J.I. Mobbs and others

X-ray crystal structure of β1- and β2-CBM in complex with gBCD

(A) Cartoon representation of β2-CBM in complex with gBCD. The ligand molecule clearly fits the density unambiguously. Blue 2F o-F c maps are at a sigma level of 1.0. (B) Cartoon representation of β1-CBM in complex with gBCD. In this case the density does not support the α1,6 branch of gBCD and so was not modelled. Positive density is represented in green. 2F o-F c maps are in blue at a sigma level of 1.0 and Fo-Fc maps in green at a sigma level of 3.0. (C) The α1,6 branch makes hydrogen bonds to the backbone of Trp99 and the side chain of Lys102 . (D) Overlay of the two proteins (β1-CBM in cyan and β2-CBM in red) with gBCD from the β2-CBM complex. Thr101 increases the pocket size of β2-CBM, allowing gBCD to make contacts with Trp99 backbone and Lys102 side chain. Simulated annealing omit maps can be found in Supplementary Figure S4.

−1 shown in Figure 5. BCD had a dissociation rate of 35.3 + − 11.9 s , more than 10-fold higher than for gBCD with a rate of 3.2 + 1.2 − 7 −1 −1 s − 1 . BCD had an association rate of (4.0 + − 2.7) × 10 7M − 1·s − 1, similar to the association rate of gBCD, (3.2 + − 1.4) × 10 M ·s (Table 4).

 c The Authors Journal compilation  c 2015 Biochemical Society

For β1-CBM, rates of binding were too high to be accurately calculated by ZZ-exchange NMR spectroscopy, so we performed CPMG relaxation dispersion experiments with sub-stoichiometric amounts of ligand at 10 ◦ C and 25 ◦ C. At 25 ◦ C, the dissociation rates for both carbohydrates are similar,

Specificity of carbohydrate binding to AMPK Table 3

253

Thermodynamic parameters obtained by ITC

C p values and errors are obtained by least-squares fit. Refer to the Experimental section for details. ND, not determined. Values and errors are reported as means and one S.D. for n = 3 experiments.

β2-CBM + BCD β2-CBM + gBCD β1-CBM + BCD β1-CBM + gBCD β2-Thr101 + BCD β2-Thr101 + gBCD β1-Thr101ins + BCD β1-Thr101ins + gBCD

Figure 4

K d (μM)

H (kcal·mol − 1 )

T S (kcal·mol − 1 )

G (kcal·mol − 1 )

C p (cal·mol − 1 ·K − 1 )*

0.98 + − 0.07 0.32 + − 0.07 4.39 + − 0.19 4.44 + − 1.09 1.15 + − 0.08 0.66 + − 0.06 2.29 + − 0.27 0.81 + − 0.12

− 7.6 + − 0.5 − 10.0 + − 0.7 − 5.7 + − 0.3 − 6.5 + − 0.4 − 8.4 + − 0.7 − 9.1 + − 0.5 − 6.7 + − 0.4 − 8.4 + − 0.4

0.6 + − 0.4 − 1.1 + − 0.8 1.6 + − 0.4 0.8 + − 0.5 − 0.3 + − 0.6 − 0.7 + − 0.6 1.0 + − 0.4 − 0.1 + − 0.3

− 8.2 + − 0.1 − 8.9 + − 0.1 − 7.3 + − 0.1 − 7.3 + − 0.1 − 8.1 + − 0.1 − 8.4 + − 0.1 − 7.7 + − 0.1 − 8.3 + − 0.1

− 98.0 + − 21.3 − 106.0 + − 20.8 − 106.0 + − 16.3 − 92.2 + − 11.8 ND ND ND ND

ITC profile of β1- and β2-CBM binding to (A) BCD or (B) gBCD

The top of each graph shows the direct heat outputs obtained from ITC and the binding isotherm is shown below. Energetics of the interactions are displayed in the inset of each graph.

−1 −1 189.6 + and 143.9 + for BCD and gBCD − 29.7 s − 25.3 s 7 respectively. Association rates were also similar, (4.3 + − 1.2) × 10 −1 −1 7 − 1 − 1 M ·s and (5.0 + − 1.1) × 10 M ·s for BCD and gBCD respectively. At 10 ◦ C the rates were approximately 5-fold lower and again between BCD and gBCD the rates were −1 29.4 + similar. Dissociation rates were 42.8 + − 26.3 − 62.5 s and 7 −1 −1 s − 1 and association rates were (1.5 + 2.2) × 10 M ·s and − 7 −1 −1 (1.4 + − 1.3) × 10 M ·s for BCD and gBCD respectively (Table 4).

DISCUSSION

Throughout our studies of the CBM of AMPK it has become increasingly clear that the muscle-specific β2CBM can bind carbohydrates with higher affinity than β1CBM [22,23]. Not only does β2-CBM bind unbranched carbohydrates more strongly, but also it has the added benefit of binding carbohydrates with a single α1,6 glucosylbranch with even greater affinity and these trends are  c The Authors Journal compilation  c 2015 Biochemical Society

254 J.I. Mobbs and others

 c The Authors Journal compilation  c 2015 Biochemical Society

Figure 5

ZZ-exchange and CPMG NMR experiments were used to determine binding rates (K on and K off )

(A) ZZ-exchange peaks observed for Val134 of β2-CBM at 10 ◦ C at various mixing times (0.02, 0.12, 0.2, 0.54 and 0.94 s) with the addition of gBCD. The blue peak represents the decaying apo peak (AA) and the red peak represents the bound peak (BB). The pink and cyan peaks represent the exchange peaks between apo and bound states (A→B and B→A). (B) The volumes of the peaks for Val134 are plotted and show agreeable fits. (C) Examples of CPMG curves used for analysis of β1-CBM + sub-stoichiometric amounts of gBCD.

Specificity of carbohydrate binding to AMPK

255

Table 4 Association (K on ) and dissociation (K off ) rates determined by a combination of CPMG relaxation dispersion NMR experiments, ZZ-exchange NMR experiments and ITC experiments β2-CBM ZZ-exchange BCD

10 ◦ C 15 ◦ C 20 ◦ C 25 ◦ C

β1-CBM CPMG gBCD

BCD

gBCD

K off (s − 1 )

K on (× 107 M − 1 ·s − 1 )

K off (s − 1 )

K on (× 107 M − 1 ·s − 1 )

K off (s − 1 )

K on (× 107 M − 1 ·s − 1 )

K off (s − 1 )

K on (× 107 M − 1 ·s − 1 )

35.3 + − 11.9

4.0 + − 2.7

3.2 + − 1.2 5.6 + − 1.4 9.9 + − 1.6 13.8 + − 0.6*

3.2 + − 1.4 4.0 + − 1.2 4.8 + − 0.9 5.3 + − 1.1*

42.8 + − 62.5

1.5 + − 2.2

29.4 + − 26.3

1.4 + − 1.3

143.9 + − 25.3

5.0 + − 1.1

189.6 + 4.3 + − 29.7 − 1.2 *K off for β2-CBM at 25 ◦ C determined by fitting to Eyring plot, see Supplementary Figure S6. K on determined using K off and ITC derived K d .

observed in the full AMPK complex [23]. As the two CBMs are near identical, with more than 80 % sequence identity and 100 % of the carbohydrate contacting residues conserved, the differences in affinity have not been easily explained. The only major change in sequence is an additional threonine residue (Thr101 ) in β2-CBM which is not hypothesized to directly contact carbohydrate. The structures solved in the present study fully explain the affinity difference of β1- and β2-CBM for gBCD. In β2-CBM the α1,6 branch sits in a pocket between the two conserved tryptophan residues (Trp99 and Trp133 ) forming hydrogen bonds to the backbone carbonyl of Trp99 and Lys102 side chain and hydrophobic contacts with the side chain of Trp133 . These interactions are most probably due to the Thr101 insertion increasing the loop length and allowing the formation of a pocket to accommodate the α1,6 branch. These extra contacts formed by the α1,6 branch of gBCD, as compared with the unbranched BCD, can be observed in the difference in binding affinities, thermodynamics and kinetics. For β2-CBM binding to gBCD the enthalpy was greater and entropy unfavourable, suggesting that these extra bonds come at a loss of dynamic freedom. As our protein dynamics studies show that motion persists in the protein [23], the loss of entropy may be more due to constraining the gBCD. The off-rate for β2-CBM to gBCD was 10-fold lower as compared with BCD, which further suggests a more stable interaction. On the other hand, the structure of β1-CBM in complex with gBCD, shows that the contact site of the α1,6 branch could not be readily determined. Even though this structure had a greater resolution of 1.7 Å and the cyclic core of the carbohydrate was obvious, the α1,6 branch could not be modelled unambiguously into the surrounding density and was therefore not built. As the resolution is higher than the β2-CBM structure, if the position of the α1,6 glucosyl-branch were uniform, one should expect to observe a similar density for the α1,6 glucosyl-branch in contact with the CBM. We therefore conclude that gBCD is binding the β1-CBM in multiple orientations and the α1,6 glucosyl-branch makes only transient interactions. The lack of a discernible bound structure for the α1,6 glucosyl-branch can be explained simply as steric restriction. As the loop (that contains the threonine insertion in β2-CBM) is smaller in β1-CBM the pocket is too restrictive to fit the α1,6 glucosyl-branch. The binding affinities of branched and unbranched carbohydrates to β1-CBM were similar (4.39 + − 0.19 and 4.44 + − 1.09 μM respectively) and only slight differences in enthalpy and entropy were observed, suggesting that both carbohydrates bind in a similar manner. The favourable entropy for gBCD–β1-CBM, in comparison with gBCD–β2CBM, may be due to the increased conformational heterogeneity

of gBCD in the gBCD–β1-CBM complex. These conclusions are supported by the thermodynamic analysis of the mutants, deletion of Thr101 from β2-CBM (β2-Thr101 ) and insertion of a threonine into β1-CBM (β1-Thr101ins ). Removing the threonine residue, β2Thr101 , reduced binding affinity, enthalpy and entropy of the interaction. While inserting the threonine, β1-Thr101ins , increased binding affinity and enthalpy, but made the entropy unfavourable. These data suggest that the threonine residue determines whether the CBM can bind single α1,6-branched carbohydrates better than unbranched carbohydrates. Whereas these structures and thermodynamic analyses have clarified the preference of β2-CBM for α1,6-branched carbohydrates, it is still perplexing why this isoform binds unbranched α1,4 carbohydrates more tightly than β1-CBM. The association rate of BCD to β2-CBM at 10 ◦ C appears to be slightly higher compared with β1-CBM, which is consistent with the additional microsecond motion observed for β2-CBM and suggests that β2-CBM is more amenable to binding [23]. Unfortunately, the difference is small and the data are not precise enough for us to be certain that this is the reason for the difference in affinity. Although still not clearly understood, there are two main hypotheses for the function of the CBM within AMPK, either it is needed for localizing AMPK to glycogen [18] or it is needed for regulating AMPK activity in response to glycogen [7]. Glycogen is a polymer of glucose units joined by α1,4glycosidic linkages with branches of α1,6-glycosidic linkages. In the most accepted model of glycogen structure (the Whelan model), glycogen exists in a series of up to 12 tiers, with the inner tiers being highly branched and the outer chains being unbranched [51]. When skeletal muscle requires energy, glycogen is degraded by the concerted effort of glycogen phosphorylase and glycogen-debranching enzyme. Glycogen phosphorylase first degrades each linear α1,4-linked chain until four residues remain. Glycogen-debranching enzyme has two catalytic activities. First, a transferase activity recognizes the maltotetraose created by glycogen phosphorylase and transfers three glucose residues to a neighbouring α1,4-linked chain. Secondly, a glucosidase activity then cleaves the α1,6-linkage by which this glucose is attached [52]. Our data suggest that β2-containing heterotrimers will bind with high affinity to the single α1,6 branch points created by the glycogen-debranching enzyme transferase activity. Although it is still unclear why AMPK associates with degraded glycogen, it is clear that AMPK binds to partially degraded glycogen via the CBM when a single α1,6 branch is presented. This action would prevent the glycogen-debranching enzyme’s glucosidase active site from accessing its substrate but the rationale for this is far  c The Authors Journal compilation  c 2015 Biochemical Society

256

J.I. Mobbs and others

from clear. The CBM can only bind tightly to α1,6 branches of a single glucose unit in length [21] and we can see now in the crystal structures that the binding pocket of the branch would not accommodate longer branch lengths. In support of the hypothesis that the CBM is involved in regulation, crystal structures show that activating compounds (A-769662 and 991) stabilize an active state whereby the CBM interacts with the KD [19]. Recent data suggest that glycogen binding to the CBM may prevent it from interacting with the KD, thereby pushing the population of AMPK to a more inactive state [20]. Our results suggest that this mechanism of glycogen interaction may be more important for the musclespecific β2-containing AMPK complexes and only at the point when glycogen presents a single α1,6 branch. In conclusion, we have biophysically characterized the binding of β1- and β2-CBM to the branched carbohydrate gBCD. The structures and supporting biochemical data show that the additional threonine residue within β2-CBM, with respect to β1-CBM, allows it to bind single α1,6-branched carbohydrates, such as partially degraded glycogen, with greater affinity than unbranched carbohydrates. Whereas β1-CBM binds branched carbohydrates in a similar manner to unbranched carbohydrates, it only transiently associates with the single α1,6 branch. We hypothesize that binding of partially degraded glycogen may be more important for the muscle-specific β2 containing AMPK complexes.

AUTHOR CONTRIBUTION Jesse Mobbs purified protein, performed the kinetic (NMR) and thermodynamic (ITC) experiments, conducted crystallization trials, acquired the diffraction data and solved the X-ray crystal structures and drafted the manuscript. Ann Koay purified protein, acquired the NMR triple resonance data for assigning the NMR spectra and solved the NMR solution structures. Michael Griffin supervised and guided the crystallization, collection of diffraction data and solved and refined the X-ray crystal structures. Michael Gorman and Michael Parker advised and assisted in the refinement of the X-ray crystal structures. Emma Petrie participated in the solution of the crystal structure of the apo-β2-CBM. Alex Di Paolo and Michael Bieri assisted in the kinetic and thermodynamics experiments. Larissa Doughty conducted initial kinetic experiments. David Stapleton and Paul Gooley conceived the project. Paul Gooley supervised NMR data collections, resonance assignment, NMR kinetic studies, NMR solution structures and thermodynamics experiments. All authors participated in the final preparation of the manuscript.

STRUCTURAL CO-ORDINATES The structural co-ordinates are deposited in the PDB for the crystal structures of free β2-CBM (4Y0G), gBCD bound to β2-CBM (4YEE) and gBCD bound to β1-CBM (4YEF); and the solution NMR structures of free β2-CBM (2LU3) and gBCD bound to β2-CBM (2LU4).

ACKNOWLEDGMENTS We thank Dr Demers, Dr Mittermaier and Dr Korzhnev for providing scripts and software. Parts of this research were undertaken at the CSIRO Collaborative Crystallisation Centre, Victoria, Australia and at the MX1 and MX2 beam-lines of the Australian Synchrotron, Victoria, Australia.

FUNDING This work was supported by the Australian Research Council [grant number DP110103161 (to P.R.G., D.S. and M.W.P.)]; the State of Victoria; the Rowden White Foundation; the Victorian Government Operational Infrastructure Support Scheme; the C.R. Roper Fellowship (to M.D.W.G.); the Australian Research Council Future Fellowship [grant number FT140100544]; and the National Health and Medical Research Council of Australia [grant number 1021645 (to M.W.P.)].  c The Authors Journal compilation  c 2015 Biochemical Society

REFERENCES 1 Hardie, D.G. (2004) AMP-activated protein kinase: a master switch in glucose and lipid metabolism. Rev. Endocr. Metab. Disord. 5, 119–125 CrossRef PubMed 2 Woods, A., Johnstone, S., Dickerson, K., Leiper, F., Fryer, L., Neumann, D., Schlattner, U., Wallimann, T., Carlson, M. and Carling, D. (2003) LKB1 is the upstream kinase in the AMP-activated protein kinase cascade. Curr. Biol. 13, 2004–2008 CrossRef PubMed 3 Hawley, S., Pan, D., Mustard, K., Ross, L., Bain, J., Edelman, A., Frenguelli, B. and Hardie, D. (2005) Calmodulin-dependent protein kinase kinase-beta is an alternative upstream kinase for AMP-activated protein kinase. Cell Metab. 2, 9–19 CrossRef PubMed 4 Oakhill, J., Steel, R., Chen, Z.-P., Scott, J., Ling, N., Tam, S. and Kemp, B. (2011) AMPK is a direct adenylate charge-regulated protein kinase. Science 332, 1433–1435 CrossRef PubMed 5 Minokoshi, Y., Alquier, T., Furukawa, N., Kim, Y.-B., Lee, A., Xue, B., Mu, J., Foufelle, F., Ferr´e, P., Birnbaum, M. et al. (2004) AMP-kinase regulates food intake by responding to hormonal and nutrient signals in the hypothalamus. Nature 428, 569–574 CrossRef PubMed 6 Steinberg, G.R. and Kemp, B.E. (2009) AMPK in health and disease. Physiol. Rev. 89, 1025–1078 CrossRef PubMed 7 McBride, A., Ghilagaber, S., Nikolaev, A. and Hardie, D.G. (2009) The glycogen-binding domain on the AMPK beta subunit allows the kinase to act as a glycogen sensor. Cell Metab. 9, 23–34 CrossRef PubMed 8 Bergeron, R., Russell, R., Young, L., Ren, J., Marcucci, M., Lee, A. and Shulman, G. (1999) Effect of AMPK activation on muscle glucose metabolism in conscious rats. Am. J. Physiol. 276, E938–E944 PubMed 9 Carling, D. and Hardie, D. (1989) The substrate and sequence specificity of the AMP-activated protein kinase. Phosphorylation of glycogen synthase and phosphorylase kinase. Biochim. Biophys. Acta 1012, 81–86 PubMed 10 Shearer, J., Fueger, P., Rottman, J., Bracy, D., Binas, B. and Wasserman, D. (2005) Heart-type fatty acid-binding protein reciprocally regulates glucose and fatty acid utilization during exercise. Am. J. Physiol. Endocrinol. Metab. 288, E292–E297 CrossRef PubMed 11 Chen, Z., McConell, G., Michell, B., Snow, R., Canny, B. and Kemp, B. (2000) AMPK signaling in contracting human skeletal muscle: acetyl-CoA carboxylase and NO synthase phosphorylation. Am. J. Physiol. Endocrinol. Metab. 279, E1202–E1206 PubMed 12 Jørgensen, S., Nielsen, J., Birk, J., Olsen, G., Viollet, B., Andreelli, F., Schjerling, P., Vaulont, S., Hardie, D., Hansen, B. et al. (2004) The alpha2–5’AMP-activated protein kinase is a site 2 glycogen synthase kinase in skeletal muscle and is responsive to glucose loading. Diabetes 53, 3074–3081 CrossRef PubMed 13 Davies, S., Carling, D., Munday, M. and Hardie, D. (1992) Diurnal rhythm of phosphorylation of rat liver acetyl-CoA carboxylase by the AMP-activated protein kinase, demonstrated using freeze-clamping. Effects of high fat diets. Eur. J. Biochem. 203, 615–623 CrossRef PubMed 14 Gwinn, D., Shackelford, D., Egan, D., Mihaylova, M., Mery, A., Vasquez, D., Turk, B. and Shaw, R. (2008) AMPK phosphorylation of raptor mediates a metabolic checkpoint. Mol. Cell 30, 214–226 CrossRef PubMed 15 Hardie, D. (2004) AMP-activated protein kinase: the guardian of cardiac energy status. J. Clin. Invest. 114, 465–468 CrossRef PubMed 16 Jessen, N., Sundelin, E.I. and Moller, A.B. (2014) AMP kinase in exercise adaptation of skeletal muscle. Drug Discov. Today 19, 999–1002 CrossRef PubMed 17 Xiao, B., Sanders, M.J., Underwood, E., Heath, R., Mayer, F.V., Carmena, D., Jing, C., Walker, P.A., Eccleston, J.F., Haire, L.F. et al. (2011) Structure of mammalian AMPK and its regulation by ADP. Nature 472, 230–233 CrossRef PubMed 18 Polekhina, G., Gupta, A., Michell, B.J., van Denderen, B., Murthy, S., Feil, S.C., Jennings, I.G., Campbell, D.J., Witters, L.A., Parker, M.W. et al. (2003) AMPK beta subunit targets metabolic stress sensing to glycogen. Curr. Biol. 13, 867–871 CrossRef PubMed 19 Xiao, B., Sanders, M.J., Carmena, D., Bright, N.J., Haire, L.F., Underwood, E., Patel, B.R., Heath, R.B., Walker, P.A., Hallen, S. et al. (2013) Structural basis of AMPK regulation by small molecule activators. Nat. Commun. 4, 3017 PubMed 20 Li, X., Wang, L., Zhou, X.E., Ke, J., de Waal, P.W., Gu, X., Tan, M.H., Wang, D., Wu, D., Xu, H.E. and Melcher, K. (2015) Structural basis of AMPK regulation by adenine nucleotides and glycogen. Cell Res. 25, 50–66 CrossRef PubMed 21 Koay, A., Rimmer, K.A., Mertens, H.D., Gooley, P.R. and Stapleton, D. (2007) Oligosaccharide recognition and binding to the carbohydrate binding module of AMP-activated protein kinase. FEBS Lett. 581, 5055–5059 CrossRef PubMed 22 Koay, A., Woodcroft, B., Petrie, E.J., Yue, H., Emanuelle, S., Bieri, M., Bailey, M.F., Hargreaves, M., Park, J.T., Park, K.H. et al. (2010) AMPK beta subunits display isoform specific affinities for carbohydrates. FEBS Lett. 584, 3499–3503 CrossRef PubMed 23 Bieri, M., Mobbs, J.I., Koay, A., Louey, G., Mok, Y.F., Hatters, D.M., Park, J.T., Park, K.H., Neumann, D., Stapleton, D. and Gooley, P.R. (2012) AMP-activated protein kinase beta-subunit requires internal motion for optimal carbohydrate binding. Biophys. J. 102, 305–314 CrossRef PubMed

Specificity of carbohydrate binding to AMPK 24 Cai, M., Huang, Y., Sakaguchi, K., Clore, G., Gronenborn, A. and Craigie, R. (1998) An efficient and cost-effective isotope labeling protocol for proteins expressed in Escherichia coli . J. Biomol. NMR 11, 97–102 CrossRef PubMed 25 Delaglio, F., Grzesiek, S., Vuister, G.W., Zhu, G., Pfeifer, J. and Bax, A. (1995) NMRPipe: a multidimensional spectral processing system based on UNIX pipes. J. Biomol. NMR 6, 277–293 PubMed 26 Reference deleted 27 Bahrami, A., Assadi, A.H., Markley, J.L. and Eghbalnia, H.R. (2009) Probabilistic interaction network of evidence algorithm and its application to complete labeling of peak lists from protein NMR spectroscopy. PLoS Comput. Biol. 5, e1000307 CrossRef PubMed 28 Herrmann, T., Guntert, P. and Wuthrich, K. (2002) Protein NMR structure determination with automated NOE assignment using the new software CANDID and the torsion angle dynamics algorithm DYANA. J. Mol. Biol. 319, 209–227 CrossRef PubMed 29 Guntert, P. (2004) Automated NMR structure calculation with CYANA. Methods Mol. Biol. 278, 353–378 PubMed 30 McPhillips, T.M., McPhillips, S.E., Chiu, H.-J., Cohen, A.E., Deacon, A.M., Ellis, P.J., Garman, E., Gonzalez, A., Sauter, N.K., Phizackerley, R.P. et al. (2002) Blu-Ice and the Distributed Control System: software for data acquisition and instrument control at macromolecular crystallography beamlines. J. Synchrotron. Radiat. 9, 401–406 CrossRef PubMed 31 Kabsch, W. (2010) XDS. Acta Crystallogr. D Biol. Crystallogr. 66, 125–132 CrossRef PubMed 32 Evans, P.R. and Murshudov, G.N. (2013) How good are my data and what is the resolution? Acta Crystallogr. D Biol. Crystallogr. 69, 1204–1214 CrossRef PubMed 33 Winn, M., Ballard, C., Cowtan, K., Dodson, E., Emsley, P., Evans, P., Keegan, R., Krissinel, E., Leslie, A., McCoy, A. et al. (2011) Overview of the CCP4 suite and current developments. Acta Crystallogr. D Biol. Crystallogr. 67, 235–242 CrossRef PubMed 34 McCoy, A., Grosse-Kunstleve, R., Adams, P., Winn, M., Storoni, L. and Read, R. (2007) Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674 CrossRef PubMed 35 Murshudov, G., Vagin, A. and Dodson, E. (1997) Refinement of macromolecular structures by the maximum-likelihood method. Acta Crystallogr. D Biol. Crystallogr. 53, 240–255 CrossRef PubMed 36 Paul, E. and Kevin, C. (2004) Coot: model-building tools for molecular graphics. Acta Crystallogr. D Biol. Crystallogr. 60, 2126–2132 CrossRef PubMed 37 Langer, G., Cohen, S.X., Lamzin, V.S. and Perrakis, A. (2008) Automated macromolecular model building for X-ray crystallography using ARP/wARP version 7. Nat. Protoc. 3, 1171–1179 CrossRef PubMed 38 Adams, P.D., Afonine, P.V., Bunkoczi, G., Chen, V.B., Davis, I.W., Echols, N., Headd, J.J., Hung, L.W., Kapral, G.J., Grosse-Kunstleve, R.W. et al. (2010) PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. D Biol. Crystallogr. 66, 213–221 CrossRef PubMed 39 Vagin, A.A., Murshudov, G.N. and Strokopytov, B.V. (1998) BLANC: the program suite for protein crystallography. J. Appl. Crystallogr. 31, 98–102 CrossRef

257

40 Renko, M., Taler-Vercic, A., Mihelic, M., Zerovnik, E. and Turk, D.A. (2014) Partial rotational lattice order-disorder in stefin B crystals. Acta Crystallogr. D Biol. Crystallogr. 70, 1015–1025 CrossRef PubMed 41 Pletnev, S., Morozova, K.S., Verkhusha, V.V. and Dauter, Z. (2009) Rotational order-disorder structure of fluorescent protein FP480. Acta Crystallogr. D Biol. Crystallogr. 65, 906–912 CrossRef PubMed 42 Reference deleted 43 Farrow, N., Zhang, O., Forman-Kay, J. and Kay, L. (1994) A heteronuclear correlation experiment for simultaneous determination of 15N longitudinal decay and chemical exchange rates of systems in slow equilibrium. J. Biomol. NMR 4, 727–734 CrossRef PubMed 44 Demers, J.-P. and Mittermaier, A. (2009) Binding mechanism of an SH3 domain studied by NMR and ITC. J. Am. Chem. Soc. 131, 4355–4367 CrossRef PubMed 45 Loria, J.P., Rance, M. and Palmer, A.G. (1999) A relaxation-compensated Carr-Purcell-Meiboom-Gill sequence for characterizing chemical exchange by NMR spectroscopy. J. Am. Chem. Soc. 121, 2331–2332 CrossRef 46 Bieri, M. and Gooley, P. (2011) Automated NMR relaxation dispersion data analysis using NESSY. BMC Bioinformatics 12, 421 CrossRef PubMed 47 Mazur, A., Hammesfahr, B., Griesinger, C., Lee, D. and Kollmar, M. (2013) ShereKhan–calculating exchange parameters in relaxation dispersion data from CPMG experiments. Bioinformatics 29, 1819–1820 CrossRef PubMed 48 Korzhnev, D., Salvatella, X., Vendruscolo, M., Di Nardo, A., Davidson, A., Dobson, C. and Kay, L. (2004) Low-populated folding intermediates of Fyn SH3 characterized by relaxation dispersion NMR. Nature 430, 586–590 CrossRef PubMed 49 Toone, E.J. (1994) Structure and energetics of protein-carbohydrate complexes. Curr. Opin. Struct. Biol. 4, 719–728 CrossRef 50 Tomme, P., Creagh, A., Kilburn, D. and Haynes, C. (1996) Interaction of polysaccharides with the N-terminal cellulose-binding domain of Cellulomonas fimi CenC. 1. Binding specificity and calorimetric analysis. Biochemistry 35, 13885–13894 CrossRef PubMed 51 Roach, P.J., Depaoli-Roach, A.A., Hurley, T.D. and Tagliabracci, V.S. (2012) Glycogen and its metabolism: some new developments and old themes. Biochem. J. 441, 763–787 CrossRef PubMed 52 Liu, W., Madsen, N.B., Fan, B., Zucker, K.A., Glew, R.H. and Fry, D.E. (1995) Effects of oligosaccharide binding on glycogen debranching enzyme activity and conformation. Biochemistry 34, 7056–7061 CrossRef PubMed 53 Chen, V.B., Arendall, III, W.B., Headd, J.J., Keedy, D.A., Immormino, R.M., Kapral, G.J., Murray, L.W., Richardson, J.S. and Richardson, D.C. (2010) MolProbity: all-atom structure validation for macromolecular crystallography. Acta Crystallogr. D Biol. Crystallogr. 66, 12–21 CrossRef PubMed 54 Shen, Y., Delaglio, F., Cornilescu, G. and Bax, A. (2009) TALOS + : a hybrid method for predicting protein backbone torsion angles from NMR chemical shifts. J. Biomol. NMR 44, 213–223 CrossRef PubMed 55 Laskowski, R.A., Rullmannn, J.A., MacArthur, M.W., Kaptein, R. and Thornton, J.M. (1996) AQUA and PROCHECK-NMR: programs for checking the quality of protein structures solved by NMR. J. Biomol. NMR 8, 477–486 PubMed

Received 2 March 2015/13 March 2015; accepted 16 March 2015 Published as BJ Immediate Publication 16 March 2015, doi:10.1042/BJ20150270

 c The Authors Journal compilation  c 2015 Biochemical Society

Copyright of Biochemical Journal is the property of Portland Press Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.