AMP-Activated Protein Kinase β-Subunit Requires Internal Motion for Optimal Carbohydrate Binding

Abstract

AMP-activated protein kinase interacts with oligosaccharides and glycogen through the carbohydrate-binding module (CBM) containing the β-subunit, for which there are two isoforms (β₁ and β₂). Muscle-specific β₂-CBM, either as an isolated domain or in the intact enzyme, binds carbohydrates more tightly than the ubiquitous β₁-CBM. Although residues that contact carbohydrate are strictly conserved, an additional threonine in a loop of β₂-CBM is concurrent with an increase in flexibility in β₂-CBM, which may account for the affinity differences between the two isoforms. In contrast to β₁-CBM, unbound β₂-CBM showed microsecond-to-millisecond motion at the base of a β-hairpin that contains residues that make critical contacts with carbohydrate. Upon binding to carbohydrate, similar microsecond-to-millisecond motion was observed in this β-hairpin and the loop that contains the threonine insertion. Deletion of the threonine from β₂-CBM resulted in reduced carbohydrate affinity. Although motion was retained in the unbound state, a significant loss of motion was observed in the bound state of the β₂-CBM mutant. Insertion of a threonine into the background of β₁-CBM resulted in increased ligand affinity and flexibility in these loops when bound to carbohydrate. However, these mutations indicate that the additional threonine is not solely responsible for the differences in carbohydrate affinity and protein dynamics. Nevertheless, these results suggest that altered protein dynamics may contribute to differences in the ligand affinity of the two naturally occurring CBM isoforms.

Introduction

AMP-activated protein kinase (AMPK) is an essential enzyme for regulating energy homeostasis (1, 2). It is able to activate fatty acid β-oxidation (3, 4), glycolysis (5), and glucose transport (6) while inhibiting lipid synthesis (7). In this way, AMPK functions as a focal point for cellular and whole-body mechanisms of energy homeostasis. AMPK also serves to control gene transcription (8) and protein synthesis (9), adapting the body to new regimes of energy supply and demand, such as during exercise (10). The enzyme has attracted widespread interest since the discovery that it is activated by metformin, a drug that is commonly used to treat Type 2 diabetes (11).

AMPK is a heterotrimer that consists of a catalytic α and regulatory β− and γ-subunits. Each subunit exists in different isoforms, with two α (α₁ and α₂), two β (β₁ and β₂), and three γ (γ₁, γ₂, γ₃) isoforms differentially expressed in various tissues in up to 12 different isoenzymes (12). The kinase activity of AMPK is regulated by phosphorylation of the conserved threonine in the catalytic domain of the α-subunit (13), by adenine nucleotide binding to the γ-subunit (14, 15) and by glycogen binding to a conserved mid-molecule carbohydrate-binding module (CBM) located in the β-subunits (16). The CBMs of AMPK adopt a β-sandwich fold with a single carbohydrate-binding site formed by one of the β-sheets and a β-hairpin (17). The amino acid residue sequence conservation between the CBMs of the two β-subunit isoforms is high (81.7% identity in humans), and inspection of the available structures (β₁-CBM in complex with β-cyclodextrin (PDB ID: 1z0m) and unbound β₂-CBM (PDB ID: 2f15)) shows that the residues in contact with carbohydrate are strictly conserved and positioned similarly in these structures (Fig. 1). In the x-ray structure of the β₁-CBM in complex with β-cyclodextrin (17), five glucosyl units make direct contact with the protein. Two tryptophans (Trp-99 and Trp-133) form a cradle that makes extensive hydrophobic contacts with three glucosyl units (G7-G1-G2) of β-cyclodextrin. The side chain of Leu-146 pierces the β-cyclodextrin ring, making additional contacts with G7-G1-G2. The side-chain groups of Lys-126, Thr-148, and Asn-150 form hydrogen bonds to OH groups of G1-G2 while the main-chain carbonyls of Gln-145 and Leu-146 hydrogen-bond G4 and G5 (17). All of these residues are similarly positioned in the unbound β₂-CBM, although the tip of the loop from Ser-144 to Gly-147, which includes Leu-146, moves slightly away from the carbohydrate-binding site, and the indole ring of Trp-133 points down. Binding of carbohydrate likely leads to small movements of these residues to ensure close packing with the ligand. Despite these marked similarities, a comparison of the carbohydrate-binding properties of the β₁- and β₂-CBMs with a range of linear, cyclic, and branched oligosaccharides showed that the β₂-CBM always binds carbohydrate more tightly with an up to a 15-fold difference in affinity (18).

Figure 1.

Structure and sequence alignment of AMPK β-subunits. (A) The x-ray structure of β₁-CBM (1z0m) bound to cyclodextrin. The β-strands are labeled. (B) β₂-CBM (2f15) has the same fold as β₁-CBM. (C) Overlay of β₁-CBM and β₂-CBM. (D) An expansion of β₁-CBM showing residues in contact with cyclodextrin. (E) A comparison of the β₁- and β₂-CBM sequences shows that the residues in contact with carbohydrate are conserved, but a major difference is a Thr insertion in β₂-CBM at position 101. The position of this Thr in the structure of β₂-CBM is indicated in B. (Sequence alignment was performed by ClustalX, and sequence numbering is according to β₁-CBM.)

A major difference between the two CBM sequences is a Thr insertion in β₂-CBM at position 101 (Fig. 1), which appears to be a recent evolutionary event for this CBM family (18). This insertion is in a loop connecting β-strand 2 and the ill-defined outer β-strand 3. Consequently, the structures of the β₁- and β₂-CBMs deviate away from each other near this loop. Inspection of the structures, however, shows that the insertion does not affect the main-chain hydrogen-bond patterns observed in the two proteins, or the position of the critical carbohydrate-binding residue, Trp-100 (Fig. 1). Hydrogen bonds involving side chains do show some differences between the two structures. For example, the side chain OH of Ser-93 and NH₂ of Asn-97 hydrogen-bond to the carbonyl of the Thr insertion (Thr-102) in β₂-CBM. In contrast, in β₁-CBM the OH of Ser-94 hydrogen-bonds to Ser-101 and the NH₂ of Asn-98 to the carbonyl of Leu-93. Of note, the side chain of the Thr insertion points away into the solvent and is not involved in any inter- or intramolecular interactions.

The importance of the protein loop conformation in defining the specificity and affinity of protein-ligand interactions is well known for a variety of systems, including antibody-antigen (19), enzymes (20), receptors (21), and signaling proteins (22). This suggests that the ligand affinity differences observed between the two CBMs may be due to the loop containing the Thr insertion. The additional residue may increase protein flexibility, allowing the β₂-CBM to sample conformational states that are not accessible to the β₁-CBM. Indeed, deletion of Thr-101 in β₂-CBM (β₂^ΔThr-CBM) resulted in a threefold lower affinity to carbohydrates of the mutant, and insertion of a Thr in the corresponding position in β₁-CBM (β₁^Thr-Ins-CBM) increased the affinity to carbohydrates by threefold compared with wild-type (WT) CBMs (18). Because there are no clear differences in the nature of the structural contacts between carbohydrate and the CBMs, it was hypothesized that dynamics may account for the affinity differences. NMR spectroscopy is a powerful method for characterizing protein dynamics on timescales ranging from picoseconds to seconds (23), and it has been applied to investigate the conformational dynamics of loops in a variety of proteins (24, 25, 26). In this work, we characterized the microsecond-to-millisecond motion of CBMs (including mutants) using NMR techniques and found significant differences between the β₁- and β₂-CBMs, both free and saturated with ligand. In addition, the carbohydrate affinity differences observed for the isolated β₁-CBM and β₂-CBM domains were found to be retained in the intact heterotrimeric enzyme complex α1β1γ1 but not in α1β2γ1, suggesting that there is an inherent difference between the two domains in the native proteins.

Materials and Methods

Cloning and protein expression

AMPK β₁- and β₂-CBM proteins were enriched with ¹⁵N via a 1-L fermenter and expressed and purified as previously described (18). Site-directed mutagenesis was performed with the use of QuikChange (Qiagen) according to standard procedures. Heterotrimers of AMPK were produced as described elsewhere (27).

NMR spectroscopy

¹⁵N-labeled NMR samples were prepared to a final concentration of 0.3 mM protein in phosphate buffer (100 mM) at pH 6.8. Glucosyl-β-cyclodextrin was chosen as the ligand because it is highly soluble and binds tightly to both CBMs (K_d = 5 μM for β₁-CBM; K_d < 0.5 μM for β₂-CBM, in similarity to reported values for β-cyclodextrin (18)). To achieve saturation of protein with ligands, 1.5 mM of glucosyl-β-cyclodextrin (Wako Pure Chemical Industries Ltd., Tokyo, Japan) was added to β₂-CBM WT and 15 mM to β₁-CBM WT and Thr insertion as well as β₂ Thr knockout. We measured the effective ¹⁵N R₂ relaxation rates using single-scanned interleaved relaxation-compensated constant-time Carr-Purcell-Meiboom-Gill (CPMG) pulse sequences (28, 29) with a constant delay of 0.08 s (T_CPMG), 24 scans per row, on a 600 MHz spectrometer (Bruker Avance III equipped with z-gradient cryoprobe) at 298 K. Samples that showed chemical exchange were further measured on an 800 MHz spectrometer (Bruker Avance II also equipped with a z-gradient cryoprobe) at 298 K. Relaxation dispersion profiles were produced by recording spectra with varying CPMG frequencies (ν_CPMG) (ν_CPMG = 50, 2× 75, 100, 2× 150, 200, 300, 500, 2× 700, 800, 900, 1000, 1500 Hz). In addition, a reference spectrum without a CPMG pulse train was recorded. Transverse relaxation rates R₂ were measured at 600 MHz by means of ¹⁵N T₂ experiments (30) on 0.3 mM β₁- and β₂-CBM samples that were free or bound to glucosyl-β-cyclodextrin. An additional experiment was performed on 0.15 mM free β₂-CBM. Spectra were acquired in a single-scan interleaved manner, with an additional CPMG block placed during recovery for heat compensation. Nine spectra were recorded to obtain R₂ (T₂ relaxation delays = 0.016, 0.032, 0.064, 2× 0.096, 0.129, 2× 0.161, 0.193 s).

Data analysis

Spectra were processed with NMRPipe (31) and analyzed with CcpNmr (32). Overlapping peaks were excluded from further analysis. Peak intensities were used to calculate the effective transverse relaxation rates ($(R_2^{eff}=-1/T\ln ({\text{I}}_{\nu \text{CPMG}}/{\text{I}}_0)$, where T is the constant CPMG time (0.08 s), and I_νCPMG and I₀ are the signal intensities in the presence or absence of the CPMG pulse, respectively). Data-fitting, model selection, and data presentation of relaxation dispersion experiments were performed with the open source software NESSY (33). ¹⁵N T₂ experiments were analyzed with relaxGUI (34).

Curve-fitting and model selection

The data were fitted to no-, fast- (35, 36), and slow- (37) exchange models (models 1, 2, and 3, respectively). In the case of fast exchange, the chemical shift difference δω and exchange constant k_ex cannot be determined uniquely. Therefore, a single value, Φ (= p_ap_bδω², where p_a and p_b are the major and minor populations, respectively), was extracted. Note that the slow-exchange model is described by the unapproximated general exchange expression, which can only be fully determined if spins are in the slow-limit exchange. The best model was chosen using Akaike information criteria with second-order correction for small sample size (AICc) (38). Errors of extracted parameters were estimated using 500 Monte Carlo simulations, by varying individual rates within determined experimental errors obtained by duplicating measured points (33). Data recorded at two static fields were simultaneously fitted to the models described above. Parameters were held constant for each residue, except for δω (model 3) and Φ (model 2), which were weighted by the heteronuclear frequency (¹⁵N).

Cluster analysis

Residues that showed chemical exchange were grouped and simultaneously fitted to model 2 or model 3. Parameters k_ex and p_b were held constant for the whole cluster, and R₂⁰, δω (model 3), and Φ (model 2) were individually determined. The best model was selected using AICc and errors of extracted parameters were estimated using 500 Monte Carlo simulations.

Sedimentation velocity analysis

Samples were analyzed using an XL-I analytical ultracentrifuge (Beckman Coulter, Fullerton, CA) equipped with an AnTi60 rotor at 25°C. Samples at 0.2 mM were added to double-sector Epon-filled centerpieces, with sodium phosphate buffer (100 mM) in the reference compartment. To avoid saturation of the detector, due to the high concentrations involved, initial absorbance wavelength spectra from 230 nm to 310 nm were acquired to select optimal wavelengths for monitoring the samples during sedimentation. Radial absorbance data were then acquired at a rotor speed of 40,000 rpm, using wavelengths of 298 and 290 nm for the β₁- and β₂-CBMs, respectively, and with radial increments of 0.003 cm in continuous scanning mode. The sedimenting boundaries were fitted to a model that described the sedimentation of a distribution of sedimentation coefficients with no assumption of heterogeneity (c(s)) using the program SEDFIT (39). Data were fitted using a regularization parameter of p = 0.95, floating frictional ratios, and 200 sedimentation coefficient increments in the range of 0.3–10 S. For fitting to the c(M) model to estimate the molar mass, identical parameters as for c(s) were used except for the fitting range of 1000–100,000 Da.

Tryptophan intrinsic fluorescence

The fluorescence of 5 μM AMPK heterotrimers was measured as previously described for the isolated CBMs (18). Triplicates were simultaneously fitted to

$$\Delta F/F_0=\left(\frac{1}{2P_0}\right)\times \left[K_d+P_0+L_0-{\left({(K_d+P_0+L_0)}^2-4P_0/L_0\right)}^{1/2}\right]$$

with the use of in-house-written scripts. ΔF is the absolute value of the fluorescence intensity corrected for protein dilution during the titration, F₀ is the fluorescence intensity in the absence of carbohydrate, P₀ is the total protein concentration, and L₀ is the total ligand concentration. Errors for the dissociation constants (Kd) were estimated using 500 Monte Carlo simulations.

Results

Peak broadening of the Trp-133 side-chain NH in bound state of β₂-CBM

The two conserved Trps of β₁-CBM, Trp-100 and Trp-133 (Fig. 2 A), form hydrophobic contacts with carbohydrate molecules (17, 40). Consequently, small but significant chemical shift differences are observed for the indole NH signals in the ¹⁵N heteronuclear single quantum coherence (¹⁵N-HSQC) spectra for both Trps of the β₁-CBM in the free state compared with the glucosyl-β-cyclodextrin-bound state, reflecting the interaction with the carbohydrate (Fig. 2 B). In the free state of β₂-CBM, both indole NH signals (Trp-99 and Trp-133) were also observed (Fig. 2C). However, the resonance for Trp-133 in the bound state showed a markedly larger shift and was broadened due to chemical exchange. Deletion of Thr-101 of β₂-CBM (β₂^ΔThr-CBM) reduced the chemical shift difference and line-broadening of the indole NH signal of Trp-133, essentially reproducing spectra similar to that of WT β₁-CBM (Fig. 2E). For the β₁^Thr-Ins-CBM mutant, insertion of a Thr introduced both a large chemical shift difference and line-broadening of the indole NH signal of Trp-133, which were similar to those observed for the β₂-CBM WT (Fig. 2D).

Figure 2.

Effect of glucosyl-β-cyclodextrin on the Trp side chain NH for the β₁- and β₂-CBM of AMPK. (A) Structure of β₁-CBM bound to cyclodextrin, showing the two Trp that contact the oligosaccharide (17). (B) Upon titration of β₁-CBM with glucosyl-β-cyclodextrin, the indole NH of both Trps shifted. (C) Both Trp signals of unbound β₂-CBM are visible in the NMR spectra, but upon ligand binding, Trp-133 shifted and line-broadened. (D) Insertion of a Thr residue in the corresponding position of β₁-CBM (β₁^Thr-Ins-CBM) introduced line-broadening effects on ligand binding. (E) Deletion of Thr-101 in β₂-CBM (β₂^ΔThr-CBM) resulted in the indole NH of Trp-133 showing reduced line-broadening on ligand binding. The unlabeled Trp indole peaks, which are essentially unaffected by the addition of glucosyl-β-cyclodextrin, are assigned to Trp-85 of β₁-CBM and Trp-84 of β₂-CBM.

Only β₂-CBM shows microsecond-to-millisecond motion in both free and bound states

The differences in line-broadening of the indole NH of Trp-133 suggest that the protein dynamics of the two modules may differ, and could partially explain the differences in carbohydrate affinity. To understand the influence of ligand binding on the protein dynamics of the AMPK β-CBMs, we conducted NMR ¹⁵N R₂ relaxation dispersion experiments for the WT and mutants of the β₁- and β₂-CBMs either unbound or bound to glucosyl-β-cyclodextrin under saturating conditions. These experiments detect motions between different conformational states whose interchange occurs on the microsecond-to-millisecond timescale (41, 42, 43). Although exchange can occur between more than two states, it is convenient to describe the exchange that occurs between a ground state A and an excited state B. The measured chemical or conformational exchange provides information about the exchange rate (k_ex) between states A and B (k_ex = k_AB + k_BA).

No chemical exchange for β₁-CBM was observed in either the unbound or glucosyl-β-cyclodextrin-bound states (Fig. 3A). The lack of chemical exchange in β₁-CBM in both states is supported by similar R₂ relaxation rates for each residue observed in T₂ spin-relaxation experiments (Fig. S2A). In contrast, a number of residues for either free or bound β₂-CBM showed elevated R₂ relaxation rates (Fig. S2B), suggesting the presence of chemical exchange in this isoform. Consistent with this observation, in the unbound state, chemical exchange was observed for several residues of β₂-CBM that are located within or near the carbohydrate-binding β-hairpin (Fig. 3 C), as measured by ¹⁵N R₂ relaxation dispersion experiments. The mobile residues in the unbound conformation showed fast exchange (k_ex = 4936 ± 351 1/s (50% trimmed mean)). Of note, residues near or within the loop where Thr-101 is located did not show a chemical exchange. Instead of quenching motion upon addition of glucosyl-β-cyclodextrin to β₂-CBM, the motion became more widespread to include residues surrounding Trp-99 and Thr-101. Although these residues also experienced fast chemical exchange, with the majority showing a reduced exchange rate (k_ex) of 1416 ± 162 1/s (50% trimmed mean) compared with those in the free β₂-CBM.

Figure 3.

Protein dynamics of AMPK β₁- and β₂-CBMs. (A) No chemical exchange was detected for β₁-CBM in either the unbound state or bound to glucosyl-β-cyclodextrin. (B) The Thr insertion into β₁-CBM (β₁^Thr-Ins-CBM) resulted in increased motion in the bound state for residues close to Trp-100. (C) In contrast, β₂-CBM has a mobile carbohydrate-binding β-hairpin in the unbound state. The carbohydrate-binding β-hairpin is stabilized upon ligand binding, whereas residues close to Trp-99 and Thr-101 gain mobility. (D) Deletion of Thr-101 in β₂-CBM (β₂^ΔThr-CBM) reduces the gain of motion observed for the bound state of the WT. Relaxation dispersion parameters for each protein in each state are summarized in Table S1 and Table S2.

Effect of Thr insertion on the protein structure and dynamics of β₂-CBM

To further investigate the role of the Thr insertion in β₂-CBM, Thr-101 was deleted from β₂-CBM (β₂^ΔThr-CBM) and inserted into β₁-CBM (β₁^Thr-Ins-CBM) at the corresponding position of β₂-CBM. Mutations may cause major changes in structure and consequently functionality compared with the WT (44). Chemical shift differences between WT and mutant β₁- and β₂-CBMs were used to estimate the effect of the Thr insertion into β₁-CBM or deletion in β₂-CBM (Fig. 4). In β₁-CBM, insertion of a Thr residue at position 102 induced chemical shift differences for nearby residues (i.e., residues Leu-93 to Ser-101) in both the free and bound states. In addition, chemical shift differences were observed for residues Tyr-125 to Phe-127 in the unbound state. Similarly, deletion of Thr-101 in β₂-CBM resulted in chemical shift differences for nearby residues Gly-94 to Ile-103 and residues Lys-126 to Gly131 in the free and bound states. The region ∼93–103 encompasses β-strand 2 and the ill-defined β-strand 3, whereas the majority of residues 125–131, which are located in β-strand 5 and a loop, are distant from the mutation (Figure 1, Figure 4). The observed differences in chemical shifts suggest that the insertion of a Thr at position 102 has a long-range effect. The region 125–131 includes Lys-126 and is near Trp-133. The latter two residues are both essential for carbohydrate binding. To ensure that the mutations did not significantly alter the structure, ¹⁵N-edited NOESY spectra of β₂-CBM and β₂^ΔThr-CBM bound to saturating concentrations of glucosyl-β-cyclodextrin were compared (Fig. S1). The NOESY spectra indicated that no significant structural changes were caused by the mutations. Strip plots for three residues that form contacts between the protein and carbohydrate (Trp-99, Lys-126 (whose ¹⁵NH is the most affected resonance that is distant from the site of mutation), and Leu-146) were essentially the same for β₂-CBM and β₂^ΔThr-CBM (Fig. S1 and Fig. 4).

Figure 4.

Chemical shift difference for β₁- and β₂-CBM compared with the mutations. (A) Insertion of a Thr in β₁-CBM, corresponding to Thr-101 in β₂-CBM, resulted in significant chemical shift differences for residues 93–101 and 125–127 in the unbound state. (B) Addition of glucosyl-β-cyclodextrin to β₁-CBM and β₁^Thr-Ins-CBM resulted in a decrease of shift differences for residues 125–127 compared with the free state. (C) Deletion of Thr-101 in β₂-CBM resulted in shift differences for residues 94–103 and 126–131. (D) In the bound state, the chemical shift differences between β₂-CBM and β₂^ΔThr-CBM were smaller than those observed in the unbound state. Chemical shift differences were calculated according to $ShiftDifference=\sqrt{(\Delta {\delta }_H^2)+{0.15}^2(\Delta {\delta }_N^2)}$. Dashed lines indicate the median and first median deviation. Dark bars indicate the most significant differences.

Relaxation dispersion experiments were also performed on β₂^ΔThr-CBM and β₁^Thr-Ins -CBM free and under saturating concentrations of glucosyl-β-cyclodextrin. Residues of the unbound β₁^Thr-Ins-CBM had no observable microsecond-to-millisecond motion, implying that the insertion is not responsible for the observed motion in unbound β₂-CBM. However, in the presence of saturating ligand, chemical exchange was observed for residues near Trp-100 and the Thr insertion, and at the base of the ligand-binding β-hairpin (Fig. 3B). The exchange rate constant for carbohydrate-saturated β₁^Thr-Ins-CBM was k_ex = 2067 ± 67 1/s (50% trimmed mean), and the increase in mobility of residues near the Thr insertion was consistent with the gain in affinity of β₁^Thr-Ins-CBM for carbohydrate when compared with WT β₁-CBM. However, the motion induced was not as widespread as that observed for WT β₂-CBM in the bound state. The mutant β₂^ΔThr-CBM in its free state showed a number of residues with microsecond-to-millisecond motion (k_ex = 2580 ± 769 1/s (50% trimmed mean)), mostly localized to the carbohydrate-binding β-hairpin, and generally similar to unbound β₂-CBM (Fig. 3D). Upon addition of glucosyl-β-cyclodextrin, most of this motion was quenched in β₂^ΔThr-CBM, with only Ser-100 (k_ex = 700 ± 48 1/s) and Ile-103 (k_ex = 4159 ± 823 1/s) experiencing exchange effects. This near-complete loss of motion in the β₂^ΔThr-CBM is similar to the absence of motion observed for bound β₁-CBM, and is also consistent with the observed reduction in affinity to carbohydrate compared with the WT β₂-CBM.

Distal mobile regions in β₂-CBM move on similar timescales

Each residue in the free or bound states was best described by the fast-limit exchange, two-state model. Therefore, we were only able to extract k_ex and Φ = p_a p_b δω² (where p_a and p_b are the fractional populations of the major and minor states, respectively, and δω is the chemical shift difference between the two exchanging states). To investigate whether motion within the CBM occurs in the same timeframe, residues that experienced chemical exchange were custered and simultaneously fitted them to either slow-limit or fast-limit exchange two-state models (Fig. 5). Clustered fits were obtained by grouping residues near Thr-101 (Gly-94, Ser-95, Asn-97, Asn-98, and Trp-99) and Trp-133 (Val-134 and His-135). The clustered residues were best described by the fast-limit exchange two-state model with k_ex = 1280 ± 519 1/s. Because the chemical shifts of the glucosyl-β-cyclodextrin bound and free states are known (18), populations were calculated by solving the equation for Φ. A major state population of p_a = 0.897 ± 0.099 and a minor state population of p_b = 0.102 ± 0.099 were obtained.

Figure 5.

Cluster analysis of β₂-CBM bound to glucosyl-β-cyclodextrin. (A) Residues 94, 95, 97, 98, 99, 134, and 135 were clustered and simultaneously fitted to slow- and fast-exchanging two-state models (Cα and N are shown as spheres for clustered residues). (B) Clustered residues fit best to the two-state fast-exchange model (solid lines and circles for data recorded at 600 MHz, dotted lines and triangles for data recorded at 800 MHz). The extracted value for chemical exchange was k_ex = 1280 ± 519 1/s. (C) The parameter Φ (Φ = p_a p_b δω²) is represented as color- and width-coded (from yellow to red and increasing width for increasing Φ-values; the backbone N of each system is shown as a sphere), and (D) plotted for spins in a static field of 800 MHz.

Isolated CBMs are present as monomers

A major concern in working with concentrated protein solutions is the presence of protein complexes that influence protein dynamics. Therefore, sedimentation velocity experiments of 0.2 mM β₁- and β₂-CBM were performed (Fig. S3). Both β₁- and β₂-CBM in the unbound state were found to be present as a homogeneous single species, both at ∼10 kDa, indicating no aggregation. In addition, transverse relaxation rates (R₂) are known to be highly susceptible to aggregation. A comparison of R₂ values of β₂-CBM in the unbound state at 0.3 and 0.15 mM shows no differences (Fig. S2C), indicating that the protein does not aggregate at the higher concentrations (0.3 mM) of the CBM used in NMR studies. Therefore, the observed differences in protein dynamics of the modules are not due to protein-protein interactions.

The β₂-subunit containing heterotrimeric AMPK has a higher affinity for carbohydrate

To determine whether the affinity differences of the β₁- and β₂-CBMs for oligosaccharides are also observed in the intact AMPK heterotrimer, three constructs of AMPK (α₁β₁γ₁, α₁β₂γ₁ and α₁β₁ΔCBMγ₁ (27)) were bacterially expressed and purified for binding studies. The affinity of the three constructs for maltoheptaose and glucosyl-maltoheptaose were measured using tryptophan intrinsic fluorescence (Fig. 6). The heterotrimer α₁β₁ΔCBMγ₁, in which the CBM domain has been deleted, showed no significant binding to either oligosaccharide. This shows that carbohydrate binding is specific to the CBM domains. The heterotrimer α₁β₁γ₁ showed weak binding to both oligosaccharides. The dissociation constant (K_d) of α₁β₁γ₁ to maltoheptaose and glucosyl-maltoheptaose was 649 ± 31 μM and 669 ± 20 μM, respectively. Higher affinity was measured for α₁β₂γ₁, in which a K_d of 74 ± 2 μM and 24 ± 2 μM was measured when it was bound to maltoheptaose or glucosyl-maltoheptaose, respectively.

Figure 6.

Dissociation constant measurements of AMPK heterotrimers. Intrinsic tryptophan fluorescence signals for (A) α₁β₁γ₁ and (B) α₁β₂γ₁ show tighter maltoheptaose binding to the AMPK trimer containing the β₂-CBM. (C) No binding to the heterotrimer without the CBM (α₁β₁ΔCBMγ₁) is observed. Similarly, glucosyl-maltoheptaose bound less tightly to (D) α₁β₁γ₁ than to (E) α₁β₂γ₁, and did not bind to (F) CBM-deleted AMPK trimer (α₁β₁ΔCBMγ₁). Each titration was performed in triplicate and the three resulting curves were simultaneously fitted.

Discussion

The isolated β₁- and β₂-CBM domains of AMPK show different affinities for carbohydrate, with β₂-CBM always binding with a higher affinity than β₁-CBM. The most marked difference observed was the 15-fold increase in affinity by β₂-CBM for glucosyl-maltoheptaose, an α1,4 linked linear carbohydrate with a single α1,6 linkage (18). The differences observed for the isolated domain are generally maintained in the AMPK heterotrimer, suggesting that AMPKs containing the β₂-subunit will be localized to glycogen more readily than those containing the β₁-subunit. Compared with the isolated CBMs, the affinity differences of the AMPK heterotrimers for carbohydrate are more pronounced, with AMPK β₂-subunit binding ∼10-fold more tightly to maltoheptaose and ∼30-fold more tightly to glucosyl-maltoheptaose. The AMPK β₂-subunit is highly expressed in skeletal muscle, contributing to α₁β₂γ₁, α₂β₂γ₁, α₂β₂γ₃ heterotrimers, whereas the β₁-subunit is absent in human muscle but is predominant in most other tissues as either α₁β₁γ₁ or α₂β₁γ₁ (45, 46, 47). The different distributions of the β-isoforms and the heterotrimers suggest tissue-specific functional roles.

Because these affinity differences are inherent to the CBM domains, the underlying molecular phenomena were investigated. Careful inspection of the available crystal structures of the isolated CBMs does not show a simple structural reason for these differences. Although the insertion of a Thr into a loop between the β₂- and β₃-strands of the β₂-CBM results in a minor distortion of the protein backbone compared with β₁-CBM (the root mean-square deviation (RMSD) of the Cα, N, and C′ for residues 79–155 of β₁-CBM to 78–100,102–155 of β₂-CBM is 0.67 Å, whereas for 79–94,103–155 of β₁-CBM to 78–93,103–155 of β₂-CBM, thus excluding the Thr loop, is 0.63 Å), the carbohydrate contact residues are structurally conserved between the two isoforms. Nevertheless, the insertion of a Thr residue into the loop suggests a difference in protein dynamics, which may account for the affinity differences. Indeed, the protein loop dynamics, especially on the microsecond-to-millisecond timescale, was previously shown to play an important role in ligand affinity and specificity. For example, a mutant of the C1B domain of protein kinase Cα binds diacylglycerols 100-fold more tightly than the WT protein, which appears to be linked to loop flexibility (26). Mutations of a loop in RNAase A distant from the catalytic site reduce both loop flexibility and product release rates (24). Mutation of a Pro to Gly for an SH2 domain increases loop flexibility and increases ligand affinity eightfold (48), and a residue deletion from a loop of a WW domain reduces flexibility and affinity (49).

The results presented here show that β₁-CBM, in either the free or carbohydrate-bound state, did not have any detectable microsecond-to-millisecond motion. In contrast, motion was present for both the free and carbohydrate-bound β₂-CBM. However, the microsecond-to-millisecond dynamics in the free β₂-CBM was not observed within the Thr loop. It was mostly observed at the base of the β-hairpin that makes critical contacts with carbohydrate. These residues include Val-134, His-135, Asn-151, and Ile-152. Upon binding to carbohydrate, the rate of exchange of the detected microsecond-to-millisecond motion in the β₂-CBM decreased, but it became more widespread to include residues Gly-94 to Lys-107, which encompass the Thr loop and a part of the outer β-strand 3. Of significance, for β₂-CBM bound to carbohydrate, the motion at the base of the carbohydrate-binding β-hairpin and the Thr loop appeared to be on the same timescale. The dynamics at the base of the carbohydrate-binding β-hairpin allows the β₂-CBM domain to access states that are amenable to carbohydrate binding and are either unavailable to the β₁-CBM or are significantly reduced in population and cannot be detected. Upon binding to carbohydrate, new regions of motion that include residues in the Thr loop are detected, suggesting that additional states become available to the β₂-CBM. These states are presumed to preexist in the free protein (50); however, binding of ligand shifts the equilibrium, and consequently these new bound states may only be detectable in experiments on bound β₂-CBM (51, 52).

Insertion and deletion of the Thr into the WT background did not fully recover ligand-binding affinity (18) or reproduce the dynamics observed for the WT proteins. Chemical shift differences between the mutants and corresponding WT CBMs showed a long-range effect between the region of the Thr insertion to residues involved in carbohydrate binding, including Lys-126 and Trp-133, suggesting that a structural network formed through the central β-strands. In β₂-CBM, Thr-101 is relatively distant from Trp-133. The Cα-to-Cα distance of these residues is 13 Å in the crystal structure (2f15), although as the large aromatic ring of Trp-133 sits over the top β-sheet, it approaches the backbone atoms of Thr-101. For example, the N of Thr-101 and the Cη2 of Trp-133 are only 7.5 Å apart. However, the deletion of the Thr from the β₂-CBM (β₂^ΔThr-CBM) did not completely remove the motion observed for residues at the base of the carbohydrate-binding β-hairpin in the free state, suggesting that other residues may also be important for this flexibility. However, on binding carbohydrate, β₂^ΔThr-CBM showed little microsecond-to-millisecond motion. In contrast, the insertion of a Thr into β₁-CBM showed no motion when it was free of carbohydrate. Once bound, residues in the threonine loop and a residue (Asn-151) at the base of the carbohydrate-binding β-hairpin of β₁^Thr-Ins-CBM showed motion. These data suggest that the Thr insertion is important for the flexibility of the CBM but is not solely responsible for enabling access to the additional minor states of the protein, which may enhance carbohydrate affinity.

In conclusion, although the naturally occurring CBM isoforms are closely related in structure, they markedly differ in affinity for carbohydrate. This difference in carbohydrate affinity is also observed in the AMPK heterotrimer. In this study, the microsecond-to-millisecond motion of the isolated CBMs was characterized, and it showed a trend for increased motion being linked to higher carbohydrate affinity. It is likely that this phenomenon also occurs in the AMPK heterotrimer.

Editor: Michael Feig.