Crystal Structures of the Adenylate Sensor from Fission Yeast AMP-Activated Protein Kinase

See allHide authors and affiliations

Science  23 Mar 2007:
Vol. 315, Issue 5819, pp. 1726-1729
DOI: 10.1126/science.1137503


The 5′-AMP (adenosine monophosphate)–activated protein kinase (AMPK) coordinates metabolic function with energy availability by responding to changes in intracellular ATP (adenosine triphosphate) and AMP concentrations. Here, we report crystal structures at 2.9 and 2.6 Å resolution for ATP- and AMP-bound forms of a core αβγ adenylate-binding domain from the fission yeast AMPK homolog. ATP and AMP bind competitively to a single site in the γ subunit, with their respective phosphate groups positioned near function-impairing mutants. Unexpectedly, ATP binds without counterions, amplifying its electrostatic effects on a critical regulatory region where all three subunits converge.

AMPK senses the onset of energy limitation and initiates adaptive responses, including regulation of key enzymes in each of the major branches of metabolism: fatty acid and sterol synthesis, sugar metabolism, protein synthesis, and DNA replication (13). Long-term regulation by AMPK is controlled by phosphorylation of transcription factors and co-activators that regulate each of these metabolic functions (4). Mammalian AMPK is activated by binding to AMP, but this activation is inhibited by ATP (13). Because cellular adenosine diphosphate (ADP)/ATP ratios remain nearly constant due to the equilibrium maintained by adenylate kinase, it is thought that AMPK activity in vivo depends primarily on the ATP/AMP ratio, the primary determinant of cellular energy charge (5). Lowered cellular energy charge can arise either from inhibition of ATP synthesis, for example, in hypoxia or starvation (5), or by increased ATP consumption, as occurs in skeletal muscle during exercise (6, 7). When energy charge is low, active AMPK inhibits numerous ATP-consuming pathways and also activates mobilization of intracellular energy stores to produce ATP. Conversely, when energy charge is high, AMPK inactivation is favored, leading to enhancement of energy storage and use. AMPK also functions in organism-level energy homeostasis by responding to systemically circulating hormones, including leptin (8), adiponectin (9), and resistin (10).

Canonical AMPKs are αβγ heterotrimers (13). Although gene fusions between the β and γ subunits are found in some plant species, the overall domain architecture of the enzyme is conserved in all eukaryotes (11). In humans, multiple genes encode isoforms of each AMPK subunit (α1, α2, β1, β2, γ1, γ2, and γ3) (1). The α subunit includes a typical serine-threonine kinase domain near the N terminus and a C-terminal regulatory domain characteristic of AMPKs. AMPK β subunits are required for trimer assembly and subcellular localization of the AMPK complex (12, 13). Many β isoforms include a glycogen-binding domain (GBD), for which structures are now known (14). The heart of AMPK regulation by adenylate binding is thought to reside in γ subunits, which are composed of four repeated cystathionine β-synthase (CBS) domains (1517). A number of CBS domain crystal structures have been determined (1820), but only one recent structure included a bound ligand (21). Biochemical and genetic studies have produced an initial view of interactions among AMPK subunits (1) whereby β subunits play a central role in heterotrimer formation through interactions with both α and γ subunits. Trimer interaction sites on both α and β subunits have been mapped to their respective C-terminal regions (22). Nonetheless, the precise molecular architecture of the AMPK heterotrimer has not been defined.

We produced a trimer fragment from the AMPK genes of the yeast Schizosaccharomyces pombe, which yielded two distinct crystal forms that diffract x-rays to 2.6 Å (for the AMP complex) and 2.9 Å (for the ATP complex) resolution. The trimer core consists of residues α440 to α576, β205 to β298, and the whole of the γ subunit, γ1to γ334 (Fig. 1). The structure omits two regions of the AMPK complex: the β subunit N-terminal GBD (14) and the N-terminal kinase domain of α, both of which have been characterized crystallographically (23, 24). Although the enzymology of S. pombe homologs has not been investigated, sequence conservation among AMPKs from different species suggests that the structures presented here will be representative for this protein class.

Fig. 1.

Overall structure of the adenylate binding region from S. pombe AMPK with bound AMP. The ATP-bound form is nearly identical (fig. S6) and reveals no global structural changes attributable to nucleotide identity. (A) Ribbon diagram of a single heterotrimer, with α, β, and γ subunits colored yellow, blue, and green, respectively. The single molecule of bound AMP is shown in CPK representation, and connections to the GBD and KD at the N-termini of the β and α subunits, respectively, are indicated. (B) View rotated 90°, highlighting the adenylate binding entrance (AXP) and phosphate binding tunnel, which is capped on the putative KD-interaction surface by a polar flap from the β subunit. The structure corresponds to a heterotrimer defined by limited proteolysis, as indicated in (C): hatched regions were excluded. Each of the two crystal forms reported here includes a dimer of trimers in the asymmetric unit (D). Analytical ultracentrifugation analysis also demonstrates a dimer of trimers configuration.

We determined crystal structures by selenomethionine multiwavelength anomalous diffraction (MAD) phasing (25) for both AMP- and ATP-bound crystal forms that each contained a dimer of trimers in the asymmetric unit (Fig. 1 and table S1). The potential importance of this dimer of trimers is underscored by a human Wolff-Parkinson-White syndrome mutation [human γ2 Asn488→Ile488 (N488I), corresponding to S. pombe γS247 (26)] that maps to the interface. Further, dimerization is consistent with the observation from Marmorstein and co-workers that mutants that disrupt kinase domain (KD) dimerization in vitro fail to disrupt sucrose nonfermenting 1 (SNF1) complex dimerization in vivo, suggesting association through other domains (23). Each AMPK trimer is roughly triangular, with a wide base formed by the γ subunit, which associates with a tight αβ complex that forms the narrower apical domain (Fig. 1, A and B). The α subunit C-terminal domain forms a compact, mixed αβ domain (fig. S1) that is topologically related to the kinase-associated domains (KADs) of microtubule affinity-regulating kinase (MARK) kinases (27). The β subunit C-terminal region lacks an independent hydrophobic core (fig. S2) and wraps wholly around the α KAD (Fig. 1, A and B, and fig. S4), forming extensive hydrophobic contacts. These features suggest an obligate nature for αβ complexes.

The γ subunit forms an elliptical disk with an aqueous pore in the center. An adenine nucleotide is bound at the interface between CBS domains 3 and 4, positioning the phosphate groups in the pore. CBS domains 1 and 2 form the interface with the αβ complex (Fig. 1, A and B, and fig. S5). The γ interface to αβ is mediated primarily by interaction of a two-stranded β sheet from the β subunit, which hydrogen bonds with the β1strand of γ to form a three-stranded intersubunit β sheet (Fig. 1 and fig. S5). This result disagrees with a recent mutagenesis study that concluded that there is no direct binding between β and γ (28). In contrast to the αβ complex, very few hydrophobic interactions are formed between β and γ, with this interface formed mainly through hydrogen bonding and salt bridge interactions, suggesting that βγ association may not be obligatory.

The γ subunit structures reported here provide initial views of CBS domains binding their regulatory ligands. A single molecule of either ATP or AMP binds at the same site, between β strands 6 and 7 of CBS3 and β strands 9 and 10 of CBS4 (Fig. 2). ATP and AMP bind in nearly identical conformations and use the same set of ligating residues. In addition to these structural elements from CBS3 and CBS4, CBS2 also contributes critical interactions important for binding the nucleotide phosphates. Remarkably, outside the localized nucleotide binding regions, no substantial structural changes attributable to bound nucleotide are observed among these four independent trimers (fig. S6).

Fig. 2.

Nucleotide binding. (A) Stereo diagram of AMP bound within the γ subunit. Adenine and ribose moieties are bound by functional groups within the CBS3-CBS4 domain pair; however, the terminal α-phosphate forms salt bridges with two Arg side chains (γR139 and γR141) donated from CBS domain 2. (B) One protomer exhibits a single ATP conformation (ATP 1), whereas the other (C) adopts multiple alternate conformations (ATP 2). ATP binds through a set of protein ligands identical to those used by AMP, accommodating the β and γ phosphates of ATP by adopting a compact helical structure for the triphosphate group (B). Electrostatic surfaces of the putative kinase-binding face, looking into the phosphate tunnel, show that the charge potential of the tunnel entrance is largely positive (blue) when AMP is bound [(D) and (E)] but negative (red) when ATP is bound [(F) and (G)].

In both ATP- and AMP-bound structures, a hydrophobic cleft formed from the side chains from CBS3 (γI216 and γP220) (26) and CBS4 (γI303 and γF292) sandwich the adenine ring. The N2 ring nitrogen hydrogen bonds to the backbone carbonyl groups of γA196 and γA218. The ribose moiety is bound in a polar pocket in CBS4, in which the ribose 2′ and 3′ hydroxyls are hydrogen bonded to side chain atoms of γT191, γD308, and γS305. In the AMP complex structure (Fig. 2A), the terminal AMP α-phosphate forms salt bridges with the side chain of the conserved residue γR290 in CBS4 and two residues in CBS2, γR139 and γR141. Excellent electron density is observed for all AMP atoms in the complex structure (fig. S7). The ATP phosphates are ligated by an identical set of arginine side chains (Fig. 2, B and C); however, γR290 ligates the terminal γ-phosphate, and γR139 and γR141 have poor density and appear to adopt multiple conformations in coordinating the α-and β phosphates.

In AMPK, unlike the vast majority of ATP-bound protein structures, no electron density for metal counterions is observed bound to ATP. To confirm this observation, we performed experiments with crystals grown in the presence of 2 mM GdCl3. The Gd3+ lanthanide ion is known to effectively substitute for Mg2+ in ATP substructures (29) and provides a large anomalous signal for crystallographic experiments. Anomalous difference maps revealed no bound ions (fig. S7). These observations suggest that ATP-bound metal counterions must be stripped before binding to AMPK, where charge balance is provided mainly by the side chainsof arginine residues. This agrees with the prior observation from Hardie and co-workers that Mg2+ is not required for ATP binding by purified γ subunits (16).

Our structural findings are in agreement with the reported relative binding affinities of AMP (∼100 μM) and ATP (∼400 μM) for a protein encoding fragments of the CBS3 and CBS4 domains of the human γ2 subunit (16). However, the apparent absence of regulation by AMP of S. cerevisiae AMPK activity (30, 31) raises the possibility that nucleotide-dependent regulation of S. pombe AMPK may differ from that of the human enzyme. Cellular ATP concentrations are higher than AMP concentrations, often by an order of magnitude. Even when ATP concentrations fall, leading to AMPK activation, ATP concentrations generally remain higher than AMP. Thus, tighter binding to AMP is a functional requirement for AMPK. The structures presented suggest how AMPK binds ATP more weakly than AMP. This appears to be accomplished, at least in part, through the necessity for stripping ATP of shielding ions for binding to the adenylate sensor. Further, the adenylate binding site is better able to accommodate the smaller AMP ligand: The α phosphate of AMP and the γ phosphate of ATP are coordinated by the same Arg side chains, and the positions of the two terminal phosphates are nearly the same (Fig. 2 and fig. S6).

The overall orientation of nucleotide binding, for both AMP and ATP, situates the nucleotide phosphates toward an internal cavity that we refer to as the phosphate tunnel, which spans the breadth of the γ subunit disk (Fig. 1B). In both ATP- and AMP-bound forms, the surface-exposed nucleotide groups include parts of the adenine ring and the phosphate-distal face of the ribose moiety. These surface-exposed groups are identical in both AMP- and ATP-bound structures, suggesting that the nucleotide-binding face of the AMPK regulatory trimer is unlikely to function as the site of activation modulated by adenylate binding. The primary differences between AMP- and ATP-bound forms of the heterotrimer lie within the phosphate tunnel. One substantial difference between the ATP and AMP complexes is in the surface electrostatic potentials at the distal exit of the phosphate tunnel, the putative kinase-interaction face (Fig. 2, D to G). This suggests the possibility that the different effects of these ligands might arise from the charge difference between the mono- and triphosphate groups of AMP and ATP.

Prior studies have identified a number of mutations within the regulatory heterotrimer that lead to impaired function of AMPK, primarily in the γ subunit (1). These mutations include an insertion in helix E at position γ91 (32), and a point mutation at γS247 (N488I in human γ2) (33), which is found within the dimer-of-trimers interface region. However, the large majority of functionally important mutations, which include changes to residues γR290 (human γ2 R531) (34), γR141 (human γ2 H383) (32), γV56 (human γ2 R302) (35), and γT162 (human γ2 T400N) (33), are all found lining the interior surface of the phosphate tunnel (Fig. 3, A and B). Residues γR290 and γR141 co-ordinate nucleotide phosphates; however, the other mutations, relative to the bound nucleotide phosphate groups, are positioned further toward the protein surface opposite to the nucleotide binding face. This face of the molecule, where α, β, and γ subunits meet, may constitute a region for KD interaction, and we thus refer to it as the putative kinase domain interaction face (Figs. 1B and 3).

Fig. 3.

[(A) and (B)] Functional mutations map within the phosphate binding tunnel, a large internal cavity that traverses the γ subunit, shown in red. The majority of known function-impairing mutants map to the surface of this tunnel, positioned between the terminal phosphate of the bound nucleotide and the putative kinase-binding face. Two orthogonal views are shown.

The phosphate tunnel traverses the γ subunit, defining a large void that is capped on the KD-binding face by a polar loop from the β subunit. We refer to this loop, which includes residues 244 to 255, as the β flap. The region of the β flap that covers the phosphate tunnel includes only polar and charged residues and makes no contacts to the hydrophobic core, suggesting the possibility for structural rearrangement. The β flap appears to be highly mobile (average B factors of 84.3 Å2 in the four independent β subunits, as compared with an overall average B factor of 51.6 Å2 for all protein atoms) and adopts slightly different conformations in the four independent copies of the structures presented here. The majority of γ-subunit mutations that affect AMPK activation are positioned within the phosphate tunnel, between the terminal phosphate of bound AXP and the β flap. Because the difference between the inhibitory (ATP) and activating (AMP) ligands is in the number of phosphates placed within the tunnel and because mutants that affect kinase activation also lie within this tunnel, it appears likely that this represents a critical regulatory region.

We have presented crystal structures that define the core heterotrimeric architecture for AMPKs. The S. pombe AMPK binds either AMP or ATP at a single site, suggesting that activating ligands such as AMP are likely to function by displacing the inhibitory ligand ATP. Nonetheless, possible binding of additional regulatory nucleotides in the context of nucleotide mixtures or the holoenzyme complex cannot be excluded. Although a detailed understanding of the mechanism of AMPK regulation will require structures of the holoenzyme, the structures presented here should provide an entry point for the rational design of AMPK-directed therapeutics.

Supporting Online Material

SOM Text

Figs. S1 to S8

Table S1

References and Notes

View Abstract

Navigate This Article