Structure of Pyruvate Dehydrogenase Kinase

Structure of the Monomer

The monomer of PDK2 folds into two well defined domains (9). The N-terminal domain extends from residue 1 to residue 179 and consists of 7 α-helices connected by flexible loops. Four of these long helices are amphipathic and form a helical bundle with a hydrophobic core (Fig. 1, a and b). The α5/α6 helices (Fig. 1b) almost form one continuous long helix except for a short loop region where Pro-120 disturbs the helical secondary structure. This “kink” in the helix orients the conserved His-115 toward the interior of the helical bundle (see Fig. 4d). The positioning of this residue is of particular interest as it is conserved throughout the PDK family of kinases and was a candidate as a possible phosphohistidine intermediate (22). Its orientation within the helical bundle, where it is not solvent-accessible, makes it unlikely that His-115 fills this role.

Within the helical bundle, His-115 is stabilized by a network of hydrogen bonds between the imidazole ring and Ser-83 on the adjacent α4 helix and via a well ordered water molecule to Tyr-80 and Arg-154 (see Fig. 4d). The result of this irregularity in the long α5/α6 helix is to expose to solvent a number of arginine residue side chains. These positive charges are concentrated in a patch on the floor of the crevice that is formed between the N- and C-terminal domains (see Fig. 5).

The C-terminal nucleotide-binding domain is folded as a mixed α/β structure (Fig. 1, a, b, and d). The core consists of five β-strands forming a single β-sheet. Four α-helices are arranged on one face of the sheet toward the N-terminal domain. A hydrophobic core formed from side chains as the helices and strands pack together stabilizes the domain. The opposite face of the β-sheet is in contact with the equivalent β-sheet of the other monomer in the asymmetric unit and forms the dimer interface (Fig. 1, a and c). There is an unusual “left-handed” arrangement of β2-α9- β3 (Fig. 1b), which has also been observed in bacterial histidine kinase nucleotide-binding domains (23). The electron density for the ADP co-crystallized with PDK was quite strong and was located between the α-helices and the β-sheet as it widens toward one end (see Fig. 3a).

The overall fold of the C-terminal nucleotide-binding domain of PDK2 is remarkably similar to prokaryotic histidine kinases and ATPases. Aligning the published structures of histidine kinases (CheA (24) and EnvZ (25)) and ATPases (DNA Gyrase B (26), MutL (27), and Hsp90 (28)) with TOP3D from the CCP4 (17) suite gives an RMSD for the Cα atoms of less than 2 Å over the bulk of the C-terminal domain (Fig. 2). The topology of the ATPase nucleotide-binding domain is different from that of PDK2 or the histidine kinases, because it has a rearrangement of the secondary structural elements, but the overall fold is preserved.

The C-terminal and N-terminal domains of PDK2 are connected by a flexible, poorly ordered loop from Ser-171 to Pro-177 (α7/α8). They are stabilized relative to each other by the antiparallel interaction between β1 and β7, as well as hydrophobic interactions between the side chains of helices α4/α5/α6 in the N terminus and α11 in the C terminus. As a result of this association, there is a rather large solvent-exposed cavity between the two domains (see Fig. 5). Helices α9 and α11 form one side of the cavity and the N-terminal helices α4/α5/α6 the other side. Due to the kink in helix α5/α6 noted above, one side of the cavity carries a positive charge because of the exposed arginine side chains.

Dimer Interactions

Biochemical evidence suggests that PDK2 is a dimer. The asymmetric unit in the crystal structure of PDK2 contains two molecules related by a strong non-crystallographic symmetry axis, and these two molecules are presumed to form the functional dimer. Protomers in the dimer are in a head-to-head orientation, with the primary interaction between the β sheets of the C-terminal domains (Fig. 1a). The interface buries 2200 Ų (18), or 12.4% of the total surface area (17,763 Ų for monomer; crystallography and NMR system (CNS) (18), 1.4 Å probe) and is stabilized by a number of hydrophobic interactions and hydrogen bonds. The remainder of the molecules extends out in a horseshoe configuration, with the two N-terminal domains in the arms and the C-terminal domains forming the dimer interface at the base. In this configuration, the two nucleotide binding pockets are located on the outer, solvent-exposed surface of the dimer and are readily accessible.

ATP-binding Site

As mentioned previously, PDK2 shows sequence homology to the histidine kinases, specifically in the G1, G2, and N-box motifs. Observing the presence of these characteristic motifs in PDK2, Bowker-Kinley and Popov (9) have suggested that PDK belongs to the ATPase/kinase superfamily, which includes histidine protein kinases (EnvZ and CheA) and ATPases (Hsp90, DNA Gyrase B, and MutL). The present study confirms this observation, because fitting of the structure of PDK2 to other members of the family, the structures of which are known, shows that the overall fold of the C-terminal nucleotide-binding domain of PDK2 is structurally homologous to other members of this family. This domain similarity is shown in Fig. 2, in which the Cα tracing of the nucleotide-binding domains of the histidine kinase CheA and the ATPases Gyrase B, MutL, and Hsp90 are superimposed on PDK2. This structural homology is remarkable given that only the ADP/ATP moiety was used to fit these molecules. In addition, x-ray structural studies have shown that there is extremely high structural homology within the ATP-binding pocket in the histidine kinases and ATPases (29).

In the case of PDK2, the electron density corresponding to the bound ADP was quite clear, as is demonstrated in Fig. 3a, and was located at the center of a cone-shaped cavity formed between the β-sheet and the α-helices of the C-terminal domain (Fig. 1b). The ADP molecule itself adopts a skewed orientation (Figs. 2 and 3a), in contrast to the more open arrangement seen in cytoplasmic serine/threonine kinases but similar to that found in DNA Gyrase B, Hsp90, and MutL. The conserved G1 motif (residues Asp-282-X-Gly 284-X-Gly 286) is involved in binding the adenine moiety of the nucleotide. In PDK2, it forms one side of the binding pocket (Fig. 3a, left) and is a portion of the loop connecting helices α10 and α11 (residues 296–320). The G1 box is the basis of the adenine specific interactions between the nucleotide and the protein (compare Figs. 3a and 4a). These include a direct hydrogen bond from the C6 amino group (N-6) of the adenine ring to the carboxylate side chain of Asp-282 in the G1 box. This same aspartate residue also provides one of the three protein ligands for the water molecule that hydrogen bonds to the N1 imino nitrogen of the adenine ring (Fig. 4a). Two other protein ligands for this water molecule come from the side chain of the invariant Thr-346 and the peptide nitrogen of the invariant Gly-286 (G1 box). These interactions provide the specificity of PDK2 for adenyl nucleotides. The N-7 of the adenine ring is bound to another highly ordered water molecule that is part of an extensive hydrogen bonding network of water molecules and ionic side chains that extends across the nucleotide-binding site (Figs. 3a and 4a).

The central portion of this loop containing the G1 box extends over the nucleotide-binding site. In the family of ATPases, this loop forms a lid that is thought to limit access of ATP to the binding site (23) and may serve a similar role in PDK2. Part of this region is often disordered in the structures of the ATPases/kinases and also in PDK2, suggesting a high degree of mobility.

The other side of the binding pocket distal to the lid corresponds to the G2 box (Gly 317-X-Gly319-X-Gly321), again a conserved motif found in bacterial histidine kinases and structurally homologous ATPases (Fig. 3a, right) (9). The G2 box curves around the α- and β-phosphate groups of the ADP. These phosphate groups are stabilized by a number of direct hydrogen bonds between their oxygen atoms and the amide nitrogen atoms of residues Gly-317, Phe-318, Tyr-320, and Gly-321 as shown in Fig. 4b. Interestingly, Gly-319 (Fig. 4b) adopts a glycine-specific ϕ angle that is essential to the curve of this loop and the formation of these stabilizing hydrogen bonds, thus explaining the “dead” kinase when this residue is mutated (9). The positioning of the phosphate groups takes advantage of the helix dipole from α11 to stabilize their polyanionic character.

Additional interactions between the ADP and the protein include a direct hydrogen bond between the 2′ hydroxyl group of the ribose and the side chain of the invariant residue Thr-302 (ATP lid). The carbonyl oxygen of Thr-302 interacts with the 3′ hydroxyl group of the ribose, which, in turn, forms a hydrogen bond to the side chain of the invariant Arg-250 protruding from helix α9 across the nucleotide-binding pocket. The position of Arg-250 is stabilized through interactions with Glu-254 (helix α9) and the solvent hydrogen bonding network.

The electron density associated with the magnesium ion coordinated to each ADP molecule is easily seen (Fig. 3b). Each phosphate group provides an oxygen atom to coordinate with the Mg²⁺ ion (Fig. 4c). The side chain oxygen of Asn-247 together with two water molecules completes the observed coordination sphere (Fig. 4c). Asn-247 is part of the conserved N-box motif. These residues (Glu-243-X-X-Lys 246-Asn-247-X-X-Ala-250) are located on helix α9 (see Fig. 3b and Fig. 4, b and c) with the conserved residues occurring on successive turns of the helix. Glu-243 is thought to act as a general base in the reaction mechanism (see discussion below), whereas Asn-247 coordinates with the magnesium ion. The presumed sixth position to finish the classic octahedral coordination of the Mg²⁺ does not contain sufficient electron density to warrant the addition of a water molecule to the model. It is likely that the γ-phosphate of ATP will occupy this position when ATP is bound (see below).

Possible Mechanisms of Catalysis and Regulation of Enzymatic Activity

Because this structure was solved with bound nucleotide, a number of conclusions may be drawn about a possible mechanism of catalysis for the phosphate-transfer reaction. The position of the nucleotide within the binding pocket predetermines the approximate location of the γ-phosphate. A least-squares fitting program from the CCP4 suite (TOP3D) (17) was used to map the coordinates of the ATP-analogue found in the structure of MutL (30), the ATP-binding site of which is structurally homologous to that of PDK2 (see previous discussion), as a rigid body onto the ADP in the PDK2 structure. The adenine, ribose, α-phosphate, and β-phosphate groups of the MutL structure line up exactly with the ADP found in PDK2, and the resulting position of the γ-phosphate makes biochemical sense. The modeled γ-phosphate oxygen atoms could make four hydrogen bonds to the peptide nitrogen atoms of the G1 loop residues (Fig. 4b). In addition, the Mg²⁺ ion is in a position to coordinate to one of the γ-phosphate oxygen atoms, thus filling the sixth position that is missing in our model. This model would still conserve the stabilization of the polyanionic ATP by the helix dipole from the N-terminal end of helix α11.

Given this active site conformation, it is reasonable to ask which residues are in a position to carry out the chemistry of phosphate transfer and if formation of a phospho-histidine intermediate is reasonable. Examination of the structure of PDK2 in the vicinity of the putative γ-phosphate binding site provides no obvious candidate for a residue to carry out a nucleophilic attack. The only histidine conserved throughout the PDK family, and therefore a candidate for a histidine-phosphate intermediate, is His-115 (22). However, it is involved in making a number of important structural hydrogen bonds in the N-terminal domain and is not solvent-accessible (Fig. 4d). Given the absence of a nucleophile, it is quite likely that PDK uses a general base catalysis to activate the hydroxyl group of the substrate seryl residue for direct attack on the γ-phosphate group. It is interesting to examine the PDK2 catalytic pocket in light of what has been concluded about the very closely related catalytic pockets of the ATPases and histidine kinases.

In the ATPases/bacterial histidine kinases, the same basic structure is used for catalysis of histidine phosphorylation or ATP hydrolysis (i.e. transfer of the phosphate to water). In the histidine kinases, the phosphoacceptor (a histidine residue in the N-terminal domain) is itself a quite reactive nucleophile, and it has been suggested that it directly attacks the γ-phosphate of ATP. The nucleophilicity of the histidine residue is further increased through interaction with a neighboring glutamate residue (His-47 and Glu-70 in CheA) (31). It is noteworthy that both of these residues lie in the N-terminal phosphate acceptor domain of their respective kinases. Thus the N-terminal domain complements the catalytic center by providing a nucleophile for phosphotransfer and an activating glutamate residue. It is evident from our structure that this N-terminal complement to the catalytic pocket is not present in PDK2.

In the ATPases (DNA Gyrase B and Hsp90), the nucleophile (water) is much less reactive and needs to be activated in order to attack the γ-phosphate. In DNA Gyrase B, the activating glutamate (Glu-42) acts as a general base and is located within the ATPase nucleotide-binding pocket (32). Also, it is polarized and aligned by the neighboring His-38. Thus, the ATPase nucleotide-binding domain contains all of the circuitry needed for catalysis. This appears to be similar to the arrangement of catalytic residues in the nucleotide-binding domain of PDK2. The glutamate residue corresponding to Glu-42 of the DNA Gyrase B is Glu-243 (N-box, Fig. 3b) in PDK2 and is uniformly conserved in all mitochondrial protein kinases identified to date. It is positioned 3 Å from a water molecule in PDK2 that occupies the likely position for the serine substrate (Fig. 4b). This suggests that the mitochondrial protein kinases might use a mechanism similar to the general base catalysis of the ATPases to activate the serine residue of the protein substrate. In addition, His-239 of PDK2 might correspond to His-38 of DNA Gyrase B and may be involved in polarization and positioning of Glu-243 (Fig. 3b). It is in line with the modeled γ-phosphate group and located one turn along the α9 helix from residue Glu-243. The imidazole side chain of His-239 is thus positioned to orient the glutamate residue to interact with the putative serine substrate position. At the same time, the presence of His-239 could modify the effective pK_a of Glu-234 to an appropriate range for the role of a general base in the reaction. It thus appears that the active site of PDK2 is capable of activating the serine substrate and transferring the γ-phosphate directly, without the necessity of a covalent intermediate.

The positive charge to stabilize the transition state, represented by Lys-246 (N-box) in PDK2 (Figs. 3b and 4b), is also present in the ATPases. Examination of sequence alignments of ATPases shows a variety of residues at the PDK2 Lys-246 position. However, the available x-ray crystal structures demonstrate that there is a positively charged residue in the catalytic pocket in all of the ATPases that is able to occupy the position in three-dimensional space of the lysine ∈-nitrogen. Examples of this would be Lys-737 in Gyrase B and Lys-307 in MutL. In other words, when these structures are aligned to the PDK2 model with TOP3D (17), they position a positive charge less than 2 Å from the ∈-nitrogen of Lys-246 in our structure. Given this catalytic similarity of the ATP-binding domain of PDK2 with the family of ATPases, it is not surprising that PDK2 has been found to also possess an ATPase activity, representing 1–3% of total kinase activity (33).

In contrast to the ATPases, an alignment of multiple histidine kinases to the PDK2 model shows that the position of the general base Glu-243 of PDK2 is occupied by a variety of residues and the positive charge contributed by Lys-246 in PDK2 is entirely absent. Thus, although the overall fold of the PDK2 is that of a histidine kinase, the chemistry of the active site is similar to that of an ATPase and is probably capable of a direct transfer of the γ-phosphate to the serine substrate.

PDK2 and the Pyruvate Dehydrogenase Complex

It is generally believed that the enzymatic activity of PDK is intimately dependent upon interactions with protein substrate and supporting structures of the PDC. Several lines of evidence suggest that the kinase is tightly bound to the lipoyl-bearing domain 2 of the E2 component of the complex (34). Association with the lipoyl-bearing domain appears to be responsible for the almost 10-fold increase in kinase activity as well as conferring the sensitivity of the kinase activity to metabolite concentrations (35). As noted above, the lipoyl domains are self-folding structures carrying a lipoic acid prosthetic group that is acetylated by the acetyl-TPP-E1 complex formed after pyruvate is decarboxylated.

The three-dimensional structure of these domains is conserved across species as a flattened β-barrel with a hydrophobic core and many exposed acidic residues (5). As noted earlier, examination of the three-dimensional structure of the PDK2 monomer revealed the existence of a rather large cavity between the C- and N-terminal domains (Fig. 5). The size and shape of the cavity appears to be complementary to the size and shape of the lipoyl-bearing domain. The cluster of positively charged residues exposed in the floor of the cavity by the discontinuity in helix α5/α6 is ideally located to interact with the negatively charged residues of the lipoyl domain. These observations bring about the intriguing possibility that the pocket between the domains of the monomer might comprise the lipoyl domain binding site. Orienting the lipoyl domain to fit into this putative binding cleft positions the lipoic acid ligand close to the nucleotide-binding pocket. This suggests that the lipoic acid could interact directly with the residues near the nucleotide-binding pocket. In this way, the state of the lipoic acid (acetylated, reduced, or oxidized) would be in a position to influence kinase activity.

The two putative lipoyl binding pockets are located on the outside of the PDK2 dimer, oriented in opposite directions. This suggests that PDK2 may bind to more than one lipoyl domain at a time. It has been shown that only a few kinase molecules are associated with any one PDC, and they must service many E1 molecules (34). This activity is presumed to involve the kinase moving over the surface of the complex without dissociation. The mechanism proposed for this movement is a “hand over hand” motion in which the kinase alternately binds and releases E2 domains while remaining bound to the complex (37). The orientation of the proposed binding sites in this model of PDK2 makes this alternating, multiple binding a very reasonable possibility.

Conclusions

The three-dimensional x-ray structure of PDK2 presented here clearly shows that this molecule shares a lineage with modern prokaryotic histidine kinases and ATPases. The fold of the nucleotide-binding C-terminal domain has been exactly conserved throughout this family to preserve the positioning of the nucleotide in the binding pocket. Variations of particular amino acids over time have allowed this domain to be adapted to three different biochemical roles: ATPase, histidine kinase, and serine kinase. In addition to this insight into molecular adaptation, the structure presented also illustrates the dimer and domain structure of PDK2. Taken together, our results suggest a mechanism whereby PDK2 can bind to the lipoyl domain of the pyruvate dehydrogenase complex and adjust its activity based on the state of the lipoic acid ligand. This has important implications for the modulation of PDC activity in normal intermediary metabolism. In addition, the inappropriate activity of PDK2 in diabetes provides a therapeutic opening for controlling hyperglycemia independent of insulin sensitivity. Clearly, future studies will involve determining the structure of PDK2 in complex with the lipoyl domain in order to better understand this important interaction.