Crystal structure of pyruvate dehydrogenase kinase 3 bound to lipoyl domain 2 of human pyruvate dehydrogenase complex

L2 binding markedly stimulates the kinase activity of PDK3

There are inconsistencies in the literature regarding the magnitude of L2 domain-stimulated PDK3 activity. The study by the Roche group showed that the L2 domain and E2p stimulate human PDK3 activity by 12.8- and 17-fold, respectively (Baker et al, 2000). These authors concluded that L2 binding is the major determinant for stimulated PDK3 activity and that the small additional activation by E2p over L2 is imparted by the co-localization of the kinase and substrate E1p to E2p. In contrast, the Popov group showed that PDK3 activity is activated by 3- and 38-fold, respectively, by the L2 domain and E2p (Tuganova et al, 2002).

To resolve this discrepancy, the kinase activity of human PDK3 was measured by E1p phosphorylation in the absence and presence of a lipoylated L2 construct (residues 126–233, designated as L2), with an intrinsic C-terminal linker (residues 209–233) identical to that used by the Roche group. A maltose-binding protein (MBP) was N-terminally linked to the kinase to produce a soluble MBP-fused PDK3 (MBP–PDK3) fusion protein. The presence of the MBP tag has no effect on the enzymatic properties of the related rat BCK (Davie et al, 1995; Wynn et al, 2000). As shown in Figure 1A and B, at saturating L2 concentrations (20–30 μM), PDK3 activity is stimulated by 11.9-fold, whereas at the 40 nM concentration of E2p used previously (Tuganova et al, 2002), PDK3 activity is maximally stimulated by 16.4-fold. Thus, the fold increases of PDK3 activity stimulated by L2 and E2p in the present study are in good agreement with those reported previously by the Roche group. The significantly lower E2p concentration required for maximally stimulated PDK3 activity presumably results, in part, from the multiple copies of L2 domain present in the 60-meric E2p core.

In a separate experiment, PDK3 activity was measured in the absence and presence of an L2 construct (designated as L2_s) containing the shorter C-terminal linker (residues 209–214) studied by the Popov group or the above L2 construct used in crystallography. With these two L2 constructs, we obtained three- and 13.3-fold stimulation, respectively, of PDK3 activity (Figure 1C). Our results, therefore, establish that the significantly smaller L2-stimulated PDK3 activity reported by the Popov group is due to the shorter L2 C-terminal linker they used. It follows that the length of the C-terminal linker is critical for the augmentation of PDK3 activity by the L2 domain. In BCK, we have shown that a long C-terminal linker (residues 85–99) of its cognate lipoyl domain is also essential for efficient binding to this kinase (Chuang et al, 2002). However, other than stabilization effects on L2 (Gong et al, 2000), the role of the C-terminal linker in PDK3 activation is not clear, since this region does not appear to interact with PDK3 in the present PDK3–L2 structure (see below).

Overall structure of the PDK3–L2 complex

We determined three crystal structures of the PDK3–L2 complex (L2 residues 126–233), either without (apo) or with bound nucleotides (ATP or ADP) (Table I). These structures are virtually identical (RMSD: less than 0.55 Å for over 370 C_α atoms), except for the ATP lid (residues 305–327). The refined PDK3 structure is similar to the published PDK2–ADP structure in the absence of L2 (RMSD: 1.07 Å for 321 C_α atoms); however, there are small, yet significant, conformational differences apparently caused by L2 binding. The structure of the cognate BCK is also similar to PDK3 (RMSD: 1.4 Å for 277 C_α atoms), despite low sequence identity (23%) between these two kinases.

In the present PDK3 structure, the N-terminal domain comprises residues 12–178 and 366–381, and the C-terminal domain encompasses residues 182–365 (Figure 2A). Structures of these domains are essentially identical to the corresponding domains in PDK2, except that certain disordered regions in PDK2 are ordered in PDK3–L2. One of the ordered regions is the loop between helices α2 and α3 of the N-terminal domain in PDK3. The lipoyl-lysine residue of L2 binds to this region, which is part of the lipoyl-binding pocket, to stabilize the loop conformation. The C-terminal region, which is not visible in PDK2, is ordered in PDK3–L2 (Figure 2A). This region joins the N-terminal domain with the formation of helix α13, and projects out from that domain in an extended conformation. The extended segment (residue 382–401), which is designated as the C-terminal tail, is critical to L2 binding (see below).

The L2 domain is a member of the biotinyl/lipoyl-carrier protein superfamily (http://scop.mrc-lmb.cam.ac.uk/scop/data/scop.b.c.baf.b.b.html). In the complex, L2 adopts a β-barrel structure comprising two antiparallel β-sheets (Figure 2A). This fold is similar, but not identical, to the NMR structure previously reported (RMSD: 2.1 Å for 65 C_α atoms) (Supplementary Figure S1B) (Howard et al, 1998). The crystal structure of L2 shows a helix (A1) in the C-terminal region, whereas in the NMR structure this region is in an extended conformation. The lipoyl-lysine residue at position 173 is located in a short loop connecting strands β4 and β5. The long side chain of the lipoyl-lysine protrudes into its binding pocket in the N-terminal domain of PDK3 (Figure 2).

Each asymmetric unit contains one PDK3 monomer bound to one L2 (Figure 2A), with two symmetry-related monomers forming a functional dimer (Figure 2B and C) (Bowker-Kinley et al, 1998; Baker et al, 2000; Korotchkina and Patel, 2001). This result is consistent with the presence of PDK3 dimers in solution, as indicated by analytical ultracentrifugation (Hiromasa and Roche, 2003) and gel filtration (JL Chuang, unpublished data, 2004). The dimer formation of PDK3 is achieved by extensive interactions in a head-to-tail fashion between two β-sheets each from the C-terminal domains of the two subunits. This dimerization mode is similar to that described for PDK2 (Steussy et al, 2001) and BCK dimers (Machius et al, 2001). The PDK3 dimer binds two L2 with no interactions observed between these L2 domains (Figure 2C, left). Remarkably, the extended C-terminal tail from PDK3 subunit II crosses over to subunit I and interacts with L2 bound to subunit I, becoming an integral part of the lipoyl-binding pocket in the latter subunit.

The present structure is consistent with earlier biochemical studies that indicated the subunit stoichiometry of PDK3:L2=1:1 (Tuganova et al, 2002; Hiromasa and Roche, 2003). Isothermal titration calorimetry (ITC) was performed to dissect interactions between MBP–PDK3 and lipoylated L2 (Figure 3). The binding isotherm gives rise to a dissociation constant (K_d) of 1.17±0.23 μM for L2 binding to PDK3. This constant is 10-fold lower than those for L2 binding to PDK1 and PDK2 previously determined by gel filtration (Tuganova et al, 2002). The close to unity (0.74) subunit stoichiometry (PDK3 monomer: L2 monomer) and the iterative-best-fit monophasic binding isotherm indicate that both L2-binding sites within the PDK3 dimer are equivalent.

The C-terminal domain of PDK3 binds ATP or ADP through conserved residues in the common motifs of the GHKL superfamily (Dutta and Inouye, 2000). The ATP-binding site is located on the sidewall of the C-terminal domain facing the cleft between the N- and C-terminal domains (Figure 2A). The γ-phosphate of bound ATP is accessible to the E1p substrate from within the cleft, which is likely to be the active site for the phosphoryl-transfer reaction. Based on previous studies with PDK2 (Tuganova et al, 2001), the invariant D247 in PDK3 located in the active-site cleft presumably serves as the general base for the activation of phosphorylatable serine residues in E1p. The bound L2 in the N-terminal domain of subunit I is located more than 30 Å away from the active-site cleft of this subunit, and is at more than 23 Å distance from the opposing active-site cleft in subunit II (Figure 2B and C). Thus, there appears no direct interaction between the bound L2 and either of the active-site clefts, suggesting that L2 exerts its effects on PDK3 through an allosteric mechanism.

Extensive interactions exist between PDK3 and L2

Figure 4A shows the 619 Ų surface area in the N-terminal domain of PDK3 subunit I that constitutes the binding interface with L2. Superimposition of electrostatic potential surfaces on PDK3 subunit I and L2 (Figure 4C) with interacting surfaces on the corresponding subunit/domain (Figure 4A) reveals two types of interactions: (1) hydrophobic interactions in the lipoyl-binding pocket and its surrounding regions, and (2) electrostatic interactions in the lower part of the N-terminal domain including helix α13. In the PDK2 structure without L2, the C-terminal region comprising α13 is largely disordered (Steussy et al, 2001); therefore, the formation of helix α13 in the complex structure is apparently induced by L2 binding.

The hydrophobic interactions occur mainly in the lipoyl-binding pocket in PDK3. The pocket is formed by the loop between helices α1 and α2 as well as helices α2 and α3, with helix α8 serving as the backwall (Figure 5). These structural components confer a 9-Å deep cylindrical pocket. The long aliphatic side chain of the lipoyl-lysine residue in L2 bends approximately 60° at the C4 atom, allowing the lipoamide group to protrude into the lipoyl-binding pocket. Hydrophobic residues L27, F32, F35, F48, L49, L164, L171, and F172, which are highly conserved among human PDKs (Supplementary Figure S1), constitute the inner lining of the pocket and appear to be essential for hydrophobic interactions with the aliphatic side chain of lipoamide. In particular, the reduced dithiolane moiety of lipoamide (Figure 5, inset) is sandwiched between the invariant F35 and F48. Crystallization of the PDK3–L2 complex was carried out in the presence of 10 mM DTT; at this DTT concentration the oxidized form of lipoamide is readily reduced (Bao et al, 2004a; Hiromasa et al, 2004); this apparently facilitated the crystallization of the PDK3–L2 complex. The reduced form of lipoamide in L2 has higher affinity for PDKs than its oxidized counterpart, as a feedback inhibitory mechanism in the PDC (Baker et al, 2000; Roche et al, 2003; Bao et al, 2004b). Hydrophobic residues in the lipoyl-binding pocket are also conserved in the related BCK (Supplementary Figure S1). The conservation of these residues strongly suggests that the recognition of the lipoyl-lysine residue by PDKs and BCK is similar. Surrounding the lipoyl-binding pocket in PDK3, highly conserved residues F22, P26, and V55 as well as methylene groups of the K51 side chain contribute to interactions with highly conserved hydrophobic residues in L2 (L140, P142, A174, and I176) (Supplementary Figures S1 and S2). The structural data explain the preliminary findings that substitutions of L140 and A174 of L2 both remove 80% of L2-stimulated PDK3 activity (Roche et al, 2003).

The second interface between the N-terminal domain of PDK3 and L2 consists of electrostatic interactions. Basic residues R21 (α1), K374, and R378 (α13), which are highly conserved in human PDKs (Supplementary Figure S1), are located underneath the lipoyl-binding pocket (Figure 4C). This basic region interacts with an acidic cluster on L2 formed by conserved residues E162, D164, E179, E182, and E183 (Supplementary Figures S1 and S2). E162 and E179 have also previously been implicated in interactions of the L2 domain with PDK3 in site-directed mutagenesis studies (Roche et al, 2003).

As described above, it will be of particular significance for diabetes therapy to design drugs that specifically inhibit elevated PDK activities. To date, two types of specific PDK inhibitors have been developed (Aicher et al, 2000; Morrell et al, 2003). One of these compounds (AZD7545) shows ATP-independent inhibition of PDK2 activity, and it is speculated that this inhibitor acts as a competitive inhibitor to prevent L2 from binding to PDK2. Thus, our structural data depicting the interactions between PDK3 and L2 may constitute a framework for designing a new generation of specific PDK inhibitors to eradicate impaired glucose utilization in type II diabetes.

Interactions of the PDK3 C-terminal tail with the L2 domain

As described above, a significant area (603 Ų) of the extended C-terminal tail from subunit II provides additional interface with L2 bound to subunit I (Figure 4B). The additional surface accounts for 49% of the total interface of L2 for PDK3. To facilitate the illustration of these interactions, the C-terminal tail is divided into three consecutive regions in the N- to C-terminus direction (Figure 6).

In the PDK3 dimer, the region of residues 386–391 in the C-terminal tail from subunit I forms an extensive hydrogen-bond network with conserved residues in subunit II (Figure 6A). Specifically, S390 and N391 from subunit I form hydrogen bonds with L369 of subunit II and D387 of subunit I, respectively. In other PDKs, S390 and N391 of PDK3 are substituted by cysteine and valine/isoleucine, respectively (Figure 6D); the side chain of cysteine is a poor hydrogen donor and that of valine or isoleucine is unable to form hydrogen bonds.

Residues 392–397 of subunit I interact with the L2 domain bound to subunit II (Figure 6B). This region largely recognizes residues from L2 flanking the lipoyl-lysine residue. R196, which is located on the back of L2 opposite to the PDK3–L2 binding interface, forms a hydrogen bond with the conserved E395 from PDK3. Another notable residue is R397 in subunit I, whose side chain makes two hydrogen bonds to D172 of L2. In other human PDKs, R397 in PDK3 is replaced by a lysine residue whose side chain is capable of forming a single hydrogen bond (Supplementary Figure S1). The present results provide the structural basis for the earlier findings that substitutions of R196 and D172 of L2 significantly diminish L2-stimulated PDK3 activity (Roche et al, 2003).

The end of the C-terminal tail from residues 397 to 401 is integrated into the lipoyl-binding pocket (Figure 6C). Side chains of R397 and A399 in the C-terminal tail make van der Waals contacts with the aliphatic lipoamide backbone. This region is further stabilized by three hydrogen bonds between: (1) R397 in subunit I and Q31 in subunit II, (2) S400 in subunit I and D38 in subunit II, and (3) an internal hydrogen bond between D398 and K401 of the C-terminal tail in subunit I. The hydrogen bond between S400 and D38 appears to occur only in PDK3 and is not present in other PDKs (Supplementary Figure S1). A C-terminal truncation (Δ392–406) of PDK3 abolishes L2 binding to PDK3 as determined by ITC (Figure 3). The data support the notion that the interactions between the C-terminal tail and L2 are essential for L2 binding to PDK3. The C-terminal regions of PDKs from Ascaris suum and Caenorhabditis elegans were previously shown to be required for the binding of these kinases to the respective nematode E2p (Chen et al, 1999).

Previous biochemical studies have shown that PDK isoforms exhibit different affinities for L2, with PDK3 possessing the highest affinity (Tuganova et al, 2002). As shown above, the PDK3 dimer binds L2 through two regions, namely the N-terminal domain and the C-terminal tail. The interface between the N-terminal domain and L2 consists of hydrophobic and electrostatic interactions (Figure 4) that occur similarly in all four PDKs based on sequence conservation (Supplementary Figure S1). These results suggest that the hydrogen-bond formation in the C-terminal tail of PDKs contains specific determinants that confer differential affinity for L2. In the PDK3–L2 complex, but not in other PDKs, the C-terminal tail is further stabilized by additional interactions with residues from the other subunit (Figure 6A and C). Moreover, the two hydrogen bonds described above occur between R397 in PDK3 and D172 in L2 (Figure 6B), as opposed to only one potential hydrogen bond in other PDKs. These structural determinants in combination likely impart the higher affinity of PDK3 for L2 than other PDKs.

L2 binding-induced conformational changes in PDK3

To decipher L2-induced conformational changes in PDK3, it is necessary to determine the structure of PDK3 free of L2. However, PDK3 alone is sparingly soluble, which prevented the crystallization of free PDK3. Since amino-acid sequences are highly conserved between PDK3 and PDK2 (65% identity), the available PDK2 structure free of L2 (Steussy et al, 2001) was compared with the PDK3–L2 complex. In the two β-sheets of the C-terminal domains, many residues that participate in the dimer formation are conserved in both the PDK2 and PDK3 structures. Therefore, the structure at the dimer interface is expected to be similar between these two PDKs.

As described above, PDK3, but not PDK2, is activated by the isolated L2 domain (Baker et al, 2000; Tuganova et al, 2002). However, recent studies by the Popov group showed that the dissociation constant (K_d) of PDK2 for Mg-ATP is increased by 10-fold in the presence of L2 (Tuganova and Popov, 2005). These authors suggest that L2 binding promotes a conformational change in PDK2, which accounts for the reduced affinity of PDK2 for ATP. The L2-induced conformational change in PDK2 was also suggested previously by the Roche group, based on the modest activation of PDK2 activity by a dimeric L2 construct and an E2p construct lacking the E1p-binding domain (Baker et al, 2000; Bao et al, 2004b). These findings provide additional rationale for using the PDK2 structure to deduce possible L2-mediated conformational changes in PDK3.

When the right subunit of PDK3–L2-ADP is superimposed onto that of the PDK2–ADP dimer, the left subunit of PDK2 swings outward by 15° relative to the corresponding subunit of PDK3 (Figure 7A). If one assumes that the PDK3–ADP structure free of L2 is similar to the PDK2–ADP structure, the crossover of the two C-terminal tails is likely to restrict subunit movements in the PDK3–L2 complex, resulting in a clamping of the space between two subunits in the structure. This clamping has long-range effects on the conformation of the active-site cleft in each monomer. This becomes apparent when the monomeric structure of PDK2–ADP is superimposed onto that of PDK3–L2–ADP. As shown in Figure 7B, helix α7 of the N-terminal domain and helix α9 of the C-terminal domain in PDK3 move outward in opposite directions by ∼2 Å each, relative to the corresponding helices in PDK2. This domain shift results in the widening of the active-site cleft in PDK3–L2–ADP compared to PDK2–ADP. In addition, the ATP lid of PDK2–ADP is partially ordered, whereas the equivalent in PDK3–L2–ADP is largely invisible (Figure 7B) (for stereo-view, see Supplementary Figure S3A). The order-to-disorder transition of the ATP lid is caused by loss of interactions between the lid and the phosphate group of the bound ADP. The disordered conformation is not due to the loss of crystal lattice interactions, because this region is not involved in crystal packing. Significant movements occur in Y326 and M304 in the ATP lid, which result in the repositioning of their side chains in the space vacated by reciprocal shifts of conserved residues V130 (α7) and I147 (α8) in the N-terminal domain of PDK3. These concerted side-chain movements are likely to contribute to the disordering of the ATP-lid conformation.

In the PDK3–L2–ATP structure, the C-terminal ordered region (residues 325–327) of the ATP lid is stabilized by forming hydrogen bonds with the γ-phosphate of bound ATP (Figure 7C) (see also Supplementary Figure S3B and Table SI). G325 in this region in turn forms a hydrogen bond with M304 in the N-terminal segment of the same ATP lid, further stabilizing the lid conformation. However, the L2-induced widening of the active-site cleft in PDK3–L2–ADP occurs similarly in PDK3–L2–ATP. The subsequent weakened interactions between the ATP lid and the N-terminal domain sidewall could conceivably render the lid in PDK3–L2–ATP more flexible than that in PDK2–ADP.

L2 binding markedly decreases affinities of PDK3 for both ADP and ATP

We have previously shown that the ATP lid is significantly ordered in the ADP-bound form of rat BCK free of its cognate lipoyl domain (Machius et al, 2001). The ordering of the ATP lid, which effectively traps ADP, accounts for product inhibition during the phosphoryl-transfer reaction catalyzed by BCK. We proposed that E2 or the lipoyl domain bound to BCK functions as a nucleotide exchange factor to promote the substitution of inhibitor ADP for substrate ATP in the nucleotide-binding site. The complete disordering of the ATP lid in PDK3–L2–ADP (Figure 7B) may be related to a reduced affinity of PDK3 for ADP induced by L2. To support this notion, ITC was employed to measure dissociation constants (K_d's) for ADP binding to MBP–PDK3 in the absence or presence of the L2 construct (Table II). PDK3 without L2 exhibits a K_d of 0.16±0.04 μM for ADP, whereas when a two-fold molar excess of lipoylated L2 is present, PDK3 shows a K_d of 1.30±0.14 μM for the same nucleotide. The data indicate a significantly (>8-fold) decreased affinity of PDK3–L2 for ADP compared to free PDK3. The good fits of monophasic binding isotherms in the presence of lipoylated L2 (data not shown) indicate that the two nucleotide-binding sites in the PDK3 dimer are equivalent and not interacting with each other, in accordance with that recently reported for PDK2 (Tuganova and Popov, 2005). Thus, the present study validates and extends the above hypothesis, in that L2 binding promotes the release of ADP from PDK3 to remove product inhibition by this nucleotide. This model provides the structural basis for recent findings that show the rate of ADP dissociation is a rate-limiting step in E2p-stimulated PDK2 activity (Bao et al, 2004a, 2004b).

As described above, the ATP lid in PDK3–L2–ATP is apparently destabilized by weakened interactions with the N-terminal domain even though the lid is still partially visible in the crystal structure (Figure 7C). The flexible ATP lid accounts for the more than five-fold decrease in affinity for ATP, when one compares the dissociation constants for ATP between free PDK3 (0.16±0.005 μM) and PDK3–L2 (0.84±0.10 μM) (Table II). Similar dissociation constants were obtained for ATPγS, with K_d=0.16 μM for PDK3 and K_d=0.96 μM for PDK3–L2 (Table II). The significantly lower affinity of PDK3 for ADP than ATP is likely to facilitate ATP/ADP exchange to remove the product inhibition by ADP. The flexible ATP lid may also explain the recent biochemical findings that K_d for ATP binding to PDK2 is elevated by 10-fold in the presence of L2 (Tuganova and Popov, 2005). Similarly, K_m for ATP of PDK2 kinase activity is increased by eight-fold by E2p (Bao et al, 2004a).

Allosteric R↔T equilibrium as the mechanism for L2-stimulated PDK3 activity

Based on the present structural and biochemical data, we hypothesize that the PDK3 dimer in the absence of L2 adopts a ‘closed' conformation similar to that in the published PDK2–ADP structure (Figure 7D, top). In this conformation, the active-site cleft is closed because the C-terminal tail of each subunit is disordered and not interacting with the lipoyl-binding pocket in the other subunit. In the closed active-site cleft, ADP binds to PDK3 with the same affinity as ATP (Table II) because the ATP lid is stabilized by interactions with the sidewall of the N-terminal domain. The high-affinity binding of ADP prevents an efficient ADP/ATP exchange, resulting in product inhibition by ADP with low kinase activity. Upon the binding of L2 to PDK3, the C-terminal tails are ordered in the crossover configuration, which promotes the widening of the active-site cleft as well as the dimer interface, which in turn leads to an ‘open' conformation in the PDK3 dimer (Figure 7D, bottom). In this open conformation, the interactions of the ATP lid with the sidewall from the N-terminal domain are significantly weakened, which renders the ATP lid largely disordered. The subsequent lower-affinity binding of ADP than ATP in the PDK3–L2 complex accelerates the release of trapped ADP to mitigate product inhibition by this nucleotide. The open conformation of PDK3 may also facilitate the access of the E1p substrate to the active-site cleft. Indeed, recent biochemical studies suggest that the K_m for E1p of PDK2 decreases significantly when the kinase is bound to an E2p construct devoid of the E1p-binding domain (Bao et al, 2004a). This situation is analogous to cyclin-dependent kinase 2 (CDK2), where a pivotal T loop is utilized to regulate its kinase activity (Jeffrey et al, 1995). In the free form of CDK2, the T loop covers the active site to block the substrate access, whereas the binding of cyclin A to CKD2 induces a conformational change in the T loop, resulting in the opening of the active site. Thus, our data suggest that L2-induced stimulation of PDK3 activity may employ the classic allosteric mechanism of R (relaxed)↔T (tight) equilibrium (Monod et al, 1965) between the L2-bound and free PDK3, with the L2-bound form favoring the R or open conformation.

Structural data deposition

The model coordinates and structural factor amplitudes have been deposited in the Protein Data Bank for structures of the apo- (1Y8N), ADP-bound (1Y8O) and ATP-bound PDK3–L2 complex (1Y8P).