Crystal Structure of a Human IκB Kinase β Asymmetric Dimer

Background: IκB kinase β is a key regulator in the NκB signaling pathway.

Results: Crystal structure of a human IKKβ asymmetric dimer shows one kinase active site phosphorylated and in the active conformation and the other unphosphorylated and inactive.

Conclusion: Depending on the phosphorylation state, IKKβ can adopt distinct dimeric geometry.

Significance: High resolution structure of hIKKβ provides structural basis for its activation and potential use of inhibitor design.

Abstract

Phosphorylation of inhibitor of nuclear transcription factor κB (IκB) by IκB kinase (IKK) triggers the degradation of IκB and migration of cytoplasmic κB to the nucleus where it promotes the transcription of its target genes. Activation of IKK is achieved by phosphorylation of its main subunit, IKKβ, at the activation loop sites. Here, we report the 2.8 Å resolution crystal structure of human IKKβ (hIKKβ), which is partially phosphorylated and bound to the staurosporine analog K252a. The hIKKβ protomer adopts a trimodular structure that closely resembles that from Xenopus laevis (xIKKβ): an N-terminal kinase domain (KD), a central ubiquitin-like domain (ULD), and a C-terminal scaffold/dimerization domain (SDD). Although hIKKβ and xIKKβ utilize a similar dimerization mode, their overall geometries are distinct. In contrast to the structure resembling closed shears reported previously for xIKKβ, hIKKβ exists as an open asymmetric dimer in which the two KDs are further apart, with one in an active and the other in an inactive conformation. Dimer interactions are limited to the C-terminal six-helix bundle that acts as a hinge between the two subunits. The observed domain movements in the structures of IKKβ may represent trans-phosphorylation steps that accompany IKKβ activation.

Introduction

The immune, inflammatory, and acute phase responses of vertebrates depend on exquisitely controlled and marshaled patterns of gene expression. Among the central regulators of these coordinated events are members of the NF-κB family of nuclear transcription factors (1, 2). Dysregulation of their function contributes to numerous diseases, including atherosclerosis, asthma, arthritis, cancer, cachexia, diabetes, euthyroid sick syndrome, acquired immune deficiency syndrome, inflammatory bowel disease, and stroke (3, 4).

In their resting or inactive state, NF-κB factors exist as cytoplasmic complexes with members of the NF-κB inhibitor protein family, IκBα, IκBβ, and IκBϵ. These interactions prevent translocation of NF-κB proteins to the nucleus and restrain them from binding to cognate sites on DNA and stimulating the transcription of target genes.

The multisubunit protein kinase IKK⁵ regulates NF-κB activation (5). Multiple stimuli such as inflammatory cytokines, bacterial, or viral products, or various types of stress lead to IKK-catalyzed phosphorylation of IκB inhibitor proteins, an event that activates the canonical NF-κB pathway. Phosphorylation permits IκB proteins to be polyubiquitinated and then catabolized by the proteasome. Liberation from their inhibitors leaves NF-κB factors free to enter the nucleus and activate transcription of genes encoding proteins that participate in the immune and inflammatory response, cell adhesion, growth control, and protection against apoptosis. A subset of inducers can also stimulate the non-canonical NF-κB pathway, in which IKK-mediated phosphorylation of the IκB-like domain in the NF-κB2 protein leads to activation of that transcription factor (6).

IKK is a multiprotein complex that contains two possible kinase subunits, IKKα and IKKβ, and the regulatory subunit, NEMO (NF-κB essential modulator) (7, 8). NEMO is essential for the activation and substrate specificity of IKK (9, 10). Both IKKα and IKKβ contain an N-terminal kinase domain (KD, residues 15–308 in hIKKβ), a leucine zipper region (residues 458–479), and a helix-loop-helix region (residues 603–642) (11). Each kinase has a NEMO binding domain at its carboxyl terminus (residues 737–742). IKKβ also has an additional ULD domain after the KD, which is absent in IKKα. The novel ULD domain is required for the functional activity of IKKβ and important for its substrate specificity (12, 13).

Two IKK-related kinases, IKKϵ (or IKK-I) and TBK1 (TANK-binding kinase) (14) also contribute to immune responses but mediate different signal pathways (15). Of the principal kinases, IKKα and IKKβ seem to have very distinct functions: IKKβ is a more potent NF-κB activator and plays a major role in the canonical NF-κB pathway responsible for immune responses, whereas IKKα is more important in the non-canonical pathway required for developmental processes (16, 17).

Because of its importance in many human diseases, hIKKβ has been viewed widely as a potential therapeutic target (4, 18–20). The first crystal structures of an IKKβ, reported for the phosphomimetic mutant of Xenopus laevis IKKβ, xIKKβ(S177E/S181E), have greatly advanced our understanding of IKK (13). Similar structural elucidation of IKKβ in an active phosphorylated state has been hampered by the inherent kinase heterogeneity due to phosphorylation. We have produced and isolated a near full-length human IKKβ wild type protein, which was phosphorylated at the activation loop and retained kinase activity. We then co-crystallized this protein with the staurosporine analog K252a. Here, we report the resulting 2.8 Å resolution crystal structure of a phosphorylated hIKKβ dimer and delineate its mechanistic similarities and distinctions from that of the unphosphorylated xIKKβ dimer. In addition, compared with the reported structures of xIKKβ determined at 3.6 and 4.0 Å resolution, respectively (13), the present hIKKβ structure resolves many new additional details at a sufficiently high resolution to permit structure based drug design. We complement our data with biochemical analysis of the dimer formation and mass spectrometric analysis of the phosphorylation state.

EXPERIMENTAL PROCEDURES

Protein Expression and Purification

A construct encoding a hexahistidine tag followed by residues 1–664 of hIKKβ was cloned into a pBacPAK vector (Clontech) to enable its expression in baculovirus-infected insect cells. As a result of cloning, the hIKKβ(1–664) sequence was extended with the vector-derived residues SPGRPLN at its C terminus. The vector was then transfected into Sf9 insect cells according to the manufacturer's instructions (Clontech), and a clonal isolate was amplified in suspension culture using shake flasks (Invitrogen). A multiplicity of infection of 0.5 pfu/cell was used to infect Sf9 cells at a density of 5 × 10⁵ cells/ml. The amplified virus was then used to infect Sf21 cells at a density of 2.0 × 10⁶ cells/ml with a multiplicity of infection of ∼2.0 pfu/cell. The cells were harvested at 72 h post infection, pelleted, resuspended in phosphate-buffered saline, pH 7.2, and then flash frozen in liquid nitrogen.

Sf21 cell pellets expressing hIKKβ(1–664) were resuspended in cell lysis buffer (50 mm HEPES, pH 7.5, 100 mm NaCl, 5% glycerol, 5 mm imidazole, 2.5 mm β-glycerophosphate, 5 mm β-mercaptoethanol, and protease inhibitor mixture from Roche Applied Science). The cells were lysed using a Parr Bomb, cell debris was removed by centrifugation, and the resulting supernatant was loaded onto a nickel-nitrilotriacetic acid column. The column was washed extensively with the lysis buffer, after which hIKKβ(1–664) was step-eluted with lysis buffer containing 200 mm imidazole. The nickel column eluate was diluted 4-fold with buffer A (25 mm HEPES, pH 7.5, 10 mm NaCl, 5% glycerol, 2 mm EDTA, 5 mm β-glycerophosphate, 5 mm DTT) and loaded onto a heparin affinity column. The column was developed with a NaCl gradient to 1.0 m, and hIKKβ-containing fractions were combined and diluted 3-fold with buffer A before being loaded onto a MonoQ HR10/10 column (Pharmacia). The MonoQ column was washed extensively with buffer A, and the column was developed with a NaCl gradient to 1.0 m. Fractions containing pure hIKKβ were identified by a SDS-PAGE 4–12% gel and concentrated before loading onto a TSK G3000 gel filtration column equilibrated with the SEC running buffer (25 mm HEPES, 250 mm NaCl, 5% glycerol, 2 mm tris(2-carboxyethyl)phosphine). The hIKKβ peak was collected and used for subsequent studies.

Crystallization and Data Collection

The His-tagged hIKKβ (1–664) protein was crystallized using the hanging drop method. Crystallization drops were formed by mixing protein at 6 mg/ml in the presence of 2-fold molar excess of the staurosporine analog K252a with an equal volume of a precipitant solution of 22% PEG 3350 and 120 mm trilithium citrate. Crystals were soaked briefly in a cryoprotectant solution of the precipitant solution containing 20% PEG 400 before being flash cooled in liquid nitrogen.

Crystals of hIKKβ were of the P2₁ space group. The best crystallographic data set was collected on a single crystal which diffracted to 2.8 Å at the Advanced Light Source (Berkeley, California) and was processed with program HKL2000 (21). Statistics for data collection are listed in Table 1. Further processing was carried out using program suite CCP4 (22).

TABLE 1.

Data collection and refinement statistics

Data collection
Space group	P21
Cell dimensions
a, b, c	110.954, 68.689, and 107.401 Å
α, β, γ	90, 107.401, and 90°
Resolution (Å)a	47.2–2.8 (2.9–2.8)
Rmerge	0.103 (0.574)
I/σ	10.0 (1.5)
Completeness (%)	95.3 (80.8)
Redundancy	2.9 (2.0)
Refinement
Resolution (Å)	47.2–2.8
No. reflections	34,922
Rwork	0.184 (0.27)
Rfree	0.239 (0.35)
No. atoms and B
Protein	10,282, 53.0 Å2
Ligands	112, 34.0 Å2
Water	253, 36.5 Å2
r.m.s.d.b
Bond lengths (Å)	0.010
Bond angles	1.14°
Ramachandran (%) (favored, outliers)	96.9, 0.2

^a Statistics in the highest resolution shell are shown in parentheses.

^b r.m.s.d., root mean square deviation.

Structure Determination and Refinement

The hIKKβ structure was solved by molecular replacement using the published xIKKβ structure (Protein Data Bank code 3QA8) as model with the program Phaser (23). Extensive model rebuilding was carried out using Coot (24), and refinements were carried out using program Buster with TLS options (25). The refinement statistics are listed in Table 1.

Dynamic Light Scattering

Dynamic light scattering experiments were conducted using a DynaPro DLS instrument for each sample to determine the hydrodynamic radius of the protein. Protein samples were centrifuged for 10 min at 14,000 rpm in a desktop microfuge at 4 °C prior to the reading at 18 °C. Measurements were based on the average of triplicate runs, with each run involving a minimum of 20 readings.

RESULTS

Analysis of Protein Constructs and Crystallization

Two protein constructs of hIKKβ were prepared and analyzed for crystallization trials: full-length N-FLAG-IKKβ (data not shown) and N-His₆-hIKKβ (1–664). Full-length hIKKβ was heterogeneously phosphorylated at 6–19 sites owing to inclusion of the C-terminal NEMO binding domain region, which is prone to heavy phosphorylation (data not shown). With the same region absent, N-His₆ hIKKβ(1–664) was phosphorylated at only one to three sites (Fig. 1A). These two recombinant forms of the kinase were shown to have similar specific activities (data not shown). In addition, several commercially available kinase inhibitors exhibited similar IC₅₀ values against the full-length hIKKβ and hIKKβ(1–664). Among these inhibitors, the staurosporine analog K252a showed the best potency. Furthermore, analytical size-exclusion chromatography (Fig. 1B) and dynamic light scattering (data not shown) indicated that both forms of the enzyme existed as dimers in solution.

FIGURE 1.

Characterizations of N-terminal His-tagged hIKKβ(1–664) used in crystallization. A, mass spectrum of protein indicated that the protein has one to three sites phosphorylated (top). These phosphate groups could be removed by dephosphorylation with alkaline phosphatase treatment (bottom). The theoretical molecular weight of N-terminal acetylated hIKKβ(1–664) is 78,112 Da (mass discrepancy of 128 ppm). B, analytical size-exclusion chromatography indicated that hIKKβ was a dimer in solution. Elution times of molecular weight standards are marked arrows.

Crystallization screens produced crystal hits for both protein forms, the full-length hIKKβ (1–742) and the truncated form N-His₆-hIKKβ(1–664). However, only the truncated form of the protein, consistent with its protein heterogeneity profile, yielded crystals that were suitable for structure determination and that diffracted to 2.8 Å resolution. We refer to this construct as hIKKβ throughout the text, except when otherwise noted.

Overall Structure of hIKKβ

There are two hIKKβ molecules in the crystal asymmetric unit, and consistent with the biochemical studies, they form a dimer around a quasi-2-fold rotational axis (Fig. 2A). Similar to its xIKKβ homolog, each hIKKβ monomer has a trimodular linear architecture: the N-terminal kinase domain, KD(1–309), the central ubiquitin-like domain, ULD(310–404), and the C-terminal dimerization domain, SDD(408–664). The KD has a typical bilobal kinase fold, the ULD has the ubiquitin fold (20), and the SDD has a helical blade structure consisting of six α-helices, α1s–α6s (Fig. 2B).

FIGURE 2.

Overall structure of hIKKβ. A, ribbon diagram of the hIKKβ dimer. The N-terminal lobe (N; residues 1–109) and C-terminal lobe of KD (C; residues 110–307) and ULD (residues 308–404) are colored in coral, yellow, and magenta, respectively. The SDD(410–664) of hIKKβ protomer B is colored in cyan and that of protomer A in light blue. In general, protomer A is more flexible than protomer B and is missing more residues. Bound inhibitor is shown in stick models, with carbon, nitrogen, and oxygen shown in magenta, blue, and red, respectively. The N and C termini, KD N-terminal lobe and C-terminal lobe, ULD, and SDD are labeled. The pseudo 2-fold axis of the hIKKβ dimer is indicated by a dashed line. The same color scheme is used throughout this work, unless noted. B, ribbon diagram of the hIKKβ protomer B, with all secondary structures labeled. C, different domains in protomers A and B in hIKKβ dimer adopt slightly different relative orientations. Compared with protomer B, the C-terminal part of the SDD of protomer A (gray) rotates 6° away from the KD. Figures were prepared using program CCP4MG (42).

The overall interdomain architecture of xIKKβ is largely maintained in hIKKβ, with the first two domains, KD and ULD, tightly aligned along the N-terminal end of the third SDD domain (Fig. 3). This arrangement is stabilized by van der Waals and hydrogen bond contacts between all three domains, with a major contribution from the KD-ULD interactions (Fig. 3). Most of the residues involved in interdomain interactions are conserved among different IKKβ species (supplemental Fig. S1), highlighting the biological significance of this arrangement.

FIGURE 3.

Interfaces between KD, ULD, and SDD of hIKKβ. Side chains of contacting residues within 4 Å are labeled and are shown in sticks with carbon atoms colored according to the domains in which they are located. A, the ULD-SDD interface. Among residues at the interface, Lys-619 and Glu-352 are highly conserved in IKKβ sequences across different species. B, the KD-SDD interface. Lys-428 and Phe-219 are semi-conserved. Arg-419 is either an Arg or a Lys in IKKβ sequences but is neither in IKKα. Phe-111 of the KD is semiconserved in IKKβ sequences. In xIKKβ, the β11 and β12 main chains are traced differently from that of hIKKβ. C, the KD-ULD interface. The hydrogen bond forming residues Asn-377 and Glu-378 are not conserved in IKKβ sequences. N, N terminus; C, C terminus.

In contrast to the xIKKβ crystal structure, in which eight independent representations of the IKKβ molecule are similar (13), the two hIKKβ protomers are less alike. In particular, protomer A exhibits a higher level of structural disorder compared with protomer B (Fig. 2C) and differs in a way that appears functionally significant, as discussed in detail below.

Pairwise comparison of the two hIKKβ protomers based on superimposition of their KDs gives the root mean square deviation of 0.58 Å for 262 Cα pairs and an root mean square deviation value of 1.22 Å for 541 Cα pairs of the whole protomer structure (Fig. 2C). The observed differences result mainly from variations in orientation of the C-terminal part of the SDD relative to the rest of the molecule, which is reflected in a slightly more bent (∼6°) structure of protomer B (Fig. 2C). Thus, the two hIKKβ protomers on the whole assume slightly different conformations, one more extended (and more similar to xIKKβ) and the other bent.

Structures of the Active and Inactive Kinase Domains in the hIKKβ Dimer

hIKKβ in our crystal form presents two versions of the kinase domain, one with the activation loop phosphorylated (protomer B) and the other not (protomer A). When the two KD structures are compared, the most significant differences occur in the N-terminal lobe and at the activation loop. The ATP-binding pockets in both KDs are occupied by the staurosporine analog inhibitor K252a. It binds in a similar fashion to staurosporine, making hydrogen bond contacts from the lactam ring to the hinge backbone and an additional hydrogen bond from the hydroxyl to the carbonyl oxygen of Glu-149 (Fig. 4A). The presence of this additional hydrogen bond explains the higher affinity of K252a compared with staurosporine (data not shown). The glycine-rich loop in protomer B is well structured with the tip of the loop protruding from the ATP site. This loop in the structure of protomer A is partially disordered (Fig. 4B). The entire N-terminal lobe of protomer A has weaker electron density in general compared with protomer B and some regions of completely missing density.

FIGURE 4.

The two KDs in the hIKKβ dimer are different. A, in both KDs, the staurosporine analog K252a is found to bind at the ATP site, and its interactions with hIKKβ are almost identical in both KDs. B, the KD in protomer B is phosphorylated, whereas in protomer A (gray), it is unphosphorylated. The activation loop (residues 166–194) in hIKKβ molecule B is colored red, and the glycine-rich loop is colored cyan. The active site of molecule A is more flexible than that of molecule B and has more disordered residues, including the glycine-rich loop and residues 170–179 in the activation loop. In molecule B, the phosphate group at Ser-181 interacts with the cationic subsite of the KD. In molecule A, the unphosphorylated activation loop adopts an inactive conformation, which is incompatible with substrate binding. Hydrogen bond and ion pair interactions are shown in dashes. C, the pSer interactions in hIKKβ protomer B stabilize the active kinase conformation, similar to that adopted by the KD of peptide substrate bound protein kinase A (Protein Data Bank code 1ATP) (32). For clarity, only the activation loop of protein kinase A (PKA; gray) and the bound peptide substrate (light blue) are shown. N, N terminus; C, C terminus.

The activation loop in protomer B (residues 166–194) assumes a conformation characteristic of an active kinase (Fig. 4C), with three residues missing (residues 174–176) due to disorder. The corresponding region in protomer A is less well defined. It has 10 residues disordered (residues 170–179) and adopts a conformation that is not compatible with protein substrate binding (Fig. 4B). Because the inhibitor binding mode is identical in both domains, we conclude that the apparent differences in kinase flexibility and conformation are solely due to the differences in phosphorylation state between the two domains. These differences are unlikely to be caused by crystal packing effects, as both protomers, A and B, make wide ranging contacts with crystal symmetry mates and pack equally tightly.

Phosphorylation of the activation loop at two sites, Ser-177 and Ser-181, was reported to be required for the activation of IKKβ (8, 11, 26, 27). In protomer B, both sites are phosphorylated. The electron density for the phosphate group at Ser-181 is strong, and at Ser-177, it is clearly observed but weaker. All three phosphate oxygen atoms of phospho-Ser-181 make direct ionic interactions with the cationic cluster formed by Arg-144 from the catalytic loop and Lys-171 from the activation loop. Additionally, one of the phosphate oxygen hydrogen bonds to the main chain amide nitrogens of phospho-Ser-181 and the preceding Thr-180 (Fig. 4, B and C). As in many active kinase structures, these phosphate-mediated interactions provide a direct link to the catalytic center, helping to stabilize the correct orientation of the catalytic base Asp-145 (Asp-135 in TBK1) and allow for the formation of the polypeptide substrate pocket (Fig. 4C). Ser-177, although phosphorylated, is completely exposed to solvent with no apparent direct role in the stabilization of the observed active conformation. Although phospho-Ser-177 may play a role in long range electrostatic interactions, differences in the dispositions of phospho-Ser-181 and phospho-Ser-177 suggest that phosphorylation of Ser-181 alone may be sufficient for hIKKβ activation. Indeed, the IKKβ S177A mutant was shown to be fully active, whereas the S177A/S181A double mutant was inactive (28). This is akin to the observation that monophosphorylation at Ser-172 in TBK1, a residue equivalent to Ser-181 in hIKKβ (supplemental Fig. S1), is necessary and sufficient for the stabilization of the active conformation of TBK1 (29, 30).

In contrast, in protomer A both Ser-177 and Ser-181 are more mobile, with fewer intramolecular interactions, and appear to be unphosphorylated. The activation loop segment on which Ser-181 resides rearranges into a short α-helical segment (residues 180–188) and relocates to a new position (Fig. 4B). Here, Ser-181 is ordered, but displaced by ∼19 Å from its phosphorylated counterpart position. There is clearly no density for the phosphate group at Ser-181 and residue Ser-177 is completely disordered. When compared with those in the peptide-bound active kinase structures (31, 32), the unphosphorylated activation loop is clearly in the “off” conformation as it sterically occludes substrate access to the catalytic site. Besides these differences, the rest of the unphosphorylated KD in hIKKβ shows features of an active kinase. As such, the αC helix is positioned correctly, placing Glu-61 (Glu-55 in TBK1) for hydrogen bonding with the catalytic lysine 44 (Lys-38 in TBK1), a bond required for catalysis (Fig. 4B).

The hIKKβ and TBK1 crystal structures (29, 30) both demonstrate that the activation loop, when unphosphorylated, adopts conformations that are incompatible with protein substrate binding, highlighting the essential role of phosphate-mediated interactions for the activities of these kinases. The structure of xIKKβ is no exception. The xIKKβ activation loop harboring the S177E/S181E double mutation appears in yet another off conformation (13). In the structures of phosphorylated hIKKβ and TBK1 (29, 30), all three oxygens of the essential phosphate phospho-Ser-181 (phospho-Ser-172 in TBK1) interact with the cationic subsite of the kinase active site. This indicates that glutamates at these positions would serve as only a partial mimic of phosphoserines and could result, as in the cases of the TBK1 S172E and the hIKKβ S177E/S181E double mutant, in much weaker activity than the phosphorylated kinases (33, 34). Altogether, these observations provide the structural basis for the lack of kinase activity of IKKβ (S177A/S181A) (26) and TBK1 (S172A) mutants (34).

In this regard, the message from the xIKKβ S177E/S181E crystal structure is somewhat surprising. On one hand, the reported glutamate substitutions render the enzyme constitutively active (35), albeit with a lower kinase activity compared with the phosphorylated wild type kinase (33). On the other hand, these mutations seem to destabilize the correct folding of the activation loop. The kinase mutant adopts a conformation in which the phosphomimetic loop and the glycine-rich loop touch each other so that the protein substrate is excluded from binding. The high similarity in the conformations of this loop seen in all crystalline copies of xIKKβ (13) suggests that this conformation also exists in solution rather than being the result of crystal lattice contacts only. It is possible that the S177E/S181E mutant could sample the more productive, active conformations only when an ATP molecule is bound.

Interestingly, there is at least one more phosphorylation site, phospho-Ser-257, that appears to be occupied, albeit partially, as judged by electron density and phosphate tetrahedral geometry. This site is occupied in protomer A but not in protomer B. It is positioned toward the C-terminal lobe end and far away from the activation loop and has no apparent impact on the inactive KD conformation of protomer A. Hence, the double phosphorylation of one hIKKβ protomer as opposed to mono-phosphorylation of the other, as visualized in the two independently crystallized kinase copies, is in accord with a one to three site occupancy profile identified by mass spectrometry (Fig. 1A).

Structural Implications for hIKKβ Inhibitor Design

A number of studies have shown that IKKβ has unusually high affinity (K_m, 100–600 nm) for ATP (26, 33, 35), posing significant challenges for ATP binding site inhibitors in view of the high levels of cellular ATP (36, 37). The binding mode of K252a to hIKKβ is virtually identical regardless of the phosphorylation state of the KD, suggesting that such an inhibitor would not differentiate between the active and inactive kinase conformations. However, outside of the ATP binding pocket, the protein conformations are distinct, and this could potentially provide some advantages in designing selective and non-ATP competitive inhibitors. One of the potential binding sites that is remote from the ATP-binding pocket and that is part of the substrate binding site is seen at the KD-SDD interface (Fig. 5). It features a 12-Å-long and 10-Å-deep channel, which is wide open in protomer B (Fig. 5B) but partially closed in protomer A (Fig. 5A). Inhibitors designed to bind to these differently folded pockets would selectively stabilize either active (protomer B) or inactive kinase form (protomer A) and can be tested in substrate competitive inhibition studies. Interestingly, studies have shown that there exist potent IKKβ inhibitors that are competitive with substrate IκB, but allosterically inhibit IKKβ (38).

FIGURE 5.

The potential allosteric ligand binding sites are different between the inactive hIKKβ and the activated hIKKβ. A, the substrate binding site is not formed in the inactive kinase (protomer A). The hIKKβ protomer is shown in surface presentation, with the KD N-terminal lobe colored in coral, the C-terminal lobe in yellow, and the SDD in cyan. Bound K252a is shown as a stick model. B, a well formed peptide binding site exists in the activated kinase (protomer B). The bound substrate peptide (blue spiral) was docked from the peptide-bound active kinase structure.

Dimerization Interface of hIKKβ

It has been demonstrated that IKKα and IKKβ have strong tendencies to form homo- or heterodimers (13, 33, 35). Our results from dynamic light scattering and analytical size exclusion chromatography experiments indicated that our hIKKβ proteins existed as dimers in solution. Furthermore, similar C-terminal SDD-mediated dimers can be identified in the crystals of both xIKKβ and hIKKβ, in which the KDs are kept apart by the N-terminal ends of the SDDs, and two kinase active sites face opposite from each other (Fig. 6A). Although the crystal structures illustrate that the hIKKβ and xIKKβ dimers are related, they differ in one respect: hIKKβ adopts an open configuration and the xIKKβ dimer a closed one (Fig. 6A). In the dimer of xIKKβ, the kinase active sites are separated by a mere 67 Å (measured by the distance between two Cα atoms of the catalytic Asp-145 residues), with SDDs stretching parallel to each other almost their entire length (Fig. 6A). In the dimer of hIKKβ, the two SDDs bend away from each other at an angle of ∼45°, similar to an open pair of scissors. The two pairs of KD move further apart, increasing the separation between the two active sites by an additional ∼35 Å. In this open arrangement, the SDDs contact each other only at the blade ends through intertwined helices (Fig. 6A).

FIGURE 6.

Similarities and differences in oligomeric structures of hIKKβ and xΙKKβ. A, comparison of hIKKβ and xIKKβ dimers and higher order oligomer structures appeared in crystals. The hIKKβ dimer adopts an open conformation with its two KDs well separated. The hIKKβ dimer also contacts a crystal symmetry-related dimer (gray) at the KD and the N-terminal ends of the SDD, forming a face-to-face KD-KD* dimer, but these two dimers do not form a higher order oligomer. The xIKKβ dimer adopts a closed conformation and forms a higher order dimer-of-dimers structure with a crystal symmetry mate (gray). The dashed line and distances indicate the separations between Cα atoms of two catalytic Asp-145 residues (highlighted by an asterisk). B, the dimerization interface of hIKKβ is conserved. At the center of the interface is the hydrogen bond network involving Ser-489, Asp-493, Gln-647, Arg-650, Gln-651, and Trp-655. For clarity, only a half-set of the symmetry-related interacting residues are labeled. C, comparison of the face-to-face KD-KD* dimer of hIKKβ with that of xIKKβ. These interfaces are extensive but not conserved and lack specific hydrophilic interactions. The activation loop (colored in red) in the inactive hIKKβ molecule is close to the active site of the activated KD. The catalytic Asp (Asp-145) in the HRD motif of the activated kinase is highlighted with an asterisk.

Each SDD contributes its α2, α3, and α6 helices to the interface in a similar, but not equivalent way to xIKKβ (Fig. 6B). The resulting helical bundle buries 1570 Ų surface area per monomer, with a minor contribution from α3. Comparison of hIKKβ and xIKKβ structures shows that, with a few exceptions, the same residues are involved. The buried surface is largely hydrophobic but utilizes also an extensive hydrogen bonding network that is not seen in the structure of xIKKβ. This hydrogen bond network is formed by residues Lys-482, Trp-655, and Gln-651 from one protomer and Ala-481, Asp-493, and Ser-489 from another (Fig. 6B). In total, there are 10 interchain hydrogen bonds between these residues. In turn, Gln-651, Asp-493, and Ser-489 form intramolecular hydrogen bonds with Gln-647 and Arg-650 (Fig. 6B). Hydrophobic side chains of Phe-485, Leu-654, and Leu-658 also contribute to maintaining this interface in both the hIKKβ and xIKKβ structures. The central role of this interface in IKKβ dimerization was previously underlined by the failure of three hIKKβ mutants L654D/W655D, W655D/L658D, and L654D/W655D/L658D to dimerize (13).

Given that the backbone positions of the interfacing residues are comparable in the two structures with just their side chain rotamers varying, the absence of the hydrogen bond network at the xIKKβ dimer interface could simply be due to the limited resolution at which the xIKKβ model was built; alternatively, it may also be due to the plasticity of the interface. Structural analysis indicates that the former most probably is the case. In both the hIKKβ and xIKKβ structures, the hydrophilic side chains of Ser-489, Asp-493, Gln-647, and Gln-651 are partially or completely buried at the interface. Formation of the observed hydrogen bonds in the hIKKβ interface can compensate for the desolvation penalty associated with the burial of polar atoms. Because of the directionality of hydrogen bonds, the observed network can also increase the complementarity and the rigidity of the dimerization interface. It is notable that the residues at the dimer interface are among the most conserved in different species of IKKβ, and in hIKKα as well, but not in IKK-I and TBK1 (supplemental Fig. S1). This observation may underlie the ability of IKKβ and IKKα to form homo- or heterodimer complexes (33) and the significant difference in dimerization interfaces between IKKB and TBK1 (6, 30, 39).

DISCUSSION

Dimerization is a necessary precondition for the phosphorylation and activation of both IKKα and IKKβ (11, 33, 40, 41). More recent studies have shown that dimerization of IKKβ is necessary for its activation but not essential for its kinase activity once the activation loop is phosphorylated (13). The mechanism through which the SDD-dependent dimerization promotes activation loop transphosphorylation is not well understood, and little is known regarding the interkinase interactions once the activation loop is phosphorylated. To better understand the interplay of these interactions, we have determined the crystal structure of hIKKβ in a phosphorylated state.

In the structure described herein, we observe an SDD-mediated hIKKβ parallel dimer in which one KD is phosphorylated and adopts an active conformation, whereas the other is unphosphorylated and in an inactive conformation. This arrangement is distinct from that seen in the structure of xIKKβ (13). First, asymmetry in the phosphorylation pattern breaks the 2-fold symmetry of the hIKKβ dimer, so that the two IKKβ protomers are no longer equivalent. Second, although in both structures, the IKKβ dimer forms through reciprocal SDD-SDD interactions, the open geometry of hIKKβ minimizes this interaction to the C-terminal helix bundle only. Superimpositions of different hIKKβ protomers (Fig. 2C) and comparison of xIKKβ and hIKKβ dimers (Fig. 6A) indicate that IKKβ protomers have certain degrees of plasticity and that the change from closed to open state occurs solely through the flexion of three helices (α2, α5, and α6) at the N-terminal end of the SDD. Hence, the six-helix bundle acts like a hinge so that the two chains can swing away or toward each other.

IKKα and IKKβ could be activated either by an upstream kinase or auto-phosphorylation (28, 33). The main feature that is common to the hIKKβ, xIKKβ, and TBK1 structures is that autophosphorylation within the dimer is precluded by steric factors (Fig. 6A). This apparently fulfills the requirement for tight control of IKKβ activation. On the other hand, it also implies that dimerization by itself is necessary but not sufficient for their autophosphorylation, making trans-autophosphorylation through oligomerization a likely mechanism for activation (13, 30). In the crystals of xIKKβ, the IKKβ dimers are structured into higher order tetramers with a symmetry that can be described as a dimer of dimmers (Fig. 6A) (13). The second dimer is an anti-parallel, face-to-face dimer, in which the active sites of two KDs face each other in proximity (Fig. 6C). It was suggested that this proximity might allow the activation loop of one protomer to reach into the active site of the other for transphosphorylation (13).

Inspection of our crystal structure reveals that, along with a different dimer geometry, the tertrameric xIKKβ crystal symmetry is also no longer maintained (Fig. 6A). However, an anti-parallel, face-to-face KD-KD* interface exists in hIKKβ. This interface resembles that in xIKKβ but is not the same (Fig. 6C). First, the hIKKβ face-to-face dimer is asymmetric because it too involves an unphosphorylated-phosphorylated kinase pair (protomer A and crystal symmetry related protomer B*). Next, it exhibits different contact geometry. In hIKKβ, the two protomers readjust their positions, enhancing the interface and bringing the two active sites closer together (Fig. 6C). In this orientation, the KD N-terminal lobes dock onto the N-terminal ends of the SDD of the dimeric partner and the KD C-terminal lobes into each other. This interaction allows for a better proximity of the fully formed active site of one KD toward the unphosphorylated activation loop of the other and hence a more proper positioning for transphosphorylation. Interestingly, although the binding interface of the second KD-KD* dimer is more extensive compared with the SDD-SDD dimer (total buried surface area of 1760 Ų versus 1570 Ų), the binding free energy, calculated with the protein interface analysis program PISA (43), is less favorable for the second dimer than for the primary one (−2 kcal/mol versus −24 kcal/mol). This high value indicates that the face-to-face dimer is less stable, consistent with the transient nature of this association for transphosphorylation.

The differences in these face-to-face dimer interfaces in hIKKβ and xIKKβ could well be due to crystal packing variations, and the differences in the oligomeric organization of two proteins could also be due to species difference. However, although the biological relevance of this anti-parallel dimer interface is unclear, we can speculate that the observed conformational differences between the two forms of IKKβ are functionally relevant. Based on our modeling analysis, it is apparent that realignment of the activation loop upon phosphorylation would impose steric strain into the face-to-face dimer of xIKKβ, forcing its rearrangement, regardless crystal packing or protein species.

The results discussed above raise the possibility that change in the overall geometry of IKKβ, similar to the loss of 2-fold symmetry, is due to the differences in activation loop status of hIKKβ and xIKKβ, and hence in the active versus inactive state conformations that they adopt. We suggest that the structure of xIKKβ describes IKKβ in its early “closed off” state that precedes the intermediate active, open state represented by our structure of phosphorylated hIKKβ. By analogy with other protein kinases that get activated by dimerization (44, 45), this change in geometry could be required to properly orient the kinase domains once an activated fraction is in play. Further studies of these enzymes will be required to confirm or dispute these findings.

Addendum

This activation mechanism based on a change in geometry does not seem to apply to the IKK-related kinase TBK1. In two independent studies that were published while our manuscript was in preparation, Larabi et al. (30) and Tu et al. (46) ruled out the possibility for dimer opening in TBK1 because the active and inactive TBK1 crystal structures both maintained compact symmetric dimer configurations, with much more extensive dimerization interfaces than observed in both xIKKβ and hIKKβ.