Molecular basis of Tank-binding kinase 1 activation by transautophosphorylation

Domain Organization of TBK1 Is Consistent with the IKKβ Tripartite Architecture.

To understand how the KD and ULD are organized within TBK1, we crystallized a fragment of human TBK1 containing both domains and a catalytic residue mutation (KU_D135N; residues Q2–E385) in the presence of an inhibitor, BX795 (26, 27). The structure was solved to 2.6-Å resolution by single-wavelength anomalous dispersion using a selenomethionine-substituted sample (Fig. S1A and Table S1). Both molecules in the asymmetric unit demonstrate a tandem arrangement of the kinase and ubiquitin-like domains, which possess the typical bilobe kinase and ubiquitin folds, respectively (Fig. 1A). In this configuration, the ULD abuts the C-terminal lobe (C lobe) of the KD. The conventional hydrophobic surface patch of the ULD, which often serves as a protein–protein interaction surface in the case of ubiquitin, faces away from the kinase domain.

A very similar domain organization is observed for IKKβ (25). Superposition of the TBK1 KU_D135N fragment onto the full-length IKKβ structure (via alignment of the kinase domains) appears to poise both TBK1 domains for interaction with the SDD (Fig. 1B). Residues identified in the KD–SDD and ULD–SDD interfaces of the IKKβ structure show moderate conservation across IKK-family members and map to comparable surfaces of the TBK1 KU_D135N fragment structure (Fig. 1B and Fig. S2); however, the orientation of the ULD relative to the KD differs slightly in the two kinase structures (Fig. 1A, Inset). In the full-length IKKβ structure, an ∼30° rotation in the IKKβ ULD relative to that of TBK1 allows the IKKβ ULD’s hydrophobic patch to bind to the adjacent SDD (Fig. S3 A and B). Similarly, an insertion into the KD fold specific to IKK-family members (residues 241–265 in TBK1) also appears to rearrange in the presence of the SDD (Fig. S3 C and D). Despite these minor differences, the TBK1 KU_D135N fragment structure appears to be compatible with the tripartite domain arrangement observed for full-length IKKβ.

TBK1 KU_D135N Structure Adopts an Activation Loop-Swapped Conformation.

Although the KD of the TBK1 KU_D135N structure generally assumes the typical kinase fold, the activation segment adopts a unique conformation in this unphosphorylated crystal form. Residues encompassing the entire activation loop (L164–G199) surprisingly extend away from the kinase core of each KU_D135N molecule in the asymmetric unit and associate intimately with the C lobe of the neighboring KU_D135N molecule (Fig. 2A). Whereas residues 191–192 and 192–198 are disordered in molecules A and B, respectively, the electron density over the exchanged region is very clear from L164–R191 in both protomers, supporting the domain-swap interpretation (Fig. S1B).

The fully formed α-helix at the activation-loop C terminus (αEF) docks onto a surface cleft between two α-helices from the opposite protomer via the HPD tripeptide motif (canonically the “APE” motif). The placement of this motif on the C lobe, albeit in trans, mirrors intramolecular interactions typically observed in active kinase structures. Moreover, the KU_D135N structure demonstrates several other features that are characteristic of active kinases (Fig. S4A).

At the activation-segment N terminus, a helix forms between D167 and L173 (αAL). Superposition of the activation loop-exchanged KU_D13N structure with the protein kinase A (PKA)–peptide inhibitor complex structure (28) (based on alignment of the kinase domains) reveals that the P+1, +2, and +3 sites inferred from the PKA complex (where P0 is the site of modification) are occupied by the αAL-helix of the neighboring protomer in the TBK1 structure (Fig. S4B). Thus, although the unphosphorylated KU_D135N structure is endowed with many active kinase traits, the domain-swapped form is not compatible with substrate binding and, therefore, would not act on a polypeptide substrate. However, S172, located in the αAL of the exchanged TBK1 activation loop, is only 6 Å away from the P0 site of the PKA substrate and less than 11 Å from the catalytic D135 residue. Elevated temperature factors in the αAL-helix underscore the dynamic nature of the TBK1 activation loop (Fig. S4C) and suggest that a local reorganization of this region, which would allow phosphotransfer to S172, may be plausible.

The extended interface created by this KU_D135N loop swapping is comparable in size and composition to bona fide protein–protein complexes, burying 2,570 Ų of surface area per molecule. However, despite this extensive interaction surface, KU_D135N was found to be monomeric in solution (Fig. 2B). KU_D135N, and a shorter construct containing just the kinase domain (KD_D135N; Q2–R308), both behaved as monomers by size-exclusion chromatography (SEC). By contrast, full-length TBK1 (FL_D135N; Q2–L729) containing the C-terminal SDD eluted as a dimer. Multiangle light scattering (MALS) confirmed the dimeric or monomeric status of each sample. Thus, the activation loop-swapped interactions in the KU_D135N structure do not appear to mediate formation of stable dimers, but may instead constitute more transient self-associations.

TBK1 Can Catalyze Autophosphorylation in Trans.

To probe whether the placement of the activation loop within the substrate-binding cleft of the KU_D135N protomers observed in the structure could also occur in solution, we tested whether active TBK1 could catalyze S172 phosphorylation. To this end, we expressed and purified three distinct TBK1 constructs that each contained the wild-type aspartic acid residue at position 135: FL_WT, KU_WT, and KD_WT. Each of these TBK1 constructs used the same domain boundaries as their equivalent D135N variant (FL_D135N, KU_D135N, KD_D135N) and showed similar SEC elution profiles as their “kinase-dead” counterparts (Fig. S5 A and B).

FL_WT, KU_WT, and KD_WT stocks treated with ATP before SDS/PAGE separation displayed a marked increase in S172 phosphorylation by anti-pS172 immunoblotting compared with stocks incubated without ATP (Fig. 3A). Quantitation of these samples by LC-MS revealed that ∼6% of the purified FL_WT stock contained pS172, whereas ∼98% was phosphorylated after a 20-min incubation with ATP (Fig. 3B). By comparison, purified KU_WT and KD_WT stocks were ∼32% S172-phosphorylated and also achieved near-complete modification after the same ATP treatment (Fig. 3B, Inset).

To determine whether TBK1 autophosphorylation could occur in trans, active TBK1 constructs were added to the purified D135N mutants and pS172 status was monitored by immunoblotting (Fig. 3C). The kinase-dead D135N variants, which showed negligible activity against a commercial peptide substrate (Fig. S5C), were unable to mediate S172 phosphorylation, whereas strong pS172 signal was observed upon the addition of preactivated FL_WT, KU_WT, or KD_WT. While transphosphorylation with prephosphorylated wild-type TBK1 constructs occurred rapidly, equivalent reactions using fully dephosphorylated KU_WT showed a substantial lag phase in the kinetics of KD_D135N (and KU_WT) S172 modification (Fig. 3D). Similar reactions using a KU variant that cannot be activated by phosphorylation (KU_S172A; Fig. S5C) were even further retarded, with transphosphorylation of KU_D135N only observable at longer time points (Fig. 3D). Further comparisons of prephosphorylated and dephosphorylated KU_WT activities using a commercial peptide substrate recapitulated the lag-phase behavior observed for transphosphorylation by pS172 immunoblotting; however, the lag was shortened with increasing concentrations of dephosphorylated kinase (Fig. S6).

Structure of the S172-Phosphorylated TBK1 KD.

To understand the structural changes that accompany TBK1 activation, we phosphorylated TBK1 KD_D135N in vitro using activated FL_WT and determined the crystal structure of the monophosphorylated KD_D135N, again bound to BX795, to a resolution of 1.8 Å using molecular replacement (Figs. S1C and S7 A–C and Table S1). The unphosphorylated and phosphorylated TBK1 KD structures, excluding the activation loops, can be superimposed with an rms deviation of 0.58 Å (for 270 Cα atoms). By contrast, the activation loop undergoes a dramatic conformational change upon phosphorylation (Fig. 4 A and B). In the monophosphorylated KD_D135N structure, the well-ordered pS172 docks onto its own kinase domain, shifting 18 Å from its position in the loop-swapped KU_D135N structure to form a network of intramolecular contacts (Fig. S7D).

The substantial refolding of the activation loop is hinged at residues L164 and G199, with remarkably few phosphorylation-induced conformational changes in the TBK1 KD outside this region (Fig. 4 A and B). All of the other structural features indicative of an active conformation present in unphosphorylated TBK1 KU_D135N are preserved in the pS172-KD_D135N structure; however, the HPD motif is now docked onto the KD in an intramolecular interface. More importantly, the backbone of the P+1 binding pocket now adopts a similar contour as PKA in the PKA–peptide inhibitor complex, and thus appears competent to bind polypeptide substrate. Closer inspection of the site suggests that TBK1 would strongly favor a hydrophobic residue at the P+1 position (Fig. S7 E and F).

Determinants of TBK1 Catalytic Activity and Substrate Recognition Appear to Be Confined to the Kinase Domain.

To assess domain contributions to TBK1 kinase activity, the ability of each active construct to phosphorylate putative peptide substrates was evaluated. Here, activated FL_WT, KU_WT, and KD_WT stocks were each incubated with a panel of biotinylated peptide substrates, including a peptide encompassing the S172 activation site, in the presence of ³²P-labeled ATP. Reaction mixtures were blotted onto a streptavidin membrane, and peptide phosphorylation was visualized by autoradiography (Fig. 5). The activation-loop peptide displayed robust phosphorylation signal that exceeded that of peptides containing established IRF3 or AKT modification sites (15, 29). In accordance with previous studies using this methodology, the TBK1 constructs were also active against a peptide sequence originating from the Dead-box helicase DDX3X (30); however, they did not mediate phosphorylation of the negative control peptide, GSK3-like. Conversely, the AKT control showed no discernable modification of the TBK1-, IRF3-, or AKT-derived peptides, but did phosphorylate its proven GSK3-like peptide substrate (31). Finally, all TBK1 constructs showed comparable activities across a given substrate, and no phosphorylation was detected in the absence of kinase. TBK1-catalyzed phosphorylation of recombinant IRF3 and AKT protein substrates, as evaluated by phospho-specific immunoblotting and LC-MS, gave similar results (Fig. S8).

Additionally, full enzymatic profiles were generated for each of the active TBK1 constructs using IRF3 as the substrate (Fig. 6). Fits of the data to the Michaelis–Menten equation yielded the enzymatic parameters V_max and K_m for each variant. Loss of the SDD alone, or in combination with the ULD, essentially had no effect on IRF3 binding, as K_m values for KU_WT and KD_WT were within experimental error of the K_m of FL_WT. The truncated variants also displayed only minor differences in V_max compared with the full-length kinase. In agreement with this detailed kinetic evaluation, FL_WT, KU_WT, and KD_WT also displayed comparable activities toward a commercial peptide substrate over a wide range of kinase concentrations (Fig. S9). Importantly, all comparisons of TBK1 constructs with IRF3, AKT, and peptide substrates were performed using TBK1 stocks that were preactivated with ATP, so as to remove TBK1 autophosphorylation events from the analyses.