Several strategies were employed to obtain sufficient protein purification yields for crystallization: (1) variation of N- and C-terminal boundaries of the Tyk2 catalytic domain (some constructs included the pseudokinase domain); (2) variation of the affinity purification tag; (3) introduction of a kinase-inactivating mutation; and (4) use of multiple orthologs. Table 1 lists the different strategies and examples employed for Tyk2 construct design. After exploring roughly 40 constructs (list shown in Additional file 1 Table 1), we prioritized a mouse construct that produced adequate amounts of soluble protein for crystallization (GST-Tev-muTyk2 (870–1170) Asp1016Ala). The human and mouse Tyk2 catalytic domain sequences are highly conserved (91% identity in catalytic domain, 78% identity overall); however, several divergent surface residues had the potential to impact protein aggregation and crystallization behavior. A glutathione-S-transferase (GST) tag was included to increase solubility during early stages of purification, and the Asp1016Ala kinase-inactive mutation was introduced to increase conformational homogeneity by preventing multiple phosphorylation states; this mutation also increased expression approximately three-fold (data not shown). Asp1016 is the conserved catalytic base that is essential for phosphotransferase activity in protein kinases 26 .

Previous attempts to purify the human Tyk2 protein using multiple chromatographic steps resulted in low yields or no detectable protein. Due to the aggregation and solubility problems seen with the human isoform, orthologs were considered and an abbreviated purification protocol was implemented. This protocol entailed batch binding to GST resin for several hours, followed by a resin wash and an “on column” TEV protease cleavage step. A critical step was to introduce the ligand (Compound 1) (Figure 1) at low protein concentrations, to prevent precipitation, and subsequently to co-concentrate the Tyk2/Compound 1 complex to a level useful for crystallization trials. Compound 1 (IC₅₀ 6 nM against human Tyk2; Table 2) was one of the few inhibitors that co-crystallized with mouse Tyk2, allowing us to determine the structure of the mouse Tyk2 kinase domain. We also present the structure of Compound 2 complexed to mouse Tyk2, which was solved using inhibitor soaking methods.

Values are mean and standard errors.

For values below the sensitivity of the assay; the number of replicates is given.

Despite not directly forming crystal contacts, we found that inclusion of an ATP-competitive inhibitor was required for formation of mouse Tyk2 crystals. To understand the importance of ligand binding to the overall stability of the enzyme, we measured the Tyk2 kinase domain’s susceptibility to proteolysis in the presence and absence of a ligand. Compound 2 (30 μM) significantly increased resistance to partial proteolysis by thermolysin (Additional file 1: Figure S1). Minor processing of the kinase domain from ~29 kDa (intact) to ~27 kDa form by thermolysin is unaffected by addition of Compound 2, suggesting that its binding in the ATP site is insufficient to prevent cleavage of one of the extreme termini of our Tyk2 kinase domain construct. However, the rate of degradation of the enzyme to smaller forms (<27 kDa) is reduced by 13-fold (Figure 2). Like all protein kinases, the ATP binding site for Tyk2 is nestled between the N-terminal and C-terminal lobes. Our proteolysis data suggest that the conformational flexibility of the kinase, other than a ~2 kDa portion of one terminus (likely the N-terminus based on the disorder seen from residue 870–884 in the crystal structure described below), is decreased by the binding of these 3-aminoindazole inhibitors. The ability of Compound 1 to enable robust Tyk2 crystallization may be related, as inhibitor-induced decreased flexibility may favorably affect entropic loss during crystal nucleation and growth.

The overall structure of the mouse Tyk2 kinase domain is very similar to that of the recently reported human Tyk2 kinase domain complexed to CP-690,550 (PDB entry 3LXN; r.m.s.d 0.5 Å) (Figure 3) 27 . Two particular sequence differences between mouse and human Tyk2 may enable the crystallization of the mouse ortholog. The structure revealed that the substitution of Glu927 and Gly928 for Ala934 and Asp935 in human Tyk2 permits Gly928 to form a close, van der Waals crystal contact. Additionally, there is a potential interaction between Glu927 and Arg1132 in an adjacent molecule in the crystal lattice. Primarily due to steric clashes, a similar crystal packing would not be possible in human Tyk2. Figure 4a illustrates the sequence alignment between the mouse and human Tyk2 catalytic domains, and Figure 4b provides a view of this crystal contact.

The mouse Tyk2/Compound 1 co-crystal structure is illustrated in Figure 5a. The 3-aminoindazole core serves as a canonical hinge binder, forming three hydrogen bonding interactions with hinge residues Glu972 and Val974. The inhibitor’s central phenyl group linker positions the sulfonamide chlorophenyl group under the glycine rich loop. Figure 4a shows that the chlorophenyl moiety occupies a distinct hydrophobic pocket proximal to the DFG pocket. The placement of this moiety is guided by the sulfonamide linkage’s stabilizing interactions with the NH backbone of Glu898 in the glycine-rich loop, and conserved residues Asn1021 and Arg1020. The structure of Tyk2 and Compound 2 is illustrated in Figure 5b. The binding mode and trajectory of the chlorophenyl is identical to that of Compound 1 and, as a result, the glycine-rich loop adopts the same conformation in both structures. The furan substituent on the hinge-binding 3-aminoindazole core was well-ordered, providing clear evidence that the inhibitor soak was successful. The furan occupies the extended hinge region, sandwiched between Arg894 and Gly977.

One notable secondary structure difference between the co-crystallized mouse Tyk2/Compound 1 complex and the recent human Tyk2/CMP-6 complex (PDB entry 3LXP) occurs at the tip of the glycine-rich loop. An overlay shows that Compound 1 induces a ~4 Å upward shift in the loop (Figure 6), resulting in a more open active site conformation. In a recent review, it was suggested that the conformational dynamics of the glycine-rich loop may differ within the Jak family 28 . This may be due to sequence diversity in the glycine-rich loops of Jak1, Jak2, Jak3 and Tyk2. Specifically, in Tyk2 and Jak1, a “collapsed” glycine-rich loop conformation may depend upon an interaction between a histidine residue and a proximal aspartate (His907 and Asp1023 in human Tyk2). These residues are absent in Jak2 and Jak3. In the mouse Tyk2 structures, complexed to either Compound 1 or Compound 2, the steric bulk of the sulfonamide chlorophenyl moiety occupies substantial hydrophobic space under the glycine-rich loop and would potentially disrupt the His/Asp “glycine-rich loop lock,” thereby creating a larger active site pocket. While there are crystal contacts near the loop, we believe, based on multiple crystal structures determined with different soaked inhibitors (data not shown), that the loop conformation is driven mainly by the ligand. We cannot rule out, however, that some differences in loop conformation between human and mouse Tyk2 may be driven by crystal packing. Despite a more open conformation, we hypothesize that mouse Tyk2 was able to crystallize with these inhibitors because the chlorophenyl moiety stabilized the flexible glycine-rich loop. Inclusion of the chloro group also improves potency by roughly 10-fold in an enzyme activity assay (data not shown).

Mouse Tyk2 (UNIPROT Q9R117) was cloned and expressed in Sf9 insect cells. The coding region of the catalytic domain of mouse (mu) Tyk2 (residues 870–1170) was PCR sub-cloned into pDONR221 using the BP reaction of the Gateway® (Invitrogen/LTI) cloning system. The muTyk2 catalytic domain was immediately preceded by a primer encoded Tobacco Etch Virus (TEV) protease cleavage site. The resulting Tev-muTyk2 (870–1170) was modified using the Quick-Change Site-Directed Mutagenesis System (Stratagene) to replace Asp1016 with Ala. After sequence confirmation, Tev-muTyk2 (870–1170) Asp1016Ala was sub-cloned into the pDEST20 expression vector (Invitrogen/LTI) using the Gateway® LR reaction to create an in-frame fusion with an amino-terminally encoded Glutathione S-Transferase (GST). The resulting expression plasmid, pDEST20 GST-Tev-muTyk2 (870–1170) Asp1016Ala, was confirmed by DNA sequencing. The entire expression cassette was then transferred to baculovirus. Virus production and amplifications were carried out according to Invitrogen/LTI Bac-To-Bac system instructions.

High titer virus stocks were made as recommended and used to infect Sf9 cells, cultured in Sf900II medium (Invitrogen/LTI) at 27.5°C, at an estimated M.O.I of 2.5 to 5.0. Infected cells were harvested by centrifugation at 48 h post-infection, which was optimal for Tyk2 protein expression.

Mouse Tyk2 (870–1170) Asp1016Ala protein pellet was suspended on ice in lysis buffer containing Buffer A (50 mM HEPES pH 7.5, 500 mM NaCl, 10% glycerol, and 1 mM adenosine) in addition to 2X protease inhibitor tablets (Roche Applied Science). The resulting mixture was sonicated three times with 20 second blasts. The mixture was then added to 10 mL of GST affinity resin for 2.5 h, centrifuged at 1,000 × g, and washed. TEV protease was added to the resin and the mixture was loaded into a column; it was incubated for 2 h at room temperature, and additionally overnight at 4°C. The protein was then washed off with Buffer A and collected as monitored by A₂₈₀. The pooled protein was concentrated and dialyzed overnight into 50 mM HEPES pH 7.5, 100 mM NaCl, 5 mM DTT, 1 mM ADP. The resulting protein was pooled (0.5 mg/mL) and used directly for crystallization trials.

Mouse Tyk2 protein (0.5 mg/mL) was incubated with Compound 1 (0.1 mM) and concentrated to 10 mg/mL. After 3–4 days, protein crystals grew using the vapor diffusion method in sitting drop plates under the following condition: 4.3–4.7 M ammonium formate, 100 mM Tris pH 8.0. Crystals were subsequently used for soaking inhibitors of interest. Compound 2 was soaked into the Tyk2 crystals by adding 1 μM inhibitor (final concentration) to a 100 μL well of harvest mother liquor. Crystals were frozen from mother liquor solution containing 20% glycerol.

X-ray diffraction data from mouse Tyk2/Compound 1 crystals were collected at the IMCA beamline 17ID at the Advanced Photon Source in Argonne, IL. The crystals were maintained at 100 K with an Oxford Cryosystems Cryostream cooler during data collection. A total of 180 frames were collected at an oscillation range of 1.0°. The data were processed with the HKL2000 suite of programs. After determining the crystal orientation, the data were integrated with DENZO, scaled/merged with SCALEPACK, placed on an absolute scale and reduced to structure factor amplitudes with TRUNCATE. Five percent of the unique reflections were assigned randomly to the “free” set, for calculation of the free R-factor (R_free) 29 . The remaining 95% of the reflections constituted the “working” set for calculation of the R-factor (R). The x-ray diffraction data and refinement statistics are summarized in Table 2.

A maximum likelihood molecular replacement solution was determined using the program PHASER 30 31 32 . One Tyk2 monomer was located in the asymmetric unit, in the space group P3₁21. The search model was a crystal structure of Jak2 reported previously (PDB entry 2B7A). Coordinates were generated based on the molecular replacement solution. The refinement of the Tyk2/Compound 1 complex crystal structure began with the molecular replacement solution coordinates. Rigid-body refinement was conducted by the program REFMAC 33 in the CCP4 suite of programs, which resulted in the following statistics at 2.6 Å: R 0.39 (R_free= 0.39). Experimental Tyk2 and inhibitor electron density was observed. Manual building of Compound 1 into the density was attempted using the molecular graphics program O 34 and examination of 2Fo–Fc and Fo–Fc electron-density maps. The refinement program REFMAC was used for iterative rounds of restrained refinement 33 . Final rounds of refinement were conducted using AUTOBUSTER (Global Phasing) 35 , which added water molecules to the final model, resulting in the following statistics: R 0.199 (R_free= 0.232). Final refinement statistics are shown in Table 3. The quality of all models was evaluated using COOT 36 . The co-crystal structure of Compound 2 complexed to Tyk2 was solved by molecular replacement using the Tyk2/Compound 1 structure as a probe. An origin shift of [0 0 ½] was applied to match the Compound 1 coordinates. DETWIN 37 was used with a twinning fraction of 0.24 to improve refinement statistics. Final rounds of refinement were conducted using AUTOBUSTER (Global Phasing) 35 . Final refinement statistics are listed in Table 3.

Mouse Tyk2 kinase domain (0.25 mg/mL, residues 870–1170; Asp1016Ala mutant) was incubated for 0 to 90 minutes with thermolysin (0.25 mg/mL) at room temperature in 50 mM HEPES pH 6.7, 150 mM NaCl, 5% glycerol, and 2.5 mM CaCl₂ in the presence and absence of 30 μM Compound 2. EDTA (final conc 100 mM) was used as stop solution to quench the proteolysis reactions. Samples were separated by use of a Caliper LC90 system and the remaining substrate and product bands were quantitated. Intact (undigested) Tyk2 ran in this system at ~29 kDa. Addition of thermolysin yielded a partial digestion product of ~27 kDa (Additional file 1: Figure 1) within 5 minutes, which was unaffected by addition of Compound 2. Subsequent degradation products were strongly influenced by the presence of Compound 2 (Additional file 1: Figure S1). Therefore the sum of the intensities of the 29 and 27 kDa peaks was used as a measure of inhibitor-dependent resistance to thermolysin digestion. Addition of Compound 2 did not alter the measurable digestion of a BSA control by thermolysin in the same buffer (Additional file 1: Figure S2a), indicating that Compound 2 was not an inhibitor of thermolysin protease activity. Thermolysin amounts were found to be unchanged over the course of the experiment, providing a convenient loading control (Additional file 1: Figure S2b).