The gene encoding full-length fliS (126 amino acids) was amplified from the genomic DNA of H. pylori strain 26695 and cloned into the GST expression vector pGEX6p-3. The gene encoding HP1076 (171 amino acids, 19.3 kD), FlaB (514 amino acids, 53.8 kD) and C-terminally truncated FlaB (FlaBc, residues 415–514, 10.8 kD) were cloned into the pAC28m vector, which expresses a 6× His tag at the N-terminal of the recombinant protein. The FlaBc fragment consists of the C-terminal D0 and D1 region based on the multiple sequence alignment with other flagellin homologs. The resulting plasmids pGEX6p-3-fliS and pAC28m-flaBc were co-transformed into Rosetta 2 cells and grown in LB broth containing selection antibiotics. Protein expression was induced by the addition of 0.2 mM isopropyl β-D-1-thiogalactopyranoside, followed by incubation at 25°C for 18 h. The cells were harvested and lysed by sonication in a lysis buffer containing 10 mM Tris (pH 7.5) and 300 mM NaCl. To purify the FliS-FlaBc complex, the clarified lysate was passed through a glutathione affinity column, after which the GST-tag was removed using Precission protease (GE Healthcare). The eluates containing FliS and His-FlaBc were purified further on a Ni-NTA affinity column and Superdex 75 size-exclusion chromatography column using a buffer containing 10 mM Tris (pH 7.5) and 150 mM NaCl. His-HP1076 was also purified on Ni-NTA affinity and Superdex 75 size-exclusion chromatography columns. To purify the FliS-FlaBc-HP1076 ternary complex, the FliS/FlaBc complex was mixed with His-HP1076 at a molar ratio of 1:1 and incubated on ice for 2 h. The protein mixture was then loaded on a Superdex 75 column. The retrieved fractions containing the FliS-FlaBc-HP1076 complex were concentrated to 6 mg/ml and subjected to crystallization studies.

The H. pylori G27 strain was cultured at 37°C on Columbia blood agar containing 5% defibrinated horse blood and H. pylori-selective antibiotics (trimethoprim, amphotericin, vancomycin, cycloheximide, cefsulodin, polymyxin, β-cyclodextrin) under microaerobic conditions (5% CO₂, 4% O₂, and 91% N₂) controlled by AnaeroGen gas packs (Oxoid). H. pylori transformant selection cultures also included 5 μg/ml chloramphenicol. Brucella broth containing 10% (v/v) fetal bovine serum (BB10) was used for liquid H. pylori culture.

The H. pylori G27 deletion strain Δhp1076 was created as described in a previous study (Zhang et al., 2017). Briefly, the kanamycin resistance cassette was amplified from a kanamycin-resistant strain of Campylobacter jejuni (a gift from KW Ling, CUHK). The PCR product was digested and inserted into a pGEM-hp1076 plasmid in which aphA3 was flanked by hp1076. The resulting hp1076:aphA3 fragment was released and cloned into the pBluescript II KS vector, and the insertion was confirmed by sequencing. A similar strategy was used to generate the fliS-null strain, except inverse PCR was used to amplify the backbone of the pBluescript vector together with the flanking region of the fliS gene. The digested aphA3 cassette was then inserted into the inverse PCR product pBlueScript-fliS to yield the pBS-fliS:aphA3 plasmid in which aphA3 was flanked by fliS. For transformation, the recombinant plasmids on the pBluescript background were methylated using endogenous methylases from a H. pylori cell lysate according to (Donahue et al., 2000). The methylated plasmids were then introduced into wild-type H. pylori using natural transformation. Briefly, freshly cultured H. pylori was mixed with 1 μg of methylated plasmids and cultured in a microaerophilic environment. The cells were then transferred onto blood agar plates supplemented with 10 μg ml^–1 kanamycin. After 3–4 days, single colonies were picked and sub−cultured on fresh plates. This selection procedure was repeated three times to obtain hp1076−null and fliS-null strains. Finally, the strains were confirmed by PCR and immunoblotting.

An anti-HP1076 antibody was produced from the serum of a rabbit that had been immunized with purified recombinant HP1076 protein (Invitrogen antibody production service). The H. pylori G27 and Δhp1076 strains were grown in Brucella broth for 1 day prior to the detection of HP1076. The cells were pelleted, resuspended in PBS and boiled in the Laemmli sample buffer to yield total cellular proteins. The membranes were incubated with the anti-HP1076 antibody to detect the protein in the lysates.

The H. pylori G27, Δhp1076 and ΔfliS strains were cultured in Brucella broth containing 0.4% cholesterol at 37°C under microaerophilic conditions for 2 days. The cells were then harvested and washed three times with phosphate−buffered saline (PBS). The cells were further resuspended in PBS to an OD600 of 10. Ten-microliter aliquots of the cell suspension were dotted onto a Formvar/Carbon coated grid (TED PELLA) and stained with 0.2% neutral phosphotungstic acid. Excess staining was removed using Whatman paper. Flagellation was examined under a transmission electron microscope (Hitachi H−7650) at 80 kV.

The H. pylori G27, Δhp1076 and ΔfliS strains were cultured on Columbia blood agar plate for 3 days. The cells were harvested and resuspended in Brucella broth containing 0.4% cholesterol, and 1-μl aliquots were dotted onto soft agar plates containing Brucella broth, 0.4% cholesterol and 0.35% Bacto agar. The plates were incubated at 37°C for 3 days under microaerobic conditions (5% CO₂, 4% O₂, and 91% N₂) generated by CampyGen sachets (Oxoid) and then the colony diameters were measured and statistically analyzed.

H. pylori G27 and fliS-null cells were cultured on blood agar plates for 3 days, harvested and washed three times with PBS. The cells were resuspended in the lysis buffer containing 100 mM Tris-HCl (pH 8.0) to an OD600 of 10 and lysed by sonication. The unbroken cells and cell debris were removed by centrifugation at 4°C and 6000 × g for 10 min. The supernatant (i.e., whole cell fraction) was subjected to ultracentrifugation at 4°C and 50,000 × g for 1 hr. The soluble fraction was collected as the cytosolic/periplasmic fraction. The pellet was washed three times, resuspended in the lysis buffer and collected as the total membrane fraction. All fractions were mixed with SDS-PAGE loading buffer and analyzed by SDS-PAGE and immunoblotting.

H. pylori G27 cells were lysed by sonication in the buffer containing 20 mM HEPES (pH 7.5), 137 mM NaCl, 27 mM KCl, 5% glycerol, 0.1% Tween-20 and 10 mM DTT. Unbroken cells and cell debris were removed by centrifugation at 4°C and 6,000 × g for 10 min.

The GST-FliS pull down experiment was performed as described previously (Lam et al., 2010). Purified GST and GST-FliS were mixed with HP1076, HP1076ΔN20, or HP1076ΔN20ΔC29 and immobilized to glutathione sepharose. After washing, aliquots of the resin were subjected to SDS-PAGE to confirm the formation of the GST-FliS-HP1076 and GST-FliS-HP1076 ΔN20 complexes. HP1076ΔN20ΔC29 was used as a negative control because this truncated fragment does not interact with FliS (Lam et al., 2010). The resin was incubated with H. pylori cell lysate for 3 h and washed three times. The resin beads were suspended in the loading buffer, and protein binding was analyzed by SDS-PAGE.

For the flagellin/HP1076 competitive binding assay, purified GST-FliS was immobilized on glutathione sepharose, washed and aliquoted. After measuring the concentration of protein on the resin, aliquots were mixed with equal amounts of H. pylori cell lysate and HP1076 with GST-FliS at different molar ratios (0, 0.5:1, 5:1, and 50:1), followed by a 3-hr incubation. The beads were washed and analyzed by SDS-PAGE. The intensities of the protein bands were quantified using ImageJ software.

The purified FliS and FliS-FlaBc complex were labeled fluorescently using the NanoTemper Monolith NT Protein Labeling Kit. Next, labeled FliS or FliS-FlaBc (40 nM) was mixed with HP1076 ranging from 6.2 μM to 0.189 nM or 3.83 mM to 117 μM with a volume ratio of 1:1, respectively, in the buffer containing 20 mM Tris (pH 7.5), 150 mM NaCl and 2 mM DTT. The reactions were run at 30% excitation power, 40% MST power and 21.5°C. MO. Affinity Analysis software was used to analyze the data and calculate the dissociation constant (K_d) values.

Crystals of FliS-FlaBc-HP1076 were obtained using the sitting-drop vapor-diffusion method at 16°C. The buffer contained 3% (v/v) PEG 20000, 0.2 M potassium bromide, 0.2 M potassium thiocyanate and 0.2 M sodium acetate (pH 5.0). The crystals were soaked briefly in the crystallization buffer containing 20% glycerol and were cooled by plunging into liquid nitrogen. The X-ray datasets were collected using beamline 13B1 at the National Synchrotron Radiation Research Center, Taiwan. The datasets were then processed using the iMosflm package (Battye et al., 2011) and were scaled and reduced using SCALA from the Collaborative Computational Project, Number 4 (CCP4) suite (Winn et al., 2011). Crystals of FliS-FlaBc-HP1076 were formed in the P6₁ space group. The structure was solved by molecular replacement, using FliS and HP1076 from H. pylori (PDB ID: 3K1H) as a search model. The molecular replacement program Phaser (McCoy et al., 2007) in the PHENIX suite and data in the resolution range were used, and the refinement and manual rebuilding steps were performed using the programs COOT and Phenix (Adams et al., 2010; Emsley et al., 2010). The statistics of data collection and refinement are summarized in Supplementary Table 1. The coordinates and structure factors of the FliS-FlaBc-HP1076 complex have been deposited in the Protein Data Bank (PDB ID: 6LEA). All figures were prepared using Chimera (Pettersen et al., 2004).

The H. pylori flagellins FlaA and FlaB each contain four domains, consistent with other flagellin homologs (Beatson et al., 2006). Although the FliS-FliC structures have been resolved in A. aeolicus, B. subtilis and S. Typhimurium (Evdokimov et al., 2003; Altegoer et al., 2018; Xing et al., 2018), recent studies have demonstrated that the C-terminal D0 domain of flagellin is critical to the ability of H. pylori to evade TLR-5-mediated host recognition (Forstnerič et al., 2017). Therefore, the atomic details of FlaA/FlaB and the mechanism of binding to FliS are important. We first attempted to express the full-length FlaA and FlaB, as well as a C-terminal fragment encompassing amino acid residues 416–510 of FlaA (FlaAc) and 416–514 of FlaB (FlaBc). Both full-length FlaA and FlaB formed aggregates when expressed and purified individually, which precluded further analysis. We next tested the co-expression of GST-tagged FliS with FlaAc and FlaBc. Notably, the presence of FliS enhanced the solubility of the flagellins, especially FlaBc. The binary complex of FliS-FlaBc was purified using glutathione sepharose and size-exclusion chromatography prior to crystallization studies. However, no protein crystal was obtained. Previously, we reported that HP1076 improved the protein solubility of FliS mutants in which the hydrophobic residues in the interior of the helix bundle had been mutated. We therefore reasoned that HP1076 might also promote the stability of the FliS-FlaBc complex and would enable molecular packing for crystallization. We formed a ternary complex by incubating the FliS-FlaBc complex with HP1076, and purified the new complex using size-exclusion chromatography.

The FliS-FlaBc-HP1076 complex was crystallized in the space group P6₁, with two ternary complexes per asymmetric unit. The crystal structure was resolved at 2.95 Å with an R factor and R_free value of 19.0% and 24.7%, respectively (Figure 1, Supplementary Table 1). The two FliS-FlaBc-HP1076 complexes in the asymmetric unit were virtually identical. The final structure comprised residues 18–122 of FliS, residues 476–514 of FlaB and residues 22–147 of HP1076. The electron density map was clearly defined throughout the structure, except for the regions comprised of residues 1–17 and 123–126 of FliS, residues 416–475 of FlaBc and residues 1–21 and 148–171 of HP1076. FliS formed a four-helix up-and-down bundle structure that resembled its apo-form (PDB ID 3IQC), with a root-mean-square deviation (rmsd) of 0.50 Å. The side chains of residues Leu23, Tyr28, Arg33, Arg56, Glu63, Asn108, and Glu119 adopted a different rotamer conformation to accommodate the binding of FlaBc. Regarding HP1076, superimposition upon our previously solved FliS-HP1076 complex (PDB ID:3K1I) (Lam et al., 2010) yielded an rmsd value of 0.67 Å, with no significant structural difference.

The structure of FlaBc comprised four helices surrounding the core of the FliS helix bundle (Figures 1, 2). FlaBc Helix α1 originated from the last two residues of the spoke region before the D0 domain. Helices α1 and α2 were oriented in a half-opened hairpin structure and interacted with helices α3 and α4 of FliS. Helix 3 was embedded in the flagellar binding pocket, which was consistent with other FliS-FliC structural models (Evdokimov et al., 2003; Altegoer et al., 2018; Xing et al., 2018). Helix 4 was located near the faces of helices α1 and α2 of FliS.

The binding interface of FliS and FlaBc was analyzed using PDBePISA (Krissinel and Henrick, 2007). This heterodimer contained an extensive buried interfacial area of approximately 1,500 Ų per molecule. The interactions of FliS and FlaBc were stabilized by extensive hydrogen bonding between the side chains and main chains, including Tyr28_FliS: Gln501_FlaBc, Arg33_FliS:Leu513_FlaBc, Arg56_FliS:Leu512_FlaBc, Leu64_FliS:Gln501_FlaBc, Ile67_FliS:Asn503_FlaBc, Asp69_FliS: Ala500_FlaBc, Gly73_FliS:Ser496_FlaBc, Tyr79_FliS:Glu481_FlaBc, Tyr84_FliS:Gln501_FlaBc, Gln87_FliS:Ser482_FlaBc and Asn108_FliS: Phe485_FlaBc, as well as by the salt bridge Glu119_FliS: K487_FlaB (Figure 2). The residues on the binding interface were highly conserved, suggesting that the interaction of FlaA with FliS would be similar. Notably, the superimposition of FliS-FlaBc on the FliS-FliC complexes from A. aeolicus, B. subtilis, and S. Typhimurium revealed structural differences. One FliS-FliC binding interface consisted of helical stacking between the α1 of FliC and helices α1 and α2 of FliS. In FliC, however, the extended helix α1 is separated into two distinct helices (α1 and α2), whereas this structure was folded in a hairpin in FlaBc (Figures 3A,B). We also examined whether crystal packing could induce the structural variation observed in FlaBc. Although we observed a symmetrical equivalent next to helix α2 of FlaBc, molecular packing only involves helix α2, and the solvent environment was sufficiently large to accommodate an extended helical structure (Supplementary Figure 1). Given the relatively high structural and sequence homology of the D0 domain of FliC across other bacterial species, it is interesting that the helical conformation of FlaB changed upon interaction with FliS in H. pylori. As the hinge of the hairpin did not contain a glycine or proline residue, we speculated that the hairpin fold was guided by the groove generated by the variable residues Tyr79 and His86 of FliS (Figure 3C). The insertion of Phe485 of FlaB into the hydrophobic pocket of FliS may further promote the binding of the hairpin structure to FliS.

A previous study demonstrated similarities in the fold of HP1076 with those of a flagellin homologue, hook associated protein and FliS (Lam et al., 2010). Given the increasing number of structures deposited into the Protein Data Bank in the last decade, we decided to re-analyze the structural homology of HP1076 using the DALI server (Holm and Rosenström, 2010). The top hits from our homology search included the flagellin from S. Typhimurium and B. subtilis in complex with TLR-5 PDB ID: 5GY2; Z-score: 11.8; rmsd: 2.4 and PDB ID: 3V47: Z-score 11.8; rmsd 2.2, respectively) and flagellin (PDB ID: 4NX9; Z-score: 11.4; rmsd: 2.4). Superimposition of the structures of HP1076 and FliC revealed that both interacted with helix α2–α3 of FliS (Figures 4A,B). However, these proteins formed hydrogen bonds with FliS at different residues, whereas Lys50 and Asn93 were the only common residues. Given the high structural similarity of HP1076 with the flagellin D1 domain and the FliS binding interface, we hypothesized that HP1076 and FlaA/B may competitively bind to FliS in H. pylori. To test this hypothesis, we performed pull-down experiments using GST-FliS and H. pylori lysates and included two truncated HP1076 fragments, HP1076ΔN20 (with the deletion of 20 residues from the N-terminus) and HP1076ΔN20ΔC29 (with the deletion of 20 and 29 residues from the N- and C-terminus, respectively), as well as full-length HP1076. Notably, HP1076ΔN20 retains the ability to interact with FliS, whereas HP1076ΔN20ΔC29 exhibited a defective ability to bind to FliS (Lam et al., 2010). We first validated the interaction between FliS and HP1076 before adding the H. pylori lysate (Figure 4C). In the absence of HP1076, the GST-FliS pull-down assay yielded a protein band with a molecular weight of approximately 66 kDa. Mass spectrometry was used to determine that this band contained FlaA and FlaB (Supplementary Figure 2). However, the intensity of this band was reduced when the pull-down assays were performed in the presence of HP1076 and HP1076ΔN20. Yet, the addition of HP1076ΔN20ΔC29 to the pull-down assay did not affect the FlaA/FlaB band intensity. We repeated the pull-down assays by titrating different molar ratios of HP1076 to demonstrate further that HP1076 and FlaA/FlaB competitively bind FliS (Figure 4D). The results indicated that the interaction between FliS and FlaA/FlaB was hindered in the presence of HP1076 in a concentration-dependent manner. Furthermore, we performed microscale thermophoresis assays to compare the binding affinities between FliS and HP1076 and between the FliS-FlaBc complex and HP1076 (Figure 4E). The Kd of the interaction between FliS and HP1076 was approximately 15 nM. However, the Kd of the interaction between FliS-FlaB and HP1076 increased by approximately 10⁵-fold to 1.6 mM. Taken together, these results suggest that HP1076 and FlaA/FlaB share a binding interface on FliS.

To understand the functional importance of HP1076, we performed a molecular genetics study wherein we constructed hp1076-null and fliS-null mutants of H. pylori strain G27. Consistent with the results of previous studies, the deletion of fliS led to non-flagellation or shorter flagellar filaments than those in the wild-type strain (Yokoseki et al., 1995). Impaired bacterial motility was also observed in the fliS-null strain (Figures 5A,B). However, the disruption of hp1076 had no effect on flagellation, and the hp1076-null and wild-type strains exhibited similar swarming behaviors, suggesting that HP1076 may not be associated directly with flagellar synthesis and export. We further investigated the functional importance of HP1076 in H. pylori by examining its cellular localization. We performed an immunoblot analysis of the various cell fractions, using heat shock protein and outer membrane protein as cytosolic and membrane protein markers, respectively. We detected HP1076 predominantly in the membrane fraction (Figure 5C). We also performed a liquid chromatography–mass spectrometry analysis (in duplicate) of the cellular localization of HP1076 to profile the proteins in the membrane and cytosolic/periplasmic fractions extracted from H. pylori G27. Notably, HP1076 peptides were identified solely in the membrane fraction (data not shown). Because HP1076 forms a stable complex with the export chaperone FliS, we hypothesized that FliS may be required for the membrane localization of HP0176. We repeated the cell fractionation assays using a fliS-null mutant strain (Figure 5C). In the absence of FliS, HP1076 did not accumulate in the cytosol. These findings indicate that the membrane association of HP1076 is independent of FliS.