The unique trimeric assembly of the virulence factor HtrA from Helicobacter pylori occurs via N-terminal domain swapping

Knowledge of the molecular mechanisms of specific bacterial virulence factors can significantly contribute to antibacterial drug discovery. Helicobacter pylori is a Gram-negative microaerophilic bacterium that infects almost half of the world's population, leading to gastric disorders and even gastric cancer. H. pylori expresses a series of virulence factors in the host, among which high-temperature requirement A (HpHtrA) is a newly identified serine protease secreted by H. pylori. HpHtrA cleaves the extracellular domain of the epithelial cell surface adhesion protein E-cadherin and disrupts gastric epithelial cell junctions, allowing H. pylori to access the intercellular space. Here we report the first crystal structure of HpHtrA at 3.0 Å resolution. The structure revealed a new type of HtrA protease trimer stabilized by unique N-terminal domain swapping distinct from other known HtrA homologs. We further observed that truncation of the N terminus completely abrogates HpHtrA trimer formation as well as protease activity. In the presence of unfolded substrate, HpHtrA assembled into cage-like 12-mers or 24-mers. Combining crystallographic, biochemical, and mutagenic data, we propose a mechanistic model of how HpHtrA recognizes and cleaves the well-folded E-cadherin substrate. Our study provides a fundamental basis for the development of anti-H. pylori agents by using a previously uncharacterized HtrA protease as a target.

The current paradigm for treatment of bacterial infection is to eradicate bacterial pathogens with antibiotics. How-ever, rapid evolution and dissemination of antibiotic resistance among pathogens pose a huge threat to human health worldwide (1). Recently, antivirulence strategies have been proposed as an alternative for the development of new antimicrobials (2,3). The strategy seeks to interfere with the bacterial virulence factors that promote infection without threatening their existence. This leads to reduced selective pressure for drug-resistant mutations. Notably, development of antivirulence drugs requires an in-depth understanding of the structures and functions of virulence factors in disease processes.
Helicobacter pylori, a Gram-negative microaerophilic bacterium that colonizes the human stomach and is the leading cause of human gastric disease, such as peptic ulcers, gastritis, and even gastric cancer (4), produces a series of virulence factors, e.g. the oncoprotein CagA (5), the vacuolating cytotoxin VacA (6), and urease enzyme (7). Some virulence factors have been recognized as potential drug targets to eradicate H. pylori infection (8,9). High-temperature requirement A protein of H. pylori (HpHtrA) is a newly identified virulence factor that helps H. pylori to efficiently break through the gastric epithelium by cleaving proteins within the epithelial tight junction (occludin and claudin-8) and adherens junction (E-cadherin) (10,11). Structural and functional characterization of this virulence factor may facilitate the design of new types of anti-H. pylori drugs.
HtrA homologs, which are widely found in prokaryotic and eukaryotic organisms, represent a class of highly evolutionarily conserved heat shock-induced serine proteases and chaperones (12)(13)(14). HtrA proteases are composed of an N-terminal signal peptide, a trypsin-like serine protease core domain, and a C-terminal PDZ (15) domain. Based on the different domain organization, HtrA family proteins can be classified into three groups. The members in group 1 contain only one protease and one PDZ domain, such as Escherichia coli DegS (EcDegS) and HtrA2 in mammals (16,17). Proteins in group 2 have one protease domain and two PDZ domains, including HpHtrA, E. coli DegP (EcDegP), and E. coli DegQ (EcDegQ) (18,19). Those in group 3 contain two protease domains and four PDZ domains, such as the recently reported Nma111p (20). EcDegP is a wellcharacterized group 2 protease. Crystal structures revealed that the inactive EcDegP is a hexamer. The basic EcDegP trimeric unit is composed of three monomers that closely interact with each other using the protease domains (21,22). The trimer was further assembled into a hexamer in a manner of staggered dimers of trimers (23,24). When it engages with the proper substrates, EcDegP forms a higher-order multimer and exhibits protease activity (18,22,25,26).
It is known that HtrA family proteases strictly act in the periplasm and play a vital role in protein quality control (27,28). However, recent studies demonstrated that H. pylori actively secretes HtrA into the extracellular environment, where it cleaves the extracellular domain of the epithelial cell surface adhesion protein E-cadherin, facilitating H. pylori transmigration across gastric epithelial cells (29 -32). HtrA in H. pylori contains a signal peptide that is important for Sec-dependent cleavage and transport of the protease across the inner membrane into the periplasm (13,14,27,33). Although HtrA has been reported to be enriched in H. pylori outer membrane vesicles, the detailed mechanism of how it is transported across the outer membrane remains unclear (34). E-cadherin is a single-transmembrane protein consisting of five extracellular (EC) domains, 3 an intracellular domain, and a transmembrane domain (35). EC domains adopt a calcium-dependent homophilic interaction to mediate intercellular adhesion between epithelial cells (36). Edman degradation and mass spectrometry-based proteomics demonstrated that the [VITA]-[VITA]XXD[DN] signature motifs located at the linker regions of the EC domains are preferentially cleaved by HpHtrA (37). Importantly, inhibition of HpHtrA activity by chemical compounds efficiently blocked H. pylori transmigration across the gastric epithelial barrier (38,39). Moreover, generation of an htrA knockout mutant in H. pylori is fatal, implying that HpHtrA is critical for H. pylori survival (40,41). Therefore, HpHtrA has been considered a potential attractive drug target, given its vital role in the pathogenesis and survival of H. pylori (42).
Here we report the crystal structure of HtrA from H. pylori. The structure reveals a trimeric HpHtrA containing a unique extended N terminus that is absent in other HtrA homologs. Importantly, the N terminus bridges HtrA monomers through domain swapping to stabilize the homotrimer, which has not been identified in other HtrA homologs reported so far. We further demonstrate that the N terminus is also essential for HpHtrA oligomer assembly and protease activity. Moreover, a key residue, Lys 328 , in the HpHtrA PDZ1 domain was identified to be essential for E-cadherin cleavage but is dispensable for unstructured ␤-casein substrate. All of these data reveal a novel type of HtrA family protease secreted by the human pathogen H. pylori.

The N terminus of HpHtrA is critical for protease activity
Sequence alignments of HtrA homologs from different bacterial species show that HpHtrA has almost the same domain composition as other members, including a signal peptide, a protease domain, and two PDZ domains (Fig. 1A). However, HpHtrA has a relatively extended N-terminal region that contains 19 residues after the signal peptide ( 18 GNIQIQSMPKVK-ERVSVDP 35 ) (Fig. 1B). To investigate the function of each domain, a series of domain truncation mutants of HpHtrA was constructed and purified (Fig. S1), including a mutant with the N-terminal 19 residues truncated (HpHtrA-⌬N), a mutant with the PDZ2 domain truncated (HpHtrA-⌬PDZ2), and a mutant with both PDZ domains truncated (HpHtrA-⌬PDZ1-2). The protease activity of the mutants was measured using a nonspecific substrate, ␤-casein, as described previously (32). As shown in Fig. 1C, proteolytic activity of the HpHtrA-⌬PDZ1-2 mutant was completely abolished. In contrast, the HpHtrA-⌬PDZ2 mutant exhibited even higher protease activity compared with the WT HpHtrA (residues from Gly 18 to Lys 475 without the signal peptide). The results are consistent with previous reports showing that the PDZ1 domain is essential for protease activity and responsible for recognizing and sequestering unfolded substrates through C-terminal residues, whereas the PDZ2 domain is mostly involved in maintaining the hexameric cage of DegP and displays an inhibitory role in chaperone and protease activity (18,43,44). Intriguingly, removal of the short N terminus also abrogated HpHtrA protease activity, indicative of an essential role of the N terminus in HpHtrA protease activity. The proteolytic activity of the mutants was subsequently measured using E-cadherin as a substrate. In contrast, all mutants showed decreased proteolytic activity against E-cadherin compared with WT HpHtrA, implying that all of these domains might participate in the E-cadherin cleavage process (Fig. 1D).

The N terminus stabilizes the trimeric form of HpHtrA
In the absence of substrate, two EcDegP trimers stack face to face to form an inactive hexamer (23); EcDegS and EcDegQ, two extensively studied HtrA family proteins in E. coli, were identified as homotrimers (19,45). To characterize the conformational state of HpHtrA in the absence of substrate, the oligomerization states of WT-HpHtrA and different mutants were examined by size-exclusion chromatography. As shown in Fig.  2A, WT HpHtrA was eluted at 11.5 ml with a calculated molecular mass of 170 kDa, consistent with a trimeric form of HpHtrA (52 kDa for a monomer). HpHtrA-⌬PDZ2 and HpHtrA-⌬PDZ1-2 were both eluted as trimers with calculated molecular masses of 120 and 90 kDa, respectively. Unexpectedly, HpHtrA-⌬N was eluted at 14.9 ml with a molecular mass of 50 kDa, indicative of a monomeric state of the HpHtrA-⌬N mutant. It has been reported that an autoproteolytic process that results in cleavage of the N-terminal region of EcDegP led to destabilization of a hexamer and formation of a monomer of EcDegP (46). However, the autocleavage process removed the N-terminal 60 -80 residues of EcDegP, in contrast to truncation of the N-terminal 19 residues in HpHtrA. The results indicate that the N terminus is essential to stabilize the HpHtrA trimer. On the other hand, the trimeric form of HtrA is likely the minimal functional unit with prominent protease activity because activation of HtrA required the substrate sensor loop from one monomer to interact with the activation loop of a neighboring monomer (28,47). Therefore, truncation of the N terminus of HpHtrA disrupted its trimeric form and significantly abolished its protease activity.

Structural characterization of HpHtrA
To further understand how the N terminus stabilizes trimeric HpHtrA, we first determined the crystal structure of fulllength WT HpHtrA at a resolution of 3.7 Å (PDB code 5Y2D). The full-length WT HpHtrA crystals grew in trigonal space group R32 and contained one monomeric HpHtrA molecule per asymmetric unit. In the full-length HpHtrA structure, only the protease and PDZ1 domain are well defined with clear electronic density, whereas the PDZ2 domain is partially visible (Fig. 2B). Intriguingly, weak but evident electron density was observed for the extended N terminus of HpHtrA, which is usually disordered and invisible in other HtrA homolog structures (22,23,26,47,48).
To obtain a structure with higher resolution, we crystallized HpHtrA-⌬PDZ2. The HpHtrA-⌬PDZ2 structure was solved by molecular replacement and refined to 3.0 Å resolution (PDB code 5Y28). The crystal form belongs to the orthorhombic P222 space group and contains one HpHtrA-⌬PDZ2 trimeric molecule per asymmetric unit. In this structure, each HpHtrA monomer is composed of a protease domain and a C-terminal PDZ1 domain. However, only two PDZ1 domains are visible with clear electron densities among three HpHtrA monomers. Typically, HpHtrA monomers are packed symmetrically around a 3-fold molecular axis as a funnel-like shape, with the protease domain forming the core and the PDZ1 domain extending outward (Fig. 2C). The HpHtrA protease domain is similar to other proteases of the trypsin family, with two perpendicular ␤ barrel lobes and a C-terminal ␣ helix. The backbone root mean square deviation between the EcDegP and HpHtrA protease domains is 0.83 Å. The catalytic triad is located in a crevice between two ␤ barrel lobes, including residues His 116 , Asp 147 , and Ser 221 (Fig. 2D). The relative orientation between the PDZ1 and protease domains is almost the same in both full-length HpHtrA and HpHtrA-⌬PDZ2 structures (Fig. S2). It has been reported that the PDZ1 domain tilts away from the protease domain upon EcDegP activation. Structural alignments of HpHtrA with inactive (PDB code 1KY9) and active (PDB code 3CS0) EcDegP demonstrate that the overall conformation of HpHtrA protease and the PDZ1 domain is identical to the inactive form, with a root mean square deviation of all Cɑ atoms of 1.6 Å (Fig. S1).
Intriguingly, the N-terminal region of HpHtrA (Gly 18 -Tyr 44 ) has clear electron density in all HpHtrA-⌬PDZ2 monomers, which allows the N terminus model to be built unambiguously (Fig. S3). The N terminus of HpHtrA is stretching out from the protease domain and is mainly composed of two short ␤ strands and two unstructured loop regions: loop1 (Gln 21 -Lys 29 ), ␤1

Unique trimerization pattern of H. pylori HtrA
(Glu 30 -Val 32 ), ␤2 (Thr 40 -Ser 43 ), and loop2 (Ser 33 -Asp 39 ). Unexpectedly, the crystal structure reveals that the extended N terminus is involved in a domain-swapping event wherein the N terminus transverses the interface to the two neighboring monomers; i.e. loop1 of monomer A interacted with the protease domain of monomer C, whereas the ␤1 and ␤2 of monomer A formed parallel ␤-sheets with ␤2 of monomer B and ␤1 of monomer C, respectively (Fig. 3A). Because of the relatively low local resolution, the side chains of residues at loop1 could not be unambiguously identified. However, loop1 has extensive contact with the neighboring monomer protease domain, as revealed by detailed protein interface analysis (Fig. S4). Biochemical data demonstrated that truncation of loop1 also abrogated the HpHtrA trimer into a monomer (Fig. S5A). The parallel ␤-sheets formed by two ␤-strands from neighboring monomers are connected by interstrand backbone hydrogen bonds. In particular, the side chain of Arg 31 from monomer A stretched out and formed a typical salt bridge with Asp 173 from monomer C. This intermolecular Arg 31 -Asp 173 salt bridge is highly conserved in all three monomers ( Fig. 3B and Fig. S6). It is worth noting that the N-terminal domain swapping of HpHtrA contributes to majority of protein-protein interfaces in HpHtrA. The average interface area between two neighbor-ing monomers in the HpHtrA trimer is 1656 Å 2 , which is significantly larger than that in EcDegP and EcDegQ. However, truncation of the N-terminal 18 residues (including loop1 and the ␤1 strand) and 26 residues (including loop1, the ␤1 strand, loop2, and the ␤2 strand) dramatically reduced the interface areas to 690 Å 2 and 446 Å 2 , respectively (Fig. 3C), which is consistent with experimental results showing that truncation of the corresponding N-terminal regions causes trimer disassembly.
The data suggest that N-terminal domain swapping is important to stabilize the HpHtrA homotrimer, which is distinct from other HtrA homologs. In previously reported HtrA homolog structures, including EcDegP and EcDegQ, formation of a homotrimer is exclusively mediated by intersubunit hydrophobic interactions involving the first ␣-helix and two ␤-strands of the protease domain rather than domain swapping (Fig. 3D) (23). Moreover, urea denaturation curves revealed that HpHtrA and EcDegQ had a denaturation midpoint of 5.0 M and 4.5 M urea, respectively (Fig. 3E), indicating that HpHtrA is relatively more stable than EcDegQ, which is consistent with the fact that HpHtrA has larger intersubunit interface areas. Therefore, HpHtrA represents a novel type of homotrimer of HtrA family proteases stabilized by unique N-terminal domain swapping.

Substrate binding triggers HpHtrA oligomer formation
It has been demonstrated that substrate binding to EcDegP and EcDegQ induces proteolytically active oligomer formation (18,19,22,48). To examine whether HpHtrA employs a similar activation mechanism, we first analyzed the HpHtrA oligomeric state in the presence of ␤-casein, which is a well-characterized unfolded model substrate for the HtrA family (49). Similar to EcDegP and EcDegQ, size-exclusion chromatography analysis revealed that incubation of HpHtrA with different molar equivalents of ␤-casein led to formation of a higher-order oligomer, i.e. a substrate-engaged 12-mer (HpHtrA12) and 24-mer (HpHtrA24) (Fig. S7A). The sizes of the formed complexes are dependent on the ␤-casein substrate concentration. At lower substrate concentration, the 12-mer is predominantly formed, whereas HpHtrA24 was observed with increasing concentration of substrate.
Time-dependent analysis of the HpHtrA proteolytic products by size-exclusion chromatography showed that ␤-casein cleavage products by HpHtrA had two major groups of elution peaks, with the first peak eluted slightly later than ␤-casein, denoted as the primary product, and several smaller peaks eluted much later, denoted as the secondary products (Fig. 4A). In the course of time, the amounts of primary cleavage product were decreased, accompanied by an increase of the secondary cleavage products. The amount of HpHtrA12 was also decreasing during the cleavage process, indicating that the proteolyti-cally active HpHtrA oligomer dissociated after substrate cleavage (Fig. S7B). In contrast, incubation of HpHtrA-⌬N with ␤-casein yielded no detectable HpHtrA oligomers or detectable cleavage products (Fig. 4B), suggesting that monomeric HpHtrA-⌬N loses the capability to assemble into the active oligomer even in the presence of excess amount of ␤-casein.
The cleavage sites of E-cadherin by HpHtrA were located at the linker regions of EC domains. In particular, a 21-residue P1 peptide (Ac-TGTLLLILSDVNDNAPIPEPR-COOH) derived from the cleavage site between E-cadherin domains EC4 and EC5 could bind directly to HpHtrA (37). Therefore, we further investigated the binding between P1 peptide and HpHtrA. No high-order oligomer complex was observed in size-exclusion chromatography when HpHtrA was incubated with an excess amount of P1 peptide, which was probably due to the low binding affinity. To better characterize the interaction, atomic force microscopy (AFM) imaging experiments were carried out to capture the oligomerization state of HpHtrA upon P1 peptide binding. Deposition of HpHtrA alone onto freshly cleaved mica, followed by AFM imaging, resulted in distribution of the protein with a typical triangular shape, which is consistent with the trimeric form, as revealed by the crystal structure (Fig. 4C). Incubation of HpHtrA with P1 peptide caused substantial changes in protein particle shape in AFM imaging, in which the assembly heights of complex particles are significantly larger than that of HpHtrA alone, indicative of oligomerization of

Unique trimerization pattern of H. pylori HtrA
HpHtrA in the presence of P1 peptide (Fig. 4C). Collectively, we demonstrate that HpHtrA is able to assemble and disassemble dynamically to form a proteolytically active oligomer, which is promoted by binding of an unstructured substrate.

A lysine residue of HpHtrA is important for E-cadherin cleavage
In contrast to ␤-casein or P1 peptide, E-cadherin is a wellfolded, rigid substrate for HpHtrA. To investigate the mechanism of E-cadherin cleavage by HpHtrA, a protein complex structural model was built by docking E-cadherin EC1-EC2 domains to the HpHtrA timer. In the docking model, the PDZ1 domains of two HpHtrA monomers work as clamps to bind the EC1-EC2 domains (Fig. S8A). Both of the EC domains have contact interfaces with the HpHtrA PDZ1 domains, which are closed to the identified PDZ1 substrate binding groove formed by ␤-strand A and ␣-helix B (28). Typically, two lysine residues, Lys 326 and Lys 328 from the PDZ1 domains, are proximal to the EC1 and EC2 domains in the complex model, implying that the two residues are possibly involved in EC1-EC2 domain recognition (Fig. S8B). To test the hypothesis, two HpHtrA mutants (HpHtrA-K326A and HpHtrA-K328A) were purified, and their substrates cleavage activities were investigated. Both mutants maintained similar ␤-casein cleavage activity compared with WT-HpHtrA. Intriguingly, the two mutants exhibited different cleavage activities for substrate E-cadherin. The HpHtrA-K326A mutant exhibited similar E-cadherin proteolytic activity compared with WT HpHtrA. In contrast, significantly attenuated E-cadherin cleavage activity was observed for the HpHtrA-K328A mutant (Fig. S9). Time-course substrate cleavage assay results also confirmed that K328A mutagenesis attenuated HpHtrA proteolytic activity for E-cadherin but not for ␤-casein (Fig. 5, B and C). The data suggest that Lys 328 of the HpHtrA PDZ1 domain is critical for E-cadherin cleavage, whereas it is dispensable for ␤-casein proteolysis. The complex structural model here may represent the first snapshot of the binding and recognition of E-cadherin substrate by HpHtrA via the Lys 328 residue of the PDZ1 domains, and binding of E-cadherin would further induce a conformational change of HpHtrA, facilitating cleavage of bound E-cadherin at the domain linker region.

Phylogenetic analysis and cellular localization of HpHtrA
In E. coli, three members of the HtrA family have been identified: EcDegP, EcDegQ, and EcDegS. The three proteases exert different bacterial physiological functions (48,50,51). In contrast, only one HtrA homolog has been identified in H. pylori, implying that HpHtrA could be multifunctional in H. pylori. Sequence alignment of HtrA family proteins from several eukaryotic and prokaryotic species revealed that a similar extended N terminus can be found in Campylobacter jejuni HtrA (CjHtrA), which is also an identified protease secreted by human pathogenic bacteria. Intriguingly, the N terminus of HpHtrA exhibits high sequence similarity with the N-terminal region of DegS homologs, which is part of the transmembrane domain (Fig. S10). Further phylogenetic analysis of HtrA family proteins also indicates that HpHtrA is more closely related to DegS than to DegP, implying a potential evolutionary relationship between DegS and HpHtrA (Fig. 5D).

HpHtrA is a virulence factor secreted by H. pylori
HpHtrA contains a signal peptide for Sec-dependent transport across the inner membrane into the periplasm (52). However, it is still unknown whether HpHtrA is simultaneously transported across the out membrane when it enters the periplasm. Therefore, we investigated the cellular localization of HpHtrA with an HpHtrA-specific antibody. As shown in Fig.  5E (Fig. S11), clear HpHtrA protein bands were visible in the bacterial culture medium and soluble fraction after cell lysis.

Unique trimerization pattern of H. pylori HtrA
The soluble fraction should include HpHtrA from both the bacterial cytoplasm and periplasm. Surprisingly, substantial amounts of HpHtrA were also identified in the inner membrane fraction, implying that HpHtrA might directly anchor to the inner membrane or bind tightly to an inner-membrane protein.
The results demonstrate that the H. pylori periplasm and inner membrane may serve as temporary reservoirs for HpHtrA before its extracellular secretion.

Discussion
It is well established that H. pylori translocates a virulence factor, CagA, into the gastric epithelial cell cytoplasm via a type IV secretion system (53,54). Such a process depends on interaction of the bacterial type IV secretion system with the host cell surface ␣ 5 ␤ 1 receptor (55)(56)(57). Recent studies demonstrated that H. pylori utilizes a novel secreted serine protease, HtrA, to cleave the host Occludin, Claudin-8, as well as E-cadherin proteins, which breaks down the E-cadherinbased adherens junctions and tight junctions between gastric epithelial cells to disintegrate the epithelial barrier. After cleavage, H. pylori can efficiently enter the intercellular space and interact with the exposed ␣ 5 ␤ 1 receptors for virulence factor translocation (11,29,37). The HtrA protein was initially identified as a protease, functioning in the bacterial periplasmic space for protein quality control. Recent studies demonstrated that HtrA-mediated host cell E-cadherin cleavage is a prevalent pathogenic mechanism for multiple Gram-negative bacterial species, indicating that bacterial

Unique trimerization pattern of H. pylori HtrA
HtrA could serve as an attractive target for the design of antibacterial agents (30,38,39).
Here we determined the crystal structure of H. pylori HtrA protease, which possesses an extended N terminus different from other homologs. In the crystal structure, three HpHtrA monomeric molecules assemble into a new type of trimer via unique N-terminal domain swapping. Importantly, the HpHtrA trimer is mainly stabilized by N-terminal domain swapping. This assembly pattern is distinct from that of the E. coli homologs DegP and DegQ, both of which are stabilized by hydrophobic interactions of the protease domains. Truncation of the N terminus completely abrogated HpHtrA trimer formation, leading to monomeric HpHtrA without detectable protease activity, indicative of the essential role of the N terminus. A recent study of HpHtrA demonstrated that autocleavage of the HpHtrA N terminus abolished HpHtrA secretion and protease activity, which is consistent with our protein structure data (58). Although it is still unknown why HpHtrA adopts such a unique mechanism for protein trimer assembly, one plausible explanation is that larger intersubunit interface areas contributed by domain swapping enhance HpHtrA trimer stability so that HpHtrA can sustain protease activity after secretion into the hostile gastric niche (59). Therefore, targeting the N terminus to abolish HpHtrA trimer formation may represent a new potential anti-H. pylori strategy.
Similar to other HtrA homologs, HpHtrA could assemble into a proteolytically active oligomer in the presence of ␤-casein substrate. Similar assembly was also observed when HpHtrA was incubated with P1 peptide, which is derived from the E-cadherin cleavage site. However, HpHtrA oligomerization is unlikely to happen when HpHtrA binds to the E-cadherin ectodomain. Unlike unstructured ␤-casein or peptide, E-cadherin is a rigid protein with a well-folded structure, which could prevent the higher multimer active cage formation of HpHtrA. Previous studies demonstrated that EcDegP oligomeric cage assembly is not required for its proteolytic activation (60). It would not be surprising if HpHtrA could cleave E-cadherin in the trimeric form. Therefore, the recognition and cleavage mechanism of E-cadherin by HpHtrA should be different from that of ␤-casein. Indeed, our biochemical studies identified that Lys 328 of HpHtrA was indispensable for E-cadherin cleavage but not for ␤-casein.
Although HpHtrA contains two PDZ domains similar to EcDegP and EcDegQ, phylogenetic analysis reveals a closer evolutionary relationship between HpHtrA and EcDegS. EcDegS is a serine protease anchored on the E. coli inner membrane and involved in the cellular response to extracytoplasmic stress via activation of the E. coli factor E (61). Cellular localization analysis revealed that a substantial amount of HpHtrA was also identified in the bacterial inner membrane. Although it is unknown how HpHtrA is attached to the inner membrane, it is possible that H. pylori adopts a strategy to preserve HpHtrA so that export of the virulence factor could occur by a triggered mechanism, which avoids unnecessary secretion. Therefore, it would be interesting to further investigate whether the cellular localization of HpHtrA is functionally relevant.

Conclusion
In summary, we report the first crystal structure of the H. pylori HtrA trimer stabilized by unique N-terminal domain swapping, which represents an unprecedented novel assembly of HtrA family proteases. The N terminus of HtrA from H. pylori is also indispensable for its protease activity. The molecular mechanism of how HpHtrA recognizes and cleaves E-cadherin was elucidated based on a structural model that is distinct from that of the conventional substrate ␤-casein. Given that HpHtrA is a promising target for the design of anti-H. pylori agents, the structure we report here could facilitate the development of a new type of anti-H. pylori agents by targeting HtrA both at the active site and N terminus.

Protein expression and purification
Details regarding expression and purification of H. pylori HtrA proteins can be found in the supporting Methods. In brief, N-terminal His-tagged H. pylori WT or mutant HtrA proteins lacking the signal peptide sequence were expressed in the E. coli BL21 (DE3) strain and purified by nickel affinity chromatography followed by gel filtration. Recombinant proteins were pooled and stored at Ϫ80°C. All the PCR primers were listed in Table S1.

Protein crystallization and structure determination
Crystals of full-length HpHtrA were obtained by sitting drop diffusion at 20°C and mixing equal volumes of the protein and the reservoir solution, consisting of 2.1 M DL-malic acid, and 0.1 M HEPES (pH 7.0). For HpHtrA-⌬PDZ2, crystals were obtained similarly, with the reservoir solution containing 20% PEG 1500, 0.1 M HEPES (pH 7.5), and 0.2 M proline. Crystals were cryoprotected with reservoir solution supplemented with 10% glycerol and flash-frozen in liquid nitrogen. Diffraction data were performed at the Shanghai Synchrotron Radiation Facility (Shanghai, China) using beamlines 17B, 17U1, and 19U1. Raw data images were processed with HKL2000 (62). Molecular replacement solution was obtained from the PHENIX program using E. coli DegP (PDB code 1KY9) as a search model (63). Subsequent model building and refinement were carried out in COOT (64) and PHENIX (Table S2). The figures were prepared using PyMOL (Schrödinger, LLC).

HpHtrA proteolytic activity
HpHtrA proteolytic assays were performed as described previously (65). For the time-course ␤-casein cleavage assay, 40 nM WT HpHtrA or mutants was incubated with 3 g of ␤-casein in reaction buffer (20 mM Tris-HCl (pH 7.5), 250 mM NaCl, and 1 mM DTT) at 37°C. At the indicated times, aliquots of the reactions were removed for SDS-PAGE analysis. For the E-cadherin cleavage assay, ϳ0.8 M WT HpHtrA or mutants was incubated with 50 ng of the N-terminal domain of E-cadherin in reaction buffer for 16 h at 37°C. The remaining substrates were blotted with an antibody to E-cadherin. The SDS-PAGE and Western blot band intensities were quantified using ImageJ software (66).

Unique trimerization pattern of H. pylori HtrA
Atomic force microscopy imaging AFM imaging was carried out in QI mode in liquid with a NanoWizard 4 microscope (JPK Instruments AG) equipped with SNL-10 cantilevers (Bruker Probes). Square images of 256 ϫ 256 pixels were collected with a scan size of 500 nm and a set point of 300 pN. For each experiment, ϳ20 l of HpHtrA protein samples (0.01 mg/ml) in the absence or presence of P1 peptide was added onto freshly cleaved mica and allowed to absorb for 15 min. After that, the mica surface was rinsed and submerged in Tris buffer (400 mM Tris-HCl (pH 7.4) and 100 mM NaCl).

Size-exclusion chromatography analysis
Size-exclusion chromatography was performed with a Tricorn Superdex 200 Increase 10/300 GL column (GE Healthcare) at 4°C. The column was calibrated with a gel filtration calibration kit (GE Healthcare) and pre-equilibrated with gel filtration buffer (20 mM Tris-HCl (pH 7.5), 250 mM NaCl, and 1 mM DTT). For oligomerization state analysis, 100-l samples containing 40 M WT HpHtrA or mutants were injected into the column. For substrate-triggered oligomer formation, ϳ100 l of protein mixtures containing 40 mM HpHtrA and 160 mM ␤-casein or E-cadherin EC1-EC2 domain was incubated on ice for the indicated time before loading onto the column.

Structure model of the HpHtrA and E-cadherin complex
An HpHtrA trimer structure model with three intact PDZ1 domains was generated using chain C of the HpHtrA-⌬PDZ2 structure (PDB code 5Y28) as a template. The protein complex structure was built by docking the structure of the human E-cadherin EC1-2 domain (PDB code 4ZT1) to the HpHtrA trimer using the ESCHER protein-protein automatic docking system with default parameters (67). The structure with the lowest energy was selected as the complex model.

Circular dichroism
CD spectra were recorded at ambient temperature on a JASCO J-810 spectrophotometer using a quartz cuvette with a path length of 0.1 cm. Urea denaturation experiments were performed as described previously (68). In brief, urea gradients were added to 20 M WT HpHtrA and EcDegQ in 10 mM Tris-H 2 SO 4 buffer supplemented with 100 mM NaSO 4 (pH 7.5). After equilibrating overnight at 4°C, CD spectra were recorded between 200 and 270 nm with an average of three scans. The absorption at 220 nm was used to characterize the unfolding state of the protein.

Sequence alignment and phylogeny analysis
HtrA family protein sequence alignment and the phylogenetic tree were done using the MAFFT online program (version 7) with default options (https://mafft.cbrc.jp/ alignment/server/). 4 All protein sequences were obtained from the UniProt database, and signal peptide sequences were removed manually.

Analysis of the cellular location of HpHtrA
The separation of inner and outer membranes was carried out according to the method described by Doig and Trust (69) with minor modifications. Details of the experiments are described in the supporting Methods. All cellular fractions were analyzed by Western blotting using an HpHtrA-specific antibody.

Accession codes
The coordinates for full-length HpHtrA and HpHtrA-⌬PDZ2 have been deposited in the PDB with accession codes 5Y2D and 5Y28, respectively.