Crystal structure of the N-terminal domain of the secretin GspD from ETEC determined with the assistance of a nanobody

Structure Determination

Initially, crystal growth of peri-GspD from several species appeared to be very difficult (See Material and Methods). Therefore we decided to follow another path for obtaining crystals, focusing on residues 1-237 from ETEC GspD (hereafter called “peri-GspD”) and exploring the power of nanobodies in promoting crystal growth. The immunization, library construction and selection were performed following procedures described in Experimental Procedures. Three successive rounds of panning yielded nine GspD-specific nanobodies (Figure S1), all of which could be purified in complex with peri-GspD by size exclusion chromatography. Very rapidly, four of these gave hits in sparse matrix crystallization screens. The initial hits of ETEC peri-GspD in complex with NbGspD-7 (hereafter abbreviated as “Nb7”) could be optimized readily and diffraction quality crystals were obtained for two crystal forms (Table 1). The crystal structure of the peri-GspD:Nb7 complex was determined at 2.80 Å resolution using single-wavelength anomalous diffraction phasing with Se-Met labeled GspD. We refined the structure of the better diffracting P3₁21 crystals, with 1271 amino acids per asymmetric unit, to an R_work of 0.192 and an R_free of 0.239 with excellent geometry (Table 1). Both crystal forms reveal the same compact (peri-GspD)₂:(Nb7)₂ heterotetramer with local two-fold symmetry (Figure 1B). ETEC peri-GspD contains eleven β-strands and seven α-helices, and consists of two globular lobes (Figure 1C). The larger N-terminal lobe can be further divided into two subdomains: N0 and N1. The second lobe contains subdomain N2 (Figure 1). The ∼20-residue long loop connecting the N0 and N1 subdomains is disordered in our structure but the linker connecting domains N1 and N2 is well defined.

The N0 subdomain and structural homologs

The first subdomain (N0) is composed of two α-helixes flanked on one side by a mixed three-stranded β-sheet and the other by a two-stranded anti-parallel sheet. DaliLite (Holm and Sander, 1996) and SSM (Krissinel and Henrick, 2004) database searches showed that the N0 subdomain has the same fold as two interesting groups of proteins (Figure 2):

1. the signaling domain of TonB-dependent receptors (Figure 2A) (Brillet et al., 2007; Ferguson et al., 2007; Garcia-Herrero and Vogel, 2005)superimposes onto the N0 subdomain of GspD with a 2.3 Å RMSD for 65 amino acids with only 14% sequence identity between the two chains;

2. bacteriophage proteins of the cell-puncturing needle (Figure 2B,C) (Kanamaru et al., 2002); PDB 3CDD; PDB 1WRU). Despite a low sequence identity of 14% over 71 residues, the GspD N0 subdomain superimposes with a 2.2 Å RMSD onto the 43 kDa tail protein of prophage MuSo2 from Shewanella oneidensis (PDB 3CDD).

The N1 and N2 subdomains and homologs

The previously reported sequence repeat in the N-terminal part of GspD (Pfam family PF03958 (Finn et al., 2006; Fig. 3B in Chami et al., 2005) is reflected in structural homology between the N1 and N2 subdomains which consist of two or three α-helices on one side of a three-stranded β-sheet (Figure 3). These two subdomains can be superimposed with a 1.6 Å RMSD and have 25% sequence identity (Figure 3A, G). The short first helix α3 of the N1 subdomain has 3₁₀-character and is followed by the longer helix α4. The 3₁₀-helix is essentially absent in the N2 domain. The fold of the N1 subdomain is similar to the eukaryotic type I KH (“hnRNP K homology”) domains (Siomi et al., 1993; Valverde et al., 2008), which are typically involved in binding RNA and DNA. The 3₁₀-helix 1 and α-helix 2 of the N1 subdomain provide a structural match with the two tandem helices of the KH domain (Figure 3B,C,D). Whereas the RNA-DNA binding surface of canonical KH-domains contains critical positively charged residues, these charges are not conserved in the N1 and N2 domains of peri-GspD (Arg116/Ser181, Arg123/Glu188; Figure 3G). This is expected since no evidence exists that the T2SS interacts with nucleic acids.

Interestingly, the first domain of the EscJ protein from the type 3 secretion system (T3SS) (Yip et al., 2005) also shows structural similarity with the N1and N2 subdomains of peri-GspD (Figures 3C,F). The N1 subdomain superimposes onto the first EscJ domain with a 3.1 Å RMSD and a same low 11% sequence identity over 54 residues, whereas the N2 subdomain superimposes with a 2.2 Å RMSD and 11% sequence identity over 46 residues. The EscJ domain is located in the periplasm and forms a 24-meric ring that is associated with the inner membrane in the “needle complex” of the T3SS. Further implications of the GspD-EscJ similarity will be discussed below.

Interactions between the N0 and N1 subdomains of GspD

The first repeat unit (N1) interacts with the N0 subdomain to form the first globular lobe of peri-GspD (Figure 1C). The N0:N1 interface (Figure 4A) is substantial with a buried solvent accessible surface of 1180 Ų and a low gap volume index of 1.84 Å (Jones and Thornton, 1996), and includes: (i) an anti-parallel pair of β-strands involving β3 of N0 and β6 of N1, resulting in an extended anti-parallel interdomain β-sheet of in total five strands; (ii) the Arg42-Glu102 salt bridge; and (iii) a cluster of hydrophobic residues including Ile41, Leu47 and Leu55 from N0 and Met103, Val107, Val134 and Met 144 from N1 (Figure 4A). These residues are hydrophobic throughout the T2SS GspD family (Figure 1A), which suggests that the compact N0-N1 lobe of peri-GspD is most likely a common feature in type 2 secretion systems.

The nanobody Nb7 and its interactions with peri-GspD

The structure of Nb7 has the typical compact VHH fold, with a CDR3 loop of 14 residues, a frequently observed length in nanobodies (Vu et al., 1997). Each Nb7 chain in the heterotetramer interacts with: (i) the N0 and N1 subdomains of peri-GspD (Figure 4B,C); and, (ii) the second Nb7 molecule in the GspD:Nb7 heterotetramer, via an anti-parallel arrangement of β-strands (Figure 1B). The N0 and N1 domains contribute 36 and 64%, respectively, to the total buried surface of 1896 A² in the nanobody peri-GspD interface. The gap volume index of 2.74 Å for the peri-GspD:Nb7 interface is better than the average gap volume index of 3.0 Å for antibody-protein complexes (Jones and Thornton, 1996). The contributions of CDR1, CDR2 and CDR3 to the buried surface are 0, 0 and 629 A², respectively. In addition, the framework buries 329 A² in its interface with peri-GspD. Framework involvement in nanobody target protein interactions is rare, but has been seen before such as in porcine pancreatic α amylase-VHH complexes (Desmyter et al., 2002).

Nanobodies and the crystallization of ETEC peri-GspD

Two nanobodies together with two peri-GspD chains form a surprisingly complex heterotetramer (Figure 1B), an arrangement which has, to the best of our knowledge, never before been seen in complexes of target proteins with antibodies or nanobodies. In addition to the nanobody-GspD contacts, interactions between two nanobodies in the tetramer bury 760 Ų of solvent accessible surface, with a tight interface as reflected in a gap volume index of 2.52 Å. Interactions of Nb7 with other molecules in the crystal do occur although these all have a much larger gap volume index. Taking all this together, the prime function of the nanobody in promoting crystal growth is probably formation of the heterotetramer (Figure 1B). Peri-GspD in the tetramer is almost certainly considerably more rigid than peri-GspD by itself given the potentially flexible linker between the N1 and N2 subdomains (Figure 1C).

The N0 subdomain of peri-GspD may interact with a β-strand from another protein

The homology of N0 with the signaling domain of the outer membrane receptor FpvA from P. aeruginosa (Brillet et al., 2007), leads to a suggestion for the function of the N0 domain in T2SS secretion and/or assembly. In apo-FpvA, this signaling domain shows an intramolecular interaction with the so-called “TonB box”. The TonB box spans ∼8 residues, adopts a β-strand conformation, and is engaged in an anti-parallel β-strand arrangement with a β-strand from the mixed three-stranded β-sheet of the FpvA signaling domain (Brillet et al., 2007). When superimposing the FpvA signaling domain plus its TonB box β-strand onto the N0 domain of ETEC-GspD, it appears that an extra β-strand could be bound by the N0 domain in a fashion similar to the interaction between the TonB box and the signaling domain (Figure 2A). That is, the main chain hydrogen bond donors and acceptors of β2 of N0 are solvent accessible and not blocked by, for instance, the tight interactions between the N0 and N1 subdomains. Hence, β2 of GspD might interact with a β-strand from another protein during the functioning of the T2SS, as will be discussed further below.

N3 subdomain structure

A sequence alignment of the N3 subdomain of GspD (Figure 1A) with subdomains N1 and N2 yields sequence identities of 16 and 17 % for 68 and 71 residues respectively (Figure 3G) (Pfam family PF03958 and the schematic Figure 3B of (Chami et al., 2005). The trypsin cleavage sites of K. oxytoca GspD reported by these authors are interesting when mapped onto our structures of N1 and N2, and onto the predicted structure of N3 in ETEC GspD. The internal “nick” in the N-terminal fragment of K. oxytoca GspD, called “PulD-N” by (Chami et al., 2005), would be equivalent to a cut in the peptide bond after Ser236 in ETEC GspD, which corresponds with the penultimate residue in our ETEC peri-GspD construct (Figure 1A). This suggests that the linker between the N2 and N3 domains is flexible. The C-terminal trypsin cleavage site which generated PulD-N, corresponds with Lys272 in ETEC GspD (Figure 1A). The homology between the N1, N2 and N3 subdomains indicates that this residue is located in a long insertion of N3 (Figure 3G) which may explain its accessibility for the protease. Hence, the proteolysis data on the T2SS secretin from K. oxytoca agree well with our structural data. It is likely that the N3 subdomain adopts a similar fold as N1 and N2, and that all three subdomains are connected by flexible loops.

Domain organization of N-terminal regions in the secretin superfamily

The superfamily of secretins differs most in the N-terminal half, as mentioned in the Introduction. The T2SS secretin starts with one N0 domain followed by three N1-like domains followed by the conserved C-terminal domain. The structural relationship we found between the N0 domain of peri-GspD and the signaling domain of TonB-dependent receptors can be combined with the sequence relationship between the signaling domain and the N-terminal domain of the T4PBS secretin (Pfam family PF07660). Like the T2SS secretins, the T4PBS secretins start with a detectable N0-like domain, but are followed by only a single N1-like domain. It appears that the T3SS secretins lack the N-terminal N0-like domain, but begin with two N1-like-domains followed by the conserved C-terminal domain. The evolutionary relationship between these different secretin families is clearly complex (Figure S2), and the implication of the domain organization for the way in which these systems function still needs to be unraveled. Yet, the presence of an N0-like domain at the N-terminus of both the T2SS and the T4PBS suggests functional similarities, such as β-strand complementation proposed for the T2SS in the next sections.

Cylindrical arrangements of periplasmic GspD domains

In order to explore the propensity of peri-GspD to form rings, we modeled cylindrical structures of the two lobes using the SymmDock server (Schneidman-Duhovny et al., 2005), assuming 12-fold symmetry based on the electron microscopy studies of the K. oxytoca T2SS secretin (Chami et al., 2005). Both the N0-N1 lobe and the N2 lobe produced models which suggest that both lobes can form an assembly with cyclic C12 symmetry (Figure 5A,B; Figure S3). It was not possible to obtain a plausible C12 model using the entire peri-GspD structure, presumably because the flexible loop connecting the two lobes adopts in the intact secretin another conformation than in the GspD:Nb7 heterotetramer. In the N0-N1 ring, there is possibly formation of an extended β-sheet involving two subunits due to the proximity of β4 from one N0 subdomain to β1 from a neighboring N0 subdomain (Figure 5D).

The outer dimension of the assembly is ∼117 Å, which agrees well with the 111 Å estimate from electron microscopy studies on the K. oxytoca GspD dodecamer (Chami et al., 2005). The inner diameter of the N0 ring is approximately 70 Å, sufficient to allow entry of molecules like the CT and LT AB₅ enterotoxins into the periplasmic cavity of the secretin, since the cross section of the folded B-pentamer of LT and CT is approximately 64 Å(Sixma et al., 1991). The same holds for the N2 ring which has an inner diameter of ∼ 62 Å. Intriguingly, the inner diameter of the N1 ring is ∼46 Å (close to the inner diameter of 54 Å in the K. oxytoca GspD dodecamer electron microscopy reconstruction (Chami et al., 2005) which would seem to be too narrow for passage of the B-pentamer (Figures 5A,B). However, the N-terminal α-helix, or another part of the pentamer, might be transiently unfolded to interact with strand β2 of the N0 domain, thereby permitting passage of the pentamer through the N1 and N2 rings. Alternatively, or additionally, the secretin undergoes conformational changes during translocation of proteins, and this might widen the N1 ring.

The electrostatic potential projected onto the surface of the N0-N1 GspD ring shows that the middle inner surface of the ring is positively charged with an excess of negative charge on both the outer membrane and periplasmic inner side of the pore (Figure S3A). However, this characteristic does not appear to be well conserved in the T2SS family of secretins (Figures S3A,B). The most clearly conserved surface area occurs in N1, at the end of helix α4 near the outer membrane on the outside of the torus (Figure S3B).

Interestingly, the EscJ homolog from the T3SS of EPEC is assumed to form a planar 24-meric ring which has been modeled on the basis of a crystallographic 6₅ screw arrangement of 24 EscJ subunits per turn with a pitch of 67 Å (Yip et al., 2005). The interactions between subunits in the N0-N1 and N2 dodecameric rings resemble those in the EPEC EscJ 24-mer from the T3SS (Figures 5C,D,E). Helix α1 of one N1 domain approaches helix α5 of a neighboring N1 domain in a manner similar to the corresponding helices in the rings of N2 and of EscJ.

The T2SS may engage in β-strand complementation

An important feature of the resulting GspD N0-N1 ring is that strand β2 of the N0 domain, which we have suggested above to be involved in interactions with a β-strand of a secreted protein or of another T2SS component, is fully accessible in the dodecameric arrangements obtained (Figure 5D, Figure S3). This was not a requirement used during generation of the model.

A fascinating question is whether the periplasmic N0 domain of GspD binds to secreted proteins or to T2SS proteins, or to both. The set of proteins translocated across the outer membrane by type 2 secretion systems is very diverse (Cianciotto, 2005) and the signal for T2SS translocation is unknown. Whereas GspD has been implicated in interaction with secreted proteins (Bouley et al., 2001; Shevchik et al., 1997), not all secreted proteins might bind with detectable affinity to GspD (Guilvout et al., 1999). It appears that different secreted proteins have different affinities for different T2SS components during translocation. Yet, some or all secreted proteins might still engage in transient and quite weak interactions with strand β2 of the first GspD domain during the initial steps of outer membrane translocation.

Among the T2SS proteins, so far the major pseudopilin GspJ (Douet et al., 2004) and the HR-domain of GspC (Korotkov et al., 2006) have been shown to interact with the periplasmic region of GspD. Interestingly, the globular head of GspG (Köhler et al., 2004) contains several β-strands at its surface, and the HR-domain of GspC is thought to have a predominant β-character (Fig. 3 in Korotkov et al., 2006). Therefore the possibility exists that these T2SS proteins interact, possibly transiently, with strand β2 of the first GspD domain. This leads to the suggestion that perhaps both secreted proteins and certain T2SS proteins compete for binding to β2 of the first domain of GspD, thereby moving a secreted protein from the periplasm into the periplasmic cavity of the large secretin, to be followed by additional steps leading to translocation across the outer membrane. This hypothesis clearly requires confirmation by further structural, biochemical and microbiological experiments.

Initial Crystallization Studies

The efforts for the structure determination of the periplasmic part of GspD started with protein from V. cholerae. This construct was only partially soluble and did not lead to successful crystallization. To improve the solubility and increase the chances for crystallization, we next used a directed evolution approach, using a C-terminal fusion with GFP as a reporter for correct folding and solubility (Waldo et al., 1999). Analysis of the mutants selected after the first round of evolution revealed that ∼80% of the mutations accumulated in the C-terminal ∼40 residues of the construct, suggesting that this region is probably responsible for the low solubility of the chosen N-terminal GspD construct. Taking into account these results of molecular evolution, we created a set of shorter N-terminal constructs, varying in length, also including proteins from V. vulnificus, V. parahaemoliticus, enterotoxigenic and enterohemorragic E. coli (ETEC and EHEC). Out of all constructs tried, only three led to crystals, which, however, diffracted anisotropically to a resolution of only 6 to 8 Å and were highly susceptible to radiation damage.

ETEC peri-GspD expression and purification

The gene fragment corresponding to the periplasmic part of GspD (residues 1-237) was PCR amplified from genomic DNA of enterotoxigenic E. coli (strain H10407) and cloned into pProEX HTb vector (Invitrogen) for expression with an N-terminal hexahistidine tag followed by a TEV protease cleavage site. The protein was expressed in E. coli BL21 (Novagen) cells at 30 C for 4h after induction with 0.5 mM IPTG. Peri-GspD was purified using Ni-NTA column (Qiagen) followed by His-tag cleavage with TEV protease and an additional pass through Ni-NTA. Final purification included ion-exchange and size exclusion chromatography using 30Q and Superdex75 columns (GE Healthcare). Se-Met labeled peri-GspD was expressed using metabolic inhibition of methionine biosynthesis as described (van Duyne et al., 1993).

Generation of nanobodies

A llama was immunized 6 times with 200μg of purified recombinant peri-GspD over a period of 6 weeks. From the anti-coagulated blood of the immunized llama, lymphocytes were used to prepare cDNA which served as template to amplify genes coding for the variable domains of the heavy-chain antibodies. The PCR fragments were ligated into a pHEN4 phagemid vector and transformed in E. coli TG1 cells. In this way, a VHH library of 5×10⁷ transformants was obtained with a VHH gene insert rate of at least 82% as determined by PCR. The VHH repertoire of this library was expressed in phages after superinfection with helper phages and selection of phage particles expressing peri-GspD binding VHH was performed. Phages were recovered by incubating the peri-GspD coated wells with 100mM triethylamine pH10 for 10 min. These peri-GspD coated wells were then washed once with TrisHCl pH 6.8 and several times with PBS and freshly grown TG1 cells were added to the wells to recover the noneluted phages. A clear enrichment was observed after three consecutive rounds of selection on solid-phase coated antigen. Twice 48 randomly chosen colonies - after the second and third round - were grown for expression of their nanobody as soluble protein. Crude periplasmic extracts were tested in an ELISA, 80 extracts were shown to be specific towards peri-GspD. From the positive clones the VHH genes were amplified by PCR and a HinfI RFLP was performed on all of them. Sequence analysis on 34 clones revealed nine different nanobodies against the peri-GspD (Figure S1). Some of them were found only in the pool of phages that were eluted with triethylamine, while others (NbGspD-5, NbGspD-6 and NbGspD-7) were found only after rescuing phages that were still bound to peri-GspD after elution. Finally, all selected nanobody genes were cloned in a pHEN6 vector for expression with a His-tag in E. coli (Conrath et al., 2001).

Solid-phase ELISA

Maxisorb 96-well plates (Nunc) were coated with purified peri-GspD overnight at 4 °C at 1μg/ml in sodium bicarbonate buffer pH 8.2. Residual protein binding sites in the wells were blocked for two hours at room temperature with 2% milk in PBS. Detection of antigen-bound nanobodies was performed with a mouse anti-haemaglutinin-decapeptide-tag (clone 16B12, BAbCO) or a mouse anti-histidine-tag (Serotec), as appropriate. Subsequent detection of the mouse anti-tag antibodies was done with an alkaline phosphatase anti-mouse-IgG conjugate (Sigma), respectively. The absorption at 405 nm was measured 15 min after adding the enzyme substrate p-nitrophenyl phosphate.

Complex purification and crystallization

Individual nanobodies were mixed with peri-GspD at a 1:0.95 ratio and binary complexes were purified by size exclusion chromatography using a Superdex 75 HR 10/30 column in 20 mM Tris-HCl pH 7.8, 200 mM NaCl, whereafter the protein complexes were concentrated to ∼3-4 mg/ml. Crystallization conditions were found using Protein Complex (Qiagen), SaltRx and PEG/Ion (Hampton Research) screens. Complexes of peri-GspD with Nb5, Nb6, Nb7 and Nb8 gave multiple crystal hits and two crystal forms of peri-GspD:Nb7 were optimized to produce diffraction quality crystals. The first (trigonal) form was crystallized by vapor diffusion using 1.0 M Na/K phosphate pH 5.0 as precipitant and crystals were gradually transferred to precipitant solution supplemented with 20% ethylene glycol for cryoprotection. The needle-like crystals of the second (orthorhombic) form were produced using 15% PEG 4000, 0.2 M Li sulfate, 0.1 M bisTris-HCl pH 5.5 and cryoprotected with 20% glycerol.

Structure determination

Diffraction data were collected at beamline BL9-2 of the Stanford Synchrotron Radiation Laboratory (SSRL) and processed using XDS (Kabsch, 1993) or HKL2000 (Otwinowski and Minor, 1997). The structure of peri-GspD:Nb7 complex in the trigonal crystal form was solved by single-wavelength anomalous diffraction (SAD) phasing with SHELXD (Sheldrick, 2008) and SOLVE (Terwilliger, 2004). All expected 28 Se atom positions were found and used to calculate 4-fold non-crystallographic symmetry operators as implemented in RESOLVE (Terwilliger, 2004). Density modification with 4-fold averaging and iterative model building by RESOLVE (Figure S4) led to an initial model of ∼1000 residues that was completed manually using Coot (Emsley and Cowtan, 2004). Nb7 model building was assisted by superposition of a homologous llama VHH domain structure, PDB code 1SJX (Dolk et al., 2005). The final structure was refined to 2.80 Å resolution with R_work=0.192 and R_free=0.239 with REFMAC5 (Murshudov et al., 1997) using TLS groups defined by the TLSMD server (Painter and Merritt, 2006). The stereochemical quality of the model was verified using Coot and Molprobity (Davis et al., 2007). Final data collection and refinement statistics are summarized in Table 1. Residues of peri-GspD corresponding to N and C-terminal residues (1-2 and 236-237) and three loops (32-34, 81-98, 192-204) had poor or missing electron density and were omitted from the final structure. The structure of the peri-GspD:Nb7 complex in the orthorhombic crystal form was solved by molecular replacement using Phaser (McCoy et al., 2007) and contains one (peri-GspD)₂:(Nb7)₂ heterotetramer in the asymmetric unit.