Structural Insights into the Secretin PulD and Its Trypsin-resistant Core*

Limited proteolysis, secondary structure and biochemical analyses, mass spectrometry, and mass measurements by scanning transmission electron microscopy were combined with cryo-electron microscopy to generate a three-dimensional model of the homomultimeric complex formed by the outer membrane secretin PulD, an essential channel-forming component of the type II secretion system from Klebsiella oxytoca. The complex is a dodecameric structure composed of two rings that sandwich a closed disc. The two rings form chambers on either side of a central plug that is part of the middle disc. The PulD polypeptide comprises two major, structurally quite distinct domains; an N domain, which forms the walls of one of the chambers, and a trypsin-resistant C domain, which contributes to the outer chamber, the central disc, and the plug. The C domain contains a lower proportion of potentially transmembrane β-structure than classical outer membrane proteins, suggesting that only a small part of it is embedded within the outer membrane. Indeed, the C domain probably extends well beyond the confines of the outer membrane bilayer, forming a centrally plugged channel that penetrates both the peptidoglycan on the periplasmic side and the lipopolysaccharide and capsule layers on the cell surface. The inner chamber is proposed to constitute a docking site for the secreted exoprotein pullulanase, whereas the outer chamber could allow displacement of the plug to open the channel and permit the exoprotein to escape.

Comparison of secretin sequences led to the definition of two major domains of approximately equal length (19). The predicted domain organization was confirmed by analysis of trypsin or proteinase K-resistant domains of PulD and XcpQ (14,17) that, in the case of XcpQ, formed channels whose conductance was similar to that of the intact protein (14). These data prompted speculation that the well conserved secretin C domain is anchored in the outer membrane by 10 -14 potentially transmembrane amphipathic ␤ strands (19) characteristic of other outer membrane proteins (22). The N domain, corresponding approximately to the first half of the protein, is much less conserved. The Erwinia chrysanthemi exoprotein pectate lyase binds to this region of its cognate secretin, OutD, indicating that it might carry a specific exoprotein recognition determinant (23). A third secretin domain sometimes present downstream from the C domain (the S domain) interacts with a protein (pilotin) that facilitates secretin targeting to the outer membrane (24 -27).
The three-dimensional structures of detergent-solubilized secretins pIV and N. meningitidis PilQ were determined by cryo-electron microscopy to resolutions of 18 and 12 Å, respectively. The pIV complex is ϳ13.5 nm in diameter and 12 nm deep and comprises three stacked rings with two large chambers separated by a central protein mass (18). The C and N domains, identified by labeling specific cysteine residues with nanogold particles, are on opposite rings (18). In contrast, PilQ complexes have a cone-like structure that is closed at the tip (12,28,29). Individual secretin subunits were not resolved in either structure, but symmetries of 14 (pIV) and 12 (PilQ) were inferred from analysis of the rotational power spectrum (18) and from regular patterns of stain accumulation visible in the reconstituted three-dimensional image (28,30), respectively.
The pullulanase T2SS of K. oxytoca is one of the most extensively studied secretons and has been completely reconstituted in E. coli. A low resolution three-dimensional structure of a purified complex of the pullulanase T2SS secretin PulD and its pilotin, PulS, revealed a cylindrical complex with 12-fold symmetry and a central open channel of about 7 nm encircled by radial spokes that we originally presumed to be the pilotin, which is absent from pIV and PilQ (16). Here, we report refined biochemical and structural analyses of intact and proteolyzed PulD multimers.
Construction of PulDhis and PulShis-To create PulDhis, a His linker made by annealing oligonucleotides 5Ј-GATCCATCACCATCAC-CATCACGC-3Ј and 5Ј-GATCGCGTGATGGTGATGGTGATG-3Ј was introduced into the BglII site in pulD carried by pCHAP3516 (24). The resulting plasmid was digested with EcoRI and HindIII, and the fragment corresponding to pulDhis was ligated into pSU18 (32) and cleaved by the same enzymes to create pCHAP3678. The ability of PulDhis to replace PulD was tested in a pullulanase enzymatic assay (33) using strain PAP105 carrying pCHAP1226 (encoding all Pul secretion factors except PulD (34)). Bacteria were grown in LB medium buffered with 10% M63 salts solution (31) and 0.4% maltose to induce expression of genes encoding pullulanase secretion factors. pCHAP3678 restored secretion to 83% of the wild-type level.

Construction and Purification of PulD-N-See the supplemental material.
Purification of Secretins-Outer membranes were isolated from PAP105 (pCHAP3516/pCHAP5506) and PAP105 (pCHAP3678/ pCHAP585) producing PulD and PulShis or PulDhis and PulS, respectively, as described previously (16). Membrane proteins were solubilized in 50 mM Tris-HCl (pH 7.5) containing 3% n-tetradecyl-N,N-dimethyl-3-ammonio-1-propanesulfonate (ZW3-14), 250 mM NaCl, and 0.1 mg/ml protease inhibitor Pefablock (Pierce) and incubated for 1 h at room temperature. Solubilized proteins were recovered after ultracentrifugation at 185,000 ϫ g and mixed with cobaltagarose resin (Talon, Clontech) previously equilibrated in 50 mM Tris-HCl (pH 7.5), 0.6% ZW3-14, 250 mM NaCl (TZN buffer) containing 5 mM imidazole for 1 h at 20°C. The resin was washed with 10 volumes of TZN buffer plus 5 mM imidazole and then poured into a column. Bound proteins were eluted with 5 column volumes of TZN buffer containing 5 mM EDTA and immediately loaded onto a HiTrap Q HP column (1 ml) (Amersham Biosciences) connected to an AKTAprime system (Amersham Biosciences) at a flow rate of 0.1 ml/min. After extensive washing, secretin complexes were eluted using a linear gradient of 250 mM to 1.25 M NaCl. To completely dissociate PulS from the complex, the fractions containing PulD (eluted at 700 -750 mM NaCl) were incubated with 50 mM reducing agent Tris[2-carboxyethylphosphine] hydrochloride (Pierce) for 2 h at 20°C. PulS was separated from the PulDhis complex by gel filtration on Sephacryl S300HR. The absence of PulS in the PulDhis fractions was confirmed by immunoblotting using antibodies directed against MalE-PulS. The fractions were concentrated on Q-Sepharose and stored in 100-l aliquots at Ϫ80°C.
Trypsin Proteolysis of PulDhis Secretin-100 g of purified PulDhis complex (0.1 mg/ml) was incubated with 10 g/ml N-p-tosyl-L-phenylalanine chloromethyl ketone-treated trypsin (Sigma) for 2 h at 18°C. Pefabloc was then added to 0.1 mg/ml, and the sample was immediately injected onto a Sephacryl S300HR column. A control experiment was performed under the same conditions except that the trypsin was omitted. In both cases aliquots from all fractions were treated with phenol to dissociate the PulD complexes (8) and then analyzed by SDS-PAGE. The Sephacryl S300HR column was calibrated using thyroglobulin, alcohol dehydrogenase, bovine serum albumin, and carbonic anhydrase. PulD-containing fractions were concentrated and stored in 50-l aliquots at Ϫ80°C for electron microscopy.
Electrophoresis, Immunoblotting, and Protein Sequencing-See supplemental material.
Circular Dichroism Analysis-Circular dichroism (CD) spectra were recorded from 190 to 250 nm on a Jasco dichrograph (J-710) equipped with a thermostatically controlled cell holder and connected to a computer for data acquisition. Purified PulD-N at 100 g/ml was dialyzed against 1 mM Tris-HCl (pH 8) containing 250 mM NaF and submitted to CD analysis in a 1-mm quartz cell (3 data sets were acquired). The PulD-C domain was dialyzed against 1 mM Tris-HCl (pH 7.5) containing 250 mM NaF and 0.6% ZW3-14 immediately before analysis in a 5-mm quartz cell at a final concentration of 10 g/ml (5 data sets were acquired). Purified PulDhis at 60 g/ml in 50 mM Tris-HCl (pH 7.5) containing 250 mM NaCl and 0.6% ZW3-14 was diluted twice in 500 mM NaF containing 0.6% ZW3-14 immediately before CD analysis in a 1-mm quartz cell (5 data sets were acquired). Peptide Analysis by HPLC, Sequencing, and Mass Spectrometry-See the supplemental material.
Scanning Transmission Electron Microscopy-A Vacuum Generators (East Grinstead) HB-5 STEM interfaced to a modular computer system (Tietz Video and Image Processing Systems) was used. Samples were prepared on 200-mesh-per-inch, gold-plated copper grids as previously (35). Grids were washed on 5-7 droplets of quartz double-distilled water to remove detergent. Isolated tobacco mosaic virus particles (kindly supplied by R. Diaz-Avalos) adsorbed to a separate grid and air-dried served as the mass standard.
For structural examination, digital 512 ϫ 512 pixel dark-field images were recorded from negatively stained (2% uranyl acetate) PulD at an acceleration voltage of 100 kV and a nominal magnification of 500,000ϫ using doses between 3450 and 10250 electrons/nm. The pixel size (around 0.33 nm) depended on the focus conditions (35).
For mass determination, digital 512 ϫ 512 pixel dark-field images were recorded from unstained freeze-dried samples of the PulD complexes at an acceleration voltage of 80 kV and a nominal magnification of 200,000ϫ. The recording dose used varied between 295 and 640 electrons/nm 2 for these main data sets. In addition, repeated low dose scans were recorded from some grid regions to assess beam-induced mass loss (35). The images were evaluated with the IMPSYS program package as described previously (35). Complexes were delimited by circles, the total mass was calculated, and the background was subtracted. The beam-induced mass loss was assessed as previously (35). The experimental mass data were multiplied by the correction factors calculated for the recording doses used and scaled according to the mass-per-length measured for tobacco mosaic virus particles. Finally, they were binned into histograms and fitted with Gauss curves. The overall experimental uncertainty of the results was estimated from the corresponding S.E. (S.E. ϭ S.D./͌n) and the 5% uncertainty in the calibration of the instrument by calculating the square root of the sum of the squares.
Metal Shadowing-A 5-l aliquot of purified sample at 0.05 mg/ml was applied to copper grids covered with a thin carbon film of uniform thickness, blotted with Whatman 2 filter paper, washed with 5 droplets of water to remove excess detergent, and quick-frozen in liquid ethane at Ϫ178°C. The frozen grids were transferred into a Balzers 300 freezeetching unit and freeze-dried at Ϫ90°C for 5 h in a vacuum of ϳ1.3 ϫ 10 Ϫ5 Pa. Platinum/carbon (1-2-nm thick) was evaporated onto the surface at an angle of 45°while the sample was rotating. Images were taken on Kodak SO163 film at a nominal magnification of 50,000ϫ with a Hitachi 7000 electron microscope operating at an accelerating voltage of 100 kV. The film was digitized at 4 Å/pixel with a Primescan D 7100 drum scanner (Heidelberger Druckmaschinen AG). EMAN boxer (36) was used to interactively select 1300 particles. Neural network-based classification, averaging, and calculation of the rotational power spectra (37) were done using the Xmipp (38) software package.
Cryo-electron Microscopy and Single Particle Analysis-A 5-l aliquot of purified sample at 0.05 mg/ml was applied to thin carbon film mounted on copper grids, washed with 5 droplets of water to remove excess detergent, and then quick-frozen in liquid ethane at Ϫ178°C. The frozen grids were transferred to a Philips CM200-FEG electron microscope using a Gatan cryo-holder. Images were recorded on Kodak SO163 film at a nominal magnification of 50,000ϫ and an accelerating voltage of 200 kV under low dose conditions. The defocus of the micrographs ranged between 2.4 and 5.1 m. The negatives were digitized at 2 Å/pixel with a Primescan D 7100 drum scanner.
Particles were selected interactively from the images with the EMAN boxer (36) software and analyzed using the SPIDER image processing programs (39). Reference-free alignment (40) and classification by correspondence analysis was employed to generate class averages. These were then used as references for iterative multi-reference alignment, with the cycles of alignment and classification repeated until no further change was observed in the class averages.
For the intact complex, the various side views were utilized to generate a rough starting model by a common line search varying only one Euler angle. A cyclic 12-fold symmetry (C12) was imposed for the starting model, which was refined by projection matching of the top and side views while still imposing the C12 symmetry for the final volume. Several cycles of projection matching with increasing numbers of reference projections followed by iterative back projection in real space were required to obtain the final volume. All particle views were included in the refinement and matched to evenly balanced projections covering the whole three-dimensional space. The mandatory contrast-transfer function correction at the defocus values employed was made by a Wiener filter using a combination of the MRC (41) and SPIDER (39) software suites. The resolution of the generated map was estimated from the Fourier shell correlation function using the 50% threshold.
Because practically only dimers of side views of the proteolyzed PulDhis complexes were available, a first volume of the dimer was computed and refined, imposing a dihedral 12-fold symmetry (D12) throughout. The two reconstructed complexes were then separated and merged; the average of the few complex monomer side views available was used to measure the dimensions of the fragment to obtain a clean separation of the two, thereby eliminating symmetrization artifacts at the interface. The contrast-transfer function was corrected as above.

Purification and Proteolysis of Intact His-tagged Secretin Multimers-The
PulD-PulS complex reported previously by Nouwen et al. (16) precipitated from ZW3-14 solutions at low temperature and at low salt concentrations, which hampered high resolution cryo-electron microscopy analysis. To circumvent this problem, we dissociated PulS from PulD without denaturing the latter. To facilitate the purification of PulD alone, a hexahistidine tag was inserted in the S domain of PulD at position 635 (i.e. in the S domain; see Fig. 3). The tagged protein, PulDhis, restored pullulanase secretion in a strain lacking PulD but producing all other secreton components ("Experimental Procedures"). When produced at high levels together with PulS, PulDhis was readily solubilized from membranes in ZW3-14 and was purified as a high molecular mass complex by cobalt and anion exchange chromatography ("Experimental Procedures"). PulS co-eluted with PulDhis but at a stoichiometry of only 0.1:1 (Fig. 1A), compared with ϳ1:1 for PulSϩPulD (16) or PulShisϩPulD (Fig. 1A), showing that the His tag in the S domain PulD did not prevent it from inserting into the outer membrane but might cause PulS to dissociate. Reduction of the disulfide bridge in PulS (42) by Tris[2-carboxyethylphosphine] hydrochloride completely dissociated the remaining PulS from the purified PulDhis complex (Fig. 1A).
The purified PulD complex and the complex remaining after limited trypsin proteolysis ("Experimental Procedures") were subjected to size exclusion chromatography (Fig. 1B). The intact complex eluted at 46 ml, corresponding to a Stokes radius of Ն12 nm. This is exactly the same elution volume as that of PulD-PulShis or PulD-PulS complexes (data not shown), indicating that the removal of PulS does not drastically modify the shape of the complex. In contrast, the protease-resistant complex eluted at 52 ml, indicating a Stokes radius of Ϸ9 nm.
STEM mass analysis (Fig. 1C, left histogram) showed the intact PulDhis complex to have a mass of 919 kDa (Ϯ109 kDa). Previous studies of PulD-PulS revealed a stoichiometry of 12 (17). Assuming that this stoichiometry is unaltered by the removal of PulS and the insertion of a His tag (see the next paragraph), the PulDhis complex comprises 12 of the 68.745-kDa monomers with a total mass of 825 kDa, the remaining 94 kDa being bound detergent. This is in good agreement with the earlier experiments on purified PulD-PulS complexes, in which the detergent contribution was estimated to be 56 kDa (17), and with the amount of detergent (75 Ϯ 20 kDa) expected to bind to the hydrophobic surface of the PulD homomultimer (112 nm 2 for a 12-nm diameter (see below), 3-nm thick (43) hydrophobic belt). STEM analysis showed the mass of the proteolyzed complex to be 537 kDa (Ϯ64 kDa; Fig. 1C, right histogram; see the next section).
To verify the stoichiometry, both complexes were freeze-etched, metal-shadowed. and observed by transmission electron microscopy. As expected, the complexes appeared as ring-like structures when viewed end-on ( Fig. 2A). Individual protomers were sometimes visible for the proteolyzed complex ( Fig. 2B) but not for the intact complex (data not shown), possibly because flexible loops averaged out the signal in the latter. The projection average obtained upon single particle analysis of the metal-shadowed proteolyzed complex has a pronounced 12-fold symmetry (Fig. 2C), confirming that the stoichiometry of the complex is not modified by removal of PulS and proteolysis.
Polypeptide Composition of Proteolyzed PulDhis-The polypeptide composition of the trypsin-resistant PulDhis multimers was determined by SDS-PAGE, mass spectrometry, and Edman degradation analyses. A single band with an apparent molecular size of 40 kDa was observed after Tris-Tricine SDS-PAGE (Fig. 3A). Successive cycles of Edman degradation indicated that this polypeptide started with QAAK, corresponding to residues 298 -301 of PulDhis.
Mass spectrometry of proteolyzed PulDhis complexes revealed the presence of molecular species with masses of 3,950 and 33,797 Da (F1 and F2, respectively) (Fig. 3A). No other peaks were detected, indicating complete homogeneity. F2 corresponds to the 40-kDa band seen on SDS-PAGE and, thus, comprises residues 298 -617 of PulDhis (calculated mass, 33,783 Da). A genetically engineered truncated PulD protein corresponding to residues 298 -616 of PulDhis also migrated more slowly than expected during SDS-PAGE (data not shown). To identify F1, which was not detected by Tris-Tricine SDS-PAGE, the proteolyzed complex was dissociated with phenol (8) and analyzed by HPLC on a C18 column. Several peptides were detected, recovered, and submitted to Edman degradation and mass spectrometry. All had a mass of 3950 Da and began with QQAT, corresponding to residues 262-265 of PulDhis. The reason why these identical peptides eluted in several peaks from the C18 column is unknown. From the precise mass of the fragment, we unambiguously identified F1 as residues 262-297 of PulDhis (calculated mass, 3936 Da). Thus, trypsin removes regions from both the N and C termini of PulDhis, leaving a nicked core structure corresponding to residues 262-617 (Fig. 3B). This protease-resistant region, which is called the C domain, is highly conserved in all secretins and comprises the third of three CD1 modules and the CD2 module (see Fig. 3B for the explanation). The C domain corresponds to the ␤ domain defined previously (17) plus peptide F1.
According to mass spectrometry, only F1 among the peptides cleaved off by trypsin remained associated with the trypsin-resistant PulDhis complex. Therefore, the latter has a mass of 452.6 kDa (12 ϫ (33,783 ϩ  3,936)), indicating that 46% of the mass of the secretin is removed by  proteolysis. Likewise, the STEM data indicate a mass of 537 kDa for the proteolyzed complex (Fig. 1C, right histogram). The difference of 84 kDa between the value determined by STEM and that calculated from the mass spectrometry data represents the amount of bound detergent, which is very close to that associated with the intact complex (see the previous paragraph). This is entirely consistent with the fact that the C domain includes the membrane-integrated part of the complex, which is presumably where the majority of the detergent binds (see "Discussion").
Secondary Structure of PulDhis-Circular dichroism was used to determine the secondary structure of the trypsin resistant C domain of PulDhis (262 to 617). The proteolyzed complex was dialyzed to replace NaCl by NaF so that a reliable signal could be recorded down to a wavelength of 190 nm. As shown in Fig. 4, the spectrum displays a single minimum at 218 nm. Deconvolution of this spectrum using the CON-TIN program on the Dichroweb site (www.cryst.bbk.ac.uk/cdweb/ html/home.html) (44,45) indicated that 27% of the polypeptide is ␤ structure. The reconstructed spectrum could be perfectly superimposed on the experimental data (not shown). The presence of ␤ strands is not surprising, since the proteolyzed complex is predicted to contain the transmembrane domain, but the value of 27% is low compared with other bacterial integral outer membrane proteins, suggesting that a large part of the C domain is not organized in a classical outer membrane ␤ barrel structure and, therefore, is probably not embedded in the membrane.
We engineered a plasmid coding for the trypsin cleaved N domain of PulD (residues 28 -266) and an N-terminal polyhistidine tag (see the supplemental material). This domain was soluble and was easily purified to homogeneity by cobalt affinity chromatography (data not shown). It eluted as a single symmetrical peak corresponding to the monomeric form of the protein upon size exclusion chromatography (data not shown). The CD spectrum of this N domain was substantially different from that of the trypsin-resistant C domain, with minima at 208 and 222 nm, typical of ␣ helixes (Fig. 4). The observed difference in the CD spectra of the isolated N and C domains of PulD clearly indicates the radically different organization of the polypeptide chain in these two regions, confirming that they are two structurally distinct domains.
As noted above, intact PulDhis complexes are very sensitive to NaCl concentration, and we could not completely replace NaCl by NaF without precipitating the protein. Thus, we could only record a reliable CD spectrum between 200 and 250 nm, and it was not possible to deconvolute this spectrum. However, the spectrum of PulD secretin and the calculated sum of the isolated N and C domain spectra could be almost perfectly superimposed. Accordingly, the spectra of the isolated N and C domains reflect their secondary structure in the native protein.
Cryo-electron Microscopy of Intact and Proteolyzed PulDhis-When frozen for cryo-electron microscopy, membrane proteins, and in particular secretin complexes, tend to cluster at the edge of the grid holes. To avoid this problem, intact and trypsin-proteolyzed PulDhis complexes were adsorbed onto thin carbon films before freezing. Both PulDhis and the trypsin-resistant complexes mainly oriented end-on, appearing as ring-like structures with a large central density (Fig. 5). PulDhis also sometimes oriented on its side, revealing a "cup and saucer" structure ( Fig. 5A) similar to that observed previously upon negative staining (17). Occasional side-view dimeric complexes interacting via their saucer side were also observed. Trypsin-proteolyzed PulDhis complexes were rarely seen in side views unless the sample had been left at room temperature before analysis. These side views were almost always dimers in which the two complexes were associated on their cup side (Fig. 5B), indicating that the physico-chemical properties of this part of the complex are modified by proteolysis. A striking common feature of native and trypsin-resistant PulDhis top views is a sharp ring-shaped density at the edge of the projected cylinder (arrowheads in Fig. 5 insets). The diameter of this sharp ring of density, 11.9 nm, is the same as the distance between the fine structures that connect the cup and saucer (marked by arrowheads in the side-view averages shown in Fig. 5 insets). In addition, native PulDhis complexes possess distinct peripheral densities with a 12-fold rotational symmetry (Fig. 5A inset). These peripheral densities are missing from proteolyzed PulDhis complexes (Fig. 5B inset). In agreement with these observations, peripheral structures were clearly visible on some of the negatively stained PulDhis complexes imaged by STEM but were not detected on proteolyzed samples (Supplemental Figs. S1 and S2).
More than 6000 intact PulDhis complexes were selected interactively from the micrographs recorded by cryo-electron microscopy, reference-free-aligned, classified, and averaged (see "Experimental Procedures"). Characteristic top-and side-view class averages are shown in the insets of Fig. 5A. All class averages were employed to generate an initial three-dimensional volume, imposing C12 symmetry. Refinement cycles using projection matching of an increasing number of particles with calculated back-projections were repeated until the three-dimensional map did not improve further (Fig. 6, A and B). The resolution of the generated map was estimated to be 1.7 nm from the Fourier shell correlation function using the 50% criterion.
Class averages of a total of nearly 3300 top views and 1500 side views of the complex dimer were similarly calculated for the proteolyzed PulDhis complex. The inset of Fig. 5B displays a typical top-and sideview class average. Because only ϳ100 side views of the complex monomer could be found, the three-dimensional map of the proteolyzed secretin was computed as the volume of the complex dimer, applying a D12 symmetry, and refined as above. The two reconstructed complexes were then separated and merged (see "Experimental Procedures"). The resolution of the final reconstruction (Fig. 6, A and D) could not be reliably determined because the volume was composed from side views of dimers.
The models of the intact (mesh) and trypsin-resistant (blue) PulDhis complexes are superimposed in Fig. 6A. The corresponding tilted com-   (TABLE TWO). The outer contours of the three stacked rings give the two structures a cup and saucer appearance (shown in upside-down orientation in Fig. 6), one of the three rings being the saucer and the other two, the cup. Connections between saucer and cup are not visible at the current resolution of 1.7 nm, although they are visible on class averages of negatively stained side views (17) and on unstained top and side views (arrowheads in Fig. 5). Although the saucer has a central hole, a plug occludes the saucer end of the chamber formed by the cup. Almost all of these features are smaller after trypsin treatment (TABLE TWO), the reduction in the height of the cup from 6.4 to 4 nm being the largest change. This means that trypsin removes the outer ring of the cup, which must, therefore, be formed by the N-terminal domain of PulDhis, the larger of the two domains that are cleaved (Fig. 2). Because the saucer is also shallower in the trypsin-resistant complex, the overall height of the structure decreases from 10.8 to 8 nm. These data suggest that the major contribution to the mass of the plug and the saucer is made by the C domain. The diameter of the saucer also decreases from 17.9 to 14.3 nm upon proteolysis through removal of the spoke-like densities. In addition, the peripheral densities that surround the native PulDhis cup disappear completely upon proteolysis. Projections of the three-dimensional maps of the native and the trypsin-resistant PulDhis along the cylinder axis (Figs. 6, C and E, respectively) facilitate visualization of these differences.

DISCUSSION
The model we propose here for the secretin PulD complex is based on biochemical analysis and cryo-electron microscopy of particles adsorbed from zwitterionic detergent solution onto carbon films. The outer contours give the structure the appearance of a cup and saucer. The essential features of the model are two rings that form large chambers on either side of the complex and a central disc sandwiched between, including the plug that occludes the channel. This is in good agreement with the model proposed earlier based on images recorded from negatively stained PulD-PulS complexes (17). The changes caused by proteolysis indicate that the N-terminal domain of PulDhis forms the outer ring of the cup, whereas the C domain forms most of the plug and the saucer with its outer chamber. The overall features of the model are remarkably similar to those proposed for the quite distantly related bacteriophage f1-encoded secretin pIV in the non-ionic detergent Triton X-100 (18) and for Salmonella enterica secretin InvG in purified type III secretion needle complexes (46). The major difference between the models proposed for PulD and pIV is that the former is clearly dodecameric, whereas pIV is a tetradecamer (15,18). STEM analysis gave a mass of 919 kDa for the native PulDhis complex, including the detergent present, thereby excluding the presence of 14 subunits for which the calculated mass is 962 kDa. Moreover, the PulD complex is highly unlikely to contain 13 subunits like YscC (13), since the detergent contribution would then only be about 25 kDa, which is incompatible with the dimensions of the hydrophobic belt (43). Changes in stoichiometry might have occurred during the evolution of the secretin superfamily to permit the transport of differently structured macromolecules (bacteriophage, pseudopili, or folded exoproteins). Only relatively small changes in subunit dimensions and packing would be required to accommodate a change from 12 to 14 secretin subunits (or vice versa). A similar scenario was recently proposed to explain the variations in dimensions of TatA complexes isolated from the E. coli (47). These complexes apparently form the channel through which specific folded proteins of varying sizes are translocated across the inner membrane. The TatA complexes might acquire or release protomers to adapt to particular substrates or might exist as a mixed population to enable the entire range of substrates to be accommodated. As with secretins, the number of protomers in the TatA complex is probably so high that the addition or subtraction of a small number of TatA monomers would involve minimal distortion of the entire complex.
The apparent structural similarity of the InvG, pIV, and PulD complexes makes it all the more remarkable that the structure proposed for N. meningitidis PilQ secretin complex is so different (29). Initial electron microscopy of negatively stained PilQ extracted in deoxycholate that was then replaced by SDS (12) or zwitterionic detergent 3-10 (28) indicated a ring-like structure similar to that of other secretins in SDS (12) and with peripheral densities similar to those of PulD in zwitterionic detergent (16,17). However, class averages of side views of deoxycholate-extracted PilQ complexes dissolved in zwitterionic detergent 3-10 and examined by cryo-electron microscopy after negative staining with ammonium molybdate revealed that the central density corresponds to the closed aperture at the tip of a cone-like structure (29) rather than the central plug observed for pIV (18) and PulD (this study) or the septum of InvG (46). The closed aperture in PilQ appears to be the site to which the type IV pilus binds (48). Although it is not inconceivable that different secretins might assemble into different structures, we note that the primary sequences of PulD and PilQ are more closely related than are those of PulD and pIV. In addition, although normally dedicated to pullulanase secretion, PulD can perform the same function as PilQ, i.e. the assembly of surface pili (49,50) including a pilus normally produced by Neisseria (7).
The central plug is a major, newly defined feature of the PulD complex (Fig. 6). The plug was not apparent in the previous three-dimensional model (16) but was observed on later, negatively stained samples (17). Its presence is consistent with the failure of PulD to form constitutively open channels (16,19). The fact the major contribution to the mass of the plug is made by the C domain and not by the N domain as originally proposed (17) explains why the low electrical conductance of the PulD homologue XcpQ in artificial planar bilayers does not increase when the N domain is removed by proteolysis (14). In addition, mutations in secretin structural genes that alter the permeability of the outer membrane invariably map to the region encoding the C domain (20,21,51,52). Fine connections presumably anchor the central plug to the perimeter of the cup in proteolyzed PulD complex. Our failure to visualize the plug in the previous model (16) might be explained by differences in preparation and analytical methods. It is worth noting that the mechanism by which the secretin channel is occluded is completely different from that used to prevent the uncontrolled movement of solutes through TonB-dependent outer membrane transporters, where the N-domain constitutes the plug (53).
The structural changes caused by proteolysis allow the membrane orientation of the PulD complex to be predicted. The N domain, corresponding mostly to the rim of the cup that disappears upon proteolysis, must face the periplasm to perform its proposed function in exoprotein recognition (23). The location of this domain at one end of the secretin is in line with gold labeling experiments performed with pIV (18). The secretin N domain forms the outer part of a large chamber into which macromolecules (e.g. exoproteins or filamentous bacteriophages) might insert before they are transported across the outer membrane. Exoprotein docking to this site might cause displacement of the plug to create a continuous channel that would remain blocked by the exoprotein during its translocation. Assembly of pseudopilins into a pseudopilus that reaches the cup of the secretin might provide the driving force to expel the exoprotein from the inner chamber (7,50). Exoprotein release and pseudopilus retraction would allow the plug to reform and the channel to close. A similar model explains the opening of InvG secretin channel to allow the assembly of the type III secretion system needle, except that the septum does not close again once it is opened (46).
Another striking feature of both PulD complexes that is clearly seen on specific class averages of top views is a sharp but faint ring-shaped density at the periphery of the cylinder (marked by arrowheads in Fig. 5  insets). This corresponds to the fine connections present between cup and saucer in side views but is not seen in the final three-dimensional map (Fig. 6). Their absence in the latter is explained by the limited resolution of the three-dimensional map (1.7 nm), which results from using all projections to reconstruct the volume rather than considering specific top and side views only. Bound detergent, not visible on the images (the ZW3-14 detergent used has a density close to 1 4 ), could explain why this region was not proteolyzed. Accordingly, these fine connections could reflect the ␤-barrel inferred from CD spectra and indicate the position of the lipid bilayer. Indeed, a small number of ␤ strands from each monomer would suffice to form a large ring structure, as in the integral outer membrane region of the E. coli protein TolC trimers (54) or the Mycobacterium tuberculosis outer membrane porin MspA (55).
If the ␤ strands are indeed located in between the cup and saucer parts of the structure, some of the saucer (Ϸ1.5 nm, see Fig. 6A) would be on the outside of the outer membrane, possibly even reaching beyond the lipopolysaccharide and other outer layers of the cell. The disc and outer ring of the cup would extend Ϸ5 nm into the periplasm, which would leave sufficient space for exoproteins to enter its inner chamber. An alternative model where the ring corresponding to the saucer is fully embedded in the outer membrane with little structure exposed on the cell surface as proposed for pIV (18) is inconsistent with the relatively 4 A. Lusting, personal communication. PulD Secretin Structure NOVEMBER 11, 2005 • VOLUME 280 • NUMBER 45 low amount of ␤ strand secondary structure in the C domain of PulD (27%). It would also be inconsistent with the irregular outer wall of the saucer (note the lateral projections in Fig. 6), around which the outer membrane lipids would not fit snugly. Furthermore, this model would imply that the rest of the structure extends through the periplasm almost to the inner membrane, which might make it difficult for exoproteins to access the inner chamber. Peripheral protein mass first observed in PulD-PulS complexes (16,17) imaged by negative staining and by cryo-electron microscopy was proposed to represent PulS bound to the periphery of the PulD complex because it was almost completely absent from top views of negatively stained, trypsin-digested PulD-PulS particles (which lack PulS and the S domain of PulD as well as the N domain) (17). However, as shown by Figs. 5 and 6, similar peripheral objects (termed peripheral densities) are also associated with native PulDhis complexes without PulS. Thus, the earlier assignment was not quite correct; the peripheral densities do not correspond solely to PulS, as proposed, but must also contain a small amount of PulD. Both the spokes of the saucer and the peripheral densities disappear upon trypsin treatment ( Fig. 5B and 6E) and are not visible on top views of negatively stained particles of trypsin-resistant PulDhis observed by STEM (see the supplemental material). Because the larger cleaved domain must correspond to the N-terminal region, the peripheral densities could be the S domains, which would be compatible with a mass loss of Ϸ6 kDa (Fig. 3) and would place them close to the periplasmic face of the outer membrane, in which PulS is probably anchored by its fatty acids (8,56).
The data and refined model presented here indicate that the trypsinresistant C domain is a substantial part of the periplasmic structure of the secretin complex. It could form a basic core structure required for stable multimerization and interaction with other envelope components such as lipopolysaccharide and peptidoglycan. This leads us to propose that the C domain, which is more highly conserved throughout the secretin superfamily than the N domain, is a scaffold onto which different functions have been grafted through evolution of the N domain. In the case of the T2SS secretin family, these functions could include interactions with the exoproteins (23), with the pseudopilus (57), and with other secreton components such as PulC or its homologues (57)(58)(59)(60).