Distinct docking and stabilization steps of the pseudopilus conformational transition path suggest rotational assembly of type IV pilus-like fibers

SUMMARY

The closely related bacterial type II secretion (T2S) and type IV pilus (T4P) systems are sophisticated machines that assemble dynamic fibers promoting protein transport, motility or adhesion. Despite their essential role in virulence, the molecular mechanisms underlying helical fiber assembly remain unknown. Here we use electron microscopy and flexible modeling to study conformational changes of PulG pili assembled by the Klebsiella oxytoca T2SS. Neural network analysis of 3900 pilus models suggested a transition path towards low-energy conformations driven by progressive increase in fiber helical twist. Detailed predictions of inter-protomer contacts along this path were tested by site-directed mutagenesis, pilus assembly and protein secretion analyses. We demonstrate that electrostatic interactions between adjacent protomers (P-P+1) in the membrane drive pseudopilin docking, while P-P+3 and P-P+4 contacts determine downstream fiber stabilization steps. These results support a new model of a spool-like assembly mechanism for fibers of the T2SS-T4P superfamily.

INTRODUCTION

Gram-negative bacteria have evolved different strategies to display proteins on their surface to impact their surroundings. T2SS is a widespread multimeric protein complex that enables bacteria to secrete specific folded proteins such as toxins, enzymes and cytochromes from the periplasm (Douzi et al., 2012; Korotkov et al., 2012; McLaughlin et al., 2012). Structural and molecular studies of this system reveal striking similarities with the T4P assembly machineries (Ayers et al., 2010; Mattick, 2002; Nunn, 1999). Furthermore, T2SSs can assemble T4P-like fibers on the cell surface when the major fiber component is overproduced (Durand et al., 2003; Sauvonnet et al., 2000). We term these fibers T2S pili (T2SP), to distinguish them from T4P. Efficient assembly of T2SP in K. oxytoca and Pseudomonas aeruginosa correlates with functional protein secretion (Campos et al., 2010; Cisneros et al., 2012a; Sauvonnet et al., 2000; Durand et al., 2011). This suggests the existence of short periplasmic fibrils, called pseudopili, which promote secretion under native conditions. Most current models assign a piston-like function to the pseudopilus in protein transport (Shevchik et al., 1997; Vignon et al., 2003).

T2SP and T4P are helical polymers of the major (pseudo)pilin subunits, designated as GspG in the T2SS and PilA in P. aeruginosa T4P. Pilins are inserted in the inner membrane (IM) as precursors with a positively charged signal anchor and are processed by the prepilin peptidase GspO (PilD) (Strom et al., 1993) on the cytoplasmic side of the IM. Pilin transmembrane (TM) segments are highly conserved and include the invariable residue E5 that is essential for fiber assembly and function (Campos et al., 2010; Strom and Lory, 1991). The hexameric ATPase motor GspE (PilB) at the base of the system (Camberg et al., 2007; Patrick et al., 2011; Robien et al., 2003) and the IM platform protein GspF (PilC) (Takhar et al., 2013) catalyze fiber assembly and are conserved in all members of this superfamily, including archaeal flagella (Albers and Pohlschroder, 2009). The additional IM proteins GspL (PilM and PilN in T4P), GspM (PilO) and GspC (PilP) link the fiber assembly complex with the outer membrane channel formed by the secretin GspD (PilQ) (Korotkov et al., 2011; Lybarger et al., 2009; Tammam et al., 2013). Upon assembly, the hydrophobic pilin segments are buried in the fiber core and the globular domains are exposed on the surface, providing pili with binding properties required for specific function in adhesion, motility, signaling, DNA or protein transport (Campos et al., 2010; Craig et al., 2006; Kohler et al., 2004). Understanding the molecular basis of these diverse functions requires mechanistic insight into fiber assembly and dynamics.

Structural information obtained by combining electron microscopy (EM) and X-ray crystallography has made it possible to study the function and assembly of these fibers (Campos et al., 2010; Craig et al., 2006; Li et al., 2008). A novel modeling approach provided the first detailed atomic structure of the PulG pilus from the K. oxytoca Pul T2SS (Campos et al., 2011; Campos et al., 2010). A cluster of 200 PulG pilus models obtained by this strategy revealed striking structural similarities with gonococcal (GC) T4P (Craig et al., 2006), in which every pilus protomer (P) interacts with three protomers above (P+1, P+3 and P+4) and below (P−1, P−3, P−4) (Campos et al., 2010) (see also Figure 3, left panel). Biochemical and functional validation of PulG pilus structure demonstrated that conserved electrostatic interactions at the interface between neighboring subunits P and P+1 play a key role in pseudopilus assembly and in secretion of the specific substrate pullulanase, PulA (Campos et al., 2010).

Figure 3. Representative PulG pilus structures in distinct transition path basins.

Left, surface model of the PulG pilus with protomers P (orange), P+1 (green), P+3 (blue) and P+4 (maroon). Right, twist-angle evolution in the transition path showing top (above) and side views (below) of representative structures in basins A, B and C.

Earlier EM analysis of PulG pili suggested their non-uniform structure (Kohler et al., 2004). Furthermore, double cysteine crosslinking used to validate PulG pilus models provided additional evidence for their structural flexibility (Campos et al., 2010). Cysteine residues at positions 16_P and 10_P+1 in the hydrophobic segments resulted in highly efficient crosslinking of PulG subunits in the assembled fibers, indicating the proximity of these residues in majority of the structures. However, lower degrees of crosslinking were also observed between cysteine residues at positions 16_P and 9_P+1 or 16_P and 11_P+1. These mutually exclusive contacts suggest an iris-like movement of the hydrophobic fiber core (Campos et al., 2010). EM studies revealed similar structural heterogeneity in archaeal pili and flagella, with different subunit arrangements and symmetries within the same fiber (Wang et al., 2008; Yu et al., 2012).

To explore the full range of pseudopilus conformations and to gain insight into fiber assembly, dynamics and function in PulA secretion, we analyzed the structural variation of PulG pili by EM. We used information obtained to generate atomic models of the full range of pseudopilus conformational states. The structural diversity produced by modeling was analyzed using a novel approach based on Self-Organizing Maps (SOM) (Kohonen, 2001), in which structures are organized in a 2D map according to their similarity. Global analysis of these conformations provided detailed molecular predictions of interactions, allowing us to test their role in fiber assembly and function.

RESULTS

The T2SS pili show variations in twist angle

Purified negatively stained PulG pili were examined by EM. The averaged power spectrum suggested a helical symmetry of ~4.3 subunits per turn of a 44 Å pitch helix. A reconstruction generated with the Iterative Helical Real Space Reconstruction (IHRSR) method (Egelman, 2000) converged to this symmetry using 19,610 overlapping segments, each 100 pixels (416 Å) long. However, the power spectrum suggested variability of these parameters, as observed previously (Kohler et al., 2004) (Figure 1A). We therefore used a multireference-based approach for sorting by generating filament models with different twists and different subunit axial rise values. The results showed that almost all of the variability was in the twist. In order to assess the robustness of the method, three subsets of filaments were selected, each with an average axial rise of 10.4 Å, but with average twists of 82.8° (n=3,995), 84.3° (n=4,645) and 85.8° (n=2,777) (Figure 1B). When these subsets were run with IHRSR, they converged to the parameters associated with the subset, showing that the sorting procedure worked. It should be emphasized that dividing the filaments into three subsets was completely arbitrary, and that there was no evidence that the twist existed in three discrete states. Rather, all EM observations and analysis suggested that the twist was continuously variable.

Figure 1. Variability of PulG filaments as seen by EM.

(A) A reference-based approach was used to sort segments of PulG filaments extracted from EM images of negatively stained samples. Nine references were generated, having an axial rise of either 8.4, 10.4 or 12.4 Å and a twist of either 82.8°, 84.3° or 85.8°. Each reference was used to generate 36 different projections, involving increments of the azimuthal angle by 10° steps, for a total of 324 reference projections. The distribution found in (A) suggests that almost all of the variability is in the twist. The reality of this sorting can be tested by looking at averaged power spectra.

(B) Averaged power spectra are shown for PulG segments selected from the central three bins of the distribution in (A). It can be seen that the near-equatorial layer line (Bessel order n=4) behaves exactly as expected for filaments with a rather fixed axial rise and a variable twist.

PulG pilus modeling reproduces the twist-angle continuum

To sample structural variations of PulG pili, we generated detailed PulG pilus models based on twist angles randomly chosen between 81.0 and 88.0°. A total of 3901 models were generated by a multistage minimization and molecular dynamics procedure, as described previously (Campos et al., 2011). We used the SOM algorithm to classify the structures according to their conformational similarities on a 2D periodic map, while preserving the topological relationship between the input conformations. Interpolating topological gaps in the conformational changes in this map allowed us to reconstruct a continuous conformational change between the different angles. The SOM was trained with the 3D coordinates of PulG heavy atoms as conformational descriptors (Experimental procedures). The map resulting from this training procedure was visualized by computing distances between adjacent elements of the SOM matrix (Figure 2A). This approach allowed us to classify PulG conformations by reducing and discretizing the input space, representing the continuous set of conformations in a 2D map topologically ordered to cluster closely related structures in contiguous areas.

Figure 2. SOM analysis of 3901 PulG pilus models obtained from continuous distribution of angles between 81° and 88°.

(A) U-matrix of the SOM clustering. The three main basins are labeled A, B and C. An ensemble of 100 possible transition paths is depicted in gray and red line indicates the corresponding mean path (as in B, C and D).

(B) Projection of the twist-angle values (in degrees, purple/green color map) and the U-matrix (contour plot).

(C)Projection of the distances between F1 and E5 in Å.

(D) Projection of the distances between D48 and R87 in Å.

(E) Evolution of the U-value (a), the twist-angle (b), relevant distances along the path between F1 and E5 (P-P+1 interface), K28 and E5 or K35 and E5 (P-P+3 interface) (c), and distances between D53 and K30 or K51 and D29 (P-P+4 interface) (d). Standard deviation is depicted as colored area around the mean curve of the 100 calculated transition paths shown in (A).

(F) Side view (left) and top view (right) of PulG tetramer colored according to the rmsf per residue in Å, from red (rigid) to blue (flexible).

We used the Unified distance matrix (U-matrix, see Experimental procedures), representation to quantitate and visualize structural inhomogeneity. The U-matrix revealed basins of homogeneous conformations separated by barriers reflecting the distance between clusters. The basins and barriers are closely related to the free energy landscape (Bouvier et al., 2014). Quite remarkably, this map suggested contours of three clusters of structural models, corresponding to free-energy basins (Figure 2A). The size of these basins indicated relative free energy, with the smallest, basin A clustering less stable conformations, and basin B populated by transition structures. The largest basin C corresponded to the most stable PulG pilus structures. The twist-angle values computed from the transformation matrix embedded in the descriptors were projected onto the SOM (Figure 2B). The models recapitulated the observed variability in twist-angle (Figure 1B) showing that the most stable structures constituting basin C correspond to the highest twist-angles, above 85.0°.

The correlation between free energy decrease and twist-angle increase in the map (Figure 2A) suggested a plausible transition path along the energy minima in the three structural basins. To derive this path we used an unbiased Markov Chain Monte Carlo (MCMC) algorithm and reconstructed an ensemble of 100 possible transition paths from basin A, to basin C via B (Figure 2A). These were then used to calculate the mean path (red line) and standard deviation (grey zone) (Figure 2A).

For each point of the path we reconstructed the full-atom structures and plotted the U-value extracted from the U-matrix (Figure 2E, panel a) and the twist angle (panel b). The mean transition path contains 115 PulG pilus models. The evolution of the U-value along the path shows a high-energy barrier crossed during the transition from basin A to basin B, which correlates with increase in twist angle (Figure 2E, panel b). The root mean square fluctuation (rmsf) per residue was computed from the ensemble of structures reconstructed along the path (Figure 2F). This calculation shows that the N-terminal part of the α-helix and the external/surface region of the globular domain are highly flexible, as observed in the animations (Movies S1 and S2). In basin A structures pilin TM segments are parallel to each other, while in basin B they come in contact. This contact coincides with the transient reversal of twist, which is overcome in the high twist-angle structures of basin C (Figure 2E, panel b and Movie S2). Based on the global evolution of conformations in the path, from low-twist, high free energy to high-twist, low free energy, we hypothesized that similar transitions could take place during PulG pilus assembly. The twist force provided by the ATPase and the IM platform could drive the assembly and effectively reduce the energy barrier of the transition between basins A and B.

Evolution of inter-protomer contacts in the transition path

We took advantage of the atomic structures along the transition path to gain insight into the molecular basis of T2SP flexibility, by following the evolution of specific inter-protomer distances involving the flexible α-helical PulG stem (Figure 2E, panels c and d). First, we focused on residue E5 that is essential for PulG pilus assembly and protein secretion (Campos et al., 2010). Craig et al. proposed that the interaction of E5_P with the N-terminal amine of protomer P+1 is involved in the docking step of T4P assembly (Craig et al., 2006). However, the evolution of this distance (Figure 2E, panel c, blue trace) and its projection onto the SOM show that these residues generally do not interact in PulG pili, except in the high-energy zone that surrounds basin A (Figures 2C and S1). Instead, in basins B and C residues K28_P and K35_P are in contact with E5_P+3 (Figure 2E, panel c, green and red trace, respectively, Figure 3 and Figure S2). The presence of P-P+3 contacts in stable structures of basins B and C suggests their role in fiber stabilization. We also analyzed the P-P+1 interface contacts between D48_P and R87_P+1 that were previously demonstrated to be essential for PulG pilus assembly and function (Campos et al., 2010). The projection of these distances onto the SOM shows a very close contact of these residues in the low twist-angle basin A (Figure 2D). These observations suggest that the D48_P-R87_P+1 contact and not the E5_P-F1_P+1, plays a critical role in the early docking step of pilus assembly.

Interactions at the P-P+4 interface show inverse correlation with changes in the twist-angle, in agreement with the previously suggested flexibility of this interface (Campos et al., 2010). The contacts between D53_P and K30_P+4 are present in low twist-angle basin A (Figure 3), while alternative contacts between K51_P and D29_P+4 are predominant in basin C (Figure 2E, panel d, blue and green traces respectively).

The P-P+1 interactions determine pilin dimerization in the membrane

Previously the existence of conserved electrostatic interactions at the P-P+1 interface between residues D48_P-R87_P+1 and E44_P-R88_P+1 was demonstrated by using single and compensatory charge inversions (Campos et al., 2010). The proximity of residues D48_P-R87_P+1 in basin A models suggested that P-P+1 contacts form early during assembly of pseudopilins still localized in the membrane. To test this prediction, we studied the interactions between full-length, membrane-embedded PulG by making use of the bacterial two-hybrid (BACTH) system (Karimova et al., 1998). Mature PulG, fused to the C-terminal end of T18 and T25 fragments of Bordetella pertussis adenylyl cyclase (CyaA) led to strong activation of the lacZ reporter expression (Figure 4A, WT, pink bar), indicating efficient dimerization of T18-PulG^WT and T25-PulG^WT chimera in the membrane (Figure 4B). This interaction was stronger than that of leucine-zipper positive control (+) (Figure 4A, grey bar) and was specific, since absent in the T18-PulG^WT-T25 (−) control (yellow bar). To avoid compensatory contacts, we substituted for Ala two residues adjacent in the PulG structure, E44 and D48 (PulG^E44A/D48A) or R87 and R88 (PulG^R87A/R88A) in T18- and T25-PulG chimera. These mutations drastically reduced or abolished lacZ expression (Figure 4A, purple and orange bar, respectively). Weak reconstitution of CyaA activity of the T18-PulG^R87A/R88A-T25-PulG^E44A/D48A pair (orange bar), statistically significant, suggested an orientation favorable to productive contacts T18 and T25 fragments (Figure 4B, orange). Similarly, these double Ala-substituted variants led to reduced LacZ activity when combined with T25-PulG^WT (red and blue bars). Quadruple Ala substitutions in of one of the partners (44-48A/87-88A) abolished the interaction (cyan bar), suggesting that these residues are crucial for PulG dimerization in the membrane. The same Ala substitutions abolished PulG pilus assembly under conditions favoring piliation (Figure 4C), as well as PulA secretion, measured at physiological expression levels (Figure 4D). Consistent with the predictions of the transition path, mutations E5A or K30E had no effect on PulG dimerization (Figure 4A, green and ochre bars). These results demonstrate that residues R87 and R88 of the last assembled protomer P in the pilus provide the docking site for the complementary charged residues E44 and D48 of the incoming P−1 protomer (Figure 4E).

Figure 4. Functional analysis of P-P+1 interface.

(A) Bacterial two-hybrid analysis graph showing mean values of β-galactosidase activity from 6 independent clones producing hybrid proteins with T18 and T25 CyaA fragments. (Ø), negative and (+) positive yeast leucine zipper control, WT, PulG^WT or its variants, with residue positions and substitutions indicated. Error bars represent SD. Unpaired t test was used for statistical analysis: p<0.05 (**), p<0.005 (***), Ψ (no significant difference).

(B) Schematic representation of T18- and T25-PulG hybrid orientation-dependent CyaA reconstitution leading to Lac+ (above) or Lac- phenotypes (below).

(C) PulG immunodetection in 0.025 A_600nm units of cells and pili fractions (C, S) of E. coli PAP7460 carrying pul genes on plasmid pCHAP8184 and pulG alleles as indicated. PulG* indicates a PulG adduct, probably resulting from oxidation.

(D) PulA immunodetection in cell extracts and supernatants (C, S) of 0.005 A_600nm units of E. coli PAP5299 carrying pul genes on plasmid pCHAP8184 and pulG alleles as indicated.

(E) Cartoon PulG pilus model showing P−1-P interactions with E44-R88 and D48-R87 side chains shown as purple spheres.

Contacts P-P+3 and P-P+4 are formed during distinct fiber stabilization steps

The basin B and C models of the transition path are characterized by two specific contacts at the P-P+3 interface, K35_P-E5_P+3 and K28_P-E5_P+3 (Figure 2E, panel c, Figure 3, basin B and C and Figure S2). To test the function of these contacts in piliation we replaced the correpsonding residues by Ala. As shown previously (Campos et al., 2010), the E5A substitution abolishes piliation (Figure 5A). Variant PulG^K28A was assembled into pili with similar efficiency to PulG^WT, while mutation K35A caused reduced piliation (Figure 5A). Combining these substitutions in variant PulG^K28A/K35A abolished piliation, consistent with the model predicting that both lysine residues contribute to fiber stability through interaction with E5_P+3 (Figure 3, basin C and Figure 5A). The role of these residues was confirmed by immunofluorescence microscopy, which showed complete absence of pili for variant PulG^K28A/K35A (Figure 5B). Hence, the requirement of the K28 and K35 residues for piliation supports the model-predicted roles of K35_P-E5_P+3 and K28_P-E5_P+3 contacts in high-twist fibers.

Figure 5. Functional analysis of P-P+3 and P-P+4 interfaces.

(A) PulG immunodetection in 0.005 A_600nm of cell extract and sheared fractions (C, S) of E. coli expressing pul genes on plasmid pCHAP8184 and pulG alleles on pCHAP8658 and derivatives (Table S1).

(B) Immunofluorescence microscopy of the E. coli strains as in (A). Cells stained with DAPI are shown in magenta and PulG pili, revealed with anti-PulG and Alexa-488 conjugated secondary antibodies, in green. Scale bar, 2 μm.

(C) PulA immunodetection in cell extracts and supernatants (C, S) of 0.005 A_600nm units of E. coli PAP5299 carrying pul genes on plasmid pCHAP8184 and pulG alleles as indicated.

(D) Percentage of secreted PulA (mean + SD) from four independent experiments like in (C).

At the P-P+4 interface, residue D53_P at the tip of the PulG α-helical stem interacts with K30_P+4 in the low twist-angle structures (Figure 2A, panel d and Figure 3, basin A). Single charge inversions in variants PulG^K30E and PulG^D53R abolished piliation (Figure 5A). Combining these substitutions in variant PulG^D53R/K30E restored surface pili (Figures 5A and 5B), demonstrating that D53_P and K30_P+4 form a salt bridge. Consistent with the role of conserved D53 in intra-molecular contacts (Figures S3 and S4), substitution D53R led to reduced PulG levels. The existence of the alternative stabilizing contact K51_P -D29_P+4 that characterizes basins B and C (Figure 2E, panel d) has been demonstrated previously by site-directed mutagenesis, showing absence of pili for the charge inversion variant PulG^K51E (Campos et al., 2010).

We analyzed the involvement of P-P+3 and P-P+4 interfaces in PulA secretion at physiological protein levels and in liquid culture conditions, where surface-exposed T2S pili are absent (Campos et al., 2010; Cisneros et al., 2012a; Sauvonnet et al., 2000). Variants PulG^K28A and PulG^K35A, as well as PulG^K28A/K35A, which does not assemble PulG pili, all promoted fully efficient PulA secretion (Figure 5C). Similarly, K30E that abolished piliation did not reduce secretion efficiency (Figure 5C). In agreement with this result, K30E substitution did not affect PulG dimerization in the BACTH assay (Figure 4A, ochre bar), indicating an intact P-P+1 interface for this variant. Similarly, our previous study shows that PulG^K51E variant, altered in the P-P+4 contacts characterizing basins B and C, supports wild type PulA secretion efficiency (Campos et al., 2010).

In summary, consistent with the predictions of the transition path, the results show that PulG pilus interfaces have distinct functions in fiber assembly. The P-P+1 contacts play a crucial role in the docking step essential for pseudopilus assembly and PulA secretion (Figure 4 and (Campos et al., 2010). Interactions P-P+3 and P-P+4 are formed downstream from the P-P+1 contacts to consolidate the fiber. Disrupting the P-P+3 and P-P+4 interface selectively abolishes PulG pilus stability, but not the process of pilus assembly and protein secretion.

Periplasmic pseudopilus assembly is followed by turnover

Although substitutions of PulG surface residues were not expected to affect its stability, we observed different protein levels for some PulG variants, like PulG^K28A/K35A (Figure 5A). We hypothesized that blocking pseudopilus assembly within the periplasm at different stages might alter the susceptibility of PulG variants to proteolysis. Therefore, we followed PulG stability after addition of antibiotics to arrest protein synthesis (Experimental procedures) in conditions of chromosomal pul operon expression that support PulA secretion. In strain PAP7232, wild type PulG undergoes turnover with a half-life of ~100 min (Figure 6A). In contrast, the assembly-defective variant PulG^E5A (strain PAP5327) was highly stable, accumulating presumably in the IM. We used similar expression conditions to test the turnover of variants PulG^K28A/K35A and PulG^K30E encoded from low copy-number plasmids in strain PAP7228, which carries a chromosomal pulG deletion. The targeting and docking-competent variant PulG^K28A/K35A showed very high turnover rates, consistent with the implication of K28 and K35 in early P-P+3 contacts critical for fiber stabilization. In contrast, the turnover of variant PulG^K30E at the P-P+4 interface was similar to that of PulG^WT (Figure 6B). We hypothesize that pilin subunits are extracted from the IM during assembly and transiently protected from proteolysis through interactions with pilins in the fiber. The high stability of PulG^E5A suggests that this variant is arrested in a step prior to membrane extraction and remains in a compartment protected from periplasmic proteases. Variant PulG^K28A/K35A enters into assembly pathway via P-P+1 contacts, but is unable to make the stabilizing P-P+3 contacts upon IM extraction and undergoes rapid degradation. Such stabilizing P-P+3 interactions are present in variant PulG^K30E thereby conferring protease resistance similar to that of PulG^WT. However, the K30E substitution at the P-P+4 interface destabilizes long surface exposed PulG pili due to a clash with residue D53 (Figure 3, basin A). Therefore, P-P+4 interface mutations affect a late stabilization step of T2SP assembly.

Figure 6. PulG turnover during protein secretion.

(A) Left, strains PAP7232 (pulG^WT) or PAP5327 (pulG^E5A) were grown to late exponential phase in LB, 0.4% maltose. Protein synthesis was arrested by the addition of chloramphenicol (170 μg.ml⁻¹) and 0.005 A_600nm of bacterial extracts taken at indicated time points (min) were analyzed by SDS-PAGE and immuno-detection with anti-PulG and anti-RbsB antibodies. Right, levels of PulG^WT (■) and PulG^E5A (□) quantified using Image J, normalized against RbsB levels and plotted as a function of time.

(B) Left, turnover PulG^WT, PulG^K30E and PulG^K28A/K35A variants produced from low copy number plasmids in strain PAP7228 (ΔpulG). Protein synthesis was arrested by the addition of spectinomycin (100 μg.ml⁻¹) and 0.005 A_600nm of bacterial extracts taken at indicated time points (min) were analyzed by SDS-PAGE and immuno-detection with anti-PulG and anti-RbsB antibodies. Right, PulG levels normalized against RbsB levels, as a function of time. PulG^WT (■), PulG^K30E (◪) and PulG^K28A/K35A (□).

In conclusion, the data support a sequential mode of PulG pilus assembly in which P-P+1 interactions drive the initial docking and assembly of pilins at the base of the growing fiber (Figure 7). The twist force, generated by PulE ATP hydrolysis, would spool the docked pilin into the fiber, leading to its partial extraction from the membrane by 1 nm and to its rotation by 84° on average. Two subsequent rounds of such elongation would lead to full extraction of this protomer (now P+3) from the membrane. Relative 250° rotation between the incoming protomer P and P+3, induced by twist force would favor interaction of K28_P and/or K35_P with E5_P+3. These contacts protect the TM segment from proteolysis, as indicated by high turnover rates of variant PulG^K28A/K35A. Although E5 plays a stabilizing role critical at this step, the high stability of PulG^E5A variant suggests an additional role of E5 in an earlier stage, essential for assembly, as discussed below.

Figure 7. The pseudopilus assembly model.

Schematic representation of T2SS components labeled using the single letter code, with the secretin (D) in the outer membrane (OM), platform proteins F, C, L and M and assembly ATPase E in the inner membrane (IM). (1) Major pseudopilin G (in green) is targeted to the minor pseudopilins HIJK (in pink), and to assembly factors E and F. Variant G^E5A (in grey) is defective in this step. (2) GspG protomer P (in blue) docks on the P+1 (in green) at the assembly site via electrostatic contacts (+ and − signs). (3) ATP hydrolysis promotes rotation of F in complex with nascent pseudopilus, spooling P into the fiber to add 10.4 Å to the polymer. (4) Top view showing ATPase E (grey circle with 6 segments) and F (crescent shape), surrounded by CLM complexes (small circles). Cycles of ATP binding, hydrolysis and release (red, yellow and white stars, respectively) drive rotation of F and assembly of G subunits at the base of the growing fiber. (5) After three cycles, P+3 is extracted from the membrane and stabilized through E5_P+3 interactions with K28 and K35. (6) The next elongation step allows P-P+4 interactions to stabilize T2S pili in high twist (via K51_P-E29_P+4) or low twist (D53_P - K30_P+4) states.

The two alternative contacts at the P-P+4 interface formed after next round of elongation, D53_P-K30_P+4 (basin A) and K51_P-D29_P+4 (basins B–C) would stabilize the long-range interactions between pilins in T2SP fibers providing them with flexibility and resistance to external force. Consistent with our model, these contacts are dispensable for initial steps of pseudopilus assembly, protein secretion and PulG stability. To analyze the effect of mutations at this interface on formation of periplasmic pseudopili, we took advantage of the double cysteine-substituted variant PulG^I10C/L16C allowing formation of covalently cross-linked fibers under oxidizing conditions (Campos et al, 2010). This experiment was performed on liquid-grown bacteria, where surface pili are absent. Similar degrees of PulG oligomerization were observed in the WT and the K30E variant, while the E5A variant predominantly showed dimers. Oligomers were not present in strains lacking the Pul T2SS or the ATPase PulE, indicating that they correspond to assembled pseudopili (Figure S5). These results suggest that mutations at the P-P+4 interface do not significantly affect the formation of short native pseudopilus fibers in the periplasm.

DISCUSSION

T2SS pseudopili, T4P and archaeal flagella belong to a conserved superfamily of dynamic fibers that play diverse roles in motility, biofilm formation, signaling and macromolecule transport. Their role in virulence raises the need for fundamental understanding of their functional and substrate specificities. In the present study of the T2SS pseudopilus EM, modeling and SOM analysis, combined with functional and interaction studies provide atomic-level insights into distinct fiber assembly steps and dynamic conformational changes.

EM analysis has provided ample evidence for flexibility of purified fibers of this class (Wang et al., 2008; Yu et al., 2012). Interestingly, externally applied force induces even more dramatic changes in T4P dimensions and organization (Biais et al., 2010). In the present study, EM revealed that the main element of variability of the PulG pili is in their twist. Such twist variability has been observed in filaments ranging from F-actin (Egelman et al., 1982) to MDA5-RNA polymers (Berke et al., 2012), demonstrating it to be more the rule rather than the exception. The interesting question is how proteins accommodate such variability at the level of atomic interactions. As the resolution of EM continues to improve, and with the modeling approach used here, new insights can be gained into the underlying mechanisms of such structural polymorphism.

To gain molecular insight into these variations, the flexible pilus-modeling method described previously (Campos et al., 2010; Campos et al., 2011) was taken a step further to generate a much wider range of pseudopilus structures. Remarkably, modeling based on multiple randomly chosen angles (Figure 2) or on three fixed angles (Figure S1) within the EM-defined range reproduced the twist angle continuum observed by EM analysis. In every case, SOM analysis defined three major low-energy basins whose size correlated with twist-angle increase. The full-atom models of these conformations provided predictions of crucial inter-protomer interactions. The large number of generated structures permitted us to analyze the conformational space that pseudopilins sample globally and revealed striking correlations between free energy, twist angle and distance parameters. The robust calculation of the transition path allowed us to follow the evolution of specific distances and analyze their cross-correlation during conformational changes. Insights into hydrophobic residue contacts in the PulG pilus core provided by the path structures fit well with the cysteine cross-linking data (Campos et al., 2010). Residues 16_P and 10_P+1 interact in all models, in agreement with the high crosslinking efficiency between cysteine residues at these positions. Contacts 16_P-9_P+1 predominate in basin A models, while 16_P-11_P+1 interact in high-twist angle basin C (Figure S6).

The information contained in the transition path structures suggests a rotation-driven mechanism of fiber assembly. The low twist-angle state coincides with specific P-P+1 contacts, which would drive fiber assembly and preset the axial rise of the helical fibers. The results of our study contradict the current T4P assembly model, which assigns the docking role to the E5_P-F1_P+1 contact (Craig et al., 2006). We show that E5_P-F1_P+1 contacts are not only absent in basin A, but also energetically highly unfavorable, corresponding to an energy barrier in the SOM-defined path (Figure 2A, C). These residues are part of the highly mobile N-terminal region of the TM segment (Figure 2F), which undergoes major conformational changes in the path. While reflecting the intrinsic flexibility of these non-covalently linked fibers, we hypothesize that these changes also indicate how pseudopilus responds to external forces that act during or upon assembly. Assembly platform components might transduce the twist force from the ATPase to pilins, favoring E5_P+3 interactions with positively charged residues K35_P and K28_P, required for the transition to more stable structures.

Ala substitutions were used to confirm that P-P+1 contacts in the globular domains form in the membrane and are crucial for subsequent assembly and function. Recent molecular dynamics simulations and interaction studies demonstrate the ability of (pseudo)pilins to tilt and form staggered complexes in the membrane (Lemkul and Bevan, 2011), driven by interactions of their periplasmic domains (Cisneros et al., 2012a). Specific dimerization of major pilins in the IM explains why PulG and the T4 pilin PpdD assemble only homo- and not hetero-polymers when co-produced in the same bacteria (Cisneros et al., 2012b). A rotational force would hence be required to assemble pilin subunits into helical fibers by spooling them into a filament and promoting their gradual extraction from the membrane (Figure 4E).

Taken together, the results presented are consistent with the one-start right-handed helix model of pseudopilus assembly proposed here (Figure 7 and Movie S3). Initiation of assembly requires minor pseudopilins GspHIJK (in pink) in the compartment defined by the secretin GspD in the OM and GspC, L, M and F components in the IM associated with the hexameric ATPase GspE. In the first step, interactions with one or more platform components (CLM in Figure 7) (Gray et al., 2011; Tammam et al., 2013) target GspG subunits to this compartment. This step might involve the essential residue E5 as proposed previously (Aas, et al., 2007; Campos et al., 2010). In step (2) an incoming protomer P (in blue) would dock onto the fiber-associated P+1 (in green) in the membrane plane via the highly conserved electrostatic contacts D48_P-R87_P+1 and E44_P-R88_P+1, to be spooled into the nascent pseudopilus in step (3). Structural analysis of GspE family members suggests the presence of three alternating conformational states: ATP bound (“ready”), ATP hydrolyzing (“active”) and ADP-bound (“resting”) (Misic et al., 2010; Satyshur et al., 2007) (Figure 7, panel 4). We propose that in the docking step (2) GspE is in the “ready” ATP-bound state. In step (3) the ATP hydrolysis (“active” state) would promote the counter clock-wise rotation of GspF associated with the nascent pseudopilus. Finally, ADP-bound “release” state would allow nucleotide exchange. Three cycles of ATP binding and hydrolysis, each promoting insertion of one pilin and fiber elongation by 1 nm, would lead to extraction of P+3 protomer from the membrane. At this stage (5), interactions of E5_P+3 with K28_P and K35_P are critical for P+3 protection against proteolysis and provide the fiber with rigidity and resistance to force. Finally, interaction of residue K51_P with D29_P+4 would complete the helical turn and consolidate the T2S pilus though the stabilizing P-P+4 interface (6). Alternative D53_P-K30_P+3 contacts would stabilize the relaxed low twist state.

Residues E5, D53, K51 and a negative charge at position 29 in interface P-P+4 are fully conserved in all T2SS major pseudopilins, further supporting our results. In addition, D48_P-R87_P+1 and D53_P-K30_P+3 pairs, and residues K28 and K35 are fully conserved in the PulG subclass, but not in the second subclass that includes Legionella Lsp and Pseudomonas Xcp and Hxc T2SSs (Figure S3). Structural differences (e.g. the conserved residue P28) could be at the origin of possible alternative stabilizing contacts that remain to be identified in the latter subclass.

The assembly model proposed here can be extended to describe the assembly of T4P on the basis of high levels of structural similarity and subunit organization between PulG and GC pili. The current three-start helix model of T4P assembly (Craig et al., 2006) proposes an energetically unfavorable extrusion of three pilins from the membrane via an upward movement, leaving a 31.5 Å gap at the pilus base. The GC model also assigns the crucial docking role to the E5 residue. In the new assembly model proposed here, sequential addition of protomers coupled to rotation would provide a smooth transition between the membrane-embedded and pilus-associated states. Although explicitly excluded in the T4P assembly model (Craig et al, 2006), rotation had been proposed earlier as a plausible mechanism to assemble helical fibers (Mattick, 2002). Furthermore, structural similarity of GspE and FlaI ATPase of archaeal flagella with ATP synthase also supports the rotation model (Streif et al., 2008; Reindl et al., 2013). Thus, we propose that pseudopili and T4P rotate during assembly/disassembly, although their higher flexibility compared to the rigid flagella has precluded the observation of this motion (Herzog and Wirth, 2012).

The results presented here have important implications for the mechanism of protein secretion. The widely accepted piston model assigns an essential role to minor pseudopilins at the tip of the fiber as exoprotein substrate binding determinants (Forest, 2008). Support for this model comes from in vitro binding of the minor pseudopilin complex to the specific substrate in P. aeruginosa Xcp T2SS (Douzi et al., 2011). However, based on the structure of the minor pseudopilin GspJ-GspI-GspK complex (Korotkov and Hol, 2008), recent molecular dynamics, interaction and functional studies established that this complex promotes initiation of pseudopilus assembly and possibly activates GspE ATPase (Cisneros et al, 2012a). Upon activation, ATPase-driven addition of GspH would place GspK at the P+4 position, defining the helical fiber geometry leading to GspG assembly during elongation. However, stabilizing contacts may not exist between the minor and the major pseudopilins, which would explain why these subunits have never been found in purified surface pili (Durand et al., 2005; Vignon et al., 2003). Mutations that disrupt “long-range” (P-P+3 and P-P+4) stabilizing interactions between major pseudopilins do not affect protein secretion, further arguing against the role of minor pseudopilins downstream of the assembly step. Indeed, such “unstable” pili are naturally found in the fully functional Hxc T2SS of P. aeruginosa (Durand et al., 2011).

Our results favor alternative models that have been proposed based on the idea of rotation-coupled helical fiber assembly (Mattick, 2002; Nunn, 1999) and show that the assembly step at the IM base is crucial for secretion. In the simplest model, direct binding to major pseudopilins could promote transport of exoprotein substrates. Polymorphic switching during pseudopilus assembly could drive substrate binding and release. Alternatively, in the absence of direct binding, rotation coupled to pseudopilus assembly could provide a torque to facilitate substrate transport through the secretin channel, as in an Archimedes’ screw (Nunn, 1999).

The combination of EM, modeling, functional and biochemical analyses provided new insight into dynamic structural fluctuations of a fiber from the T2SS-T4P superfamily and established an improved and testable model for assembly of this class of fibers. A similar approach could provide insight into T4P retraction, which shows very different dynamic and force range compared to pilus assembly (Clausen et al., 2009). The structure-based modeling and SOM analysis described here could be a general tool to study structural basis of allosteric changes in other proteins and protein complexes.

EXPERIMENTAL PROCEDURES

Electron microscopy

The PulG pili were purified as described previously (Vignon et al., 2003). The PulG pili sample was applied to freshly glow discharged carbon-coated grids and negatively stained with 2% [wt/vol] uranyl acetate. EM images were taken on an FEI Tecnai12 operated at 80 kV with a nominal magnification of 30,000. Micrographs were scanned on a Nikon Coolscan 9000 at a raster of 4.16 Å/pixel. The SPIDER software package (Frank et al., 1996) was used for most of the processing, including the implementation of the IHRSR (Egelman, 2000) algorithm. Filament images were extracted from the micrographs using the helixboxer routine within EMAN (Ludtke et al., 1999).

Molecular modeling and model analysis

We calculated the models of the pili with protocols described in detail previously (Campos et al., 2011), the only difference being that rather than imposing the same twist angle (84.71°) for all models, we chose a twist angle randomly between 81.0° and 88.0° with uniform probability. Self-Organizing Maps (SOM) were used to classify 3901 models from the continuous twist-angle simulation. Conformations were encoded by the 3D coordinates of the heavy atoms of one PulG monomer. The 4×4 transformation matrix containing the symmetry information of the pilus fiber was added at the end of the descriptor to take into account the variation in twist-angle. The resulting matrix of size 3901×3721 was then used to train the SOM. A modified python implementation of the SOM algorithm (Bouvier et al., 2014) was used.

A periodic Euclidean SOM containing 50×50 neurons was constructed as described in Supplemental Experimental procedures. The analysis of large SOM was performed with tools developed in house and implemented in a python library. The Unified distance matrix (U-matrix) was computed as described in Supplemental Experimental procedures. Such a matrix allowed us to compute possible transition paths along the different basins revealed by the U-matrix. 100 paths were computed with a Markov Chain Monte Carlo (MCMC) algorithm. The starting point was defined in basin A and the end point was defined at the global minimum in basin C.

As the neurons of the trained map contained all the information to build a full atom PulG pilus structure, we used it to build structures for all 2500 neurons. The characteristic distances and twist-angles were then calculated for the 2500 structures and projected onto the map. The structures along the path were extracted to obtain an atomic representation of the transition path.

Bacterial strains, plasmids and molecular biology techniques

The plasmids used in this study are listed in Table S1. E. coli strain PAP7460 [ΔmalE444 malG501 F’(lacI^QΔlacZM15 pro+Tn10)] was used for pilus assembly assays and PAP5299 [araD139 Δ(argF-lac)U169 rpsL150 relA1 flb5301deoC1ptsF25 thi pcnB::Tn10) (F’lacI^Q)] for PulA secretion assays as described previously (Cisneros et al., 2012a). PulG stability was assessed in isogenic strains PAP7232 (pulS, pulAB, pulCDEFGHIJKLMNO), PAP7228 (PAP7232 ΔpulG) and PAP5327 (PAP7232 pulG^E5A). Strain and plasmid construction is described in Supplemental Experimental Procedures.

Bacterial two-hybrid assay

Bacterial two-hybrid assay was performed using the cya mutant E. coli strain DHT1 (Dautin et al., 2000) as described previously (Cisneros et al., 2012a). Details are described in Supplemental Experimental Procedures.

PulG pilus assembly and PulA secretion assays

PulG pilus assembly and PulA secretion assays were performed as described previously (Campos et al., 2010). Briefly, E. coli grown on LB agar plates containing 0.4% maltose were collected and resuspended in LB medium at 1 A_600nm ml⁻¹ and vortexed to release pili. Cell and pili fractions were processed for immunodetection with PulG antibodies. For PulA secretion assays, overnight cultures of E. coli were inoculated into LB containing 0.4% maltose and 1 mM iso-propyl β-D-thiogalactoside. Cultures grown to early stationary phase and fractionated by centrifugation. Equivalents of 0.05 A_600nm of cell and supernatant fractions were analyzed by SDS-PAGE and immunodetection using PulA antibodies. Details are described in Supplemental Experimental Procedures.

Immuno-fluorescence microscopy

Bacteria were grown 16 h at 30°C on LB plates containing 0.4 % maltose and appropriate antibiotics, gently resuspended in 1 ml of phosphate buffer saline (PBS) at 1 A_600nm and immobilized on poly-L-lysine-coated coverslips. Samples were fixed with 3.7% formaldehyde, blocked with 1% BSA in PBS and incubated with anti-PulG primary antibodies (1:2,000) and secondary Alexa Fluor 488-coupled anti-rabbit IgG (Invitrogen). Samples were examined with Axio Imager A2 microscope (Zeiss). Images were taken using Axiovision (Zeiss).

Highlights.

• Flexible modeling of PulG pili reproduces twist-angle variability observed by EM

• Conformational transition path shows correlation between pilus stability and twist

• Distinct inter-protomer interfaces determine fiber assembly and stability

• The results support a spool-like mechanism of T2SS and T4P fiber polymerization