Structure of the Neisseria meningitidis Type IV pilus

Neisseria meningitidis use Type IV pili (T4P) to adhere to endothelial cells and breach the blood brain barrier, causing cause fatal meningitis. T4P are multifunctional polymers of the major pilin protein, which share a conserved hydrophobic N terminus that is a curved extended α-helix, α1, in X-ray crystal structures. Here we report a 1.44 Å crystal structure of the N. meningitidis major pilin PilE and a ∼6 Å cryo-electron microscopy reconstruction of the intact pilus, from which we built an atomic model for the filament. This structure reveals the molecular arrangement of the N-terminal α-helices in the filament core, including a melted central portion of α1 and a bridge of electron density consistent with a predicted salt bridge necessary for pilus assembly. This structure has important implications for understanding pilus biology.

N eisseria meningitidis (Nm) is one of the very few extracellular bacterial pathogens able to cross the blood brain barrier after having invaded the bloodstream from the nasopharynx. In the bloodstream an unusually tight interaction between the bacteria and the brain microvasculature endothelial cells is established, leading to cortical plaque formation that results in opening of the blood brain barrier and bacterial invasion of the brain [1][2][3][4][5] . This interaction is mediated by Type IV pili (T4P), long thin polymers of pilin proteins that interact with two endothelial cell receptors, CD147 and the b2 adrenergic receptor (b2AR). The interaction with CD147 is responsible for bacterial adhesion 6 and the interaction with the b2AR induces signalling in endothelial cells 7 . While the host cell signalling events that lead to disruption of the blood brain barrier are somewhat well-defined, the mechanism by which T4P initiate this process is poorly understood and would benefit from a detailed structure of the N. meningitidis T4P.
X-ray crystal structures of full-length Type IV pilin proteins have been obtained by dissociating intact pilus filaments with detergent, revealing a canonical structure with an extended gently curving 53-amino-acid a-helix (a1), the C-terminal half of which is embedded in a globular C-terminal domain [8][9][10][11][12] . The N-terminal half of the a-helix, a1N, protrudes from the globular domain and is comprised of hydrophobic residues, with the exception of a threonine or serine at position 2 and an invariant glutamate at position 5. Two helix-breaking residues, Pro22 and Gly42, which are conserved in Type IV pilins of the IVa class, introduce kinks in a1 and are responsible for its curvature. An additional glycine at position 14 is also conserved. a1N has dual functions in T4P biogenesis: it anchors the globular domain in the inner membrane before pilus assembly 13 , and it interacts with adjacent a1Ns in the assembled pilus, forming a staggered helical array in the filament core 8,9,14,15 . The conserved Glu5 is critical for T4P assembly 14,[16][17][18][19][20][21][22] and models of T4P, based in part on the crystal structure of the full-length pilin subunit from Neisseria gonorrhoeae, PilE (Ng PilE), place this side chain in a salt bridge with the positively charged N-terminal amine of its neighbouring molecule in the otherwise hydrophobic core of the filament 8,9,11,14,23 .
N. meningitidis PilE (Nm PilE) is 78% identical to Ng PilE and 100% identical in a1 (Fig. 1a,b). Both pilins share a B20 aminoacid hypervariable region, located between two conserved cysteines, imparting antigenic variability to these pili [24][25][26] . In N. gonorrhoeae PilE the hypervariable region is located on a b-hairpin that lies on top of the globular domain b-sheet 9,11 . A cryo-electron microscopy (cryoEM) reconstruction of the Ng T4P was determined at 12.5 Å resolution and an atomic model was built by fitting the full-length Ng PilE structure into the cryoEM density 9 . The PilE globular domains fit well into the reconstruction, exposing the hypervariable region prominently on the pilus surface and burying the N-terminal a-helices in the filament core. PilE subunits are related in the Ng T4P reconstruction by an axial rise of 10.5 Å and an azimuthal rotation (twist) of 100.8°. The subunits are held together largely by hydrophobic interactions among the N-terminal a-helices but these helices were not resolved due to the limited resolution cryoEM map. The importance of the hydrophobic interactions was corroborated by biochemical studies, which showed that Ng T4P require detergent to dissociate into pilin subunits 9,11,14 . The interactions between subunits are very strong, requiring temperatures of 60°C to denature the filaments 14 , consistent with their ability to withstand tensile forces of 100 pN or more [27][28][29][30][31] . Yet both Ng and Nm T4P have been shown to undergo a forceinduced conformational change that exposes an epitope, EYYLN, along the length of the pilus 29,32 . EYYLN is located at the end of a1 (resides [49][50][51][52][53] in Ng PilE, a site that is buried by subunit:subunit interactions in the Ng T4P model and only accessible at the tips of the pili 9 . Consistent with this model, anti-EYYLN antibodies normally bind to the tips of Nm and Ng T4P but not along its length 33,34 . Yet when Ng T4P are placed under force using optical or magnetic tweezers or molecular combing they become longer and narrower 29 . This force-induced conformational change exposes the anti-EYYLN epitope all along the length of the pilus, allowing antibody to bind. Similarly, Brissac et al. showed that anti-EYYLN antibodies can bind along the length of Nm T4P when Nm are adhered to endothelial cells, a property that is dependent on expression of the minor pilin, PilX 32 . It is unclear how these T4P extend in a reversible manner without completely disrupting the extensive hydrophobic interactions among the N-terminal a-helices in the core of the filament. Another poorly understood phenomenon of T4P from at least one species, Geobacter sulfurreducens, is their ability to facilitate long-range electron transport, a property that is thought to occur via overlapping p-p orbitals of aromatic residues in a1 (refs 35,36). Higher resolution structures may provide a molecular basis for these and other remarkable properties of T4P.
Here we report the 1.44 Å X-ray crystal structure of Nm PilE and a cryoEM reconstruction of the Nm pilus at higher resolution, B6 Å, than any T4P structure reported previously. Portions of the N-terminal a-helices are clearly resolved in this reconstruction, allowing their precise placement, which provides evidence of a salt bridge between Phe1:N and the Glu5 side chain, and reveals a melting of the central portion of a1N. This structure provides a molecular framework for understanding key aspects of T4P biology.

Results
Structure of the major pilin PilE from N. meningitidis. N. meningitidis DN-PilE, comprising residues 29-161 of PilE from the high adhesive SB variant of N. meningitidis strain 8013 (ref. 37) was expressed with an N-terminal His-tag. This PilE variant, when expressed as a maltose-binding protein fusion on Staphylococcus aureus, induces recruitment of the b2AR in endothelial cells 38 . DN-PilE crystals were grown in the P2 1 2 1 2 1 space group and the structure was solved by molecular replacement with the full-length PilE structure from N. gonorrhoeae (Ng PilE, 2HI2 9 ), which shares 78% sequence identity with N. meningitidis PilE (Nm PilE; Fig. 1a). Data collection and refinement statistics are reported in Table 1. Nm DN-PilE is a globular protein with an N-terminal a-helix corresponding to a1C, the C-terminal half of a1 (Fig. 2a), embedded in a four-stranded antiparallel b-sheet (b1 to b4, residues 77-121). Between a1C and the b-sheet is the ab-loop (residues 54-76), an extended segment with a single-turn a-helix and a 3 10 helix that forms one edge of the globular domain. Two residues in the ab-loop, Ser63 and Ser69, are post-translationally modified in the native PilE 39,40 . On the opposite side of the b-sheet, the polypeptide chain exits b4 and forms a b-hairpin (b5-b6) that lies atop the b-sheet, followed by an irregular segment that protrudes from the globular domain surface then wraps under it, ending in an extended C terminus at the edge opposite the ab-loop. Conserved disulfide-bonded cysteines at the end of b4 (Cys120) and close to the C terminus (Cys154) define the boundaries of the D-region containing the b-hairpin. The most protruding segment of the D-region, the b-hairpin loop and b6 plus the beginning of the loop that follows b6, is highly variable in sequence among N. meningitidis Type IV pilins and is referred to as the hypervariable region (residues B127 to 146) 38 . The b-hairpin loop hooks over the top of the b-sheet and across the b1-b2 loop; Ala129 at its apex forms a main chain hydrogen bond with Trp57 at the beginning of the ab-loop.
Another N terminally truncated PilE structure, from N. meningitidis strain MC58, is available in the Protein Data Bank (PDB ID 4V1J). MC58 PilE is 89% identical in sequence to the SB pilin variant of strain 8013 PilE described here, differing mostly in the hypervariable region. PilE 4V1J crystallized in a different space group than ours, P6, and was solved to 1.43 Å resolution. This structure superimposes well on our strain 8013 PilE structure (root mean square deviation for main chain atoms, 1.1 Å, Supplementary Fig. 1). Small differences in the backbone conformations occur mainly in the hypervariable region at the end of the b6 strand of the b-hairpin and in the loop following this strand. The MC58 pilin has an Ala142 in place of the very prominent Lys140 seen in our 8013 PilE structure. Electron density is absent for a single residue, Ala131, at the tip of the b-hairpin loop in 4V1J, and the b3-b4 loop conformation differs very slightly between the two pilins.
Not surprisingly, the Nm PilE structure is highly similar to Ng PilE (Fig. 2a, b), superimposing with a root mean square deviation of 3.1 Å for all main chain atoms. Since the N-terminal 54 residues are identical in both proteins the missing 29  N-terminal residues in Nm DN-PilE monomer are also expected to be a-helical, forming a continuous a-helix with a1C, as seen in the full-length Ng PilE structure. The conserved Pro22 and Gly42 introduce kinks in Ng PilE, giving a1 a gentle S-shaped curve (Fig. 2b). Both proteins share essentially the same secondary structure elements in the globular domain (Figs 1a and 2). The only notable main chain deviations are in the position of the b3-b4 loop, which bends away from the b-strand in Nm PilE to a greater degree than it does in Ng PilE, and in the b-hairpin, which is more twisted in Nm PilE and has a two-residue insertion in its loop that allows it to contact the ab-loop.
CryoEM reconstruction of the N. meningitidis T4P. To better understand the role of N. meningitidis T4P (Nm T4P) in host cell adhesion and signalling, we generated a cryoEM reconstruction. Not only are the Nm T4P filaments quite flexible, but image analysis indicated substantial variability, mainly in the twist (Methods). The Iterative Helical Real Space Reconstruction (IHRSR) 41 approach was used to obtain the reconstruction, which after sorting converged to a rise of 10.3 Å and a rotation of 100.8°f or the subset of segments used. The cryoEM map, with an estimated resolution of B6 Å, shows well-defined density for the PilE globular domain, with a central depression and a protruding ridge on one side (Fig. 3a). Clear connectivity between the globular domains is observed between subunits in the right-handed 4-start ( þ 4) helix, and to a lesser extent in the left-handed 3-start ( À 3) and right-handed 1-start ( þ 1) helices. Rod-like density is present in the core of the filament consistent with the protruding N-terminal a-helical segments of PilE, a1N (Fig. 3b, c). This density shows that the a1N segments are staggered in a helical array in the filament core and partially tilted relative to the filament axis, with their N termini close to the centre of the filament and their C termini lying at a larger radius. Surprisingly the a-helical density is not continuous, being considerably weaker just before it connects with the globular domains (red asterisks in Fig. 3c). This weak density indicates loss of a-helical order and increased flexibility in a central segment of a1 in the intact pilus.
Atomic model of the N. meningitidis T4P. As a first step in building an Nm T4P model, the full-length Ng PilE structure was fit as a rigid body into the Nm pilus cryoEM density map to determine a global fit for Nm PilE that included a1N. While the Ng PilE globular domain, which includes residues B28-53 of a1 (a1C), fits nicely into the map, the kink in a1N at Pro22 drives the N-terminal end of a1N well away from the rod-like density corresponding to this subunit (Fig. 3c). Thus, we built a model for the Nm T4P by separately fitting the Nm DN-PilE structure (residues 29-161) and a1N from the full-length Ng PilE structure, separated into two segments, a1:1-14 and a1:24-28, into the cryoEM map. These a-helical segments were then joined by an extended non-helical segment, a1:15-23 (Fig. 3d). The Nm T4P model was built iteratively as described in Methods. The final model was selected based on its fit to the electron density map, minimal steric clashes between subunits, and close proximity of the Glu5 side chain with the N-terminal amine of Phe1 in adjacent subunits. Refinement statistics are shown in Table 2. The diameter of the Nm T4P reconstruction and filament model is B 60 Å. The Nm DN-PilE crystal structure within the pilus model fits nicely into the cryoEM map (Fig. 4a, Supplementary Movie 1), having undergone only minor conformational adjustments during fitting and refinement including movement of the b3-b4 loop to bring it into proximity with an adjacent subunit (Fig. 4b). The ridge of density located on one side of the globular domain is occupied by the hypervariable segment of the D-region, the b-hairpin loop, b6 and the loop following b6, with Lys140 prominently displayed at its most protruding point. Lys140 is critical for pilus bundling and T4P-mediated bacterial aggregation, a phenotype associated with the high adhesive variant of N. meningitidis strain 8013 (ref. 42). The long axes of the globular domains run along the 4-start helix, each strand accentuated by the hypervariable ridges (Fig. 4a). The hypervariable region is implicated in binding to and remodelling of endothelial cells and was predicted based upon homology with Ng PilE to be surface-exposed 38 . Four key residues on Nm PilE mediate interactions of Nm T4P with b2AR and activation of endothelial cells: Thr130, Lys140, Asp143 and Lys144 (ref. 38). Thr130 lies on the b-hairpin b5-b6 loop near the top of the protruding hypervariable ridge whereas Lys140, Asp143 and Lys144 are located near the bottom of this ridge (Fig. 4a,b). The distance between these two binding sites on one PilE subunit (B26 Å) is approximately equal to the distance between these sites in neighbouring subunits in the 4-start helix. Thus, b2AR may bind to individual pilin subunits but may also recognize a conformational epitope spanning two subunits.
Pilin subunits are connected in the Nm T4P by contacts between the globular domains, between the globular domains and the a1s, and between the a1s themselves. The staggered globular domains along the 1-start helix have limited contact, between the b3-b4 loop on one subunit and the ab-loop on the next (Fig. 4c). The bottoms of each globular domain fit into a gap between two globular domains in the next turn of the 1-start helix: the b2-b3 loop of the upper subunit contacts the tip of a1 of one of the lower subunits, and the D-region of the upper subunit contacts both the top of a1C and the ab-loop of the other lower subunit (Fig. 4c). These interactions, mediated by mostly polar and charged residues, connect the globular domains in each consecutive turn of the 1-start helical strand, and would block antibody access the EYYLN epitope at the end of a1C. The interactions in the core of the filament are more extensive and involve mainly hydrophobic residues. The primary contact points between the a1s occur along the 1-start helical strand: the continuous a-helical segments a1:24-41, which are an integral part of their respective globular domains, alternate along the 1-start helix of the pilus filament with the ordered segments of a1N (residues 1-14) from subunits in the next turn up in the 1-start helix (Fig. 4d, e). Each a1:1-14 inserts between two a1:24-41s of the 1-start helical turn below it, contacting both a1:24-41 and the globular domain for one of the subunits and a1:24-41 alone for the adjacent subunit. The most N-terminal portion of a1:1-14 extends to the helical turn below this one to contact the C-terminal residues of a1 at the EYYLN epitope (Fig. 4e). Thus, each subunit contacts three consecutive turns of the 1-start helix-the top one via globular domain interactions, the one below that via a1:1-14 interactions with a1:24-41 and globular domains, and the one below that via a1:1-14 interactions with the EYYLN epitope of a1. The hydrophobic interactions in the a-helical core of the T4P provide the remarkable stability of the pili, which are resistant to heat and chemical denaturation and require treatment with detergent in order to dissociate them 14 .
The a-helical core of the N. meningitidis T4P. In our Nm T4P model the a1 segment 24-53 is curved but runs roughly parallel to the filament axis while a1:1-14 is tilted with its N terminus oriented toward the centre of the filament (Fig. 5a-c). This orientation places the side chain of Glu5 in close proximity (o5 Å) to the N-terminal amine of Phe1 (Phe1:N) on the neighbouring a1 in the 1-start helix, consistent with a salt bridge, which matches perfectly the well-defined channel connecting the rod-like densities at their termini (Fig. 5b,c). Glu5 is invariant in Type IV pilins and Type II secretion pseudopilins and is also present in most minor (pseudo)pilins. Although the Phe1-Glu5 salt bridge has been proposed previously to neutralize these charges in the hydrophobic core of the filament 11 and to drive subunit docking into a growing pilus 9 this is the first structural evidence of such an interaction in a T4P. Neither the rod-like density for a1, nor the connection between these densities, were resolved in the 12.5 Å Ng T4P structure 9 . Thr2 is also positioned to form an intermolecular hydrogen bond with Glu5. A hydroxylcontaining threonine or serine is conserved at position 2 and is the only polar amino-acid apart from Glu5 in the first 25 residues of the Type IV pilins. Glu5 is essential for efficient T4P assembly 14,[16][17][18][19][20][21][22] and Type II secretion pseudopilus assembly 43 .
We have proposed that Glu5 facilitates docking of incoming pilin subunits into the growing pilus through electrostatic interactions with Phe1:N of the preceding subunit, neutralizing these two charges in the otherwise hydrophobic environment of the inner  membrane 9 (Fig. 6a). This salt bridge appears to be maintained in the assembled pilus and may also contribute to filament stability.
The cryoEM map shows that a1 is non-helical between the helix-breaking residues Gly14 and Pro22. The loss of a-helical order at a1:15-23 compared with the all-helical conformation seen in the Ng PilE crystal structure lengthens a1 and effectively removes the Pro22-induced kink. By eliminating the a-helical structure for a1:15-23 we were able to place a1:1-14 into the rod-like EM density, at a position quite different from that predicted for Ng T4P (Fig. 3c,d). Yet the a-helices are not particularly tightly packed, leaving gaps in the filament core and a channel B11 Å in diameter that winds through the filament (Figs 3b and 5a). This channel could accommodate flexibility in a1:15-23 and in the pilus filament itself.

Discussion
Our 1.44 Å crystal structure of Nm DN-PilE, together with our B6 Å resolution cryoEM reconstruction of the Nm T4P allowed us to generate a model for this pilus that closely resembles our earlier 12.5 Å Ng T4P structure ( Supplementary Fig. 2) but reveals new structural features that inform pilus biology. The non-helical conformation revealed for residues 15-23 of a1 is surprising given that this region is a-helical not only in the Ng PilE monomer structure but also in the other full-length Type IV pilin structures from Pseudomonas aeruginosa 8 , Dichelobacter nodosus 10 and G. sulfurreducens 12 all of which possess S-shaped a-helices with kinks at Pro22 and Gly/Pro42. These full-length Type IV pilin structures were obtained by dissociating purified pili into pilin subunits with the detergent octyl b-Dglucopyranoside (bOG), which disrupts hydrophobic interactions and solubilizes a1N. In bacteria a1N is anchored in the inner membrane prior to pilus assembly 13 . The a-helical conformation would shield the polar nitrogens and oxygens in the a1N backbone from the hydrophobic phase of the lipid bilayer  (and also from the acyl chains of bOG during crystal growth). Thus, the melting of the helix at a1:15-23 appears to be induced on integration of the pilin subunits into the pilus filament, perhaps to facilitate packing of a1N (Fig. 6a). The conserved helix-breaking residues Gly14 and Pro22 may destabilize the a-helical secondary structure in this region, allowing it to unfold during pilus assembly. Interestingly the Type II secretion pseudopilins also have Gly14 and Pro22 (Fig. 1b) and thus may also adopt a non-helical conformation in this region within the pseudopilus. Similarly, although the Type IVb pilins lack the proline at position 22 they have glycines at positions 11, 14 and 19, which may destabilize this region in a manner similar to that of Gly14/Pro22. In contrast, archaeal flagellins contain an N terminus that is homologous to a1N in the Type IV pilins 44 but possess no proline in this segment, and glycines are not in conserved positions (Fig. 1b). A B 4 Å resolution cryoEM reconstruction of an archaeal flagellar-like filament from Igniococcus hospitalis was recently determined, which shows well-defined density corresponding to a straight, fully a-helical a1N (ref. 45). Thus, the loss of helical order seen for Nm T4P may be a shared feature of the T4P and Type II secretion pseudopili but is not necessarily a universal feature of filamentforming proteins containing a Type IV pilin N-terminal domain.
The new Nm T4P model with its non-helical a1 segment provides a molecular basis with which to understand the reversible force-dependent polymorphism observed for both Nm and Ng T4P 29,32 . Ng T4P can be stretched to approximately three times their length when subjected to forces of B 100 pN, with a corresponding narrowing of the filament diameter 29 . This force-induced conformational change exposes the EYYLN epitope at the end of a1, which is normally buried in the pilus and only accessible at the tip (Figs 4c-e and 6). Similar exposure of this epitope is induced for Nm T4P upon binding to endothelial cells 32 . The mechanism by which Nm and Ng T4P can extend in a reversible manner was not apparent from the lower resolution Ng T4P model, in which the a1s are all-helical and twisted around each other in the filament core 9 . The new Nm T4P model can be envisioned as a coil-like structure: each turn of the coil, which follows the 1-start helical path, is comprised of the globular domains connected by a small number of electrostatic/polar interactions and by extensive hydrophobic interactions with intervening a1Ns (a1: [1][2][3][4][5][6][7][8][9][10][11][12][13][14] from the subunits in the next turn of the coil; each consecutive turn of the coil is held together axially by electrostatic and polar interactions between the globular domains and by a1, which acts as a tether between each turn (Figs 4d,e and 6b, left panel). Shear forces on the pili may disrupt the weaker axial interactions between globular domains in consecutive turns of the coil, and extend residues a1:15-23, stretching the pilus while maintaining the lateral interactions between globular domains, a1:24-41 and a1:1-14 along the coil (Fig. 6b, right panel). Thus, force would produce a spring-like motion separating the coils to elongate the pilus. Such a motion was described recently for the P pilus of the donor chaperone pilus family based on the 3.8 Å resolution cryoEM reconstruction 46 . The architecture of the P pilus is markedly different from that of the T4P: elongated P pilin subunits are connected along the coils by donor-strand exchange of their N-terminal extensions, which insert into a groove in the next subunit to complete an Ig fold 47 ; polar interactions hold the coils of the P pili together in the axial direction. Yet P pili, like the Nm and Ng T4P, can extend under force in a reversible springlike manner, providing an interesting example of convergent evolution.
In our Nm T4P model a1:15-23 spans B 18 Å. In a fully extended conformation these nine residues could potentially span B32 Å, which would take the axial rise per subunit from 10.3 Å to B 24 Å, an increase of B 230%. The melting of a few residues flanking 15-23 could further increase the filament length. Such an extension might separate the globular domains between consecutive turns of the coil enough for antibody to access to the EYYLN epitope, while maintaining the lateral hydrophobic interactions along the coil, between the globular domains, a1:24-41 and a1:1-14. Whereas the accompanying increase in rise per subunit can occur in the P pili by simply tilting the subunits, the more complex interactions holding the Type IV pilins may also shift relative to each other, changing the register of the a1 interactions to increase the rise per subunit, transiently resulting in thinner longer pili. Upon release of force, a1N would collapse to the more compact and stable conformation seen in the Nm T4P model and interactions between the coils would reform. This conformational elasticity would allow N. meningitidis to remain attached via T4P to endothelial cells in the brain vasculature, and N. gonorrhoeae cells to remain attached to the lining of the urethra under shear forces from blood and urine flow, respectively. Of note, the force-induced stretching of Nm T4P would separate the b2AR-binding residue Thr130 from the Lys140/Asp143/Lys144 patch in adjacent subunits (Fig. 4b). Such a change may provide a signal to activate b2AR and cortical plaque formation, leading to loosening of the endothelial cell junctions and opening of the blood brain barrier. The conformational flexibility of the Nm T4P is likely responsible for the structural heterogeneity observed during processing of the cryoEM images, which precluded a higher resolution reconstruction.
The Nm T4P structure may also be relevant in considering the mechanism for charge transfer in the related T4P from G. sulfurreducens. The G. sulfurreducens Type IV pilin is unique (2) Pilin subunits dock into a growing pilus filament, attracted in part by charge complementarity between Glu5, represented by the ( À ) charge, and the positively charged N-terminal amine ( þ ) on the terminal subunit in the growing pilus. (3) The pilus filament is extruded out of the membrane a short distance by the pilus assembly machinery, opening up a gap at the base of the pilus for a new subunit to dock. As the newly added pilin is extruded out of the inner membrane a1:15-23 (red) melts in order for a1:1-14 to pack into the filament core in an a-helical conformation. (4) Another subunit docks into the gap at the base of the pilus. The EYYLN epitope at the top of a1 is shown in orange. This epitope is exposed at the tip of the pilus and buried along its length. (b) Model for the force-induced conformational change of Nm and Ng T4P. Left panel: in its relaxed T4P conformation subunits are held together by lateral interactions between the globular domains along the þ 1-start helix and alternating a1Ns (a1:1-14) from the next turn up, as well as by axial interactions among the globular domains in consecutive turns of the þ 1-start helix (represented by the double-headed arrows in the schematic of the coil). The central portion of a1, a1:15-23 (red), is non-helical but compact. The EYYLN epitope (orange) is only exposed at the tip of the pilus. Right panel: under stress, the axial interactions are disrupted and a1:15-23 becomes fully extended but the lateral interactions between the globular domains and a1:1-14 are maintained for the most part, allowing a spring-like extension of the pilus to expose the EYYLN epitope all along its length. in its absence of a globular domain: this 61 amino-acid pilin has an extended curved a1, as seen for Ng T4P, followed by an unstructured 8 amino-acid C terminus 12 . This pilin shares 50% identity with Nm PilE in its N-terminal 30 amino acids, including Gly14 and Pro22 (Fig. 1b), thus it may also have a non-helical a1 segment when incorporated into a pilus. The G. sulfurreducens pilin contains 6 aromatic residues, Phe1, Tyr24, Tyr27, Tyr32, Phe51 and Tyr57, five of which are also present in Nm PilE. These residues are implicated in charge conduction in G. sulfurreducens T4P, as a mutant, Aro-5, with alanine substitutions for all aromatics except Phe1, makes pili but shows a 10-fold reduction in conductivity 36,48,49 . The authors proposed metallic-like charge conductivity through closely spaced p-p stacked aromatic side chains, similar to electron transport through polyaniline or polyacetylene nanostructures 48 . This model is supported by X-ray microdiffraction data for wild type conductive G. sulfurreducens T4P, which show a 3.2 Å peak representative of a repeating feature within the pilus; this peak is absent in the Aro-5 mutant 35 . Since the spacing of the aromatic residues in existing T4P models 8,9,14 is considerably larger than 3.2 Å, new computational models for the G. sulfurreducens T4P have been proposed that bring the core aromatic residues Phe1, Tyr24 and Tyr27 within p-p stacking distance (3-4 Å) to facilitate long-range electron transport through the pilus 35,50 . Such energy-minimized models have a comparable rise to Ng T4P (B10 Å) but a much smaller twist (55-60°) and are more compact 50 . While feasible, these models assume an all-helical a1.
In our Nm T4P model with its non-helical a1 segment, the aromatic residues are not within p-p stacking distance, with most being 48 Å away (Cg-Cg) from their closest neighbour. Phe1, Tyr24 and Tyr27 form an aromatic band that spirals up the filament axis (Fig. 5d), whereas residues 32, 51 and 57, corresponding to the remaining aromatic amino acids in the G. sulfurreducens pilin, occupy more peripheral positions. Phe1 lies at approximately the same level along the Nm T4P filament axis as the Tyr27 of three subunits preceding (protomers P and P-3, respectively) and B 6 Å from Tyr24 (Fig. 5d), distances that are incompatible with p orbital overlap and the 3.2 Å diffraction peak seen for G. sulfurreducens T4P. The spacing of the aromatic side chains is constrained in Nm and Ng T4P by the presence of the globular domain, which places a lower limit on the subunit rise and rotation. The absence of a globular domain for the G. sulfurreducens Type IV pilins means that they can pack more closely in the pilus, with smaller rise and rotation values. These parameters, together with a non-helical, partially extended portion of a1, may provide more realistic G. sulfurreducens T4P computational models that aromatics except Phe1, Tyr24 and Tyr27, and perhaps other aromatic side chains, within 3-4 Å of each other to form a continuous chain of p-p stacked aromatic side chains for charge transfer. Interestingly, reduction of the solution pH from 10.5 to 2 results in a 100-fold increase in electrical conductivity for G. sulfurreducens T4P and a corresponding increase in the intensity of the 3.2 Å diffraction peak 35 . A possible explanation for this apparent increase in order is that the low pH disrupts the Phe1-Glu5 salt bridge, freeing the Phe1 side chain to optimize its orientation for charge transfer.
In conclusion, the 6 Å structure of the Nm T4P provides a framework for understanding Nm pilus interactions with host cell receptors as well as more general aspects of T4P biology including assembly, function and biophysical properties. The highly conserved T4P N-terminal domain is all-helical in X-ray crystal structures of full-length Type IV pilins, which most likely represents its state in the inner membrane. On the basis of these crystal structures and low resolution EM reconstructions it has been assumed until now that a1 is all a-helical in the intact pilus as well. We have shown that the N-terminal segment is actually partially melted in the assembled pilus. The melting of this region helps to explain pilus heterogeneity, flexibility and elasticity, including T4P's ability to stretch under force, and in the case of G. sulfurreducens T4P, to conduct charge. N-terminal sequence similarity between Nm PilE and other Type IV pilins and Type II secretion pseudopilins suggest that this a1N melting may be a conserved feature, yet one that does not extend to the archaeal flagellins. The Nm T4P cryoEM reconstruction shows the limitations of secondary structure predictions, which can never be more than 80-90% accurate as they fail to account for contextual influences on protein structure. It also illustrates the value of obtaining the highest resolution possible for macromolecular assemblies, as key structural features can be missed with rigid body fitting of atomic structures of isolated subunits. Such features can have important biological implications for the protein complexes.
DN-PilE crystallization, data collection and refinement. DN-PilE crystals were grown by the hanging drop vapour diffusion method at 293 K. Initial crystallization conditions were obtained from the high throughput screening laboratory at the Hauptman-Woodward Medical Research Institute 51 then optimized in-house. Diffraction-quality crystals grew in 4 ml drops containing a 1:1 ratio of protein (15 mg ml À 1 ) and reservoir solution (0.1 M ammonium sulfate, 0.1 M MES pH 6, 34% (W/V) PEG 4,000). The needle shaped crystals were cryocooled in mother liquor with 30% (v/v) glycerol for storage and transport to the Stanford Synchrotron Radiation Lightsource (SSRL) for X-ray diffraction data collection.
Diffraction data for DN-PilE were collected at the SSRL Beamline 7.1 by remote access using Blu-Ice software 52 . Raw data were processed and scaled and structure factor amplitudes were determined by data processing suite XDS 53 . Matthews coefficient calculations using the CCP4 program MATTHEWS_COEF 54,55 indicated that the asymmetric unit contains one DN-PilE molecule with 39.3% solvent content. The crystal structure was solved by the molecular replacement method using the program PHASER 56 in CCP4i 54,57 . The Ng PilE crystal structure (PDB ID 2HI2 9 ), with the N-terminal 29 residues removed, was used as the model. Molecular replacement yielded an unambiguous solution with z-scores RFZ ¼ 9.8, and TFZ ¼ 14.5. The model residues were changed to those of Nm PilE in COOT 58 where necessary and rigid body refinement was carried out followed by 20 cycles of restrained refinement using REFMAC 59 , which brought R work and R free to 0.271 and 0.317, respectively. The DN-PilE model was examined and adjusted in COOT using difference maps. The modified model was further refined using the 'Improvement of maps by atoms update and refinement' tool in ARP/wARP 60 , bringing the R work to 0.197 and the R free to 0.266. A composite annealed omit map was calculated using CNS 61 and all the residues and water oxygens were validated with the omit map, to give R work of 0.211 and R free of 0.238. At this point the coordinates were validated and deposited in the protein data bank (PDB ID 4XNP). However, after employing the PHENIX suite 62 to refine the Nm T4P model we used this program to re-refined the DN-PilE structure with options 'Automatically add hydrogens to model' and 'Update waters', reducing the R-factors to 0.187 (R free ) and 0.210 (R free ). These DN-PilE coordinates were deposited to the PDB under the accession number 5JW8. Data collection and refinement statistics are reported in Table 1.
Preparation of N. meningitidis T4P for cryoEM. N. meningitidis 8013 cells were grown on GCB plates 63 , resuspended in solubilization buffer (20 mM ethanolamine, pH 10.5) and vortexed to shear the pili from the cells. Cells were removed from the sheared sample by two rounds of centrifugation at 4,000g and pili were precipitated by the addition of ammonium sulfate to 10% saturation and 16 h incubation at 16°C. Pili were collected by centrifugation (13,000g, 60 min, 4°C) and resuspended in solubilization buffer. A second round of purification was performed by addition of ammonium sulfate to 10% saturation and precipitation. The final protein concentration was 0.4 mg ml in 20 mM ethanolamine, pH 10.5.
CryoEM and image processing. Samples (2 ml) were applied to plasma-cleaned lacey carbon grids and vitrified using a Vitrobot Mark IV (FEI, Inc.). Images were collected at an operating voltage of 300 keV on an FEI Titan Krios equipped with a Falcon 2 direct electron detector. The magnification used provided a sampling of 1.05 Å per pixel. A total of 2,295 images were collected from a single grid using the EPU automated system on the Krios, and these were reduced to 541 images after discarding those with no filaments or extensively aggregated filaments, poor contrast transfer function (CTF) and so on. Images were stored containing seven 'chunks', where each chunk represents a set of frames corresponding to a dose of B20 electrons per Å 2 . The full dose image stack was used for the estimation of the CTF, using CTFFIND3 (ref. 64), as well as for boxing filaments using the e2helixboxer routine within EMAN2 (ref. 65). The SPIDER software package 66 was used for almost all other operations. Phase reversals in the images were corrected by multiplying each image by the calculated CTF, which is a Wiener filter in the limit of a very low SNR.
Approximately 100,000 overlapping pilus segments, each 384 px long and shifted by 15 px (B 1.5 times the axial rise per subunit) were cut from the long boxes, and these were reduced to 69,621 after discarding those with a large out-of-plane tilt. The IHRSR method 41 was used to generate a global reconstruction, and this was then deformed corresponding to different values of the axial rise and rotation (twist) of the subunits to generate an ensemble of structures used for multi-reference sorting. This sorting suggested that most structural variability existed within the twist, and a new sorting was done keeping the axial rise fixed at 10.3 Å with a twist ranging from 98.8 to 102.8°. The central bin from this sorting, corresponding to a rotation of 100.8°, contained 15,586 segments which were used for the final reconstruction. The amplitudes of the reconstruction were corrected for the CTF by dividing them by the sum of the squared CTFs, since they had been multiplied by the CTF twice: once by the microscope, and once computationally in the phase corrections. The reconstruction was filtered to 5.0 Å and sharpened with a negative B-factor of 150 for preparation of figures.
Atomic model of the Nm T4P. The Nm DN-PilE structure (residues 29-161) was docked into the pilus reconstruction using the 'Fit in Map' tool in Chimera 67 . Residue positions in the tight loop at the C-terminal end of the ab-loop (73-78) and the b3-b4 loop (112-115) were adjusted to optimize their fit in the map. Next, the full-length Ng PilE crystal structure was superimposed on Nm DN-PilE to obtain coordinates for part of the protruding segment of a1, a1N. This superposition provided coordinates for a1N residues 24-28 but residues N-terminal to these fit poorly into the cryoEM map due to the kink induced by Pro22, which directs most of a1N away from the rod-like density in the filament core. COOT 58 was used to fit varying lengths of the most N-terminal portion of the Ng PilE structure into the rod-like density (for example, residues 1-12, 1-13, 2-13, 2-14 and so on). The angle of rotation along the helical axis of this N-terminal a-helical segment was varied to minimize the distance between the Glu5 side chain and the Phe1:N in adjacent subunits in the þ 1 helix. Residues linking this a-helical segment to a1:24-28 were built in an extended conformation due to the weak density in this region. In models where residues 1 and 2 were included in the rod-like density, ideal a-helical phi/psi angles were imposed using COOT as these residues are extended in the Ng PilE structure. Bond lengths and angles of this full-length Nm PilE subunit starting model were idealized using REFMAC 59 . This PilE model, docked into the EM density, was designated Subunit A. A second PilE model was fit into the adjacent subunit density and designated Subunit B. The symmetry operation (rise and rotation) relating these two subunits was determined by LSQKAB in CCP4, and used to generate Nm T4P models of 10 or 21 subunits with CCP4 PDBSET. The cryoEM map in BRIX format was converted into CCP4 format using MAPMAN, and the Nm T4P models were refined against this map using the PHENIX suite 62 . B-factors were fixed during refinement due to the limited resolution of the reconstruction. Refined models were examined within the cryoEM map in Chimera to identify steric clashes between subunits, which were removed by altering side chain conformations and then additional refinement cycles were run. The modelling proceeded iteratively, with each new model examined in CHIMERA, modified to improve its fit in the map, minimize steric clashes and the Phe1:N-Glu5 distance followed by refinement cycles. Refinement statistics are shown in Table 2.
The resolution of the final reconstruction was estimated at B6 Å by Fourier Shell Correlation (FSC) between the map and the atomic model ( Supplementary  Fig. 3). This avoids several problems with the traditional map:map FSC, such as the fact that it can be completely artifactual for helical structures 68 .
Data availability. The Nm DN-PilE was deposited in the PDB under accession number 5JW8. The Nm T4P reconstruction was deposited in the Electron Microscopy Data Bank under accession number EMD-8287, and the corresponding atomic filament model was deposited in the PDB under accession number 5KUA. The data that support the findings of this study are available from the corresponding authors on request.