Cryo-EM Structure of the Type IV Pilus Extension ATPase from Enteropathogenic Escherichia coli

ABSTRACT Type 4 pili (T4P) are retractable surface appendages found on numerous bacteria and archaea that play essential roles in various microbial functions, including host colonization by pathogens. An ATPase is required for T4P extension, but the mechanism by which chemical energy is transduced to mechanical energy for pilus extension has not been elucidated. Here, we report the cryo-electron microscopy (cryo-EM) structure of the BfpD ATPase from enteropathogenic Escherichia coli (EPEC) in the presence of either ADP or a mixture of ADP and AMP-PNP. Both structures, solved at 3 Å resolution, reveal the typical toroid shape of AAA+ ATPases and unambiguous 6-fold symmetry. This 6-fold symmetry contrasts with the 2-fold symmetry previously reported for other T4P extension ATPase structures, all of which were from thermophiles and solved by crystallography. In the presence of the nucleotide mixture, BfpD bound exclusively AMP-PNP, and this binding resulted in a modest outward expansion in comparison to the structure in the presence of ADP, suggesting a concerted model for hydrolysis. De novo molecular models reveal a partially open configuration of all subunits where the nucleotide binding site may not be optimally positioned for catalysis. ATPase functional studies reveal modest activity similar to that of other extension ATPases, while calculations indicate that this activity is insufficient to power pilus extension. Our results reveal that, despite similarities in primary sequence and tertiary structure, T4P extension ATPases exhibit divergent quaternary configurations. Our data raise new possibilities regarding the mechanism by which T4P extension ATPases power pilus formation.

of ATP would allow completion of catalysis culminating in occupancy by the nucleotide preferred by each subunit, depending on its position in the catalytic cycle. Examination of BfpD-ANP by negative staining in the presence of a reducing agent showed the expected toroid structure, and further reference-free alignment showed 2D averages with 6-fold symmetry with protruding edges (Fig. S1).
After confirming structural integrity of the sample, cryo-EM and high-throughput data collection yielded 7,207 good movies and 313,223 good particles for BfpD-ADP. For BfpD-ANP, a data set of 9,853 good movies and 424,708 good particles with welldistributed orientation yielded two classes, with 214,673 and 58,944 particles, respectively. Reference-free 2D averaging demonstrates unambiguous 6-fold symmetry in both nucleotide states, as only 6-fold symmetric 2D averages were obtained without imposing any symmetry (Fig. S1). This result was reproduced for both data sets by cryo-EM, as illustrated by the reference-free 2D averages corresponding to the top views shown in Fig. S2, S3. Subsequent 3D reconstruction with C6 symmetry yielded resolutions of 3.1 Å, 3.0 Å, and 3.7 Å for BfpD-ADP, BfpD-ANP class-1 and BfpD-ANP class-2, respectively ( Fig. 1 and Figs. S2 and S3). BfpD-ANP class-1 has an outward N2D compared to BfpD-ANP class-2. The dimensions of BfpD are 130 Å maximum width, 68 Å height, with a central pore measuring 30 Å in diameter on one side and twice this diameter on the other.
Overall BfpD structure, active site, and hexameric association. De novo models were built for residues 107 to 219, which represents the N2D, and 232 to 534, which represents the CTD ( Fig. 2A and B). N1D and the eight residues intervening the two domains were not visualized, probably owing to intrinsic flexibility, as has been reported for other T4P extension ATPases (15)(16)(17)(18). The catalytic site is formed by the interface between the Arginine finger (Arg 217) of N2D, the Walker A motif (260-267), and the catalytic loop of the CTD (294-310) of each subunit ( Fig. 2 and 3). For the BfpD-ANP data set, despite the addition of ADP, ATP, and AMP-PNP, the active site of The diameter of BfpD toroid center expands by ;5 Å (arrows) in the presence of ANP compared to the diphosphate. Each subunit is colored differently, and the nucleotide is represented in cyan. The EM density for the loop harboring catalytic Glu295 (magenta) is well defined in the ADP structure, but not in the ANP structure (red oval).
BfpD Structure mBio all monomers was fully occupied by electron density consistent with AMP-PNP, suggesting that ultimately ANP occupied the active sites and remained there. The presence of ANP in the catalytic site results in an outward N2D movement, stabilized by interaction between Arg 217 and the g-phosphate. Both structures show catalytic glutamate residue Glu295 and Walker B motif Glu338, 5.3 to 11.6 Å and 6 Å, respectively, from the phosphorus atom in the terminal nucleotide phosphate ( Fig. 2 and 3). The loop (294-310) carrying the catalytic Glu295 has a well-defined cryo-EM density and is positioned closer to the nucleotide in BfpD-ADP compared to BfpD-ANP, assisted by a hydrogen bond between ADP and Tyr296 ( Fig. 2 and 3). Additionally, in the BfpD-ADP map, Mg 21 is seen between Ser267 (from the Walker A motif) and the beta-phosphate of ADP. As in other PilB-family members, Zn 21 is observed far from the catalytic site and from the subunit interface (Fig. 2B), coordinated by the zinc-finger-like tetracysteine motif (Cys403, Cys406, Cys445, Cys446).  The electrostatic map does not reveal a striking difference in overall charge between the two toroid surfaces (Fig. 4A, B), and, therefore, it is not possible from this analysis to make predictions regarding which side faces the membrane, as described for PilB from G. metallireducens (GmPilB) (15). Each subunit forms a chevron-like structure pointing to the right (1 subunit) when seen sideways with the N2D on top (Fig. 1, 4C). At the intersubunit interface, the outer corner of the chevron, formed by the CTD, points toward the The detail shows changes in the interaction network associated with the conformational change that occurs upon transition from dinucleotide to trinucleotide bound. Arg217 repositions to stabilize and place the g-phosphate of ANP at a farther distance from Glu295. Hydrogen bonds, in Å, are shown with green dashed lines, nonbonded interactions are shown by red arcs. Tyr296 and Arg217 are highlighted to emphasize their important role in the indicated conformations.
inner corner of the neighboring (1) subunit, interacting by the N2D/CTD and CTD/ CTD1 interfaces, with ;1400 Å 2 buried surface area. The ATP binding site is not involved in the intersubunit interaction. Several residues face each other within 3.8 Å, suggesting potential salt bridges or hydrogen bonds between adjacent subunits. Starting from the N terminus of the subunit in the left, the pairs of residues likely forming hydrogen bonds are R328-Q203, N286-Y213, N286-K129, and those likely forming salt bridges are D330-R140, D332-R140, K383-D376, D386-R475 (Fig. 4C), where the second residue is from the (1) subunit.
Presence of trinucleotide induces an expansion in the AAA+ ATPase while preserving the 6-fold symmetry. Comparison of the BfpD-ADP and BfpD-ANP structures reveals a subtle outward shift in position of the N2D with respect to the CTD such that the mobile innermost N2D loops delineate a circle of 65 Å in diameter (distance between diagonally located Ala172 residues) in the presence of ADP, and 70-73 Å diameter in the presence of ANP class 1 and 2, respectively ( Fig. 1). This change is accompanied by a slight change in angle of 1.1°for the N2D and 1.6°for the CTD between each monomer and the vertical axis for the BfpD-ANP structure compared to the BfpD-ADP structure (Movie S1), which is consistent with the outward shift in the center of mass described in the much lower resolution TtPilB cryo-EM structure (17). The conservation of 6-fold symmetry in both BfpD structures suggests that the subunits may work in a concerted manner to translate chemical to mechanical energy.
The N2D-CTD domains of BfpD have an intermediate rotation state. We compared BfpD with the four other T4P extension ATPases homologues, each of which has 2-fold symmetry, to try to understand the mechanistic basis of force generation.
The TtPilB-ATPgS crystal structure (PDB ID: 5IT5) is an elongated hexamer with 2-fold symmetry (16). In TtPilB, all six subunits are bound to ATPgS. However, one pair of opposing subunits in the hexameric ring is in the N2D-in, or "closed" conformation, and the other four subunits are in the N2D-out "open" conformation. The four open N2D-out subunits are superimposable, as are the two closed N2D-in subunits (Fig. 5A). With the C-terminal domains superimposed, we found that the open subunits have a 57°outward rotation of their N2D domain, i.e., away from the center of the toroid (N2D-out), compared to the closed subunits (Fig. 5A). Comparing BfpD-ANP to TtPilB-ATP indicated an intermediate conformation: the BfpD N2D is rotated by ;42°from the TtPilB closed conformation and 18°from the TtPilB open conformation (Fig. 5B). Similar results were obtained for BfpD-ADP, as the rotation of N2D between the two nucleotide states was ;1°.
The GmPilB-ANP and GmPilB-ADP crystal structures are also 2-fold symmetric hexamers, wherein four subunits are bound to ANP or ADP, respectively (15). The other two subunits in GmPilB-ANP and GmPilB-ADP structures bind to ADP and are empty, respectively.  BfpD-ANP showed ;46°rotation in its N2D domain from the GmPilB-ANP closed subunits. Consistent with the three distinct kinds of subunits, the nucleotide-binding pockets of GmPilB at the N2D-CTD domain interfaces have different occupancy by ATP and thus varied affinity to the nucleotide. The N2D-in subunits have 100% occupancy of ANP and are proposed to be the active ATP hydrolysis interface. The catalytic Glu395 at this interface is located closest (at ;6 Å) from the phosphorous atom of the ANP terminal phosphate. The other two pairs of subunits with ANP and ADP form an open and closed ATP binding pockets with N2D twists and 0°and 49°, respectively, and partial occupancy of 58-60% ANP and 44-54% ADP, respectively. Larger g-P distances of terminal phosphates, ;10 Å at the ADP and 7-8 Å at the ANP interface, are suggestive of ATP binding and ADP/ATP exchange sites. Furthermore, the G. sulfurreducens PilB (GsPilB) apo structure (PDB ID: 5ZFR) was also found to possess three distinct N2D conformations (18). Two pairs of the open N2D subunits have 66°and 59°rotations from the closed (N2D-in) pair of subunits (Fig. 5A). BfpD-ANP showed a ;42°rotation in its N2D domain from the GsPilB-ANP closed subunits (not shown).
The moderate twist of the N2D domain in the BfpD structure is suggestive of a partially open conformation. The N2D twist is similar to that of the pair of subunits in the GmPilB-ANP structure with a N2D twist of 48°, nevertheless with a distinct rotation axis. As Glu338 and the catalytic glutamate Glu295 are 5.3-11.6 Å away from the g-phosphate of ANP, it appears that the ATPase captured by cryo-EM is a prehydrolysis intermediate between the ADP/ATP exchange and ATP hydrolysis steps of the catalysis cycle. The single rotation state of the N2D domain with respect to the CTD in the case of BfpD concurs with its 6-fold symmetry.
Insertions in the BfpD sequence might account for its 6-fold symmetry. Amino acid sequences from five type-4 pilus extension ATPase homologues, including three that form 2-fold symmetric structures and two from pathogens that have extension ATPases closely related to those of the thermophiles, were compared with BfpD. BfpD showed conservation in the catalytic glutamate, Walker A/B motifs, and arginine fingers, which are required for nucleotide binding, stability, and hydrolysis (Fig. 6). However, several insertions were observed when the N2D and CTD were aligned separately. The N2D contains four conserved arginines, two of which are bound by ATP in the pair of active hydrolyzing subunits in TtPilB (PDB ID: 5IT5). We observed insertions in BfpD around Arg183, Arg217, the linker, and in the catalytic loop that harbors the critical catalytic residue, Glu295 (Fig. 6). We hypothesize that a longer linker and a shift in arginine fingers might be responsible for a different N2D twist seen in BfpD. Furthermore, half of the intersubunit contacts in BfpD-ANP and BfpD-ADP structures were not conserved among its 2-fold symmetric homologues. Taken together, the above sequence differences in the linker and oligomerization interface might be implicated in the observed 6-fold symmetry in BfpD.
BfpD ATPase activity and kinetics. According to molecular modeling with another AAA1-family ATPase (23), the putative catalytic E295 residue in our cryo-EM BfpD model seems to be positioned further from the nucleotide than required for catalysis ( Fig. 3 and Movie S2). We therefore examined BfpD activity, using enzyme that had been purified by metal affinity and size exclusion chromatography. We varied the concentration of ATP in the reaction and determined that the apparent V max and K m of BfpD at 0.5 mg/mL are 2.69 6 0.34 mmole min 21 and 239.31 6 117.63 mM, respectively (Fig. 7A). At an ATP concentration of 3 mM, the specific activity of BfpD was 3.16 6 0.60 nmol min 21 mg 21 (Fig. 7B), which is consistent with published results for other PilB family extension ATPases (5)(6)(7)(8). To assess the role of residues mapped to the catalytic site, we expressed and purified BfpD variants that had mutations in the catalytic glutamate (BfpD E295C ), and in both that residue and the conserved Walker B glutamate (BfpD E295C E338Q ). The single and double glutamate mutations reduced the catalytic rate by 5-and 12-fold, supporting their important role in ATP catalysis. However, in each case, we were able to measure specific activity above background spontaneous hydrolysis (Fig. 7A). The specific activity of BfpD E295C was 0.69 6 0.28 nmol min 21 mg 21 , and that of BfpD E295C E338Q was 0.38 6 0.10 nmol min 21 mg 21 . Importantly, while a plasmid encoding wild-type BfpD was able to complement a bfpD null mutant to restore bacterial autoaggregation characteristic of EPEC expressing BFP, plasmids encoding either the E295C or the E338Q BfpD variants were not (Fig. 7C). Thus, the residual activity we were able to detect is insufficient for function.

DISCUSSION
Using cryo-EM, we determined the structure of BfpD, the extension ATPase of the EPEC bundle-forming T4P. The structure is noteworthy for a number of reasons. It is the first near-atomic structure of a T4P PilB homologue in its native, frozen-hydrated state. It is the first such structure from a pathogenic bacterium. It is an enzyme that is distantly related to its homologues solved to date, all of which were from thermophiles (2,15). Despite its phylogenetic distance, the BfpD monomer is similar to those of other PilB structures (15)(16)(17)(18). As has been the case with all PilB structures solved to date, the full BfpD N terminus (N1D) was not visualized. In addition, the eight residues intervening the N2D and CTD domains, where flexible linkers exist in the PilB structures, were not resolved here. BfpD was determined either in the presence of ADP or a  RMSD values that do not exceed 1.13 Å and 0.96 Å for the N2D and CTD domains, respectively. The isolated CTD domain of BfpD superimposes well with those determined for other PilB structures (see Fig. 5), and the same applies to the isolated N2D domains, while the relative N2D-CTD position is intermediate between the two more extreme open and closed conformations manifested by the other PilB structures.
In contrast to the similarity among monomers of PilB structures, the BfpD quaternary structure revealed unambiguous 6-fold symmetry, while those of the PilB enzymes from thermophilic bacteria show 2-fold symmetry. The 2-fold symmetry is the basis for a symmetric rotary model coupling ATP hydrolysis to large domain shifts, proposed to be translated via the conserved polytopic membrane protein PilC in the center of the toroid, to lift pilin from the membrane (15,16). In contrast, a 6-fold symmetry is more compatible with a concerted model, in which all catalytic sites act synchronously. While it is possible that the differences observed in symmetry and, by extension, in mechanism of activity, are due to differences in the state of the molecules when observed after crystallization versus while hydrated, we believe for reasons discussed below that it is likely that these differences are fundamental to differences between these distantly related enzymes. Of note, the crystal structure of a catalytically inert C-terminal proteolytic fragment of TtPilB containing the N2D and CTD also displayed 2-fold symmetry (17), as does the crystal structure of the N2D-CTD fragment of GsPilB, which was solved in its apo form (18). Thus, ATP hydrolysis is not required for the 2-fold symmetry.
TtPilB was also examined by cryo-EM, both bound to AMP-PNP and without exogenous nucleotide, achieving resolution of only ;8 Å, but, for the first time, the second and third of three predicted N1Ds were observed (17). Comparison of the hexamers in the presence of AMP-PNP and without added nucleotide showed an outward shift in the center of mass of the AMP-PNP structure relative to the structure purified without addition of nucleotide (15,16). They also observed evidence of displacement of the N1D hexamer by 10-13 Å and suggested an alternative model for energy transduction in which the N1D displacement is linked to pilin through a transmembrane complex of essential T4P biogenesis proteins called PilM, PilN, PilO, and PilP.
The BfpD structures in the presence of ADP and AMP-PNP, while not compatible with the symmetric rotary model, show subtle differences in the position of the N2D with respect to the CTD that are similar to those seen in the TtPilB cryo-EM structures (17). These shifts may represent evidence for transduction of mechanical energy at the periphery of the toroid, rather than through PilC at its center. Given that structures of the N1Ds of T4P ATPases have yet to be solved, it is possible that these differences are amplified by conformational changes and transmitted through PilM and PilN to the outer surface of the cytoplasmic membrane. Support for this alternative model is found in the structure of the related type 2 secretion ATPase N1D in complex with the PilM homologue (24) and in the complex formed by a T4P pilin with the PilM, PilN, PilO transmembrane assembly (25).
The conformation of the BfpD nucleotide binding site raises interesting questions about its catalytic activity, as the distances of the putative catalytic Glu295 and of the conserved Walker B Glu338 are approximately 5.3 to 11.6 Å and 6 Å from the phosphorus atom in the ANP g-phosphate, respectively. This appears to be too far for activation of the water molecule that is responsible for hydrolysis (23). By comparison, the corresponding distances of 6 to 6.5 Å at the tightest interface in the GmPilB-ANP structure suggest that all subunits in the BfpD-ANP structure are in a prehydrolytic conformation. Nevertheless, we were able to demonstrate that BfpD has ATPase activity that is comparable to that reported for several other such enzymes (5)(6)(7)(8)(9). Surprisingly, we were also able to measure ATPase activity above background in a BfpD variant that has mutations in both Glu295 and Glu338, which suggests that mere binding of ATP to the open pocket facilitates hydrolysis. However, the residual activity is insufficient for pilus biogenesis, as demonstrated by expressing these variants in a bfpD null mutant. We further suggest that neither the in vitro activity we were able to measure in wild-type BfpD, nor that of any other PilB homologue, is sufficient to power T4P extension. BfpD specific activity of about 3 nmol min 21 mg 21 corresponds to 0.02 ATP molecules hydrolyzed per second per hexamer. In contrast, given an axial rise of 10.5 Å, corresponding to 952 subunits per micron in a T4P (26), real-time video demonstrates that the rate of T4P extension equates to 333 to 1,072 subunits added per second (11)(12)(13). Assuming that each subunit added requires at least one ATP, these in vitro enzyme activity measurements fall short of the values required in vivo by more that 10,000-fold. If six ATPs are required per subunit, this shortfall multiplies correspondingly. Modifications to the in vitro assays, such as addition of phospholipids or partner proteins (27) or imposed hexamerization (28), as well adjustments to the assumptions, would do little to alter this arithmetic. A similar conclusion was reached by the authors of the study that described a thermophilic PilB enzyme with specific activity of 700 nanomoles mg 21 min 21 (10). They suggested that enzyme activity must be higher in vivo, where partner proteins are in optimal orientation and concentration. Interestingly, these authors described complex enzyme kinetics, including substrate inhibition at concentrations exceeding 1.5 mM ATP and a nonlinear relationship between substrate concentration and catalytic rate below that concentration. They interpreted these results as consistent with the symmetric rotary model of catalysis that had been proposed for TtPilB. In contrast, BfpD displayed simple Michaelis-Menton kinetics compatible with a concerted model. The fact that all enzyme subunits were occupied by the nucleotide triphosphate or triphosphate analogue in the presence of ADP, ATP, and ANP is also consistent with a concerted model.
In summary, we report the first 3D structure of a T4P extension ATPase from a human pathogen. The BfpD structure determined under cryogenic conditions in a native state, unparalleled in its detail, in the presence of ADP and a nonhydrolysable ATP homologue, lays a structural foundation to understand similarities and differences of T4P machinery among different clades. The study expands our understanding of mechanisms of catalysis, activation, and energy transduction of the PilB family of T4P extension ATPases.

MATERIALS AND METHODS
Site-directed mutation of putative BfpD active site residues. All bacterial strains and plasmids used in this study are listed in Table S2. A codon-optimized bfpD gene in plasmid BfpD-Hcp (GenScript, USA, a kind gift from Kurt Piepenbrink, Fig. S5) was amplified using PCR with primers BfpDNcoI and BfpDXhoI (Table S3), creating a Leu to Val substitution at codon two to allow cloning into a pET30a plasmid vector at NcoI/XhoI sites and creating plasmid pJZM005. Fast-Cloning (29) was used to introduce substitutions for glutamate codons at amino acid positions 295 and 338 of BfpD in plasmids pJZM005 and pJZM032, respectively. In short, a pair of primers (Table S3) was designed such that each has a complementary sequence, including the mutated codon and divergent sequences overlapping with bfpD. Codons specifying neutrally charged hydrophilic amino acids of similar size to glutamate were chosen to minimize disruptions to structure, and those that could be expressed and purified were studied further. The PCR products were digested with DpnI and subsequently transformed into E. coli DH5a competent cells. The plasmids pJZM032 and pJZM042 encoding BfpD E295C and BfpD E295C E338Q , respectively, were confirmed by sequencing, expressed, and purified from E. coli BL21(DE3) as described below. To make plasmids for in vivo complementation, we first constructed a plasmid harboring wild-type bfpD in low-copy-number cloning vector pWKS30 (30). To do this, bfpD with its N-terminal His tag and S tag, was isolated from pRPA405 (31) as an XbaI-SacI fragment and inserted into pWKS30. The resultant plasmid pEMM1 was later discovered to lack its native stop codon, while a stop codon on the vector was noted downstream, thus adding an elongated nonnative C terminus to the predicted protein. We used FastCloning to restore a TAG stop codon in its original position in pEMM1, and the new plasmid was named pJZM031. Thereafter, pJZM031 was used as template to introduce E295C and E338Q mutations to obtain pJZM032 and pJZM036, respectively. FastCloning was used to introduce the E295C mutation. For the E338Q mutation, we first introduced the mutation using overlap extension PCR (32), which was later cloned into pJZM031 (XbaI/SacI digested). Auto-aggregation and disaggregation were assessed as previously described (31). Briefly, overnight cultures of E2348/69, UMD926, and VCU019 containing pWKS30, pJZM031, pJZM032, or pJZM036 were diluted 100-fold in Dulbecco's modified Eagle's medium (Corning) and grown for 4 h at 37°C before examination by phase-contrast microscopy.
Negative staining and 2D averaging. BfpD in 20 mM Tris (pH 7.6), 0.1 M NaCl, 5 mM MgCl 2 , 2 mM DTT, and either a mixture of ADP (1 mM), ATP and ANP at a ratio of 2:4:5 or 1 mM ADP alone was diluted to 0.01 to 0.03 mg/mL and stained with 0.75% uranyl nitrate using an established protocol (33). Images were acquired at Â50,000 magnification in low dose mode with a Tecnai F20 microscope operated at 120 kV. Reference-free 2D averages were obtained with the SPIDER program (34).
Cryo-EM grid preparation and data acquisition. The cryo-EM sample buffer consisted of 20 mM Tris (pH 7.6), 0.1 mM NaCl, 5 mM MgCl 2 , 2 mM DTT, and 0.5 mM CHAPSO, and additionally either a mixture of ADP (1 mM), ATP, and ANP at a 2:4:5 millimolar ratio (termed the BfpD-ANP data set) or 1 mM ADP alone. Usage of CHAPSO improved BfpD orientation in vitreous ice, which enabled successful 3D reconstruction. The 300mesh UltraAufoil 21.2/1.3 holey-gold (Quantifoil, Germany) grids were cleaned with a customized protocol (35) prior to glow discharge. Purified BfpD was diluted to 4.5 mg/mL and 3 mL was applied to a glow-discharged 300 mesh UltraAufoil 21.2/1.3 holey-gold (Quantifoil, Germany). Grids were blotted for 2 s with ash-free Whatman Grade 540 filter paper in a Vitrobot Mark IV (Thermo Fisher Scientific) and plunged into liquid ethane. Sample quality and distribution was assessed on a Tecnai F20 (Thermo Fisher Scientific) electron microscope. Data acquisition was carried out in a Titan Krios transmission electron microscope (Thermo Fisher Scientific) operated at 300 kV and counting mode, with a Gatan K3 detector and a 10 eV slit width Gatan Quantum Energy Filter (GIF). Data sets were collected in automated mode with the program Latitude (Gatan) with cumulative electron dose of 60 e-/Å 2 applied over 40 frames.
Single-particle image processing. Movie stacks collected for BfpD-ANP and BfpD-ADP data sets were processed in cryosparc2.15. Gain-normalization, movie-frame alignment, dose-weighting, and full and local motion correction were carried out with the patch motion correction. Global and local contrast transfer function values were estimated from nondose weighted motion-corrected images using patch CTF module. Bad micrographs with ice and ethane contamination and poor CTF fits were discarded. Subsequent image processing operations were carried out using dose-weighted, motion-corrected micrographs. 2D-class average images obtained from 1.000 manually picked particles were used to pick about 2.83 and 1.04 million particles in ANP and ADP data sets. Extensive 2D-classifications of 4-8 rounds yielded 424708 and 313223 pure particles which led to 2.98 and 3.1 Å consensus maps. The reported resolutions of the cryo-EM maps are based on FSC 0.143 criterion (36). Sixfold rotational symmetry (C6) confirmed from the 2D class averages was applied during 3D refinement. The density of ANP was confirmed at the interphase of N2D and CTD, though 2.5:1 millimolar mixture of ANP and ADP was used in the mixed nucleotide data set. A summary of image processing of the ADP and ANP data sets can be found in Fig. S2 and S3, respectively.
De novo model building and structure refinement. A crude model for a single BfpD subunit was built using the ADP map with the Phenix.map_to_model tool (37), which was improved by chain tracing. The regions from 107 to 223 (N2D) and 231 to 534 (CTD) were built. The density for the N terminus (N1D; 1 to 106) and flexible linker connecting N2D and CTD (224 to 230) was not visualized. A clear density for a loop spanning 294 to 310 in the ADP structure was obtained, which is disordered in the ANP structure. The monomeric model was expanded to a hexamer model. Local density fit of the modeled sequence was improved over an iterative process of amino acid fitting in Coot (38) alternated with real space refinement in PHENIX (39). Real space refinement was carried out with NCS constraints and secondary structure and Ramachandran restraints. Comprehensive model validation was carried out with PHENIX and a PDB validation server at https://validate-rcsb-2.wwpdb.org/ and are summarized in Table S1. Surface charge was calculated with chimera (40) and the binding interface buried area was calculated using PISA (41). Figures were generated with the PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC (https://pymol.org), and chimera.
ATPase activity. ATPase activity was measured using a previously established method (42,43) with slight modifications. In brief, stocks of BfpD purified by cobalt affinity chromatography were prepared in assay buffer (150 mM Tris-HCl, pH 7.6, 300 mM NaCl, 1 mM MgCl 2 , 2 mM DTT) to achieve 0.5 mg/mL in the final reaction and mixed with various concentrations (50 mM to 1,500 mM) of freshly prepared ATP in the same assay buffer. The reaction mixtures (in triplicate) were aliquoted to 96-well plates (one plate for each time point) and incubated at 37°C. At the defined time points, the reaction was stopped by adding 100 mL of the assay reagent (a 3:1 mixture of freshly prepared 0.045% malachite green hydrochloride in water and 4.2% ammonium molybdate in 4 N HCl, along with 1% Triton X-100) and followed by 20 mL of citrate solution (34%). The absorbance at 655 nm was measured on a Clariostar Monochromator Microplate Reader (BMG Labtech). The ATPase activity was extrapolated from a standard curve of a set of defined phosphate concentrations present in two columns of each plate. To measure apparent K m and V max , various ATP concentrations and the reaction was terminated at multiple time points. Released PO 4 concentration was plotted as a function of time for the different ATP concentrations, and the slope was used as initial velocity (V 0 ). Data from seven biological replicates were fit by nonlinear regression to the Michaelis-Menten equation with SigmaPlot (Systat Software, Inc.) to calculate apparent K m and V max . The specific activity of BfpD was measured by various BfpD concentrations (0.125 to 2.0 mg/mL) in the assay buffer (above) with 3 mM ATP and a 30 min reaction time at 37°C. The rate of T4P extension in microns per time was estimated from several studies with various methods (11)(12)(13) and from an estimate of 952 T4P subunits per micron in the pilus (26).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. MOVIE S1, MPG file, 5.