PilN Binding Modulates the Structure and Binding Partners of the Pseudomonas aeruginosa Type IVa Pilus Protein PilM*

Pseudomonas aeruginosa is an opportunistic bacterial pathogen that expresses type IVa pili. The pilus assembly system, which promotes surface-associated twitching motility and virulence, is composed of inner and outer membrane subcomplexes, connected by an alignment subcomplex composed of PilMNOP. PilM binds to the N terminus of PilN, and we hypothesize that this interaction causes functionally significant structural changes in PilM. To characterize this interaction, we determined the crystal structures of PilM and a PilM chimera where PilM was fused to the first 12 residues of PilN (PilM·PilN(1–12)). Structural analysis, multiangle light scattering coupled with size exclusion chromatography, and bacterial two-hybrid data revealed that PilM forms dimers mediated by the binding of a novel conserved motif in the N terminus of PilM, and binding PilN abrogates this binding interface, resulting in PilM monomerization. Structural comparison of PilM with PilM·PilN(1–12) revealed that upon PilN binding, there is a large domain closure in PilM that alters its ATP binding site. Using biolayer interferometry, we found that the association rate of PilN with PilM is higher in the presence of ATP compared with ADP. Bacterial two-hybrid data suggested the connectivity of the cytoplasmic and inner membrane components of the type IVa pilus machinery in P. aeruginosa, with PilM binding to PilB, PilT, and PilC in addition to PilN. Pull-down experiments demonstrated direct interactions of PilM with PilB and PilT. We propose a working model in which dynamic binding of PilN facilitates functionally relevant structural changes in PilM.

are structurally and functionally related (8,9). GspL and BfpC are bitopic membrane proteins that appear to be the functional equivalents of PilM and PilN. GspL and BfpC bind to and activate their PilB equivalents (24 -27) and bind PilC equivalents (24,27). These data suggest that PilM may bind PilB and/or PilC. However, GspL and BfpC deviate substantially from PilM. These proteins are more than 100 residues shorter and consequently lack the ability to bind ATP (28), thereby limiting the extent to which we can generate testable hypotheses for PilM based on the structures of GspL and BfpC.
Despite the similarities among the T4aP, T4bP, and T2S systems, PilM is structurally more similar to FtsA, a key component of the bacterial divisome (15). FtsA forms polymers that can be visualized by negative stain electron microscopy and as a consequence of the packing interactions in FtsA crystal structures (29,30). FtsA binds to FtsN, and this interaction is thought to lead to dissociation of the FtsA polymers, exposing protein-binding sites that allow for recruitment and activation of downstream cell division proteins (31)(32)(33). PilN and FtsN share 14% sequence identity (28% similarity) and have similar topology, including conserved cytoplasmic N-terminal residues (32), a single transmembrane helix, and a structurally similar circularly permutated ␤␣␤␤(␣␤) ferredoxin-like fold (RMSD C␣ ϭ 1.9 Å over 43 residues, PDB codes 1UTA and 4BHQ). Furthermore, identification of key residues involved in the FtsN-FtsA interaction suggests that FtsA binds FtsN using a binding pocket analogous to that used by PilM to bind PilN (33). Similar comparisons have been made elsewhere (32,34), leading us to hypothesize that binding of PilN to PilM causes structural changes in PilM that dictate the nature of its binding partners.
Herein we describe the structures of PilM and of PilM C-terminally linked to the first 12 residues of PilN (PilM⅐PilN(1-12)).
Crystallographic and functional characterization revealed that PilM binds its own N terminus to form a dimer and that binding of ATP facilitates PilN binding. The binding of PilN was found to cause long range conformational changes and to lead to PilM monomerization. Furthermore, we show that PilM binds to PilB, PilT, and possibly PilC and that binding of PilN may modulate the interaction of PilM with PilB, PilT, and/or PilC.

Experimental Procedures
Primers, Strains, and Plasmids-The primers, strains, and plasmids used in this study are listed in Table 1.
For PilM and PilM⅐PilN (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12), cell pellets were subsequently resuspended in 30 ml of buffer A1 (50 mM Tris, pH 7.0, 150 mM NaCl, 30 mM citrate, 10% (v/v) glycerol, 2 mM tris(2-carboxyethyl)phosphine, 5 mM MgCl 2 , 5 mM ATP, and 50 mM imidazole). ATP was omitted when PilM was used in the ATPase assays. Resuspended cells were lysed by passage through an Emulsiflex-c3 high pressure homogenizer, and the cell debris was removed by centrifugation for 30 min at 40,000 ϫ g. The resulting supernatant was passed over a column containing 5 ml of pre-equilibrated Ni-NTA-agarose resin (Life Technologies, Inc.). The resin was first washed with 10 column volumes (CV) of buffer A1 and then 10 CV of buffer A2 (buffer A1 adjusted to pH 6.0) before elution with 5 ml of buffer A2 plus 300 mM imidazole. The protein was then further purified by size exclusion chromatography on a HiLoad TM 16/600 Superdex TM 200pg column pre-equilibrated with buffer A2 without ATP or imidazole. Purified PilM constructs, with the exception of PilM ⌬2-8 , which precipitated at room temperature, were treated with thrombin (EMD Millipore) for 16 h at room temperature and separated from the tagged protein using a second round of nickel affinity purification. PilA assembly into a pilus fiber is referred to as pilus extension, whereas PilA fiber disassembly is referred to as pilus retraction. PilB has been implicated in pilus extension, whereas PilT is thought to be involved in pilus retraction. The schematic is not to scale. OM, outer membrane; PG, peptidoglycan; IM, inner membrane. The single letters represent protein names (e.g. PilM (M)).
For the interaction studies between PilM, PilB, and PilT, cell pellets were resuspended in 30 ml of buffer B1 (100 mM Tris, pH 7.0, 150 mM NaCl, 30 mM citrate, 150 mM (NH 4 ) 2 SO 4 , 15% (v/v) glycerol, 5 mM MgSO 4 , 50 mM imidazole). Resuspended cells were lysed by passage through an Emulsiflex-c3 high pressure homogenizer, and the insoluble components were pelleted by centrifugation for 60 min at 40,000 ϫ g. With the exception of PilB, the resulting supernatant was passed over a column containing 5 ml of pre-equilibrated Ni-NTA-agarose resin (Life Technologies). The Ni-NTA-agarose resin was then washed with 10 CV of buffer B1 and then 10 CV of buffer B2 (200 mM Tris, pH 6.0, 300 mM NaCl, 60 mM citrate, 300 mM (NH 4 ) 2 SO 4 , 10% (v/v) glycerol, 10 mM MgSO 4 , and 50 mM imidazole) before adjusting the pH with 10 CV of buffer B3 (buffer B1 adjusted to pH 7.0). The protein was eluted in 5 ml of buffer B3 plus 300 mM  imidazole and then further purified by size exclusion chromatography on a HiLoad TM 16/600 Superdex TM 200pg column  pre-equilibrated with buffer B3 without imidazole. For PilB, after the cell pellets were homogenized and centrifuged, insoluble inclusion bodies were solubilized by gentle stirring in 40 ml of buffer B3 plus 8 M urea for 45 min at room temperature. Components that remained insoluble were removed by centrifugation for 30 min at 40,000 ϫ g. The resulting supernatant was passed over a HisTrap HP 5-ml (GE Healthcare) nickel affinity column. The resin was washed with 4 CV of buffer B3 plus 8 M urea before elution with a gradient of buffer B3 plus 8 M urea and 300 mM imidazole. Eluted fractions containing PilB were then concentrated to ϳ12 mg/ml in an ultrafiltration device, and 300 l were added (in 50-l increments every minute for 6 min) to 50 ml of gently stirring renaturation buffer (buffer B3 without imidazole plus 5% (v/v) glycerol, 0.5 M arginine, 2 mM tris(2-carboxyethyl)phosphine, and 1 mM ATP) and incubated at 4°C for 16 h. Then the 50-ml solution was exchanged for buffer B3 without imidazole and concentrated to 500 l over an ultrafiltration device. The protein was then further purified by size exclusion chromatography on a HiLoad TM 16/600 Superdex TM 200pg column pre-equilibrated with buffer B3 without imidazole. Purified proteins were stored at 4°C for less than 2 days before use.
Crystallization, Data Collection, and Structure Solution-For crystallization, purified PilM and PilM⅐PilN(1-12) were concentrated to 15 mg/ml at 3000 ϫ g in an ultrafiltration device, and ATP or AMP-PNP (Sigma-Aldrich) was added to a final concentration of 5 mM as detailed below. Crystallization conditions were screened using the complete MCSG suite (MCSG 1-4) (Microlytic), using a Gryphon LCP robot (Art Robbins Instruments). Crystal conditions were screened and optimized using vapor diffusion at 20°C using Art Robbins Instruments Intelli-Plates (96-3 Shallow Well Hampton Research) with equal amounts of protein and reservoir solution. For PilM, the reservoir solution was 27% (w/v) PEG3350, 0.2 M MgCl 2 , 0.1 M BisTris, pH 7.0. For PilM⅐PilN (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12), the reservoir solution was 2 M (NH 4 ) 2 SO 4 and 0.1 M Tris, pH 8.0. The crystals were cryoprotected for 5 s in reservoir solution supplemented with 20% (v/v) ethylene glycol before vitrification in liquid nitrogen. Single wavelength anomalous diffraction or native data, as appropriate, were collected on Beamline X29 at the National Synchrotron Light Source (see Table 2). The data were indexed and scaled using HKL2000 (36). Native data were also collected on PilM with AMP-PNP at the x-ray diffraction facilities at the Hospital for Sick Children on a Bruker D8 Venture machine. These data were indexed and scaled with SAINT (Bruker AXS Inc., Madison, WI). PHENIX Autosol (38) was used to identify the 10 selenium atom positions in the asymmetric unit of SeMet PilM. The resulting electron density map was of high quality and enabled PHENIX AutoBuild (39) to build ϳ90% of the protein. The remaining residues were built manually in COOT (40). The structures of native PilM and PilM⅐PilN(1-12) complexed with ATP or AMP-PNP were solved by molecular replacement using the SeMet-PilM struc-ture as a template in Phaser (41). Through iterative rounds of building/remodeling in Coot (40) and refinement in phenix.refine (42), the structures of the SeMet-PilM⅐ATP, native PilM⅐ATP, native PilM⅐AMP-PNP, and native PilM⅐PilN(1-12)⅐ATP were built and refined. All structures were refined with individual B-factors except for the PilM⅐AMP-PNP structure, where the resolution (3.5 Å) of the data necessitated the use of grouped B-factors during refinement. Progress of the refinement in all cases was monitored using R free .
ATPase Assays-ATPase assays were performed as described previously (43) in two independent trials done in triplicate. Briefly, in a transparent 96-well plate, 50 l of buffer A1 (with 5 mM ATP) without imidazole was incubated with 200 M PilM for 16 h at 37°C. Eighty l of 1 mM malachite green, 10% (v/v) Tween 20, and 6 mM H 2 SO 4 was then added. Finally, 80 l of 50 mM ammonium molybdate and 3.4 M H 2 SO 4 was added, and the amount of phosphate released was determined by measuring absorbance at 630 nm with a SpectraMax M2 microplate reader. Comparison with a dilution series of phosphate-buffered saline permitted the quantification of the released inorganic phosphate.

Size Exclusion Chromatography Followed by Multiangle Light Scattering (SEC-MALS) Analysis-A Malvern Viscotek
GPCmax system connected to a Superdex 200 Increase 10/300 GL size exclusion column (GE Healthcare) was used for SEC-MALS analysis of PilM and PilM⅐PilN (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12). The column was equilibrated with buffer A2 without ATP. The intensity of light scattered by the column eluate was recorded using a Malvern 270 dual detector, and the refractive index was monitored using a VE 3580 detector. Bovine serum albumin at a concentration of 1 mg/ml was used for detector normalization. Molecular weights were calculated from Zimm plots with a protein refractive index increment (dn/dc) of 0.185 ml/g using OmniSEC version 5.10 software (Malvern Instruments Ltd.).
Bioinformatics Analysis of the N Terminus of PilM-Using BLAST (44), protein sequences with significant homology to PilM were identified. Sequences that were not annotated as a pilus or fimbrial protein, duplicate sequences, and sequences with duplicate species identifiers were removed, leaving 2231 sequences that were aligned in MEGA (45). We compared the alignment with our structure of PilM to identify sequences with N termini of greater than 7 residues, which was the size of the a R Sym ϭ ͚͚͉I Ϫ (I)͉/͚͚I, R Pim ϭ ͚͌(1/(n Ϫ 1) ͚͉I Ϫ (I)͉/͚͚I, and CC* ϭ ͌ (2CC1 ⁄ 2 /(1 ϩ CC1 ⁄ 2 )), where CC1 ⁄ 2 is the Pearson correlation coefficient of two half-data sets, as described elsewhere (69).
where F o and F c are the observed and calculated structure factors, respectively. R free is the sum extended over a subset of reflections (5%) excluded from all stages of the refinement. c As calculated using MolProbity (37). d Maximum likelihood-based coordinate error, as determined by PHENIX (42). motif bound by PilM. We searched these N termini for the hXXhhXϩ motif (where h is a hydrophobic residue, X is any residue, and ϩ is a positive residue) and the hX(X)hh(X)ϩ motif (where (X) represents an optional single residue deletion) using the SMS program (46). The motif sequences were extracted by the SMS program, and the consensus motif was visualized using WebLogo (47).
Negative Stain Electron Microscopy-EM grids (400-mesh Cu/Rh, Electron Microscopy Sciences) were overlaid with continuous carbon and subjected to glow discharge before PilM, diluted to 0.04, 0.4, or 4 g/ml, was applied. Excess solution was blotted away after a 2-min incubation at room temperature, and the grids were rinsed three times with water. Uranyl acetate stain (0.5% (w/v)) was applied, excess stain was blotted away, and the grid was dried. Grids were examined with an FEI Technai F20 electron microscope operating at 200 kV. Images were collected with a Gatan Orius 832 CCD camera.
In Vitro Pull-down Experiments-For pull-downs, 100 M PilM or PilM⅐PilN(1-12) was incubated with 100 M PilB or PilT in 50 l of buffer B3 for 1 h at 20°C. As described above, the PilM constructs were treated with thrombin, and the His tag was removed while both PilB and PilT were His-tagged. The proteins were then applied to 50 l of pre-equilibrated Ni-NTA-agarose resin, washed with 35 ml of buffer B3 plus 10 mM arginine, and eluted in 50 l of buffer B3 plus 300 mM imidazole. Eight microliters were then separated on SDS-PAGE, and Western blot analysis was performed using rabbit anti-PilM (16), anti-PilB (10), and anti-PilT (17) antibodies. Goat anti-rabbit HRP conjugate (Bio-Rad) was used to detect PilM, PilT, or PilB by chemiluminescence (Thermo Fisher Scientific).
Isothermal Titration Calorimetry of PilM with ATP and ADP-The dissociation constant (K d ) for nucleotide binding was determined using isothermal titration calorimetry (ITC) using a MicroCal Auto-iTC200 (Malvern Instruments Ltd.) at 10°C. PilM was used directly from the SEC column without further concentration to avoid buffer mismatch. 200 l of 96 M PilM, 73 M PilM⅐PilN(1-12), 77 M PilM K52A, or 55 M PilM S19A were injected into the cell. 2 l of SEC buffer or 1.5 mM ATP or 1.5 mM ADP resuspended in the SEC buffer were injected into the cell containing PilM. A total of 25 injections were performed with an equilibration time of 150 s between injections. The data were fitted with Origin 7 ITC analysis software using a single-site binding model.

Results
Improving the Solubility of P. aeruginosa PilM-Initial E. coli expression and purification of P. aeruginosa PilM revealed that the protein was poorly expressed, found predominantly in high molecular weight aggregates, and that it precipitated rapidly following purification. The crystallizability of T. thermophilus PilM (PilM Tt ) was improved by incubating it with a peptide corresponding to the cytoplasmic N-terminal portion (residues 1-15) of PilN Tt , although only residues 1-8 were identified in the resulting electron density (15). Similarly, in the crystal structure of cytoplasmic GspL, residues 1-12 of its PilN-like portion could be modeled (51). In an attempt to improve the solubility of P. aeruginosa PilM, we fused the cytoplasmic portion of PilN (residues 1-12) via a 3-residue linker to the C terminus of PilM (PilM⅐PilN(1-12)). Although PilM⅐PilN(1-12) also precipitated after purification, we exploited this property in a solubility screen to find a more compatible purification buffer. The solubility of PilM⅐PilN(1-12) was improved in 50 mM Tris and 30 mM citrate, pH 6.0, 150 mM NaCl, 10% (v/v) glycerol, 2 mM tris(2-carboxyethyl)phosphine, 5 mM MgCl 2 , and 5 mM ATP. Conveniently, this buffer also improved solubility and reduced aggregation of native PilM.
Structure Determination of PilM and PilM⅐PilN(1-12)-To study potential structural changes upon PilM-PilN interaction, we determined the crystal structures of PilM and PilM⅐PilN (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12). PilM and PilM⅐PilN(1-12) crystallized in a number of different conditions. Those that favored the growth of large single crystals of PilM and PilM⅐PilN(1-12) were identified after optimization ( Table 2). The crystals of PilM and PilM⅐PilN (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12) indexed to space groups P2 1 2 1 2 and P321, respectively, and diffracted to 2.4 Å. Because we were unable to solve the structure of PilM or PilM⅐PilN(1-12) by molecular replacement using the available structure of PilM Tt ⅐PilN Tt (1-15), a SeMet derivative of PilM was crystallized, and its structure was determined using the single wavelength anomalous diffraction technique to 2.5 Å resolution. The structures of native PilM and PilM⅐PilN (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12) were subsequently solved by molecular replacement using our SeMet-PilM structure as a template, and both were found to have two PilM molecules in the asymmetric unit. Model building and refinement produced final structures of SeMet-PilM, PilM, and PilM⅐PilN(1-12) with R work /R free values of 20.7/26.2, 22.4/25.7, and 21.1/23.8%, respectively. Some residues could not be modeled due to poor electron density. The entire PilM sequence could be modeled into both molecules in the asymmetric unit of the SeMet PilM structure, except for residue 9 of one molecule. In both molecules of the asymmetric unit in the native PilM structure, the entire PilM sequence could be modeled, except for residues 8 and 9. In one molecule of the PilM⅐PilN(1-12) structure, resi- The 1B and 2A domains also create an interface that binds inter-or intramolecular peptides in all superfamily members identified herein. For PilM, this peptide is the cytoplasmic N terminus of PilN. For GspL and BfpC, this peptide corresponds to the cytoplasmic PilN-like portion of GspL or BfpC, respectively. For FtsA, this peptide is expected to correspond to the N terminus of FtsN (32, 34), as mentioned above, although as yet there are no corroborating structural data. This peptide in DnaK corresponds to the linker region between the actin-like portion of DnaK and its substrate-binding domain. For hexokinase, this peptide corresponds to the linker region between an N-terminal domain and the actin-like portion of hexokinase. For actin, this peptide corresponds to the WH2 motif of Dnase I, and several other peptides bind actin at this same interface (53). Peptide binding at this interface is thus a conserved feature of PilM paralogs.

The N Terminus of PilM Contains a Conserved Motif Necessary for PilM Dimerization-Structural comparison of PilM⅐
PilM(1-7) binding with PilM⅐PilN(1-12) binding indicated that the N-terminal residues Met-1, Leu-4, Ile-5, and Lys-7 play a key role in the binding of the N-terminal peptide to the protein. If the binding of PilM(1-7) had functional implications, we hypothesized that a similar pattern of residues should be found in other PilM orthologues. PilM orthologues from unique species were aligned to our structure of PilM to define the flexible N terminus from the globular portion of PilM. This analysis revealed that approximately two-thirds of PilM orthologues have at least an extra 7 flexible N-terminal residues. Due to variations in length and sequence, these PilM N termini align poorly. Despite this, 54% of N termini have an hXXhhXϩ motif. Allowing optional single residue deletions, denoted by (X) in hX(X)hh(X)ϩ, allows this motif to be found in 80% of the sequences examined. Although our motif search was for any hydrophobic or positive residues in the hXXhhXϩ pattern, there was a clear sequence preference for an MXXLFXK motif, which closely matches the sequence of P. aeruginosa PilM(1-7). Following the MXXLFXK motifs are usually positive residues interspersed with polar residues. Using this information, we identified potential motifs in the N termini of PilM proteins from well studied model systems (Fig. 3C).
Similar to FtsA, the actin-like interface of PilM suggested that it could potentially form higher order oligomers. However, negative stain EM revealed no PilM oligomers. To further explore the dimer interface, we deleted the first 8 N-terminal residues of our PilM construct (PilM ⌬2-8 ). SEC-MALS analysis showed that PilM ⌬2-8 was monomeric in solution (Fig. 3D), indicating that the N terminus of PilM was important for homodimer formation. These data are consistent with PilM (1-7) binding to the PilN binding pocket in the crystallographic dimer. These data are also consistent with the monomeric structure of PilM⅐PilN(1-12) because the binding of PilN to PilM would prevent the PilM self-interaction.
Lys-52 and Ser-19 Are Important for but Not Essential to ATP Binding-Examination of the electron density map revealed the presence of nucleotide in both the PilM and PilM⅐PilN (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12) structures. Although we added ATP to our crystallization buffer, we found that modeling ADP and magnesium best fit the electron density observed in the PilM structure (Fig. 4A). This suggests that either the terminal phosphate was disordered or that ATP was hydrolyzed during the crystallization process. If the latter is the case, hydrolysis was likely to be non-enzymatic, because we and others (15) have been unable to demonstrate in vitro ATPase activity for either PilM or PilM⅐PilN(1-12). How-ever, aligning the ATP-binding site of PilM with homologous ATPase DnaK (PDB code 4B9Q) revealed a high degree of conservation in the residues necessary for ATP catalysis, including Lys-52 and Glu-171 of PilM, which align with the catalytic residues Lys-70 and Glu-171 of DnaK (55) (Fig. 4B). In the PilM structure, Lys-52 and Glu-171 are shifted away from the ␥-phosphate of ATP by 3 Å compared with corresponding Glu-171 and Lys-70 of DnaK. Thus, PilM could potentially have ATPase activity if a conformational change brought Lys-52 and Glu-171 closer to the ␥-phosphate.
To model the ␥-phosphate, we crystallized PilM in the presence of AMP-PNP (Fig. 4C). These crystals diffracted to lower resolution (3.5 Å), but valuable information could still be derived. Residue Lys-52 was disordered in the PilM⅐ADP structure but ordered in PilM⅐AMP-PNP structure. Lys-52 and Ser-19 appear to facilitate coordination of the ␥-phosphate of ATP. To confirm that these residues participate in ATP binding, we made S19A and K52A PilM variants and evaluated their ATP-binding affinity using ITC (Fig. 4E). The binding of ATP to PilM was enthalpy-driven, with a K d of 4.63 Ϯ 0.05 M, whereas binding of ADP to PilM was too weak to be quantified. Compared with wild type, the PilM S19A and PilM K52A variants have ϳ6and 4-fold reduced affinity for ATP, respectively. These data suggest that Ser-19 and Lys-52 are each important but not essential for ATP binding.

PilN Binding Causes Tertiary Structure Changes in PilM-
PilM Bridges PilN, PilB, PilT, and PilC-Based on data from the homologous T2S and T4bP systems, we hypothesized that in addition to ATP and PilN, PilM will bind to additional components of the T4aP system. BACTH assays (57) have been used previously to analyze the connectivity of cytoplasmic T4aP proteins in Neisseria meningitidis. These data suggested interac-tions between PilM and PilT2, a PilT paralog not present in P. aeruginosa (58). To determine whether additional binding partners could be identified in P. aeruginosa with our PilM constructs, we performed BACTH analysis with both PilM and PilM⅐PilN(1-12) using two assay outputs, MacConkey-maltose and the ␤-galactosidase assay.
Although the interactions observed for the cytoplasmic proteins in P. aeruginosa are similar to those reported for N. meningitidis, there are several important differences (Fig. 6, A and  B). As were seen in N. meningitidis, using the MacConkeymaltose assay, we identified interactions between PilC and PilA, PilC and PilO, PilC and PilT, PilT and PilU, PilT and PilB, PilM and PilN, PilN and PilO, and PilO and PilA as well as PilA/B/C/ M/O/T/U self-interactions. We also observed an PilC-PilU interaction and PilM-PilB, -PilT, and -PilC interactions. It is not clear whether these additional interactions are absent in the N. meningitidis T4aP system or if P. aeruginosa proteins are more amenable to BACTH analysis. Interactions between PilM-PilC and PilM-PilB equivalents in the T2S system have been described previously (24,27). Consistent with our SEC-MALS data, we detected PilM self-interactions but not PilM⅐PilN(1-12) self-interactions. Our BACTH analysis broadens our understanding of the connectivity of the cytoplasmic T4aP components and highlights the central position of PilM.
Comparing the BACTH signals using the ␤-galactosidase assay suggests that whereas PilM and PilM⅐PilN(1-12) bind PilB, PilT, and PilC, their affinities for these proteins are different (Fig. 6C). Specifically, PilM⅐PilN(1-12) interacts preferen-  tially with PilB, and PilM interacts preferentially with PilC and PilT. The ␤-galactosidase assay also suggested an interaction between PilA and PilM, which was not identified using the MacConkey-maltose assay, underscoring the necessity for secondary validation of interactions identified by BACTH.
Validating these differential interactions in vitro has not been possible. Purified PilB and PilT each have ATPase activity, suggesting that they are folded properly, but PilM and PilT aggregated in our biophysical instruments, impeding analysis. However, we could validate a direct interaction between PilM and PilB or PilT using in vitro co-purification studies. Whereas PilM showed some nonspecific binding to the Ni-NTA resin, additional PilM was co-purified with His-tagged PilB or PilT (Fig. 6D). PilM⅐PilN(1-12) did not nonspecifically bind to the Ni-NTA resin and more clearly co-purified with His-tagged PilB and PilT, providing evidence that PilM interacts directly with PilB and PilT in vitro.

Discussion
In this study, we characterized two distinct PilM conformations using crystallographic, biophysical, biochemical, and BACTH analyses. We identified conserved peptide motifs that bind to each of these conformations, supporting the relevance of both. Further, we have extended the PilM protein interaction network and demonstrated using two different methods that PilM binds to PilB and PilT.
An intriguing feature of our PilM crystal structure is the interaction we observed between PilM(1-7) and the adjacent PilM protomer. This peptide occupies the PilN binding groove and facilitates the formation of a PilM homodimer. Although we have not determined whether PilM homodimerizes in vivo, comparison with other species revealed that the residues involved in this interaction are a conserved feature of PilM. PilM(1-7) and PilN(1-8) residues make similar contacts in the binding pocket, and both are followed by positive residues interspersed with polar residues. The N terminus of FtsN bound by the PilM homolog FtsA also has a conserved motif followed by interspersed positive residues (32) extending the number of conserved features between PilN and FtsN. PilN (1)(2)(3)(4)(5)(6)(7)(8) and PilM(1-7) have amphipathic characteristics, reflecting the nature of the PilN binding pocket of PilM. These amphipathic qualities and membrane proximity suggest that PilN(1-8) and PilM(1-7) could be membrane-associated when not bound to PilM. The PilM homologues, FtsA and MreB, have membrane-associated C-and N-terminal amphipathic helices, respectively (59,60). The amphipathic C-terminal helix of FtsA maintains FtsA in a non-polymerized conformation until FtsA is adjacent to a lipid bilayer (61). Thus, PilM(1-7) might keep PilM in a dormant conformation until PilN binding, or it may keep PilM stable when temporarily disengaged from PilN.
The structural analysis in this study revealed that peptide binding at an interface created by domains 1B and 2A is highly conserved in actin/Hsp70/hexokinase superfamily proteins. This interface corresponds to the PilN binding pocket of PilM and may be a conserved binding site because it is located at a hinge point that can rotate upon ATP binding (62). Hence, where it has been evaluated, peptide binding to this interface can be modulated by ATP binding and/or catalysis (53,63). This is consistent with our finding that ATP binding to PilM increased PilN affinity ϳ2 fold. In DnaK, ATP binding increases affinity for the interdomain linker peptide, 389 VLLL 392 (63). When DnaK binds the 389 VLLL 392 peptide, a substrate peptide bound to a different interface is released (63). Hence, substrate release is dependent on 389 VLLL 392 binding, which is itself dependent on ATP binding. Mutating Lys-70 in DnaK prevents ATP-dependent release of substrate peptide (64). Thus, it seems likely that Lys-70 in DnaK enables ATP to increase the affinity of DnaK for the 389 VLLL 392 peptide. PilM Lys-52 aligned structurally with DnaK Lys-70, and mutation of PilM Lys-52 similarly nullified the ATP-mediated increase in PilN binding. Lack of the 2B domain necessary to bind ATP in GspL and BfpC of the T2S and T4bP systems (28) suggests that GspL and BfpC do not need to regulate association to their PilN-like portions, consistent with fusion of these domains.
We found that the half-life of the PilM-PilN(1-12) interaction is ϳ10 s, suggesting that PilN may disengage from PilM on a time scale of relevance to type IV pilus dynamics. For example, the pilus of P. aeruginosa is expected to extend or retract for ϳ1-14 s, because pili lengths vary from 1 to 7 m, and the mean extension and retraction rate is 0.5 Ϯ 0.2 m/s (65,66). In other words, the PilM-PilN interaction may be just long enough for a cycle of extension or retraction and thus could have functional implications for these or related processes.
At the time of manuscript submission, these analysis provided the first evidence for any T4aP system that PilM directly interacts with PilB and PilT and provided BACTH data to suggest that PilM might bind PilC. Before manuscript publication, it was demonstrated that PilM directly interacts with PilB in Myxococcus xanthus (67), corroborating our results. Both BfpE, the PilC homologue in the T4bP system, and GspE, the PilB homologue in the T2SS, have been shown to bind to the PilM loop 1B equivalents in BfpC and GspL, respectively (24,51). We predict that PilM interacts with PilB and potentially PilC using a similar interface. Comparing the structures of PilM and PilM⅐PilN (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12) revealed that residues in loop 1B move by 7 Å in the presence of the PilN peptide. Likewise, our BACTH data suggested that interaction of PilM with PilN differentially impacts the binding of PilM to PilB, PilT, and PilC. Based on these analyses, we predict that the binding of PilN to PilM dynamically regulates PilM binding to PilB, PilT, and/or PilC  (1-7), shown as a darker line connected to PilM) or PilN (the N terminus of which is depicted in the cytoplasm). The identification of a conserved motif in PilM(1-7) bound by PilM as well as transient binding kinetics suggest that PilM is not always bound to PilN. In addition to PilN, PilM binds PilB, PilT, and PilC, and BACTH results suggest that these binding events could be modulated by the PilN-dependent conformation of PilM. (Fig. 7). One hypothesis might be that PilN detachment from PilM is part of a molecular switch regulating a T4aP process, such as extension or retraction. The newly identified proteinprotein interactions described herein suggest that PilM is not simply a passive scaffold but potentially a key regulator of T4aP dynamics.