Membrane platform protein PulF of the Klebsiella type II secretion system forms a trimeric ion channel essential for endopilus assembly and protein secretion

ABSTRACT Type IV pili and type II secretion systems (T2SS) are crucial for bacterial adaptation, virulence, and environmental impact. A common mechanism underlying their multiple functions involves assembly of dynamic plasma membrane-anchored filaments—the (endo)pili. The cytoplasmic ATPase motor GspE/PilB is thought to energize pilus assembly via the membrane assembly platform protein GspF/PilC, but platform protein structure and its molecular role remain elusive. Here, to dissect the GspF/PilC architecture and mechanism, we generated all-atom models of the Klebsiella T2SS platform protein PulF in different oligomeric states. Comprehensive modeling, molecular dynamics (MD) simulations, cysteine crosslinking, and biochemical analyses support the trimeric state of PulF. In the trimer, the transmembrane segment TMS2 and the nonessential cytoplasmic N-domain are peripherally located, while TMS1 and TMS3 form a 6-helix bundle delineating a central transmembrane channel. Polar and proline residue pairs in these segments, conserved in all GspF/PilC homologs, define the channel constriction that can accommodate sodium ions or protons. Remarkably, obstructing this channel via Cys crosslinking abolished endopilus assembly and protein secretion, shedding light on previous findings showing that dissipating the membrane potential with ionophores reversibly abolished T2SS function. The trimeric PulF shows an excellent fit with the PulE ATPase hexamer, building a complex with structural similarities to the V-ATPase. MD simulations of PulF inserted in an Escherichia coli membrane model reveal strong binding and enrichment in cardiolipin, the phospholipid known to stimulate ATPase activity of GspE/PilB. We propose that GspF/PilC cooperates with the ATPase to energize (endo)pilus assembly using the ion motive force. IMPORTANCE Type IV pili and type II secretion systems are members of the widespread type IV filament (T4F) superfamily of nanomachines that assemble dynamic and versatile surface fibers in archaea and bacteria. The assembly and retraction of T4 filaments with diverse surface properties and functions require the plasma membrane platform proteins of the GspF/PilC superfamily. Generally considered dimeric, platform proteins are thought to function as passive transmitters of the mechanical energy generated by the ATPase motor, to somehow promote insertion of pilin subunits into the nascent pilus fibers. Here, we generate and experimentally validate structural predictions that support the trimeric state of a platform protein PulF from a type II secretion system. The PulF trimers form selective proton or sodium channels which might energize pilus assembly using the membrane potential. The conservation of the channel sequence and structural features implies a common mechanism for all T4F assembly systems. We propose a model of the oligomeric PulF—PulE ATPase complex that provides an essential framework to investigate and understand the pilus assembly mechanism.

The second primer in a pair, where indicated, is the reverse complement of the first one.

Figure S1 .
Figure S1.The PulF topology and secondary structure predictions generated by the PSIPRED server (http://bioinfadmin.cs.ucl.ac.uk/).A. MEMSAT_SVM prediction of PulF topology and boundaries of the three transmembrane segments.TMS3 is predicted as a pore-lining helix.B. The PsiPred (6) prediction showing PulF primary sequence, with a-helix (pink rectangles), b-sheet (yellow rectangles) and loop (black lines) regions indicated.The a-helix and b-sheet secondary structure elements are numbered and indicated in blue.Per-residue prediction confidence levels, ranging from low (white) to high confidence (dark blue), are shown above the protein sequence.

Figure S2 .
Figure S2.Modeling of the PulF N-domain.A. Pairwise residue contact map (cutoff < 6Å) from the Nterminal PulF model generated by Robetta (grey dots) and contact prediction from Gremlin (blue dots).B. Structural model of N-domain obtained from evolutionary couplings (green) superimposed with the Robetta model (magenta).Side chains are shown as licorice.

Figure S3 .
Figure S3.AlphaFold2-multimer models and scores for PulF in different oligomeric states.The best scoring (highest multimer score) models for monomeric, dimeric, trimeric and tetrameric (top to bottom rows) PulF are shown as cartoon and colored by chains (left column) or by predicted local Distance Difference Test values (plDDT, 2 nd column from the left).The corresponding predicted aligned error maps (PAE), colored from blue (low PAE) to red (high PAE) are presented in the 3 rd column from the left.Each chain block is labeled by the chain id with the corresponding color as shown on the left column.The last three columns on the right report the various confidence scores from AF2: average plDDT, predicted TM-score (pTM) and Interface pTM (ipTM).The trimer PulF model exhibits the highest pTM and ipTM scores and the lowest inter-chain PAE.

Figure S4 .
Figure S4.AlphaFold2-multimer scores for the five models of PulF in different oligomeric states.Left column: Predicted local Distance Difference Test (plDDT) values along the PulF sequence for the five ranked models (the different chains in the oligomer are concatenated).Right columns: predicted aligned error (PAE) maps for the five ranked models and colored from blue (low PAE) to red (high PAE).Below are reported the average plDDT, pTM and ipTM scores for each model.The ipTM values for all trimeric PulF models are higher than the best scoring dimeric or tetrameric models.

Figure S5 .
Figure S5.Comparison of oligomeric PulF models predicted by AlphaFold2-multimer. A. Contact maps of dimeric (lower-triangle) and trimeric (upper-triangle) PulF models.Observed intra-molecular contacts are colored in grey and inter-molecular contacts are colored in orange and magenta for the dimeric and trimeric models, respectively.Predicted evolutionary couplings are colored in dark blue.The overlap between predicted evolutionary contacts and inter-molecular contacts is higher for the PulF trimer model.B. Amount of inter-chain residue pairs with a predicted aligned error (PAE) < 8 Å along the PulF sequence in the dimeric, trimeric and tetrameric PulF models.Positions of PulF domains are indicated on top.The trimeric PulF model displays a significantly higher number of high-confidence interchain contacts compared to the dimeric and tetrameric models.

Figure S6 .
Figure S6.The PulF Cys substituted variants are functional.A. PulA secretion promoted by PulF variants with single Cys substitutions in PulF TM segments as indicated.Secretion of PulA was assayed in E. coli strain PAP5207 mimicking the chromosomal expression levels of pul genes.Cell and supernatant (SN) fractions from an equivalent of 0.1 OD600nm of bacteria were analyzed by SDS-PAGE and Western blot with anti-PulA antibodies.B. Assembly of PulG pili promoted by indicated PulF variants in strain PAP7460 strain carrying the pul genes on plasmid pCHAP8252 complemented with pCHAP8259 derivatives carrying wild type or mutant pulF as indicated.Cell fractions (CF) and sheared fractions (SF) from an equivalent of 0.1 OD600nm of bacteria were analyzed by SDS-PAGE and Western blot with anti-PulG antibodies.Cell fractions were further probed by anti-PulFN antibodies and PulF protein levels are shown at the bottom of each panel.SlyD indicates the cross-reacting band revealed by the anti-PulF sera.Relevant Mw markers (in kDa) are indicated on the left.The PulF L372C variant labeled in red caused degradation of PulA and PulG.

Figure S7 .
Figure S7.The PulF tetramer model is incompatible with the cross-linking data.A. The cartoon representation of the PulF tetramer model (top view) generated by AF2 algorithm, with monomers highlighted in different colors.The side chains of residues 168, 171, 370 and 371 are shown as sticks.B. A zoom on the PulF tetramer model showing the channel residues 168 and 370 (in blue), as well as interface residues 171 and 371 (in red), in stick representation.The distance between 171 and 371 side chains is compatible with crosslinking, similar to the trimer model; however, no tetrameric forms of PulF were detected upon crosslinking of variant PulF V171C-P371C , arguing against this model.The distance between the Cb atoms of the channel residues Y168 and E370 in the tetramer model is too large and is thus incompatible with crosslinking of dimers and trimers.

Figure S8 .
Figure S8.Comparison of protein dynamics between simulations in E. coli membrane versus pure POPC bilayer. A. Per-residue RMSF values mapped on to the structure of the protein showing a more mobile N-domain in simulations with POPC bilayer.B. Backbone RMSD for full-length PulF (left) or PulF variant lacking the N-domain (DND, right), highlighting the increased flexibility of the N-domain in simulations with POPC bilayer.Solid lines show average values from three simulations, and standard deviations are illustrated by the shaded areas.

Figure S9 .
Figure S9.Secondary structure preservation during MD simulations.A. The total number of residues with secondary structural elements (α-helix, β-sheet, turn and bend) throughout the trajectories in an E. coli membrane model (black) and a POPC bilayer (red).Solid lines show average over two monomers and three simulations, while shaded areas indicate standard deviations.Dotted line indicates the number of residues with secondary structure predicted by AlphaFold2.B. PulF monomer colored based on the secondary structure (left) and the conservation of the secondary structure throughout a simulation in an E. coli membrane model (right).

Figure S10 .
Figure S10.Spontaneous oxidation of Cys residues in total bacterial extracts.Bacteria of strain PAP7460 containing plasmid pCHAP8252 and plasmid pCHAP8259 or its derivatives were cultured as indicated in Fig. 9 of the Main text.Total extracts were analyzed on 4-15 % SDS gradient gels and Western blot with anti-PulFN antibodies.The migration of PulF monomers, dimers and trimers are indicated on the right.

Figure S11 .
Figure S11.Comparison of PulF structures predicted with AF2-multimer.A. Superposition of AF2 models for the PulF trimer alone (PulF3, grey) or in complex with PulE (PulE6:F3, orange).The two structures are virtually identical with an RMSD of 1.5 Å for all residues.B. Superposition of a PulE2:PulF sub-complex from PulE:PulF complex models predicted by AF2 with either dimeric (PulE6:F2) or trimeric (PulE6:F3) PulF.The PulF subunit structure is nearly identical as is the orientation of the PulE subunits and the PulE-PulF interface between the Cyto1/2 domains and neighboring N2D domains (overall RMSD 2.0 Å). C. Model of a PulE:PulF complex predicted with AF2-multimer with 6:2 stoichiometry and colored by chains.The two PulF subunits are colored in lilac and orange.The left view is rotated by 90° showing the two PulE subunits not interacting with PulF (cyan and magenta), a space occupied by a 3 rd PulF subunit in the predicted PulE:PulF complex with 6:3 stoichiometry.

Figure S12 .
Figure S12.Comparison of PulE hexameric structures predicted with AF2-multimer.A. Superposition of AF2 PulE hexamer model (lilac) with VcGspE structure with C6 symmetry (grey).B. Superposition of AF2 PulE hexamer from the PulE6:PulF3 model (orange) with VcGspE structure with C6 symmetry (grey).When predicted alone, the PulE hexameric structure (PulE6) is closer (RMSD 2.1 Å) to the VcGspE C6 structure than when modeled as a complex with trimeric PulF (PulE6:F3, RMSD 6.8 Å).The view is oriented with the N2D in front.C. Superposition of AF2 PulE hexamer model (PulE6, grey) with VcGspE structure with C2 symmetry, in which the open, open' and closed subunits are colored in dark green, light green and red, respectively.The bottom view is rotated 180° showing the CTD in front.D. Same as C but superposing the AF2 PulE hexamer from the PulE6:PulF3 model (PulE6:F3, grey).The minimum RMSD between opposite subunits (same state) is found for the open' subunits in the PulE6:PulF3 model (3.6 Å).

Figure S13 .
Figure S13.Domain orientation in PulE/GspE hexamers.A. PulE domain organization.The flexible domain N1 is shown in white, the N2 in red and CTD in cyan.B. Two domains are shown for each GspE/PulE subunit: N2D (red) and CTD (cyan).The CTD of one subunit interacts with the N2D of the neighboring subunits (N2D +1 , pink).The construction block formed by CTD:N2D +1 is pseudo-rigid while the orientation of N2D with regard to CTD is variable for a single subunit, resulting in open and closed states depending on the internal symmetry of the GspE/PulE hexameric structure (C6 or C2).The view on the right is rotated by 90°.The position of the α2 helix is labeled.C. Superposition of a PulE subunit from the best AF2 model of PulE:PulF complex with 6:3 stoichiometry (yellow) with closed (red) and open (green) subunits in the Vibrio cholerae (Vc) GspE hexameric structures with C6 symmetry (PDB id 4kss) and C2 symmetry PDB id 4ksr), respectively.The orientation of the N2D in the best AF2 PulE hexamer model is similar to the open state observed in VcGspE with C2 symmetry.Only residues from the CTD were used for superposition.The view is oriented as in B (right panel).D. Same as C, but showing the PulE subunit in all five AF2-multimer models for PulE6:PulF3.

Figure S15 .
Figure S15.The PulF Cyto1 structural homologues.A. Homologs from the GspF/PilC family superimposed on the PulF Cyto1 model (green): N-terminal cytoplasmic domains of T. thermophilus PilC (cyan, PDB id 2WHN), Vibrio cholerae TcpE (magenta, PDB id 4HHX) and Bacillus subtilis SpoIIIAB (yellow, PDB id 6BS9).B. Superposition of PulF Cyto1 model onto the second domain of the d subunit from the T. thermophilus V-ATPase (PDB id 1R5Z).Each 6-helix domain of the V-ATPase D subunit is colored differently and PulF Cyto1 is shown in green.The orientation of the structures with regard to the membrane is indicated.

Figure S16 .
Figure S16.The final snapshots of three PulF trimer simulations in E. coli membrane.A top view is shown zooming on the polar channel residues.The protomers are shown in green, magenta and cyan with Y168 and E370 residues shown as sticks.Their positions deviate from the initial symmetric model changing the distance between the side chains and affecting the probability of direct contacts and crosslinking when replaced by Cys residues.