Purification of natively assembled CupE pili

To study the role of CupE pili in P. aeruginosa, we engineered a strain with increased expression of CupE to facilitate isolation and structural analysis. Therefore, we used a previously described methodology of studying strains in which other common cell surface filaments such as Type IV pili and flagella had been deleted [37], which are otherwise abundant in pilus preparations. In addition, the gene encoding the MvaT repressor was deleted, as this had previously been shown to result in overexpression of CUP pili in P. aeruginosa [36]. Thus, all further experiments were performed using cells with a ΔpilA ΔfliC ΔmvaT background. During the bioinformatic analysis of cup operons in P. aeruginosa, we noted that the classical-type cupA operon also encodes a phosphodiesterase (PA2133, hereafter referred to as cupA6). As suggested by a previous study [39], we reasoned that deletion of this gene may increase cellular cyclic di-guanylate (c-di-GMP) levels, a biofilm master regulator, and thus further increase the expression of CUP pili. Indeed, deletion of the cupA6 phosphodiesterase resulted in increased expression of thin cell surface filaments, as quantified via negative stain electron microscopy, with 188 fibres detected on cells of a ΔcupA6 strain versus 60 fibres detected in a control over 30 micrographs acquired randomly (S1A Fig and Methods). This strain with increased expression of surface filaments was used to obtain preparations of purified filaments by shearing from the surface of cells scraped from plates (Methods). In the purified specimen, we observed long, curved filaments on cryo-EM grids (Fig 1A), with a different morphology than previously observed for classical-type CUP filaments [25], suggesting that these pili may correspond to archaic-type CupE pili rather than classical-type, tubular-shaped CupA-D pili. Indeed, mass spectrometric peptide fingerprinting confirmed that the sample contained CupE1 protein. These observations suggest that the combined deletion of genes encoding the MvaT repressor and the phosphodiesterase CupA6 caused increased expression of CupE filaments. To further verify that the observed filaments are CupE pili, we deleted cupE1-E2 in the ΔcupA6 background and found that the long, curved filaments with an apparent diameter of 60 Å were no longer present in fractions sheared from the cell surface (S1B and S1C Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. CupE1 subunits within the CupE pilus are arranged in a zigzag architecture.

(A) Raw micrograph and 2D class average of CupE pili show zigzag-shaped filaments. (B-C) A 3.5 Å resolution cryo-EM structure of the CupE pilus reveals how CupE1 subunits are arranged in a zigzag pattern, with the donor strand of the (n+1)_th subunit enveloped by the incomplete Ig-like fold of each n_th subunit. Five CupE1 subunits form a longer repeat of ~155 Å.

https://doi.org/10.1371/journal.ppat.1011177.g001

Atomic structure of CupE pili using cryo-EM helical reconstruction

After confirming the presence of natively assembled CupE pili in our preparation, we proceeded with cryo-EM analysis. Cryo-EM micrographs and 2D classes of CupE showed a zigzag appearance of the pili (Fig 1A), with a diameter of approximately 60 Å. Interestingly, some pili spontaneously associated into large mesh-like bundles (S1D Fig). After deducing the symmetry of the pilus from two-dimensional class averages of single filaments (Fig 1A), we performed helical reconstruction and obtained a 3.5 Å-resolution cryo-EM density, from which an atomic model of the CupE pilus could be built (Figs 1B, 1C and S2 and S1 Table and S1 Movie). The atomic model reveals an arrangement of CupE1 subunits in a zigzag architecture (215° right-handed rotation per subunit, Fig 1B and 1C). In agreement with classical CUP proteins, the N-terminal β1-strand of each (n+1)_th subunit completes the Ig-like fold of the n_th subunit (Fig 2A and 2B). This donor strand interacts with the n_th subunit through both β-sheet hydrogen bonding as well as hydrophobic interactions, filling a hydrophobic groove in the subunit core (Figs 2C and S3). The structure also shows two key cysteine residues (C41-C85) forming a disulfide bridge within a β-sheet from which the donor strand extends (Fig 2D). Bioinformatic analysis shows that these cysteines are highly conserved (S4 Fig), suggesting they may play an important role in maintaining subunit stability.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Architecture of the CupE1 subunit within the CupE pilus.

(A) The donor strand from the (n+1)_th subunit completes a β-sheet in the n_th subunit by providing a 13-residue β-strand. (B) Subunit topology demonstrating β-sheet architecture of the CupE1 subunit. A yellow line denotes a disulfide bridge. (C) Extensive hydrophobic interactions anchor the donor strand of the (n+1)_th subunit into the Ig-like fold of the n_th subunit. (D) Compared to the extensive interactions of the donor strand with the complemented subunit, the globular subunit:subunit interface between Ig-like folds shows few strong interactions at the interface. A disulfide bridge is observed in each CupE1 subunit (marked).

https://doi.org/10.1371/journal.ppat.1011177.g002

The 13-residue donor strand of the (n+1)_th subunit represents the majority of the interaction with the n_th subunit. The donor strand is connected to the core of each subunit through a Glycine-Alanine-Glycine (GAG) sequence (Fig 2D). The interface between the Ig-like domains is relatively small in comparison to the extensive donor-strand interactions, and is mainly maintained by the interaction of a loop with the next subunit, with two interacting valine residues V69 and V48 at the core of the interaction (Fig 2D). In a recent publication on the related Csu pilus, interactions between zigzag-arranged subunits within an archaic CUP pilus were termed ‘clinch’ interactions [40]. While the subunit arrangement is overall similar in the CupE pilus, there are significantly fewer interacting residues between subsequent subunits in the CupE pilus compared to the Csu pilus, despite the high structural and sequence similarity between the pilins (RMSD = 1.55 Å; S4 and S5 Figs). Thus, the CupE pilus shows a relaxed arrangement of subunits compared to the tight packing of the Csu pilus, resulting in a higher rise per subunit than in the Csu pilus (S5B Fig). The relaxed subunit arrangement suggests the possibility that the CupE pilus may be more flexible than the Csu pilus, which has been described as laterally stiff [40]. Indeed, increased lateral curvature within CupE pili was observed in a subset of two-dimensional class averages (S3B Fig), supporting some degree of flexibility around the subunit-subunit interface that allows the orientation of subunits to vary.

Visual inspection of the CupE pilus along the helical axis (S3C Fig) showed a serine and threonine-rich loop (TTTTSST) extending outward from the helical axis of the CupE pilus, constituting the part of the subunit that is most exposed to the environment (S3C Fig). Sequence alignment of archaic pilin subunits shows that this unusual serine and threonine-rich sequence is unique to P. aeruginosa (S4 Fig). In the EM map, we noticed extra density near the loop that could not be attributed to the atomic model of the CupE1 protein subunit and that we suspected might correspond to a post-translational modification. Mutation of three threonine residues within the loop to Alanine-Glycine-Alanine, followed by cryo-EM structure determination, resulted in a 4.1 Å resolution map with a reduced density in this region (S3D and S3E Fig and S1 Table), confirmed by calculating difference maps with the wild-type pilus structure (S3F Fig). While disassembly into monomers followed by intact mass spectrometric analysis allowed us to detect a molecular species 206 Da larger than the expected mass of the CupE1 monomer (S3G Fig), predominantly unmodified CupE1 protein was observed. Thus, the chemical identity of the extra density near the threonine-rich loop cannot be ascertained at this stage.

Imaging of CupE pili in situ on the surface of P. aeruginosa cells

CupE pili have been shown to promote the mature architecture and mushroom shape of P. aeruginosa biofilms [28]. To find out how CupE pili support cohesion between cells within the biofilm, we imaged CupE pili directly on the P. aeruginosa cell surface. To this end, we deposited cells from colonies of the ΔcupA6 strain used above for obtaining pure CupE preparations onto grids for electron cryotomography (cryo-ET) imaging. In the resulting cellular electron cryotomograms (Fig 3), we observed CupE pili with the same characteristic zigzag architecture seen in vitro extending from the cell surface (Figs 3A, 3B and S6). To confirm that these were the same CupE pili, we extracted subtomograms along the length of the pili in tomograms and performed subtomogram averaging [41]. Our subtomogram averaging structure of the cell surface filaments, produced from an unbiased cylindrical reference, recaptures the CupE pilus’ zigzag appearance (S6 Fig), which was observed even without symmetrization, validating the identity of these filaments as CupE pili. Interestingly, CupE pili attached to the cell could adopt variable curvatures in situ (Figs 3, S6C and S6D and S2 Movie). In small P. aeruginosa cell clusters, pili were repeatedly found to extend from and fold back onto the originating cells (Fig 3C and 3D). We estimated the local curvature of CupE pili in cellular tomograms and discovered local deviation from the helical axis of up to 15° per subunit (S6E Fig). This observed ability of CupE pili to locally adopt higher curvature contrasts with classical, tubular CUP pili, which are comparatively rigid assemblies [25]. At the same time, however, it also differs from γ-clade linear Chaperone-Usher fibres, such as the Caf1 and Saf pilus that were found to have extremely low persistence lengths and even higher angles between subunits [42,43]. Adopting curvature may be required to support efficient interactions of the pilus in the extracellular matrix of the biofilm, allowing the cells to embed themselves in a matrix rich in filamentous molecules. Here, some flexibility of the pilus may help promoting cell-cell adhesion within the biofilm.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Cryo-ET imaging of CupE pili on P. aeruginosa cells.

(A-B) Tomographic slice of cells expressing CupE pili. CupE pili adopt significant curvature on cells. The pilus can be seen going in and out of plane in (B). Inset: Atomic model of CupE fitted in the subtomogram averaging map produced from cellular cryo-ET data. (C-D) Segmentation of cell membranes and CupE pili illustrates how CupE pili adopt variable curvatures. See also S2 Movie.

https://doi.org/10.1371/journal.ppat.1011177.g003

Structure- and sequence-based bioinformatic analysis of CupE pili

To place the structural and imaging data on CupE pili described above into the context of P. aeruginosa biofilm formation and CUP pili regulation, we performed computational analyses, facilitated by recent developments in methods for protein structure prediction [44] and sequence analysis. While our cryo-EM structure is in agreement with previous studies that showed CupE1 to be the major CupE pilin subunit [28], the cupE operon also encodes two additional predicted pilin subunits, CupE2 and CupE3, which exhibit high pairwise sequence similarity to CupE1, 39% and 30%, respectively (S4 Fig). To model the oligomers formed by the CupE2 and CupE3 subunits, we used AlphaFold-Multimer, which has been shown to predict the structures of monomeric and multimeric proteins with atomic-level accuracy [45]. Our modelling suggests that CupE2 and CupE3 can form donor-strand-exchanged filaments similar to CupE1, being able to form homofilaments as well as heterofilaments with CupE1 (S7A and S7B Fig). As a further validation of our modeling, we found that AlphaFold2 correctly predicted the CupE1 subunit arrangement consistent with our cryo-EM data (S7 and S8 Figs), with the subunit-subunit interface differing only slightly to our experimental structure. Finally, we also used AlphaFold2 to predict a structural model of the putative complex formed between CupE1 and the tip adhesin subunit CupE6. In the obtained model, CupE6, like the tip adhesin CsuE of A. baumannii [19], is composed of two Ig-like folds. While the C-terminal domain is incomplete and caps the CupE1 filament by accepting a donor strand from CupE1, the distal end of the N-terminal fold contains a highly hydrophobic surface patch that likely plays a role in adhesion to target substrates (S7C Fig).

To investigate the conservation and co-occurrence of cup gene clusters in isolates of P. aeruginosa at the genomic level, we performed a comprehensive sequence search of the Pseudomonas Genome Database [46] and the NCBI RefSeq database. We searched for the occurrence of CupA-D and CupE systems with the usher as the query sequence (CupA3-CupD3 and CupE5), as the usher is the most conserved protein among the genes encoded by these clusters [6]. We detected complete cupE gene clusters with conserved gene order in 228 out of 233 strains of P. aeruginosa in the Pseudomonas Genome Database, showing the widespread occurrence of this archaic CUP cluster. Notably, cupE is absent in the strain PA7, a highly divergent isolate of P. aeruginosa [47]. Of the four classical cup clusters in P. aeruginosa (CupA-D), we identified the cupB cluster in all 233 strains, whereas the cupA and cupC gene clusters were absent in only a few strains (S5 Table and S1 Data). In contrast, the cupD gene cluster, previously described to be located on the pathogenicity island PAPI-1 [38,48], was found only in 11 strains. Interestingly, both the cupA and the cupE clusters are missing in most of the strains containing the cupD cluster. Given the high sequence similarity between the corresponding protein subunits of the CupA and CupD systems (>65% pairwise sequence identity), we speculate they serve very similar functions and are, therefore, often mutually exclusive. Taken together, three different classical cup gene clusters and one archaic gene cluster were found to be widespread in P. aeruginosa strains.

The success of our strategy to upregulate CupE pili by deleting a phosphodiesterase-encoding gene in the cupA operon (cupA6) suggests that CUP pili expression may be interdependent on other CUP genes, suggesting possible co-regulation. This agrees with the findings of our bioinformatics analysis, as CupA and CupE co-occur in almost all strains of P. aeruginosa. Notably, the phosphodiesterase CupA6 is encoded after the chaperone CupA5 with a sequence overlap of four nucleotides in every strain containing CupA (Fig 4A), indicating a likely translational coupling between them [49].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Co-occurrence of cupA and cupE gene clusters.

(A) Bioinformatic analysis reveals co-occurrence of the cupA and cupE gene clusters with preserved gene order in most P. aeruginosa strains. Overlapping sequences of different CUP genes indicate possible translational coupling; of the usher gene cupE5 to the adhesive tip subunit gene cupE6, of the two minor pilin subunits cupE2 and cupE3, and of all genes cupA2-cupA6. A phosphodiesterase (CupA6) is encoded after cupA5 in every strain in which cupA occurs. (B) Outside of P. aeruginosa, examples of co-occurrence can be found in pathogenic proteobacteria such as P. fluorescens, Y. pestis, A. baumannii, and B. pseudomallei. Gene order in CUP clusters outside P. aeruginosa is not preserved; e.g. in P. fluorescens ATCC_13525, an additional pilin is encoded, and in B. pseudomallei K96243, the cupA6-like gene is encoded on the opposite strand. Accession data and gene loci are provided in S5 Table and supplementary file S1 Data.

https://doi.org/10.1371/journal.ppat.1011177.g004

Outside P. aeruginosa, cup gene clusters are more divergent and do not share a conserved gene order, exhibiting missing, additional, or swapped genes. In other bacteria of the genus Pseudomonas, cupE-like gene clusters are more widespread than cupA-cupD, and the co-occurrence of cupA and cupE is rare (Fig 4); for example, while P. putida NBRC_14164 lacks cupA, P. fluorescens ATCC_13525 contains both cupA and cupE. The co-occurrence of cupE-like and cupA-like gene clusters was also identified in some species of other gammaproteobacteria (e.g., Yersinia pestis, A. baumannii) as well as betaproteobacteria (e.g., Burkholderia pseudomallei) (Fig 4B).

Construction of P. aeruginosa mutants

P. aeruginosa deletion mutants were created as described previously utilising the suicide plasmid pKNG101 [54]. Briefly, to engineer gene deletions in the PAO1 strain, 500 bp DNA fragments of the 5’ (upstream) and 3’ (downstream) ends of the gene of interest were obtained by PCR using PAO1 chromosomal DNA as a template. The upstream fragment was amplified with the oligonucleotides P1 and P2 while the downstream fragment was amplified using P3 and P4 (S2 Table). A third PCR step using P1 and P4 resulted in a DNA fragment with the flanking region of the gene of interest. The gene fragment was then cloned into pCR-BluntII-TOPO (Invitrogen), the sequence confirmed and sub-cloned into the pKNG101 suicide vector (S3 Table). The pKNG-derivatives were maintained in E. coli strain CC118λpir (for strain descriptions, see S4 Table) and conjugated into PAO1 using E. coli 1047 harbouring the conjugative plasmid pRK2013. pKNG101 was conjugated into P. aeruginosa as described in [54]. After homologous recombination, colonies were streaked onto agar containing 20% (w/v) sucrose and grown at room temperature for 48 hours to select for colonies that have lost the plasmid backbone. Gene deletions were verified by PCR using external primers P5 and P6 (S2 Table).

Chromosomal substitution of TTTTSST to AGATSST in CupE1 was achieved by reintroduction of the previously deleted cupE1-2 fragment into PAO1 ΔpilA ΔfliC ΔmvaT ΔcupA6 ΔcupE1-2. In brief, the cupE1-2 DNA fragment was amplified using primers called ΔcupE1-2 P1 and P4 (S2 Table), subcloned into pCR-BluntII-TOPO (Invitrogen) and subjected to site directed mutagenesis using CupE1-AGATSST Fw and Rev primers (S2 Table). After the sequence was confirmed, the fragment was cloned into pKNG101 and conjugation was conducted as described above. Mutation was verified by PCR using external primers ΔcupE1-2 P5 and P6 and clones with cupE1-2 fragment reintroduced were sequenced.

Isolation of CupE pili

An overnight culture of P. aeruginosa PAO1 ΔmvaT ΔcupA6 ΔpilA ΔfliC, grown in lysogeny broth (LB) medium at 37°C with agitation at 180 revolutions per minute (rpm), was used to plate lawns on agar plates, and incubated overnight at 37°C. Bacterial lawns were scraped and resuspended in 1x phosphate-buffered saline (PBS). The resulting suspension was vortexed for 90 seconds to promote dissociation of pili from the cell surface. Cells were then centrifuged at 4,500 relative centrifugal force (rcf) for 20 minutes, and the supernatant centrifuged again 3–4 times at 16,000 rcf to remove remaining cells and cellular debris. 500 mM NaCl and 3% (w/v) PEG-6000 were added to the supernatant, and pili were precipitated on ice for 1 hour. Precipitated pili were collected via centrifugation for 30 minutes at 16,000 rcf. For cryo-EM, precipitated pellets were combined and precipitated again in the same manner and resuspended in 1x PBS to produce the final product.

Negative stain electron microscopy

2.5 μl of sample was applied to a glow-discharged carbon support grid (TAAB), blotted, washed three times with water, and stained using three 20 μl drops of 2% (w/v) uranyl acetate and allowed to air-dry. Negatively stained grids were imaged on a Tecnai T12 microscope.

Cryo-EM and cryo-ET sample preparation

For cryo-EM grid preparation of purified CupE pili, 2.5 μl of the sample was applied to a freshly glow-discharged Quantifoil R 2/2 Cu/Rh 200 mesh grid and plunge-frozen into liquid ethane using a Vitrobot Mark IV (ThermoFisher) at 100% humidity at an ambient temperature of 10°C. For tomography sample preparation of PAO1 ΔpilA ΔfliC ΔmvaT ΔcupA6, a bacterial lawn from an overnight LB agar plate incubated at 37°C without antibiotics was resuspended in PBS, and 10 nm Protein-A-gold beads (CMC Utrecht) were added as fiducials prior to plunge-freezing.

Cryo-EM and cryo-ET data collection

Cryo-EM data was collected in a Titan Krios G3 microscope (ThermoFisher) operating at an acceleration voltage of 300 kV, fitted with a Quantum energy filter (slit width 20 eV) and a K3 direct electron detector (Gatan). Images were collected in super-resolution counting mode using a physical pixel size of 1.092 Å/pixel for helical reconstruction of CupE pili and 3.489 Å/pixel for cellular tomography data. For helical reconstruction of CupE, movies were collected as 40 frames, with a total dose of 45–46 electrons/Ų, using a range of defoci between -1 and -2.5 μm. For the wild-type CupE pilus dataset, 11,584 movies were collected; for the 111-113_AGA dataset, 4,665 movies were collected. Cryo-ET tilt series of PAO1 ΔpilA ΔfliC ΔmvAT ΔcupA6 cells were collected using a dose-symmetric tilt scheme as implemented in SerialEM [55], with a total dose of 121 electrons/Ų per tilt series and defoci of -8 to -10 μm, and with ±60° tilts of the specimen stage at 1° tilt increments.

Cryo-EM processing

Helical reconstruction of CupE pili was performed in RELION 3.1 [56–58]. Movies were motion-corrected and Fourier-cropped using the RELION 3.1 implementation of MotionCor2 [59], and CTF parameters were estimated using CTFFIND4 [60]. Initial helical symmetry of CupE pili was estimated through indexing of layer lines and counting the number of visible subunits along the pilus. Three-dimensional classification was used to identify a subset of particles that supported refinement to 3.5 Å resolution. For final refinement, CTF multiplication was used for the final polished set of particles [57,61,62]. Symmetry searches were used during reconstruction, resulting in a final rise of 33.12 Å and a right-handed twist per subunit of 214.56°. Resolution was estimated using the gold-standard Fourier Shell Correlation (FSC) method as implemented in RELION 3.1. Local resolution measurements were also performed using RELION 3.1.

Model building and refinement

Manual model building of the CupE1 subunit was performed in Coot [63] as follows. A homology model based on the structure of A. baumannii CsuA/B (RCSB 6FQA) was calculated using MODELLER [64] and this homology model was fit into the cryo-EM density as a rigid body. Residues of the homology model that were inconsistent with the density, including the N-terminal donor strand, were deleted and manually rebuilt. The initially built model was subjected to real-space refinement against the cryo-EM map within the Phenix package [65,66]. Five subunits of CupE1 were built and used for final refinement. Non-crystallographic symmetry between individual CupE1 subunits was applied for all refinement runs. Model validation including map-vs-model resolution estimation was performed in Phenix.

Tomogram reconstruction

Tilt series alignment via tracking of gold fiducials was performed using the etomo package as implemented in IMOD [67]. Tomograms were reconstructed with WBP in IMOD or SIRT in Tomo3D [68,69]. Deconvolution of tomograms using the tom_deconv.m script [70] was performed for visualisation purposes.

Subtomogram averaging

Subtomogram averaging of pili on cells was performed in RELION 4 [41], employing helical reconstruction [56]. A cylindrical reference was used to avoid bias. Helical symmetry was applied to enhance the signal during particle alignment. The map presented in Figs 3 and S6 is unsymmetrized.

Data visualisation and quantification

Cryo-EM images were visualized in IMOD. Fiji [71] was used for bandpass and Gaussian filtering, followed by automatic contrast adjustment. Atomic structures and tomographic data were displayed in ChimeraX [72]. Segmentation of tomograms was performed manually in IMOD. Quantification of cell surface filaments in the ΔcupA6 mutant was performed through manual annotation in 30 randomly acquired negative stain EM images targeted on cells located at low magnification. Atomic models are shown in perspective view, except for S3C Fig, which is shown in orthographic view. Hydrophobic surfaces were calculated in ChimeraX using the in-built mlp function. Difference maps were calculated using EMDA [73] with maps lowpass-filtered to the same resolution (4.2 Å). Angles between subunits (S6E Fig) were determined by spline-interpolating points every 33.1 Å along the path of the pilus, as segmented in IMOD, calculating normal lines at each point, and determining the angles between the normal of each point.

Bioinformatic analysis

Sequence data were downloaded from the Pseudomonas Genome Database version 20.2 [46] and filtered to exclude incomplete genomes. Searches were performed against every single strain using PSI-BLAST [74], with the CupA3 (PA2130 NCBI locus tag), CupB3 (PA4084), CupC3 (PA0994), and CupE5 (PA4652) proteins from the reference strain P. aeruginosa PAO1 as queries. Since the CupD system is missing in the strain PAO1, the CupD3 usher (PA14_59735) from strain UCBPP-PA14 was used as the query. To obtain unambiguous assignment of genes to CUP proteins, output data was further filtered with custom scripts and probable sequencing errors were corrected.

Structure predictions were performed with AlphaFold-Multimer version 2.1.1 [44,45], with sequences from the reference genome PAO1 as queries. Filaments were predicted without the signal peptide, which was predicted using SignalP [75]. The multiple sequence alignments (MSAs) used for the structure inference were built with the standard AlphaFold pipeline and the “reduced_dbs” preset. Template modelling was enabled and structures were inferred with eight MSA recycling iterations and all five different model parameter sets. After prediction, the models were ranked by the pTM score and only the highest-ranking model was selected. The PAE-value-plot for each structure is shown in S8 Fig. All predictions were performed using the high-performance computer “Raven”, operated by the Max-Planck Computing & Data Facility in Garching, Munich, Germany. The multiple sequence alignment shown in S4 Fig was obtained by first calculating an initial alignment using PROMALS3D [76] in default settings and then curating it manually.