Pilin antigenic variants impact gonococcal lifestyle and antibiotic tolerance by modulating interbacterial forces

We sought to understand how pilin antigenic variation affects the pilus-mediated interactions between bacteria and to characterize the phenotypic changes including growth, antibiotic resistance, and antibiotic tolerance that result from altering these interactions. In the first step, we cloned the pilE genes from the N. gonorrhoeae clinical isolates Ng24, Ng17, and Ng32 into the background strain, wt*, in place of the native pilE gene. wt* is N. gonorrhoeae MS11 with the G4 motif deleted to prevent further antigenic variation of pilE [37]. The pilS sequences of the clinical isolates are similar to the respective sequences of the wt* strain (Fig i in S1 Text, S1 Data). Specific pilS sequences can be mapped nearly contiguously to the pilE sequences (Fig ii in S1 Text), indicating that the pilE variants of the clinical isolates have arisen by pilin antigenic variation. To understand the implications of differences in pilE sequences among variants, we examined their pilin sequences and predicted monomeric pilin structures and T4P filament structures. The amino acid sequences of the pilin variants were compared to that of the wt* strain (Fig 1A). Secondary structure and features were assigned based on alignment with the N. gonorrhoeae C30 PilE crystal structure [38], which is 90% identical in sequence to wt* PilE and 99% identical in the first 120 amino acids. The sequence of the N-terminal α-helix, α1, is identical for all pilins, with the exception of a glycine instead of glutamate at position 49 of Ng17 PilE (PilE₁₇). PilE_wt and PilE₂₄ share a serine at position 63, which is posttranslationally modified with a glycan in N. gonorrhoeae MS11 and C30 [38]. PilE₃₂ and PilE₁₇ have charged residues at this position. Some sequence variation is observed in the αβ-loop, which lies between α1 and strand β1 of the β-sheet, in β1 and in the β1-β2 loop. The most variable region is the hypervariable β-hairpin that follows the β-sheet and the C-terminal tail, which have amino acid changes but also insertions and deletions. We determined the overall pilin sequence identities of the variant to that of MS11 (Table ii in S1 Text). PilE₂₄ is 90% identical to PilE_wt, whereas PilE₃₂ is 86.6% identical and PilE₁₇ is 80.6% identical.

Pilin models for each variant were generated using AlphaFold [39]. The continuous N-terminal α-helix, α1, seen in the crystal structure of PilE [38] and in the AlphaFold predictions was replaced with the partially melted α-helix seen in the cryo-electron microscopy reconstruction of the intact pilus [40] (Fig 1B). All variable residues are located on the face of the PilE globular domain that is exposed in the pilus filament, as expected for antigenic variation. PilE₁₇ shows the greatest degree of variability, particularly in the protruding β-hairpin and C-terminal tail.

Pilus filament models were generated by superimposing each pilin model onto a single subunit in the cryoEM structure of the N. gonorrhoeae T4P and applying its symmetry parameters (Fig 1C). Filament models are shown in Fig 1C. The T4P surface is undulating: the β-hairpin and the C-terminal tail regions, which show the highest sequence variability, together form protruding “knobs”; and deep cavities or grooves lie between the globular domains, lined with the variable C-terminal tail on one side and the αβ-loop and β1-β2 loop on the other. Accordingly, the conformation of both the knobs and the cavities/grooves (“holes”) vary considerably among the variants, with wt* and Ng24 T4P being most similar. The electrostatic surface potential also differs substantially from one pilus variant to the next (Fig iii in S1 Text), with positively charged residues framing the holes of PilE_wt and PilE₂₄ and negatively charged residues lining the edges of the PilE₁₇ and PilE₃₂ holes. The knobs of each pilin differ in their distribution of charged residues, with PilE₃₂ having the most negatively charged knob. Pilus:pilus interactions likely require both structural and chemical complementarity between these surface features to allow the knobs in one filament to fit into the holes in another. Thus, these marked differences in stereochemistry among the pilus variants could impact pilus:pilus interactions and bacterial aggregation.

Fig 1. Sequence alignment and structure predictions for PilE variants and pilus filaments.

(A) PilE amino acid sequence alignment for wt* and variant strains. Amino acids of the PilE variants that are identical to those in PilE_wt are shown as black bars with gray shading, conservative amino acid changes are indicated with orange shading and non-conserved residues have red shading. The secondary structure and other hallmark features of Neisseria Type 4 pilins are indicated based on alignment of PilE_wt with C30 PilE (Protein Data Bank ID 2HI2). (B) Pilin models were generated using AlphaFold [39,67]. Residues that differ from those in PilE_wt are shown in stick representation, with carbon colored as in Fig 1A, nitrogen in blue and oxygen in red. Glycan carbons are green. The conserved disulfide-bonded cysteines that delineate the D-region are blue. Variable residues are located on the face of the C-terminal globular domain. (C) Filament models were generated by superimposing the pilin predictions on the N. gonorrhoeae T4P structure (EMD-8739) and applying its symmetry parameters. The wt* T4P model is shown on the left in cartoon representation and all models are shown in surface representation, colored as in Fig 1B. The hypervariable β-hairpin and C-terminal tail together form a protruding “knob” on each subunit (dashed lines) and deep cavities or “holes” lie at the interface between subunits.

https://doi.org/10.1371/journal.pbio.3003022.g001

Pilin antigenic variation can lead to loss of T4P function, in particular to reduced levels of piliation [9,10]. To test for piliation, we assessed the density of T4P for each strain using negative-stain transmission electron microscopy (TEM). All 4 strains show high levels of piliation (Fig 2A). Quantitatively, wt_pilE32 has elevated number of pili per cell compared to the other strains which have comparable piliation levels (Fig iv in S1 Text).

Fig 2. Replacing the native pilE sequence by antigenic variants does not reduce T4P density and maintains twitching motility.

(A) Representative TEM images of pilin variants. Scale bar: 2 μm. (B) Trajectories of motile bacteria. Representative trajectories of pilin variants over 30 seconds (tracks made via Trackmate Image J [68]) (gray: wt* (Ng150), dark gray: wt_pilE24 (Ng242), purple: wt_pilE17 (Ng240), green: wt_pilE32 (Ng230)). Scale bar: 5 μm.

https://doi.org/10.1371/journal.pbio.3003022.g002

We next compared T4P dynamics among the variants. T4P dynamics correlates quantitatively with twitching motility; the faster T4P retract the faster gonococci move on BSA-coated glass [20,21]. To find out whether variations in the PilE sequence affect T4P dynamics, we compared the trajectories of gonococci moving at the glass surface. All variants exhibit twitching motility on BSA-coated glass surfaces. A representative track for each pilE variant is shown in Fig 3B. A detailed analysis of the trajectories using a correlated random walk model [21] (described in Methods, Fig vi in S1 Text) showed that strain wt_pilE17 is the most motile; both the velocity v_corr and the correlation time τ were significantly higher than for the wt* strain (Fig v in S1 Text). The correlation time is a measure of how long the bacterium moves without changing direction. We conclude that all PilE variants have comparable or even higher levels of pili and generate dynamic T4P.

Fig 3. Attractive force between pairs of bacteria depends on the pilin sequence.

(A) Sketch of dual laser tweezers setup. (B) Probability distribution of rupture forces (number of interacting cell pairs: N_wt = 71, N_wtpilE24 = 59, N_wtpilE17 = 30, N_wtpilE32 = 54). The linearity of the laser trap is limited to 80 pN. All rupture events exceeding this force were grouped into a single bin shown at 100 pN. The data underlying this figure can be found in S1 Data. (C) Representative brightfield images of gonococci in liquid culture show aggregated microcolonies for wt and wt_pilE24 and planktonic cells for wt_pilE17 and wt_pilE32. Scale bar: 10 μm.

https://doi.org/10.1371/journal.pbio.3003022.g003

We examined the effects of the variations in pilus surface stereochemistry on pilus:pilus interaction and aggregation. In this study, the term pilus:pilus interaction is used to describe the interaction between T4P of adjacent gonococci. Using a dual laser tweezers assay, we investigated the attractive forces generated by the pilin variants. For each strain, we trapped pairs of bacteria and measured the T4P-mediated interaction forces between them (Fig 3A). When bacteria in different traps interact via their T4P, and at least one T4P retracts, the cell bodies approach each other. Bacteria are deflected by a distance d from the centers of the traps. As the deflection increases, the restoring force of the laser traps increases as well leading to a rupture event when the optical force exceeds the (rupture) force that the pilus:pilus bond can sustain.

The force at which a pilus:pilus bond ruptures was used as a measure of the attractive force between gonococci. F_rupture is defined as the maximal force attained before the bond breaks and the bacteria move back to their equilibrium positions. We did not observe significant differences between wt* and wt_pilE24 (Kolmogorow–Smirnow test), where the mean rupture forces with standard errors were = 40.9 ± 0.9 pN and = 38.3 ± 1.0 pN, respectively (Fig 3B). The deviation from previously published results for the wt* [27] most likely results from a different distance of the traps. Interestingly, the wt_pilE17 and wt_pilE32 variants exhibit much lower rupture forces of pN and pN, respectively. Moreover, the probability that a pair of bacteria shows attractive interactions is lower for all pilE variant strains compared to wt* (Fig ix in S1 Text).

The attractive force between bacteria initiates aggregation into microcolonies. In this study, the term “microcolony” describes spherical aggregates formed by gonococci in liquid or at the interface between the surface and liquid (Fig 3C). By the combined action of twitching motility, pilus:pilus attraction, and microcolony-fusion, wt* gonococci self-assemble within several minutes into spherical colonies comprising thousands of cells [22,24,41]. We investigated whether the different pilE variants support microcolony formation in this time frame. In agreement with the dual laser tweezers experiments, we found that only the strains that mediate the stronger attractive forces, wt* and wt_pilE24, are capable of microcolony formation (Fig 3C). After 1 h of incubation, colonies are mostly spherical (S1 Data) with a broad distribution of colony sizes (Fig viiA in S1 Text). While many colonies are small, more than 90% of the cells reside within colonies exceeding 1,000 cells (Fig viiB in S1 Text). We note that the colony size distribution is highly dynamic because colonies are motile and can fuse [22]. By contrast, the strains wt_pilE17 and wt_pilE32 remain planktonic (Fig 3D and S1 Data). Since the number of pili per cell of the planktonic strains is not lower than the density of the aggregating strains (Fig iv in S1 Text), we attribute the different interaction properties to T4P stereochemistry.

We next addressed the question whether wt* cells and strains expressing pilin variants self-assembled into mixed microcolonies. To this end, we mixed mcherry-expressing wt* cells (wt*_red) with gfp-expressing pilin variant strains. The 2 aggregating strains wt*_red and wt_{pilE24 green} formed colonies in which both strains segregate (Fig 4A). The most common pattern is one where each strain occupies one half of the colony. According to the differential strength of adhesion hypothesis, this phenotype is indicative of strong intrastrain interactions and weaker inter-strain interactions [28,11]. Strains wt_{pilE17 green} (Fig 4B) and wt_{pilE32 green} (Fig 4C) show comparable colony pattern when mixed with wt*_red cells. Both form a shell around the wt*_red colony made up from a single-celled layer. This morphology indicates that wt* T4P attract wt_pilE17 and wt_pilE32 pili, but less strongly than they attract wt*. By contrast, wt*_red colonies do not attract unpiliated ΔpilE_green bacteria (Fig 4D). These results show that while the planktonic piliated strains cannot form colonies by themselves, they can attach to existing colonies.

Fig 4. Pilin variants segregated in mixed microcolonies.

wt*_red (Ng170) were mixed at a 1:1 ratio with (A) wt_{pilE24 green} (Ng309), (B) wt_{pilE17 green} (Ng308), (C) wt_{pilE32 green} (Ng310), (D) ΔpilE_green (Ng081), (E) wt*_green (Ng105). Scale bar: 10 μm. Additional biological culture replicates can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003022.g004

We investigated which region of the pilin governs pilus:pilus interaction. The pilin structure models show high variability between the knobs and the cavities of the different pilin variants (Fig 1). We predicted that swapping the C-terminal sequence of wt_pilE17 and wt_pilE32, which aggregate poorly, with the corresponding sequence of wt*, which aggregates well, could restore strong interactions and aggregation to these variants. The variable knob of wt* PilE is defined approximately by residues T136 to the C-terminus, thus this segment, designated “T136”, was used to replace the corresponding C-terminal segments in the wt_pilE17 and wt_pilE32 PilE proteins (Fig 5A). These hybrid pilins were predicted by AlphaFold to have very similar folds to that of wt PilE (Fig viii in S1 Text). The hybrid pilins were used to generate filament models that closely resemble the model for the wt T4P (Fig 5B). The hybrid strains were strongly piliated, although the mean T4P number was slightly lower compared to the wt* strain (Fig iv in S1 Text). Rupture forces were determined for the hybrid strains wt_{pilE17_T136} and wt_{pilE32_T136} and found to be comparable to the wt* strain (Fig 5C and 5D). Furthermore, these strains aggregate and form spherical colonies (Fig 5E and S1 Data), consistent with the C-terminal segment of PilE defining the strength of pilus:pilus interactions. We note, however, that these colonies had a different size distribution compared to wt* (Fig vii in S1 Text).

Fig 5. Swapping the C-terminal pilin region restores aggregation.

(A) Sequences of the hybrid strains. (B) Alphafold 3 predictions of pilin monomers were used to generate filament models. The regions of PilE_wt that were inserted into PilE₁₇ and PilE₃₂, T136-160 (T136) or K155-160 (K155) are shown in blue. K137 in PilE_17-K155 is colored magenta; the corresponding threonines in the other models are indicated. Probability distribution of rupture forces (C), gray: wt* (Ng150), dark purple: wt_pilE17 (Ng240), purple: wt_{pilE17_T136} (Ng293), light purple: wt_{pilE17_T155} (Ng305), and (D) gray: wt* (Ng150), dark green: wt_pilE32 (Ng230), green: wt_{pilE32_T136} (Ng295), light green: wt_{pilE32_T155} (Ng307). (Number of interacting cell pairs: N_wt = 71, N_{wtpilE17_T136} = 45, N_{wtpilE17_K155} = 54, N_{wtpilE32_T136} = 33, N_{wtpilE32_K155} = 42.) The data underlying this figure can be found in S1 Data. The linearity of the laser trap is limited to 80 pN. All rupture events exceeding this force were grouped into a single bin shown at 100 pN. (E) Typical brightfield images of gonococci in liquid culture. Scale bar: 15 μm. Additional biological culture replicates can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003022.g005

Though there is considerable amino acid sequence variability in the C-terminal region of PilE, the C-terminal tail following the conserved disulfide bond is particularly disparate between the aggregating and planktonic strains. While only the last amino acid of the C-terminal tail is different between the aggregating strains wt* and wt_pilE24, the tails of wt _pilE24 and wt_pilE32 are shorter by several residues and bear no sequence identity to each other or to PilE_wt or PilE₂₄ (Fig 1A). To determine whether this difference affects pilus:pilus interactions, we generated the pilE hybrid strains wt_{pilE17_K155} and wt_{pilE32_K155}, each with their C-terminal residues replaced with residues K155-K160 of PilE_wt (Fig 5A and 5B). This replacement did not significantly affect the number of T4P per cell (Fig iv in S1 Text). We tested their pilus rupture forces and aggregative abilities. Whereas the pilE hybrid strain wt_{pilE32_K155} showed strong pilus:pilus interactions comparable to wt*, interactions for wt_{pilE17_K155} are considerably weaker (Fig 5C and 5D). wt_{pilE17_K155} showed a stronger tendency to aggregate than wt_pilE17, but the contours of the aggregates were not well defined (Fig 5E, S1 Data). wt_{pilE32_K155} formed spherical aggregates whose size distribution was shifted towards intermediate size colonies comprising (100 to 1,000) cells (Fig vii in S1 Text). To understand why the rupture forces of wt_{pilE17_K155} and wt_pilE17 are comparable, but wt_{pilE17_K155} forms small aggregates, we addressed the question whether wt_{pilE17_K155} interacted more frequently in the double laser trap. We found that the fraction of randomly picked pairs of cells that exhibited pilus:pilus interaction was indeed higher for strain wt_{pilE17_K155} (Fig ix in S1 Text). These results show that the variable tail plays an important role in pilus:pilus interaction and aggregation, but this is not sufficient for restoring aggregation in all pilin variants.

We have previously shown that pilin posttranslational modification can affect pilus:pilus interaction [11,25]. The glycan attached to Ser63 in PilE_wt, and likely PilE₂₄, is positioned to impact the size and chemistry of the holes that govern the aggregative property of the T4P. PilE₁₇ and PilE₃₂ lack the Ser63. Therefore, we addressed the question whether loss of pilin glycosylation inhibits colony formation. To this end, we deleted the gene pglF encoding the flippase required for pilin glycosylation [42] and found that the colony phenotype was unchanged in all strains (Fig x in S1 Text). This finding shows that the loss of posttranslational modification at Ser63 is not the reason for loss of interaction of strains wt_pilE17 and wt_pilE32.

We tested whether the 2 distinct phenotypes, aggregating and planktonic, correlate with bacterial growth. First, we imaged the cells at different time points during growth. The wt* strain formed spherical colonies whose size increased with time due to growth and fusion (Fig 6B). Over time, the wt* colonies formed networks while strain wt_pilE17 remained planktonic (Fig 6C). We characterized the growth kinetics by determining the colony forming units (CFUs) as a function of time for strains wt* and wt_pilE17 (Fig 6A). The aggregating strain shows a detectable lag phase, whereas the planktonic strain resumes growth immediately after inoculation. After 2 h, both the planktonic and the aggregating strains exhibit exponential growth. After approximately 10 h, the planktonic strain enters into the stationary phase, while the aggregating strain continued to grow up to 19 h. The growth kinetics of wt_pilE24 was comparable to wt* and the kinetics of wt_pilE32 was reminiscent of wt_pilE17 (Fig xi in S1 Text). To evaluate whether the exponential growth rate depends on the lifestyle, we determined the growth rates of aggregating strains (pooled for wt* and wt_pilE24) and planktonic strains (pooled for wt_pilE17 and wt_pilE32) (Fig xi in S1 Text). The aggregating strains grow at a rate of h^-1, which is significantly lower than the rate of planktonic strains h^-1.

Fig 6. Growth kinetics of planktonic and aggregating strains.

(A) CFU of the wt* strain (gray) and the wt_pilE17 strain (purple). Shown are mean values and standard errors of 3 to 4 biological culture replicates. Typical brightfield images of (B) wt* and (C) wt_pilE17 populations during growth in liquid culture. Samples were pipetted from the bottom of the microtiter plates and transferred to microscopy plates for imaging. Scale bar: 100 μm.

https://doi.org/10.1371/journal.pbio.3003022.g006

It has been reported previously that loss of T4P enhances both the growth rate of N. gonorrhoeae [15] and transcription of metabolic genes [43]. This increase can be caused by the fact that pilin generation or pilus biogenesis consumes energy, reducing the growth rate [11] or by the fact that piliated gonococci form colonies in which central bacteria are growth arrested [44]. In this study, all strains are similar in their piliation levels, yet the aggregating wt and wt_pilE24 demonstrate very different growth behavior from the planktonic wt_pilE17 and wt_pilE32. These data indicate that aggregation strongly affects growth.

Next, we assessed whether variation of pilE affects antibiotic susceptibility, i.e., the ability to grow at elevated levels of antibiotics. We determined the minimal inhibitory concentrations (MICs) of antibiotics with different targets, in particular cell wall synthesis (ceftriaxone), DNA gyrase/topoisomerase (ciprofloxacin), and the ribosome (kanamycin) (Fig 7A). Despite the differences in aggregative behavior, there is no significant difference in MICs between the variant strains.

Fig 7. Effects of microcolony formation on antibiotic resistance and tolerance.

(A) MICs for all pilin variants and antibiotic treatments were determined as the modal value from 3 biological culture replicates. (B) Fraction of viable cells (CFU normalized to CFU at the start of treatment) as a function of time during treatment with ceftriaxone (4.8 μg/ml), starting at 10 h of growth. Combined p-values (see Methods): p_wt-wtpilE24 = 0.34, p_wt-wtpilE17 = 8.8 × 10⁻⁵_, p_wt-wtpilE32 = 0.00035, p_wt-ΔpilE = 5.3 × 10⁻⁵. (C) Fraction of viable cells as a function of time during treatment with ceftriaxone (4.8 μg/ml), starting at 6 h of growth. p_wt-wtpilE17 = 0.00025. (D) Fraction of viable cells during treatment with ciprofloxacin (2.4 μg/ml). p_wt-wtpilE17 = 8.28 × 10⁻⁵. (E) Fraction of viable cells during treatment with kanamycin (240 μg/ml). p_wt-wtpilE17 = 0.69. Gray: wt, dark gray: wt_pilE24, purple: wt_pilE17, green: wt_pilE32, blue: ΔpilE. For (B–E), mean and standard error over 3 to 5 biological culture replicates are shown. The data underlying this figure can be found in S1 Data. CFU, colony forming unit; MIC, minimal inhibitory concentration.

https://doi.org/10.1371/journal.pbio.3003022.g007

To characterize antibiotic tolerance, i.e., the ability to survive antibiotic concentrations higher than the MIC for extended periods of time, we examined the killing kinetics during treatment with lethal doses of antibiotics. Ceftriaxone is currently recommended for treatment of gonorrhea [45] and, therefore, we started by investigating its effects on survival during the planktonic and microcolony lifestyles. We let gonococci grow for 10 h in liquid media as described above and then added ceftriaxone at 4.8 μg/ml, which is 600× the MIC. The corresponding growth and survival curves in the absence of antibiotic treatment are shown in Fig 6. We found that the planktonic strain wt_pilE17 was killed significantly faster than the aggregating strain wt* (Fig 7B). To determine whether this difference was caused uniquely by aggregation or whether the T4P per se played a role in tolerance, we characterized the killing kinetics in a pilE deletion strain. ΔpilE cells were killed slightly but significantly faster than the wt_pilE17 strain, indicating that T4P or pilin have a protective role against ceftriaxone mediated killing. The killing kinetics of strain wt_pilE24 was comparable to wt* and that of wt_pilE32 and was comparable to wt_pilE17 (Fig 7B).

To verify that the antibiotic concentration is not a limiting factor for killing, we investigated the time-kill kinetics at 3 different ceftriaxone concentrations (2.4 μg/ml, 4.8 μg/ml, and 9.6 μg/ml) and found that the aggregating strains are more tolerant than the planktonic strains independent of the ceftriaxone concentration (Fig viiA–C in S1 Text). Concluding, tolerance is not due to the aggregating cells having a lower antibiotic dose per cell, as their cell density is in fact lower after 10 h of growth than that of the planktonic cells, meaning their antibiotic dose per cell is even higher (Fig 6A). This suggests that we might underestimate the protective effect of aggregation. Since the killing kinetics were qualitatively independent of the antibiotic concentration, we conclude that the antibiotic dose per cell plays a minor role in this assay. Next, we tested whether the growth phase affects the protective effect of aggregation (Fig 7C). We treated the planktonic strain, wt_pilE17, and the aggregating strains, wt*, at 6 h of growth, i.e., in the middle of exponential growth with ceftriaxone at 600× MIC (4.8 μg/ml). We observed that the protective effect of aggregation, observed for wt*, was even stronger compared to treatment after 10 h of growth. We conclude that aggregation makes gonococci more tolerant to ceftriaxone treatment while the presence of pilin has a minor effect.

Finally, we addressed tolerance against bactericidal antibiotics with different cellular targets, in particular DNA gyrase/topoisomerase (ciprofloxacin), and the ribosome (kanamycin) at 10 h of growth (Fig 7D and 7E) [46,47]. We tested ciprofloxacin at 600× MIC (2.4 μg/ml) and kanamycin at 15× MIC (240 μg/ml) as the latter is not soluble at concentrations corresponding to 600× MIC. For ciprofloxacin (Fig 7D), the difference between the aggregating and the planktonic strain is even more pronounced than for ceftriaxone, with aggregating wt* exhibiting substantially higher viability than planktonic wt_pilE17 cells. Notably, the planktonic cells showed a bi-phasic killing curve reminiscent of persister cells [32]. Under kanamycin treatment, the killing kinetics of aggregating and planktonic strains were comparable (Fig 7E). Nonetheless, the observed differences in tolerance to ceftriaxone and ciprofloxacin show that survivability of N. gonorrhoeae depends on its pilin antigenic variants, because the variation modulates T4P-mediated aggregation.

We used the same growth conditions described in previous studies [27]. Gonococcal base agar was made from 10 g/l dehydrated agar (BD Biosciences, Bedford, MA), 5 g/l NaCl (Roth, Darmstadt, Germany), 4 g/l K₂HPO₄ (Roth), 1 g/l KH₂PO₄ (Roth), 15 g/l Proteose Peptone No. 3 (BD Biosciences), 0.5 g/l soluble starch (Sigma-Aldrich, St. Louis, MO), and supplemented with 1% IsoVitaleX (IVX): 1 g/l D-glucose (Roth), 0.1 g/l L-glutamine (Roth), 0.289 g/l L-cysteine-HCL x H₂O (Roth), 1 mg/l thiamine pyrophosphate (Sigma-Aldrich), 0.2 mg/l Fe(NO₃)₃ (Sigma-Aldrich), 0.03 mg/l thiamine HCl (Roth), 0.13 mg/l 4-aminobenzoic acid (Sigma-Aldrich), 2.5 mg/l β-nicotinamide adenine dinucleotide (Roth), and 0.1 mg/l vitamin B12 (Sigma-Aldrich). GC medium is identical to the base agar composition but lacks agar and starch.

All strains used in this study (Table i in S1 Text) were derived from the opa- selected VD300 strain [57] and carry a deletion of the G4 motif required for pilin antigenic variation [58]. Colonies were grown overnight on agar plates and each colony used for the experiments was inspected using a stereomicroscope to ensure that it maintained its opa- phenotype.

In order to construct isogenic strains that differ solely in the pilE sequence, the pilE sequences of the clinical isolates were cloned into the ΔG4 background strain (Ng150, derivative of MS11, [59]) replacing the native pilE gene. To avoid further genetic modification, the pilE genes were introduced by ermC-rpsL_s based clean insertion as described [18].

The process for the construction was identical for strains wt_pilE17, wt_pilE24, and wt_pilE32 except for the respective primers. First, the 5′ UTR region of pilE was amplified from gDNA of strain Ng150 using primers sk159 and sk160 (Table iii in S1 Text). Second, the pilE gene was amplified from gDNA of the different clinical isolates NG17, NG24, and NG32 with primers sk161 and sk175, sk161 and sk167, and sk161 and sk162, respectively. Third, the 3′ UTR of pilE including the ermC-rpsL_s was amplified from gDNA of strain Ng225 using primers sk163 and sk158. The 3 PCR products were fused and the final product was spot transformed into strain Ng150. After selection on erythromycin, insertions were controlled via screening PCR with primers sk32 and sk45. The respective strains were named ΔG4 pilENG17/24/32 clean insertion step 1 (Ng239, Ng241, Ng229, respectively).

Next, the fusion construct for the counter selection was generated. To this end, the 5′ UTR region including the newly introduced pilE genes was amplified from gDNA of strains ΔG4 pilENG17/32/24 clean insertion step 1 (Ng239, Ng229, Ng241) with primers sk159 and sk176 (Table iii in S1 Text), sk159 and sk168, sk158 and sk165, respectively. The 3′ UTR of pilE was amplified from gDNA of strain Ng150 with primers sk164 and sk158. The 2 products were joined in a fusion PCR and transformed into the respective strain of the first step (Ng239, Ng241, Ng229). During selection on streptomycin, the ermC–rpsL_s construct is spliced out and the mutants are isogenic to the parental strain (Ng150) except for the pilE sequences. Insertions were controlled via screening PCR with primers sk32 and sk45 and subsequently checked via sequencing with primer sk129 (Eurofins). The final strains are referred to as wt_pilE24 (Ng242), wt_pilE17 (Ng240), and wt_pilE32 (Ng230).

pilE_T136 hybrids: The N-terminal part of the pilE gene of strain wt_pilE17 (Ng240) and wt_pilE32 (Ng230) was fused with the C-terminal part of pilE of the wt* (Ng150) starting at amino acid T136. In detail, amino acids K137-E160 of pilE₁₇ and T135-P158 of pilE₃₂ were replaced by T136-K160 of pilE_wt. To this end, the N-terminal part unil K137 of pilE₁₇ and T136 pilE₃₂ including 567 bp upstream were amplified from wt_pilE17 (Ng240) or wt_pilE32 (Ng230) using primers sk384 and sk385 or sk384 and sk388, respectively (Table iii in S1 Text). The T136 C-terminal part including the ermC rpsL_s cassette and the downstream part of pilE were amplified from strain T126C step1 (Ng225, [18]) using primers sk386 and sk390 (pilE₁₇) or sk389 and sk390 (pilE₃₂). Both PCR products were fused with the GXL-polymerase (TaKaRa). The final fusion product was transformed into either wt_pilE17 (Ng240) or wt_pilE32 (Ng230). Selection was achieved by plating transformants on agar plates containing erythromycin. Correct insertion was checked via screening PCR (sk5 and sk32) and sequencing (sk5 or sk32). The mutants were named wt_{pilE17_T136} step1 (Ng292) and wt_{pilE32_T136} step1 (Ng294).

pilE_K155 hybrids: The C-terminal tail of the pilE gene of strains wt_pilE17 (Ng240) and wt_pilE32 (Ng230) were replaced with the C-terminal tail starting at K155 of pilE of strain wt* (Ng150). The ermC-rpsL_s cassette was amplified from strain T126C step1 [18] and fused with the corresponding pilE upstream and downstream regions. The fusion product was transformed into either wt_pilE17 (Ng240) or wt_pilE32 (Ng230) and selected on erythromycin.

The pilE_17/32 gene including 567 bp upstream were amplified from wt_pilE17 (Ng240) or wt_pilE32 (Ng230) using primers sk384 and sk391 or sk384 and sk393, respectively (Table iii in S1 Text). The K155 C-terminal tail including the ermC rpsL_s cassette and the downstream part of pilE were amplified from strain T126C step1 (Ng225) using primers sk392 and sk390 (NG17) or sk394 and sk390 (NG32). Both PCR products were fused with the GXL-polymerase (TaKaRa). The final fusion product was transformed into wt_pilE17 (Ng240) or wt_pilE32 (Ng230) and after transformation selection was performed on plates containing erythromycin. Correct insertion was checked via screening PCR with primers sk5 and sk32 and sequencing (sk5 or sk32) resulting in strains wt_{pilE17_K155} step1 (Ng304) and wt_{pilE32_K155} step1(Ng306).

Deletion of the flippase pglF strongly reduces pilin glycosylation by interrupting the membrane translocation of lipid-attached carbohydrates [42]. To delete pglF, the respective strains were transformed with genomic DNA of strain Ng156 and selected on kanamycin [11].

Genomic DNA of strain Ng105 [11] was used for transformation to insert GFP into the chromosome of the respective strains, generating strains wt_{pilE24 green} (Ng309), wt_{pilE17 green} (Ng308), and wt_{pilE32 green} (Ng310) (Table i in S1 Text). Transformants were selected on plates containing erythromycin.

The pilE gene was interrupted with a kanamycin resistance cassette. Three individual DNA fragments were amplified via PCR and fused; 5′ pilE including the upstream pilE region was amplified with primers sk45 and sk46 from gDNA of strain Ng150. kanR was amplified with primers sk47 and sk48 from genomic DNA of strain Ng052 [60]; 3′ pilE and the downstream region of pilE was amplified using primers sk49 and sk50. The PCR products were fused and the fusion construct was transformed into wt* strain. Transformants were selected on kanamycin.

To verify that the pilE variants are products of pilin antigenic variation, we first determined the pilS copies in the genome of the clinical isolates. pilS were identified using annotated pilS copies of strain N. gonorrhoeae MS11 (NCBI, CP003909.1) as reference. End and start sites of pilS copies were determined either by matching blast results using the SnapGene software (www.snapgene.com) or blastn (BLAST, NCBI) of the respective MS11 orthologs. The blast results were verified manually by identifying the conserved cysteine regions cys1 and cys2 in each pilS copy. The sequence identity of pilS copies from clinical isolates to pilS copies from strain N. gonorrhoeae MS11 were obtained via blastn (BLAST, NCBI).

Next, we tracked each bacterium over the whole measurement time. From each track, we calculated the MSD from the displacements [61]. As the bacteria follow a corrleated random walk, we fitted ) + A for the first 5 s to determine the correlation time τ_c and velocity v to the data of a single track (Fig vi in S1 Text). The variable A accounts for the tracking error. We took the mean and standard variation of the fit parameters to determine the average correlation time τ and velocity v_corr for the pilE variants. Statistical analysis of the distributions of the fit parameters for single bacteria tracks was performed via the Mann–Whitney U test.

AlphaFold [39] was used to generate models of PilE_wt and the PilE varients. Models were superimposed upon Chain A of the N. gonorrhoeae T4P cryoEM reconstruction [40] in Chimera (Petterson and colleagues, PMID 15264254) and residues 1–48 of the model were replaced with that of Chain A to replace the continuous α1-helix of the AF models with the melted helix seen in the filament. The helical symmetry parameters of N. gonorrhoeae T4P (10.1 Å rise, 100.8° rotation) were imposed on the PilE models to generate 18-mer filament models. Electrostatic potential was generated using the APBS tool of PyMol [62].

The distribution of aggregate sizes was determined by incubating the respective strains at an initial OD₆₀₀ of 0.033 in an Ibidi 8-well plate attached to a cover glass for 1 h and imaging using brightfield imaging (Fig vii in S1 Text). Aggregates were segmented using the Fiji segmentation tool and the radius was determined from the area of the segments assuming that aggregates were spherical. Strains wt_pilE17, wt_pilE32 did not form aggregates. Strain wt_pilE17K155 had a stronger tendency to aggregate than wt_pilE17, but the aggregates did not show clearly defined colony shapes, and therefore, we did not attempt to determine the radii of these aggregates (Fig 5E). Therefore, these 3 strains were excluded from the analysis. All strains were characterized in biological culture replicates on at least 3 different days.

We estimated the number of cells per aggregate assuming that the aggregates are spherical and that the bacteria are spheres with a radius of 0.5 μm. The volume fraction of gonococci is Φ≈0.5 [63].

The interaction forces of pilE variants were determined via a dual laser trap. The experimental setup and analysis is already published [22]. In short, we resuspended a few bacterial colonies from overnight GC agar plates in liquid GC medium. We added 1:1,000 ascorbic acid (500 mM). Next, we inoculated the bacteria on a BSA coated cover slip (1 mg/ml) and sealed the slide with VALEP (Vaseline, wool fat and paraplast in a ratio of 1:1:1). The major building blocks of the laser tweezers setup consist of a microscope equipped with a thermo-box at 33°C, an IR-laser (1,064 nm) and an acousto-optical deflector which creates 2 time-shared optical potentials. The trap distance was set to 2.64 μm. We acquired videos of interacting bacteria with a framerate of 50 Hz. After 15 min, we stopped the measurement as we observed decreased activity of the gonococci. All strains were characterized in different samples on at least 3 different days.

We detected the displacements d of the bacteria from the equilibrium positions via a Hough transformation algorithm. From the displacement tracks, we determined the forces (F ~ d) and identified the interaction states, as described earlier in [22]. The potential of each trap was approximated to be harmonic for forces up to 80 pN.

At 100% laser intensity, the traps showed a trap stiffness of k_100% = 0.1 pN/nm, whereby the laser intensity I is proportional to the stiffness of the trap k ~ I. We note that it was not possible to conduct the experiments with the same laser intensities for all variants, since deflections of the wt_pilE32 and wt_pilE17 strains were infrequent. This indicated that these 2 strains have lower interaction forces as they could not overcome the trapping potential of the traps at 100% laser power. Therefore, we adjusted the intensities of the laser. The measurements for the wt* and wt_pilE24 could be performed at 100% laser power while for the other variants, the laser intensity needed to be decreased to 10% for strain wt_pilE32 and to 5% for strain wt_pilE17. Under these conditions, the probability of pilus:pilus binding (Fig ix in S1 Text) was high enough for characterizing T4P mediated attractive forces.

To determine the number of CFUs during 19 h of growth, we performed the same protocol as described above and additionally transferred a whole well to a 1.5 ml reaction tube every 1 to 3 h. Next, we harvested the bacteria by centrifugation (5,000 g, 3 min) and resuspended them in 500 μl GC media. Subsequently, we vortexed for 30 s and added 500 μl of MQ-water to initiate the disassembly of gonococcal aggregates. Then, we again vortexed the suspension for 2 min. The prolonged time of vortexing was already shown to be sufficient to shear pili [64]. Additionally, we ensured reproducibility for each strain and for strains showing the same lifestyle. We performed 1:10 dilution series with vortexing inbetween and plated 50 μl of different dilutions on non-selective GC agarplates. After 48 h of growth (37°C with 5% CO₂), we counted the CFUs.

To determine the growth rates, we plotted the growth curves from 2 h to 8 h. Then, we performed a linear regression fitlm for the log-plotted data via Matlab [61]. The slopes are defined as the growth rates. The significance analysis was performed via a pairwise ANOVA test for the linear regression models which includes an interaction term for the different pilE variant strains [61].

Statistical analysis of the killing kinetics was performed via a combined p-values method for discrete data [65]. We have used the Mann–Whitney U test to compare single time points followed by the combination of the p-values via Mudholkar and George combining method [66].