Pilus Production in Acinetobacter baumannii Is Growth Phase Dependent and Essential for Natural Transformation

Rapid bacterial evolution has alarming negative impacts on animal and human health which can occur when pathogens acquire antimicrobial resistance traits. As a major cause of antibiotic-resistant opportunistic infections, A. baumannii is a high-priority health threat which has motivated renewed interest in studying how this pathogen acquires new, dangerous traits.

type of HGT called natural competence for transformation, A. baumannii is able to take up extracellular DNA from its environment and incorporate it into its own genome by homologous recombination (4)(5)(6)(7). We recently reported that such transformation events can lead to frequent exchanges of genomic regions greater than 100 kbp in the naturally competent bacterium Vibrio cholerae, which could explain how bacteria such as A. baumannii acquire new DNA stretches including resistances. However, few reports have addressed natural competence directly in A. baumannii, instead extrapolating its behavior based off that of the soil bacterium Acinetobacter baylyi (8,9).
The few studies on transformation in A. baumannii have focused mainly on mild competence inducers such as serum albumin and Ca 21 , on transforming materials, and on pH (10)(11)(12). However, transforming protocols vary wildly between studies (13)(14)(15), including the use of different solidifying agents for transformation scoring on surfaces (16). Additionally, only few isolates of this species, such as the strains A118 (13) and M2 (recently reclassified as Acinetobacter nosocomialis) (15,17), were previously found to be naturally competent, though recent studies are showing that a plethora of clinical and wildlife/livestock A. baumannii isolates are likewise naturally transformable (12,14,18,19). Therefore, the process of natural competence in A. baumannii needs to be better studied and recorded.
Additionally, a number of studies have shown the induction of transformation potential in a surface-dependent manner, suggesting a correlation between natural transformability and the movement of A. baumannii on (wet) surfaces (14,15). This correlation is thought to be based on the bacterium's ability to produce type IV pili (TFP) (15), which are cell appendages that frequently constitute the central part of the DNAuptake machinery (7) (Fig. 1A). Known to play a main role in the DNA uptake complex, the regulation of TFP production is often linked to the bacterium's competence program (6). For instance, in V. cholerae, the TFP, which enhances the bacterium's natural transformability, is produced only when the bacterium grows on chitinous surfaces (20)(21)(22)(23)(24). Considering that TFP also mediate other functions that include adhesion, motility on surfaces (i.e., twitching motility), and surface sensing (25), it makes sense that the TFP would be the link between transformability and mobility in A. baumannii.
To better understand the dual roles of this dynamic TFP, it is important to understand when it is produced in the bacterium. The TFP is composed of major (PilA) and minor pilin subunits (different kinds in diverse bacteria), and extension and retraction events are energized by the cytosolic PilB and PilT/PilU ATPases, respectively. Upon or after TFP retraction, the incoming transforming DNA enters the periplasmic space through the PilQ secretin where it is bound by the DNA-binding protein ComEA. This protein is thought to act as a Brownian ratchet, which leads to the accumulation of long stretches of DNA within the bacterium's periplasm (22,26). After degradation of one strand, which, in Gram-negative bacteria, is attributed to an unidentified nuclease or a periplasmic extension of the ComEC protein, the single-stranded DNA translocates across the inner membrane through the ComEC channel aided by the ComF protein (Fig. 1A). Once in the cytosol, the single-stranded DNA is bound by the single-stranded binding protein Ssb, the RecA-loading protein DprA, and RecA. RecA ultimately fosters recombination if the transforming DNA is homologous with the chromosomal DNA of the cell. ComM (a RadA homolog) assists in this process, especially if heterologous DNA is present between the homologous flanks (7,27,28).
There are different types of TFP regulatory systems which exist in both competent and noncompetent bacteria. A major TFP regulatory system that has been studied in Pseudomonas aeruginosa (a bacterium that was only recently reclassified as naturally transformable [29]) is the two-component system (TCS) PilSR (30)(31)(32). The current model of this TCS suggests that extended pili are sensed due to the lack of inner membrane-associated PilA subunits that would otherwise interact with the atypical sensor histidine kinase PilS. The lack of interaction stimulates PilS's kinase activity, which leads to activation of the response regulator PilR and, ultimately, pilA transcription (33). This PilS-mediated pilin inventory and its impact on PilR activity also influence other coregulated processes involved in virulence and surface-associated bacterial behavior (34). The Pil-Chp chemosensory system also plays a role in TFP assembly and twitching motility (35,36). This system is composed of 10 or more components, which are mostly encoded by the pil-chp operon (pilGHIJK-chpABC) in P. aeruginosa (37,38), and was shown to react to mechanosensing signals upon surface attachment (39). The main components of the system are a transmembrane chemoreceptor (PilJ), a histidine kinase (ChpA), and the response regulators PilG and PilH, all of which share homology with flagellar chemotaxis proteins (40). The response regulators modulate intracellular cAMP levels by regulating the activity of the adenylate cyclase CyaB (41,42). Increased intracellular cAMP levels subsequently activate the virulence regulator Vfr (43). Importantly, PilG and PilH also regulate TFP dynamics in a cAMP-independent manner by modulating PilB-driven pilus extension and PilT-driven pilus retraction, respectively (37,41).
In this study, we characterized aspects of the natural competence program of A. baumannii. We showed that the pathogen's transformability varies significantly during the different growth phases. This variability is due to growth phase-dependent production of its TFP. From genetically engineered mutants and pilus visualization, we demonstrated that pilus-related genes are essential for the bacterium's transformability and its surface motility, while pilus-unrelated competence genes do not interfere with the motility phenotype. Based on their homology to their P. aeruginosa counterparts, we identified several conserved TFP regulators and showed that these regulators affect cellular piliation status and influence natural transformation in A. baumannii.

RESULTS AND DISCUSSION
Transformation of A. baumannii occurs mostly during exponential growth. To begin, we sought the optimized conditions for transformability for strain A118 (13) ( Table 1) by adapting variations from different available protocols. We found the ideal conditions to be after aerobic growth in broth followed by DNA uptake on agar surfaces (see Materials and Methods). We next tested the impact of the growth phase during the liquid culturing time on the bacterial transformability. As shown in Fig. 1B, the highest levels of transformation were observed for bacteria grown to the exponential phase (i.e., ;90 min), while the strain's transformability fell below detection after A. baumannii's dilution into fresh medium or upon entry into the stationary phase ( Fig.  1). This behavior is distinct from that of A. baylyi, which can reach transformation frequencies up to 0.7% of all cells in an autotrophy/prototrophy transformation experiment (44) and is transformable throughout all growth phases with varying efficiencies (45,46). This underscores the importance of studying A. baumannii in its own right and not relying on assumptions made from model bacteria.
To better understand this rather tight transformation window, we scored transcript levels of selected competence genes as a function of time. To identify competence genes in A. baumannii strain A118, we investigated its previously published genome sequence, where Ramirez et al. reported 186 scaffolds at least 500 bp in length (47). However, the overall high number of contigs (1,647 contigs, accession number AEOW01000000) complicated the search for genes and the determination of operons. Therefore, we resequenced the strain using long-read PacBio technology and de novo assembled its genome (see Materials and Methods).
With the 3,750,370 bp closed genome of A. baumannii at hand (Table 2), we then identified all known competence genes based on homology to the competent model organisms A. baylyi and V. cholerae (21,48) (Table 3). Next, we scored the transcript levels of six selected competence genes using qRT-PCR: pilA, pilQ, pilT, comEA, comEC, and dprA. The encoded products of these genes (e.g., the major pilin subunit PilA, the outer membrane secretin PilQ, and the retraction ATPase PilT) contribute to the assembly of the central TFP of the DNA uptake machinery and are key for DNA uptake in Gram-negative bacteria. They are also important for DNA translocation across the outer (ComEA) and inner (ComEC) membrane or in recombination-mediating activity (DprA) inside the

MB#8960
This study cytosol (7) (Fig. 1A). As shown in Fig. 1C, the transcript levels varied at time points before, during, and after the growth phase in which transformation levels were the highest (Fig. 1B). In general, the pil genes specifically peaked around 90 min after dilution ( Fig. 1C), which correlated with the beginning of the transformation window (Fig.  1B). This differed from previous reports on A. baylyi where Porstendörfer et al. showed that the expression of the pilA homologues major pilin gene comP decreased immediately after dilution into fresh medium, with the lowest levels in the mid-exponential phase. In their study, maximal comP expression levels were observed in the late stationary phase (46). Peak transformation coincides with type IV pilin production. As transcript levels do not necessarily reflect the cellular protein levels, we next tested the production of the major pilin protein PilA. To accomplish this, we translationally fused PilA to a short FLAG tag, which gave a strain with the pilA-FLAG allele at the native pilA locus that retained suboptimal but still high transformation levels (Fig. 1D). We grew this genetically engineered strain and quantified the PilA-FLAG protein levels in cells over time by Western blotting. From these data, we concluded that no PilA-FLAG was detected early on in growth. We observed that the major pilin level peaked at approximately 90 to 105 min postdilution (optical density at 600 nm [OD 600 ] of ;0.6 to 0.7), declined after 120 min of growth (OD 600 of ;1.0), and ultimately disappeared later during growth (Fig. 1E). To control for surface-exposed pili that might have been lost through shearing during the harvesting process, we repeated the same experiment in a pilQ deletion strain in which pili cannot cross the outer membrane. We observed a similar protein production pattern over time (Fig. 1F) to that of the parental wild-type (WT) background strain (Fig. 1E). From these data, we concluded that A. baumannii produces its TFP solely during the early exponential phase under liquid growth conditions. This is in stark contrast to A. baylyi, in which the major pilin is primarily produced in the stationary phase and is either absent or only present at low levels during the exponential growth phase (46).
Pilus visualization elucidates diverse phenotypes. We extended our study beyond quantification of the major pilin subunit by visualizing the TFP of A. baumannii.
To do so, we aimed at using thiol-reactive maleimide-conjugated dyes to cysteine-containing cell appendages, an approach that has been used for the flagellum of Bacillus subtilis (49) and the tight adherence (Tad) pilus of Caulobacter crescentus (50). To determine the proper location for a site-directed cysteine knock-in into the major pilin PilA, we followed the same protocol as that described for V. cholerae (23). Briefly, we predicted the surface-exposed ab-loop of PilA using the Phyre2 web portal (51) (Fig. 2A) and found that the overall structure prediction for A. baumannii's PilA was similar to that of V. cholerae ( Fig. 2A) by length and homology in the protein's N-terminal region. From this analysis, we selected an alanine-to-cysteine mutation at position 61 (A61C),   (23,52). We tested the transformation efficiency of the new A118-PilA(A61C) strain relative to that of the parental WT strain. As shown in Fig. 2B, the strain's transformation efficiency was reduced by approximately 18-fold compared to that of the WT, but it nonetheless maintained robust transformability. To potentially improve the transformation efficiency of cysteine knock-in mutants, we next designed and tested six additional site-directed mutants with amino acid exchanges in the predicted surface-exposed region (G60C, V62C, T64C, T72C, T75C, and S77C). None of these variants resulted in significantly higher transformation levels compared to those of the strain producing PilA(A61C) (Fig. 2B). Therefore, we proceeded with the PilA (A61C) strain given that our previous work on V. cholerae had identified this region as appropriate for the pilus labeling process (23).
To visualize the cells' pili, we next grew the PilA(A61C) strain to an optical density of ;0.6 to 0.7, as pilin levels were highest at this growth stage (Fig. 1E). At this time point, we added the thiol-reactive dye (Alexa Fluor 488 C5 maleimide) to the culture and then incubated, washed, and mounted the bacteria onto agarose pads. As shown in Fig. 2C, the bacteria produced pili that, for some cells, extended far beyond the outer membrane. Notably, such extended pili were not restricted to one side of the cell body as recently described for A. baylyi in a preprinted study (53). The bodies of the cells were also stained by the dye in a significant fraction of the population, which suggested incorporation of the pilin subunit into the inner membrane. Another subfraction appeared as fully unlabeled.
We considered two possible explanations for this unlabeled fraction. First, production of the pilus is a heterogeneous and potentially bistable phenotype in A. baumannii similar to competence development in B. subtilis (54)(55)(56). Second, the absence of cell body labeling might be due to the production of pilus proteins that were not surfaceexposed during the short staining time window. While we previously demonstrated for V. cholerae that cysteine-containing PilA subunits in the inner membrane were labeled even in the absence of the residual TFP components (23), this was not the case for C. crescentus (50). In C. crescentus, only extended pili and membrane-incorporated pilin subunits from pilus retraction events were labeled. It was argued that this might be due to the size-based exclusion of the AF488-maleimide dye by the outer membrane (50). While outer membrane permeability and/or porins might differ between V. cholerae, C. crescentus, and A. baumannii, we speculated that in the latter two organisms, cell-surrounding capsular polysaccharides might also impede the dye's access to the inner membrane-incorporated pilin subunits. These capsular polysaccharides are considered a major virulence determinant of A. baumannii (57) and could explain the lack of cell body staining in our images. Upon further inspection of the A118 genome sequence, we identified a 22,663 bp genome stretch between the genes fkpA and lldP,  Bars show the average transformation frequency of three independent experiments (6SD). Statistical analyses were performed on log-transformed data using a one-way ANOVA followed by Sidak's multiple-comparison test. Each mutant strain is compared to the WT strain. #, under detection limit in at least one experiment, in which case the detection limit was used for the calculation of the average value and statistical analyses. **, P , 0.01; ***, P , 0.001; (Continued on next page) Transformation and Pili in A. baumannii Journal of Bacteriology which usually flank capsular biosynthesis clusters (57). BLAST analysis of this cluster showed a 100% conservation of the gene order (18 genes in total) and a 99.2% pairwise sequence identity to the KL51 capsule biosynthesis gene cluster of A. baumannii isolate WM98c (GenBank accession number MN148384 [58]), suggesting that strain A118 is indeed encapsulated. To test our second hypothesis, we imaged a pilT-mutant derivative of strain A118-PilA(A61C). Our underlying rationale was that pili that did extend would remain surface-exposed in this strain in the absence of the retraction ATPase PilT. Indeed, we observed many cells with multiple pili or bundles/clusters thereof (Fig. 2C). Interestingly, the cell bodies themselves were not labeled, which is consistent with the notion that pilus retraction was required to reinsert labeled pilin subunits into the inner membrane after they were pulled through the capsular material and the outer membrane secretin PilQ. However, this strain still showed labeling heterogeneity, which suggested that the pilus production was a phenotype in only a subfraction of the population. Interestingly, A. baumannii strains are known for a number of phasevariable controlled phenotypes, including, for strain AB5075, cell morphology, biofilm formation, and surface motility (59), which is frequently linked to pili. Thus, we can speculate that the pilus-producing and nonproducing bacteria within the population are phase variants that foster diverse phenotypes.
While pilT mutants are frequently hyperpiliated (25), including a pilT mutant tested for A. nosocomialis M2 (15), the discovery of this in A. baumannii was somewhat surprising, as previous work noted the absence of pili in a pilTU mutant of strain ATCC 17978 (60). However, we must note that the absence of pili was scored by transmission electron microscopy in that study, which does not distinguish between diverse cell appendages, and that the fixation process might shear off pili. Interestingly, previous work had demonstrated that other A. baumannii strains produced both thin (;4 nm wide) and thick (;7 nm) pili (14). Notably, Wilharm et al. demonstrated that the thick pili were rarely observed on WT cells (only ;1 in 25 to 50 cells exposed such a pilus) though were frequently witnessed in a pilT deletion strain where each cell exposed one or several pili (14). These findings are consistent with the hyperpiliation phenotype described in this study. Indeed, our preliminary scanning electron micrographs of strain A118 supported the observation of multiple cell appendages of different widths (some of those even in a pilA-minus strain), which motivated us to adapt the pilus-specific thiol-labeling approach to unambiguously score only PilA-composed TFP.
Competence and pili mutants are impaired in transformation. To further investigate the link between pilus production and the strain's natural transformability, we next generated a set of defined deletion strains. In this context, we also generated knockout strains of conserved competence genes that we considered to encode parts of the DNA uptake machinery based on common knowledge in other naturally competent bacteria (7,61). Precisely, we generated mutants where the gene products were required for the main steps of the DNA uptake and recombination process, namely, the dynamic TFP (pilA, pilQ, pilT, and pilU), DNA uptake and translocation (comEA and comF), and ssDNA binding and recombination (dprA and comM) (Fig. 1A and Table 3). We tested these mutants for natural transformation abilities and observed abrogation of this process for all but two mutants, namely, the DpilU and DcomM mutants (Fig.  3A). The lack of transformation defects in the absence of comM was in line with the minor impact observed for similar mutants in diverse V. cholerae strains (62,63), given that the current transformation assay was based on genomic DNA as transforming material that contained a resistance marker for transformant scoring. In contrast, there was a significant decrease of the transformability in comM mutants of V. cholerae and A. baumannii when PCR-amplified fragments served as transforming material, which led to the exchange of single nucleotides or short stretches of DNA (28,64). PilU, on the other hand, is a secondary retraction ATPase (PilT is the primary enzyme) that is present in many Gram-negative bacteria, including P. aeruginosa, in which it is essential for TFP-mediated twitching motility (65). Related to natural competence, we previously showed that PilU of V. cholerae was dispensable for DNA uptake and transformation (21), which is consistent with our present findings (Fig. 3A). Interestingly, we and others recently demonstrated that V. cholerae's PilU works solely in conjunction with PilT instead of compensating for the lack of PilT (52,66), which suggests that it enhances the retraction force instead of behaving as an independent retraction enzyme, as was also suggested for P. aeruginosa (52,66,67).
We next established complementation assays to show causality between the lack of pilus or competence genes in this set of knockout strains. Toward this goal, we cloned the genes downstream of the arabinose-inducible P BAD promoter that was located on a miniTn7 transposon (23,63). These genetically engineered constructs and the parental transposon without any gene inserted downstream of P BAD (TnAraC) were then sitedirectly integrated into the respective strains (Table 1) and tested for their natural transformability. As shown in Fig. 3B, complete or high levels of complementation were observed under arabinose-inducing conditions for strains lacking pilT, comEA, comF, or dprA, compared to those of the WT parental strain and its transposon-carrying  Fig. 1). For complementation, the strains carried a transposon without (control, TnAraC; no gene downstream of P BAD promoter) or with (TnXXXX) the complementing gene on their chromosome and were grown in the absence or presence of 0.2% (B) or 2% (C) arabinose. For all bar plots, transformation frequencies of three independent experiments are plotted as mean values (6SD). ,, below detection limit (DL); #, under DL in at least one replicate (DL used for calculation of mean value). Log-transformed data were used for statistical analysis. When no transformants were obtained, the mean of the detection limit was used for statistical analyses. (D) TFP mutants are nonmotile on solid surfaces. Surface motility of the mutants described in panel A is depicted on the y axis based on the occupied area on the motility plates. Four biological experiments with three technical replicates are shown for each strain (n = 12). Images from one experimental set are shown below the graph. Statistical analyses: panels A to C, one-way ANOVA, using Sidak's multiple comparisons test; panel D, Brown-Forsythe and Welch ANOVA tests with unpaired t test with Welch's correction. *, P , 0.05; ****; P , 0.0001; n.s., not significant. The strains were compared to the WT (A and D) or the most appropriate control strain (boxed strain name in panels B and C). derivative (grown in the absence or presence of arabinose). In contrast, complementation of the pilA-and pilQ deletion strains was less efficient but could be boosted by an increase of the inducer arabinose (Fig. 3C). Collectively, these data confirmed that the transformation defects were caused by the gene-specific deletions and not secondary defects that might have occurred during strain engineering.
Pilus mutants are impaired in their surface motility. Recent studies are reporting considerable variations across A. baumannii strains that might also impact and explain their different surface mobilities. For example, considerable variations in amino acid sequences and glycosylation patterns of PilA proteins have been observed (strain A118 lacks the O-glycosylated C-terminal serine residue that is conserved in several other strains) (68)(69)(70). Moreover, X-ray crystallography highlighted another key difference between A. baumannii PilA proteins, namely, surface electrostatics, which could determine whether pili repulse or adhere to one another (70). Electrostatic adherence could trigger bundle formation, which would influence the strain's preference for biofilm formation over surface-dependent twitching motility. Given the correlation between TFP biosynthesis and natural transformation of strain A118 (Fig. 1), we queried whether deletion of the pilus and core competence genes would affect such surface-dependent movement. As shown in Fig. 3D, we observed large variations in and between independent experiments, which could again be attributed to the species' phase variability that is known to affect surface motility (59). Nonetheless, motile versus nonmotile strains were easily differentiable. We demonstrated that all pil mutants abrogated surface motility, while deletion of nonpilus competence genes did not impair surface movement. This finding contrasts previous reports by Wilharm et al., who showed that inactivation of both pilT and comEC abolished transformation and twitching-like motility in A. baumannii strains 07-095 and 07-102 (14). Interestingly, the pilU mutant had significantly impaired motility (Fig. 3D). This discrepancy between the pilU mutant's efficient transformability (Fig. 3A) and its inefficient surface motility (Fig. 3D) can be explained by the difference in requisite retraction force. While DNA uptake is unlikely to majorly constrain pilus retraction, the friction between the bacterial cell and the surface material likely requires enhanced force generation by the retraction motor PilT, which is accomplished through recruitment of and assistance by PilU. The absence of surface motility of the pilU mutant observed here for A. baumannii phenocopies that of P. aeruginosa (65). Despite this commonality between the two organisms, a recent study by Nolan et al. demonstrated that TFP were dispensable for low levels of natural transformation of P. aeruginosa (29), which contrasted with our findings. Here, we demonstrated that apart from PilU, the pilus components were essential for A. baumannii's natural transformability (Fig. 3A).
TFP regulators impact transformation in A. baumannii. Given the parallels and discrepancies between P. aeruginosa and A. baumannii, we next considered the impact of pilus-specific regulators on surface-dependent motility and transformation of A. baumannii. We were especially interested in the PilSR TCS and the Pil-Chp chemosensory systems (Fig. 4A), as homologs for these regulatory proteins exist in A. baumannii (Table 3). Notably, Leong et al. recently showed that, while important for twitching motility, these signal transduction-related systems (e.g., PilS, PilR, and PilG) were fully dispensable for natural transformation in A. baylyi (strains ADP1 and BD413) (48), which was recently confirmed for the Pil-Chp system (53). Given the differences we found in natural transformability and TFP production between A. baumannii and A. baylyi, we were interested in these systems in A. baumannii. In contrast to A. baylyi, deletion of pilS or pilR as well as chpA or pilG in A. baumannii completely abolished transformation, while the absence of pilH did not change the strain's transformation efficiency (Fig. 4B). Complementation restored natural transformability in the mutants at regular (Fig. 4C) or elevated induction levels (Fig. 4D), except for the pilS deletion strain, which could not be complemented under the tested conditions. However, when we tested a pilSR double mutant, we restored its function through production of a phosphomimetic PilR variant (D56E), while the native and unphosphorylated PilR was insufficient for restoring transformation in the mutant (Fig. 4E). In contrast, both variants efficiently Journal of Bacteriology complemented a pilR single knockout in which the PilS protein was maintained to restore piliation (Fig. 4F) and natural transformation (Fig. 4E). Therefore, these data confirm that PilS is required for natural transformation in A. baumannii due to its phosphotransfer to PilR. Moreover, these data also suggest that the observed TFP production heterogeneity is not based on the level of phosphorylated PilR given that the phosphomimetic version should be expressed from the P BAD promoter in the whole population. In addition, and similar to the situation for the WT (Fig. 1B), no transformants were detected for the complementing PilR(D56E) variant for samples taken after 240 min of growth (n = 2), suggesting that an absence of phosphorylated PilR is not the reason for the lack of transformation at this later time point during growth. Given the strong transformation phenotype of most of the regulatory mutants, we next tested their pilA transcript and PilA protein levels. For the pilS and pilR mutants, pilA expression levels were significantly reduced (Fig. 4G), and tagged PilA proteins were undetectable (Fig. 4H). On the contrary, the absence of chpA, pilG, or pilH significantly impacted neither the pilA transcript levels nor the PilA protein levels ( Fig. 4G and H). We therefore speculated that the pilus assembly/disassembly might be affected in the chpA and pilG mutants, which could explain their nontransformability (Fig. 4B). Hence, we combined the cysteine knock-in pilA allele with the various gene deletions and imaged potential pili in otherwise WT or pilT-minus background strains. Consistent with the absence of the major pilins in the Western blot analysis (Fig. 4H), neither pili nor the retracted inner membrane-localized pilin subunit were detectable in the pilS and pilR mutants (Fig. 4I). Extended pili were also undetectable in the absence of ChpA and PilG. This is consistent with previous work in P. aeruginosa that showed production of the major pilin but no surface-exposed and therefore shearable pili when similar mutants were investigated. Nonetheless, in our experiments, fluorescent puncta were observed close to the cell surface, which suggests that these strains might initiate pilus elongation but are unable to assemble full-length pili. This phenotype was partially reversed when the retraction enzyme PilT was absent, in which case extended pili/pilus bundles were visible (Fig. 4I). The labeling of the cell bodies of the chpA and pilG mutants by the thiol-reactive maleimide dye supports our idea that the initiation of pilus assembly takes place but that the very short pili seen as puncta in our images are quickly retracted to transport the dye-labeled pilins back into the inner membrane. This phenotype might be caused by an imbalanced PilH protein, which is known to enhance the function of PilT in P. aeruginosa (37). These data are also in line with a previous observation that chpA and pilG mutants of P. aeruginosa remained susceptible to pili-specific phage infections, which was not the case for an extension-defective pilB mutant (37). The pilH mutant, on the other hand, had surface-exposed long pili but contained fewer bacteria with labeled cell bodies than the WT conditions (Fig.  4I). We therefore hypothesized that pilus elongation is favored over retraction through an imbalanced PilG phosphorylation in this mutant.
Lastly, we tested the regulatory mutants for their ability to move on solid surfaces.

FIG 4 Legend (Continued)
promotes autophosphorylation of ChpA, which subsequently phosphorylates PilG and/or PilH, resulting in increased or decreased cAMP levels, respectively. PilG and PilH were also proposed to foster T4P extension or retraction. (B to E) Transformability of TFP regulation mutants without or with complementing constructs 6 inducer, as indicated. Details as in Fig. 3. The WT and WT-TnAraC served as controls. Transformation frequencies are shown as mean value (6SD) from three independent experiments. ,, below detection limit (DL). For statistical analyses, a one-way ANOVA with Sidak's multiple comparisons test was performed on log-transformed data and the different strains' values were compared to the WT (B) or to the most appropriate control strain (boxed name in panels C to E). #, under detection limit in at least one replicate. As shown in Fig. 4J, all mutants of the two systems showed strong motility defects relative to the WT. Interestingly, we observed that the pilH mutant exhibited strongly impaired surface motility despite being piliated and fully transformable. This discrepancy between fully functional transformation but a lack of proficient motility phenocopies the pilU mutant. These results suggest that PilH might play a role in the recruitment or activation of PilU, potentially to enhance high-force pilus retraction.
Conclusion. In this study, we sought to better understand how the poorly studied human pathogen A. baumannii engages in horizontal gene transfer and, precisely, in natural competence for transformation. We showed that A. baumannii has a tight time window of natural transformability under the tested conditions, which correlated with the bacterium's TFP production. Consistent with this finding, we demonstrated that this pilus is essential for the pathogen's natural transformation and that regulatory circuits resembling those of P. aeruginosa play an important part in the production and assembly of the TFP and therefore also natural transformation. Finally, we also supported previous notions that retraction forces can be enhanced by the secondary retraction ATPase PilU, which is dispensable for the DNA uptake process.
Collectively, our data highlight that natural transformation dynamics and pilus regulation in A. baumannii strain A118 vastly differs from those in the model organism A. baylyi, which suggests that common generalizations among these organisms should be taken with caution. Because A. baumannii is a current and urgent threat to human health due to its frequent multidrug resistance, it is important to be able to predict how this pathogen is likely to mutate and the best ways of dealing with infections when it does. Overall, this work sheds important light on mechanisms by which A. baumannii can acquire foreign DNA, and therefore antimicrobial resistance genes, and will assist researchers in better understanding the evolution of A. baumannii as a public health threat.

MATERIALS AND METHODS
Bacterial strains, plasmids, and culture conditions. The bacterial strains and plasmids used in this study are listed in Table 1. Bacterial strains were grown aerobically in lysogeny broth medium (LB; 1% tryptone, 0.5% yeast extract, 1% sodium chloride; Carl Roth, Germany) or on LB agar plates at 37°C. For a selection of Acinetobacter baumannii strains after bi-and triparental mating, CHROMagar Acinetobacter plates were prepared following the manufacturer's instructions (CHROMagar, France).
For transformation assays, phosphate-buffered LB agar plates (pH 6.0) were used. For induction of the P BAD promoter, L-arabinose was added to the medium at a final concentration of 0.2% or 2%, as indicated.
Long-read whole-genome sequencing (PacBio) and de novo assembly. The isolation of genomic DNA (gDNA) and whole-genome sequencing was performed as previously described (63) with minor modifications. Briefly, A. baumannii strain A118 (ATCC BAA-2093) was back diluted 1:100 from an overnight culture and grown aerobically in 25 ml LB medium for 4 h at 37°C (optical density at 600 nm [OD 600 ] 2.5 to 3.0). After 15 ml of this culture was harvested by centrifugation, gDNA was isolated using Qiagen's Genomic DNA buffer set combined with a 500/G Genomic-tip. The alcohol-precipitated DNA was ultimately dissolved in 10 mM Tris-Cl buffer (pH 8.0). Further sample processing, sequencing, and de novo assembly were performed by the Genomic Technology Facility of the University of Lausanne as described (63). The assembled genome was annotated upon submission to the NCBI database using their Prokaryotic Genome Annotation Pipeline (PGAP). The sequencing results are summarized in Table 2.
Genetic engineering of Acinetobacter baumannii. A. baumannii mutants or variants were constructed using a standard allelic exchange approach with the counterselectable suicide vectors pGP704-Sac28 or pGP704Sac-kan (71,72). Derivatives of these plasmids containing deletion constructs of the respective genes or site-directed allele exchanges (of pilA) were transferred to A. baumannii through mating with E. coli S17-1lpir. The deletion constructs were based on the PCR amplification of the flanking regions of the desired genetic regions and, when needed, the aph cassette (Kan r ) as the selection marker using PWO polymerase (Roche) or Q5 polymerase (BioLabs). Site-directed changes in the pilA allele were inserted using modified primers. The amplified fragments were joined by overlap extension PCR or Golden Gate assembly (73) and were cloned inside the suicide plasmids. The correct cloning products were screened for by colony PCR of the E. coli transformants (with GoTaq polymerase; Promega) and ultimately confirmed by Sanger sequencing (Microsynth, Switzerland).
Construction of inducible genes-of-interest was accomplished by placing the respective gene under the control of the arabinose-inducible promoter P BAD inside the miniTn7 transposon TnAraC (23, 63). The resulting transposons were transferred to the respective A. baumannii strains through a standard triparental mating approach (74).
Natural transformation assay. The transformation protocol used in this study was adapted from Harding et al. (15). Briefly, the bacteria were grown overnight in LB medium, back diluted 1:100 into 2 ml of LB medium with or without 0.2% or 2% L-arabinose, and further cultured aerobically until they reached an OD 600 of approximately 0.65 (or, for the time course experiment, the OD 600 value indicated in Fig. 1). Next, 20 ml of the bacterial culture was mixed with 1 mg of isolated genomic DNA of strain A118Dhcp1::kan, and 20 ml of this mixture was spotted onto pH-buffered LB agar plates (pH 6; without or with 0.2% or 2% arabinose). The plates were incubated for 2 h at 37°C. Following this incubation period, the cells were scraped off from the plate with a sterile loop and resuspended in 200 ml of LB medium. Serial dilutions were spotted in duplicate on kanamycin-containing LB agar plates to select the transformants and on plain LB agar plates to assess the total number of CFU. Transformation frequency was calculated as the CFU number of transformants divided by the total number of bacteria. Averages of at least two biologically independent experiments are provided. Transformation frequencies were logtransformed (75) and statistically significant differences were determined as described in the figure legends. When no transformants were recovered, the transformation frequencies were set to the detection limit to allow the calculation of the average value of all biologically independent replicates and for statistical analyses using the program Prism (GraphPad software; San Diego, USA).
TFP-dependent surface motility assay. The bacterial strains were streaked on LB agar plates and incubated at 37°C overnight to obtain single colonies. Bacterial material was retrieved from a single colony with a toothpick and gently applied onto a freshly prepared 1% agar-only plate. The plates were sealed with parafilm and incubated at 37°C for 4 days. Following the incubation, pictures of the plates were taken with an M80 Stereo Zoom microscope equipped with an MC 170 HD camera (both from Leica). The motility area was determined using ImageJ software (imagej.nih.gov/ij/). Results of four biological replicates consisting of three technical replicates for each strain are provided. Significant differences were determined using Prism, as indicated in the legends to Fig. 1 and 4.
Gene expression determination using quantitative reverse transcription-PCR (qRT-PCR). Backdiluted cultures were grown at 37°C in several parallel 2 ml LB medium-containing tubes for the indicated amount of time (for the time course experiment) or up to an OD 600 of 0.65. The bacterial cells (1.5 ml to 8 ml, depending on the growth phase) were harvested by centrifugation for 3 min at 4°C. The cell pellets were washed once with ice-cold PBS, resuspended in TRI reagent (Sigma-Aldrich), vortexed for a few seconds, and then flash-frozen on dry ice before their storage at -80°C. The RNA extraction, DNase treatment, reverse transcription, and quantitative PCR (qPCR) were performed as previously described (76). Relative gene expression values were calculated based on normalization against the transcript levels of the housekeeping gene gyrA. Averages of at least three independent biological experiments are shown. Data were log-transformed (75) and statistical differences were assessed using the Prism software. Details on the statistical analyses are provided in the figure legends to Fig. 1 and 4.
SDS-PAGE and Western blotting. To verify the production of PilA-FLAG proteins, cell lysates were prepared as described previously with minor modifications (77). Briefly, after overnight growth, the bacterial culture was back diluted 1:100 in 2 ml of LB and grown at 37°C for the indicated amount of time (for time course experiments) or up to an OD 600 of ;0.65. At the time of harvesting, the bacteria were centrifuged for 3 min, and the pelleted cells were resuspended in 2Â Laemmli buffer, whereby the volume was adjusted according to the total number of bacteria (100 ml buffer per OD 600 unit). The resuspended samples were incubated at 95°C for 15 min. Proteins were separated by SDS-PAGE using 15% resolving gels and blotted onto\polyvinylidene difluoride (PVDF) membranes as previously described (76). Primary monoclonal antibodies against the FLAG tag (ANTI-FLAG M2; Sigma-Aldrich) were used at 1:2,000 dilution, and goat anti-mouse antibody-horseradish peroxidase (HRP) served as the secondary antibody (diluted 1:5,000; Sigma-Aldrich). Sigma70 was detected as a loading control using Direct-Blot anti-E. coli Sigma70-HRP-conjugated antibodies at a dilution of 1:10,000 (BioLegend, USA distributed via Brunschwig, Switzerland). Lumi-Light PLUS Western blotting substrate (Roche) served as the HRP substrate. Luminescent signals were detected using a ChemiDoc XRS1 station (Bio-Rad).
Microscopy. Bacteria were grown until an OD 600 of ;0.65 and then spotted onto thin agarose pads that were mounted onto glass slides (1.2% agarose dissolved in 0.5Â PBS). The cells were illuminated with an HXP120 lamp and imaged through a Plan-Apochromat 100Â/numerical aperture 1.4 Ph3 oil objective by an AxioCam MRm camera attached to an Axio Imager M2 epi-fluorescence microscope (Zeiss). The Zeiss software ZEN 2.6 and ImageJ (imagej.nih.gov/ij/) were used for image acquisition and analysis, respectively.
Type IV pilus labeling. Pilus labeling was performed as previously described with minor modifications (23,50). Briefly, control strains or genetically modified bacteria with modified pilA alleles (encoding site-specific amino acid changes to knock-in a new cysteine residue) were pregrown overnight at 37°C, back diluted 1:100 in 2 ml LB (supplemented without or with L-arabinose, as indicated), and grown at 37°C until they reached an OD 600 of approximately 0.65. Alexa Fluor 488 C5 maleimide (AF-488-Mal; Thermo Fisher Scientific) was added to 100 ml of the bacterial culture at a final concentration of 25 mg/ ml, gently mixed, and incubated for 15 min at room temperature in the dark. Cells were subsequently harvested by centrifugation (5,000 Â g for 1 min.), washed once in 1Â PBS, and resuspended in 30 ml of PBS before being imaged as described above.
Data availability. The assembled genome sequence and all raw data were deposited into NCBI under GenBank accession number CP059039, BioSample number SAMN15507634, and the Sequence Read Archive (SRA) accession number SRX8711359.