Copy Number of an Integron-Encoded Antibiotic Resistance Locus Regulates a Virulence and Opacity Switch in Acinetobacter baumannii AB5075

Acinetobacter baumannii remains a leading cause of hospital-acquired infections. Widespread multidrug resistance in this species has prompted the WHO to name carbapenem-resistant A. baumannii as its top priority for research and development of new antibiotics. Many strains of A. baumannii undergo a high-frequency virulence switch, which is an attractive target for new therapeutics targeting this pathogen. This study reports a novel mechanism controlling the frequency of switching in strain AB5075. The rate of switching from the virulent opaque (VIR-O) to the avirulent translucent (AV-T) variant is positively influenced by the copy number of an antibiotic resistance locus encoded on a plasmid-borne composite integron. Our data suggest that this locus encodes a small RNA that regulates opacity switching. Low-switching opaque variants, which harbor a single copy of this locus, also exhibit decreased virulence. This study increases our understanding of this critical phenotypic switch, while also identifying potential targets for virulence-based A. baumannii treatments.

A cinetobacter baumannii is a Gram-negative nosocomial pathogen. This bacterium causes a range of opportunistic infections, including pneumonia, urinary tract infections, meningitis, bloodstream infections, and wound infections (1). A. baumannii infections are frequently difficult to treat due to widespread antimicrobial resistance in this species, which has prompted the WHO to name carbapenem-resistant A. baumannii as its top priority for research and development of new antimicrobials (2). This highlights the need for the identification of new targets for virulence-focused therapeutics for A. baumannii. Although a number of virulence factors have been identified in A. baumannii (reviewed in reference 3), a complete understanding of the regulation of the pathogenesis of this organism is lacking. Many strains of A. baumannii, including isolate AB5075, are capable of undergoing a high-frequency switch between virulent opaque (VIR-O) variants and avirulent translucent (AV-T) variants (4)(5)(6)(7)(8). Of these, only the VIR-O variant is capable of causing disease in mice and Galleria mellonella waxworms, and this variant is selected for in vivo (4,6). This virulence switch represents an attractive target for pathogenesis-focused therapeutics in A. baumannii, but currently, the mechanisms underlying switching are incompletely understood.
Opacity variation in A. baumannii is influenced by a number of gene products. The OmpR/EnvZ two-component system negatively regulates VIR-O to AV-T switching (7), whereas this switch is positively affected by the ArpAB efflux pump (6). The TetR-type transcriptional regulator ABUW_1645 appears to be involved in maintenance of the AV-T state, as this gene is upregulated in the AV-T variant, stimulates conversion of VIR-O cells to the AV-T form when overexpressed, and slightly increases the rate of AV-T-to-VIR-O switching when disrupted (4). ABUW_1645 does not transcriptionally regulate ompR-envZ or arpAB and the hyperswitching phenotype of an ompR mutant does not require ABUW_1645 (4). The VIR-O-to-AV-T switch also requires relA, encoding (p)ppGpp synthetase (9).
p1AB5075 is the largest plasmid carried by A. baumannii AB5075. It is nearly 84 kb in size and harbors resistance island 2 (RI-2) (10). RI-2 consists of two miniature inverted-repeat transposable elements (MITEs) flanking a composite integron (10). This composite integron is composed of two fused class 1 integrons and therefore encodes two highly homologous copies of the integrase gene intI (one of which is actually a pseudogene) (10). The two intI alleles flank the resistance genes aadB, cmlA, aadA1, and strAB (10). We previously showed that the genes between the two copies of intI undergo spontaneous, RecA-dependent high-copy amplification, which results in increased resistance to tobramycin and gentamicin due to increased expression of the aminoglycoside adenylyltransferase gene aadB (11). These resistance genes are presumably expressed from the integron cassette promoter P c , which is upstream of aadB (12)(13)(14)(15). Class 1 integron cassette promoter regions also harbor binding sites for the DNA binding proteins FIS, LexA, IHF, and H-NS, a small open reading frame encoding ORF-11, a nonfunctional peptide whose translation enhances translation of cassette genes, and the integron attI site (12,16,17). These promoter characteristics appear to be conserved in RI-2.
In bacteria, small RNAs (sRNAs) are generally 50 to 500 nucleotides (nt) in length and act as posttranscriptional regulators by base pairing to mRNA targets (18). sRNAs typically base pair with their targets over fairly short sequences, usually between 7 and 12 nt (19). sRNAs can act as either positive or negative regulators of mRNA activity (18,20,21). These transcripts can be derived from intergenic regions or 5= or 3= untranslated regions (UTRs) of coding genes and can be independently transcribed or generated by RNase cleavage of larger transcripts (18,(22)(23)(24)(25).
In this paper, we report that A. baumannii produces a third opacity variant along with AV-T and VIR-O. Designated the low-switching opaque (LSO) variant, this third subpopulation exhibits dramatically reduced levels of switching to AV-T relative to that for the VIR-O. Switching frequencies in the LSO and VIR-O variants are controlled by the copy number of a region between the two copies of intI on RI-2 within p1AB5075. We demonstrate that the element controlling switching is encoded at the 5= end of the aadB gene within RI-2 and that this element is likely an sRNA. Finally, we also show that if the conversion between variants was due to the generation of a random point mutation, suggesting that the LSO colonies do not arise by this mechanism.
The rate of switching in different opaque subpopulations is correlated with copy number of an antibiotic resistance locus. To determine the genetic differences between the VIR-O and LSO subpopulations, representatives from each were subjected to whole-genome sequencing (WGS) using PacBio technology. WGS revealed the VIR-O and LSO cells differed by an approximately 6-kb duplication present within a composite integron (RI-2) on the large plasmid p1AB5075 (10). The duplicated region consists of five genes flanked by two highly homologous copies of the integrase gene intI, which share 99% nucleic acid identity across 1,062 bp ( Fig. 2A). In the VIR-O variant, a tandem duplication of the entire locus has occurred, presumably through recombination between the two copies of intI ( Fig. 2A). The duplication in this region was confirmed through Southern blotting and quantitative PCR (qPCR) (Fig. 2B) (set A). For the Southern blot, genomic DNA (gDNA) was digested with ScaI, an enzyme that cuts outside the duplication, and the blot was treated with a probe specific for the duplicated region. Results were further confirmed by using qPCR to measure the copy number of aadB, the first gene within the duplication. Results for both of these assays confirmed that the VIR-O variant contains a duplication of this region. Aside from intI, all of the genes in the duplicated region mediate resistance to aminoglycosides or chloramphenicol. As expected, this duplication resulted in modest increases in resistance to some of these antimicrobials (see Table S1).
To determine whether copy number of this region is actually correlated with opacity switching frequency, we measured copy number of this region in a second indepen- This duplication comprises part of a composite integron encoded on the plasmid p1AB5075 and consists of five antibiotic resistance genes flanked by two highly homologous copies of intI. (B) Southern blotting and qPCR analysis of the duplicated region were performed with two sets of LSO and VIR-O isolates (set A samples were previously used for PacBio sequencing). For the Southern blot, gDNA was digested with ScaI and probed with digoxigenin-labeled DNA specific for aadB. The table indicates copy number of aadB in each sample as calculated by qPCR. Copy number was normalized to aacA4, a gene harbored on p1AB5075 outside the duplicated region. Data represent averages and standard deviations (SDs) from three independent samples. (C) Switching frequencies to AV-T were measured for set B VIR-O and LSO colonies after 16 h of growth (differences not significant by one-way analysis of variance [ANOVA] with Tukey's posttest). (D) The AB5075 isolate hetR-O2 was previously reported to harbor 17 to 20 copies of the duplicated region (11). Switching to AV-T of hetR-O2 and VIR-O was measured for colonies after 16 h of growth. *, P Յ 0.05, unpaired two-tailed t test. (E) An AV-T isolate harboring two copies of the duplicated region was used to derive two isolates harboring a single copy of the region, designated AV-T-56 and AV-T-80. Switching from AV-T to VIR-O was measured for colonies after 24 h of growth (differences not significant by one-way ANOVA with Dunnett's posttest). In panels C to E, data represent the averages and SEMs from five or six colonies collected in two independent experiments. dent set of interrelated LSO and VIR-O variants. Starting with a VIR-O variant that was independent from that used for WGS, we isolated and stocked a second independent LSO variant. This LSO variant was then used to isolate a new independent VIR-O (VIR-O2). These three related samples were designated "set B" samples to distinguish them from the "set A" samples used for WGS. Southern blot analysis with the set B LSO and starting VIR-O variants gave band sizes similar to those seen for the set A samples (Fig. 2B). Interestingly, set B VIR-O2 yielded a larger band than the other two VIR-O samples, suggesting further amplification of this region. qPCR analysis confirmed the Southern blot results for the set B VIR-O and LSO samples and indicated that set B VIR-O2 harbors four copies of the duplicated region (Fig. 2B). To determine whether VIR-O2 exhibited higher levels of switching due to this further increase in copy number, switching to AV-T of the set B samples was quantified. Switching was measured after 16 h of growth, a time point before VIR-O colonies typically exhibit high levels of switching. VIR-O2 colonies exhibited an 86-fold increase in switching relative to that of the original VIR-O, although the results were variable between samples and differences were not statistically significant (Fig. 2C). However, this trend toward increased switching with increased copy number suggests that amplification of this region does indeed affect switching rate from VIR-O to AV-T.
Our laboratories previously reported that high-level RecA-dependent amplifications of the same duplicated region are responsible for tobramycin heteroresistance in AB5075 (11). To determine whether colonies exhibiting high-level amplification of this region also exhibit hyperswitching from VIR-O to AV-T, we quantified switching in the previously characterized tobramycin-resistant isolate hetR-O2, which harbors 17 to 20 copies of the duplicated region (11). After 16 h of growth, hetR-O2 exhibited a significant 15-fold increase in switching relative to that of the control VIR-O (Fig. 2D). This result further suggests that the VIR-O to AV-T switching rate is controlled by the copy number of this region.
We further evaluated whether copy number of the duplicated region also correlated with an increased AV-T-to-VIR-O switching frequency. To identify AV-T isolates with one copy of the duplicated region, an AV-T isolate harboring two copies was screened for colonies with decreased tobramycin resistance. The resulting colonies (designated AV-T-56 and AV-T-80) were confirmed to have a single copy of aadB by qPCR (data not shown). Quantification of switching in these isolates and the original AV-T demonstrated that switching frequency from AV-T to VIR-O was not affected by copy number of the duplicated region (Fig. 2E). This result provides further evidence that AB5075 employs distinct mechanisms to switch from VIR-O to AV-T and vice versa, as suggested by previous research (6).
To more directly examine the influence of the duplicated region on VIR-O to AV-T switching, the LSO background was used to generate a strain that was cured of the p1AB5075 plasmid, designated LSO Δp1AB5075. LSO Δp1AB5075 exhibited similar levels of switching to that of its parent variant (Fig. 3A), indicating that the plasmidencoded region is not required for basal levels of VIR-O-to-AV-T switching. To test whether p1AB5075 plasmids harboring two copies of the duplicated region are sufficient to stimulate increased levels of switching, LSO Δp1AB5075 was transformed with total DNA isolated from either the LSO or VIR-O variant to move p1AB5075 back into these strains. Transformants that had taken up p1AB5075 were selected for using tobramycin. Transformants generated using LSO DNA maintained low levels of switching (Fig. 3A). Transformation with VIR-O DNA yielded a mix of low-switching and normally switching colonies (Fig. 3A). When qPCR was used to measure aadB copy number in these transformants, it was determined that a low-switching transformant (transformant A) had taken up a plasmid containing one copy of the duplicated region, whereas a switching transformant (transformant B) had taken up a plasmid containing two copies (Fig. 3B). This experiment demonstrates that uptake of p1AB5075 harboring two copies of the duplicated region is sufficient to increase VIR-O-to-AV-T switching in AB5075.
Overexpression of the aadB gene leads to increased opacity switching, but AadB protein is dispensable for this phenotype. To identify the gene(s) responsible for stimulating switching within the duplicated region, portions of this region were overexpressed in the LSO variant, and switching to AV-T was examined. We first used plasmid pWHaadB, consisting of the aminoglycoside resistance gene aadB and upstream DNA cloned into plasmid pWH1266; this plasmid exhibits an average copy number of 8.63 Ϯ 5.73. This plasmid stimulated switching Ͼ800-fold in the LSO background (Fig. 4). A plasmid construct was also generated containing all five resistance genes in the duplicated region; this stimulated similar levels of switching in an LSO background as that of a construct harboring aadB alone, suggesting that all of the stimulatory activity in the duplicated region was derived from aadB (see Fig. S2).
AadB mediates aminoglycoside resistance by adenylylating the antibiotic; this enzyme is not known to act on cellular substrates. Therefore, it was unexpected that overproduction of this protein would have an effect on VIR-O-to-AV-T switching. To

FIG 4
Overexpression of the aadB region is sufficient to stimulate switching. The aadB gene was cloned along with its native promoter into the vector pWH1266 (60) and transformed into the LSO variant. Switching to AV-T was measured for colonies after 24 h of growth. Data shown represent averages and SEMs from at least five colonies collected in two independent experiments. ****, P Յ 0.0001 by unpaired two-tailed t test.
determine if the AadB protein was mediating the effect on switching, a nonsense mutation was introduced at the fifth amino acid of aadB (pWHaadBstop). This allele was still capable of stimulating switching Ͼ800-fold when overexpressed in the LSO background (see Fig. S3), demonstrating that AadB is not responsible for regulating the rate of VIR-O-to-AV-T switching.
The element responsible for controlling switching rate is encoded at the 5= end and upstream of aadB. To determine where the element affecting switching was localized in the pWHaadB plasmid insert, we mutagenized this plasmid with the EZ-Tn5 ϽKan-2Ͼ transposon. Mutant plasmids were transformed into the LSO variant, and two screens were performed. Insertions that blocked the ability of the plasmid to cause switching were identified by screening for transformants that no longer exhibited the highly sectored colony phenotype stimulated by pWHaadB. Insertions in aadB that did not abolish switching were also isolated by screening for plasmids that had lost the ability to confer tobramycin resistance in Escherichia coli followed by a screen for plasmids that still stimulated switching in the LSO variant. When both groups of insertions were mapped onto the aadB region, all of the insertions blocking switching localized to the 5= end and upstream of aadB, while most of the insertions within the aadB open reading frame (ORF) did not affect switching (Fig. 5A). Insertions blocking switching were relatively evenly distributed between the 5= portion of aadB and the P c promoter, suggesting that the element affecting switching is encoded in this region.
To confirm that the element responsible for stimulating switching is encoded toward the 5= end and upstream of aadB, truncated fragments of the aadB region were overexpressed in the LSO variant. As depicted in Fig. 5A, truncated insertions extended by differing lengths into the aadB gene. Plasmids pWHaadB-T (harboring the longest truncated insert), pWHaadB-TA, and pWHaadB-TB all appeared to stimulate switching in The insert in plasmid pWHaadB contains part of the 5= end of the intI ORF (gray rectangle), as well as the promoters for intI (P intI , gray bent arrow) and aadB (P c , green bent arrow). To determine which part of the insert was responsible for stimulating switching, pWHaadB was mutagenized with EZ-Tn5 ϽKan-2Ͼ. Insertions that blocked the stimulation of switching are indicated by closed triangles, whereas insertions that did not affect sectoring are indicated by open triangles. Plasmids were then generated containing truncated inserts of different lengths to further test for the stimulation of switching in the LSO variant. The remaining length of each plasmid insert is indicated in parentheses. Plasmids pWHaadB-T (B), pWHaadB-TA (C), pWHaadB-TB (D), and pWHaadB-UP (E) were transformed into the LSO variant, and switching to AV-T was measured for 24-h-old colonies. Data shown represent averages, SEMs, and individual switching frequencies from six colonies collected from at least two independent experiments. ****, P Յ 0.0001, one-way ANOVA with Tukey's posttest.
Regulation of Opacity Variation in A. baumannii ® the LSO variant, although not to the same degree as the full-length pWHaadB plasmid ( Fig. 5B to D). For unknown reasons, colonies overexpressing these plasmids exhibited high variability in their levels of switching, resulting in differences in switching between these transformants not being significantly higher than those of the empty vector controls. However, these plasmids were still classified as being biologically active, as transformants clearly exhibited more sectoring/switching than empty vector controls on average. The shortest truncated insert, pWHaadB-UP, did not stimulate sectoring and appeared similar to the empty vector control in quantitative switching assays (Fig. 5E). The results of these truncation studies confirm those obtained through transposon mutagenesis of pWHaadB. The fragment encoded by pWHaadB-UP terminates upstream of some of the transposon insertions that blocked switching by pWHaadB, whereas the shortest truncated plasmid to still stimulate switching, pWHaadB-TB, includes the insertion sites for all transposons that blocked switching. Taken together, these results indicate that the element responsible for stimulating switching is encoded at the 5= end and upstream of aadB.
Possible role for an sRNA in VIR-O-to-AV-T switching. We hypothesized that the element affecting switching was likely a regulatory sRNA or a small peptide encoded upstream of aadB, but our data could also be explained by the presence of a protein binding site titrating a regulator required for controlling switching. To discriminate between these possibilities, site-directed mutagenesis was used to introduce a twobase pair change in the Ϫ35 region of P c (TTGACA¡TTGAGT) in plasmid pWHaadB. This mutation decreased transcription 25-fold from that of the P c promoter (see Fig. S4A) and the mutated plasmid was no longer able to stimulate switching when overexpressed in the LSO background (Fig. S4B), indicating that transcription from P c was required for the upstream element to increase switching. This result indicates that the upstream element is likely an sRNA or a small peptide. While there are a number of small putative ORFs encoded within the upstream region of aadB, transposon insertions affecting switching stimulated by pWHaadB do not localize to any of these potential peptides, suggesting that none of them are responsible for controlling the switching rate (see Fig. S5).
Based on the above-described results, the possibility that an sRNA controlled the rate of switching was investigated. To determine whether any small transcripts were produced from the P c promoter, Northern blotting was performed on RNA separated on a 10% acrylamide urea gel. Initial Northern blots were conducted with probe oSA134 (see Table S2; Fig. S6A), a digoxigenin-labeled single-stranded DNA oligonucleotide probe that only binds transcripts derived from the P c promoter. This probe hybridizes starting 40 nt downstream of the reported transcriptional start site for P c (26). Northern blotting was performed using LSO Δp1AB5075 overexpressing pWHaadB, pWHaadB-TA, pWHaadB-TB, and pWHaadB-UP along with an empty vector control (Fig. 6). Strains containing the three longer inserts all exhibited the same five major RNA products, Small RNA species are produced from the 5= end and upstream of aadB. Northern blotting was performed using oSA134 to examine RNAs produced from the P c promoter on overexpression plasmids transformed into LSO Δp1AB5075. An ϳ300-nt RNA appears to be missing or reduced in sample pWHaadB-UP (black arrowhead).
which were roughly 300, 150, 110, 80, and 70 nt in size. All of these transcripts are too small for the full aadB gene (the aadB ORF is 534 bp). While a larger transcript encoding the aadB gene is not evident on this blot, transcripts in excess of 300 nt are present on other blots using another probe for this region (Fig. S6). Interestingly, in the strain overexpressing pWHaadB-UP, the ϳ300-nt RNA was largely absent. Since this plasmid was not capable of stimulating switching in the LSO background, this result suggests that the ϳ300-nt RNA may be involved in stimulating switching.
To identify the endpoint of the ϳ300-nt sRNA, additional Northern blotting was performed using probes that bind toward the 3= end of this transcript. Probe oSA190 was designed to bind 209 nt downstream of the end of oSA134 (Table S2; Fig. S6A). Probe oSA191 binds immediately 3= of oSA190, and oSA197 binds immediately 3= of oSA191 (Table S2; Fig. S6A). Northern blotting of LSO Δp1AB5075 pWHaadB with these probes revealed that both oSA190 and oSA191 bind strongly to the ϳ300-nt RNA (as evidenced by the production of dark bands on the Northern blot), whereas oSA197 binds this band only weakly (Fig. S6B). Dot blotting of gDNA with oSA197 confirmed that this probe is not defective for binding in general (Fig. S6C), suggesting that the ϳ300-nt RNA terminates within the region bound by oSA197. This would make the ϳ300-nt RNA between 339 and 368 nt in length, assuming that the transcript begins at the reported start site for P c (26).
To determine whether the ϳ300-nt RNA was produced via RNase processing, we overexpressed aadB in single mutants defective for all nonessential ribonucleases obtained from the AB5075 transposon library (10). VIR-O mutants were qualitatively evaluated for their degree of switching to AV-T when aadB was overexpressed. Mutants that did not exhibit clearly increased switching were examined by Northern blotting. None of the evaluated mutants showed consistent differences in production of the ϳ300-nt band, while minor differences in the intensities of some other bands were observed for some ribonucleases (see Fig. S7). The results of these experiments are summarized in Table 1.
VIR-O and LSO subpopulations exhibit differences in virulence and global gene expression. We hypothesized that because the VIR-O variant is more virulent than the  (Fig. 7). Surprisingly, the LSO variant exhibited at least a 3-log decrease in the geometric mean of CFU in all three organs, demonstrating that this variant is less virulent than VIR-O.
To obtain a global view of gene expression differences between VIR-O and LSO variants, and to determine why the LSO variant might be less virulent, RNA sequencing (RNA-seq) was performed on each subpopulation in triplicates. Results of the RNA-seq are detailed in Data Set S1. Using a fold change cutoff of 1.5 (log 2 -fold 0.7) and a P value cutoff of 0.05, we identified 55 upregulated genes and 87 downregulated genes in the LSO variant. As expected, the genes within the duplicated region were all downregulated in the LSO variant. Interestingly, several genes involved in pathways that were previously implicated in A. baumannii virulence were found to be downregulated in the LSO variant, including the phospholipase C gene plc1 (27), genes involved in cysteine metabolism/sulfur assimilation (cysW, cysT, and cysN) (28), and genes involved in the phenylacetic acid catabolic pathway (paaK, paaJ, and paaI2) (29,30). The LSO variant also showed decreased expression of genes from the csu operon, which is involved in the production of pili important for surface adherence and biofilm formation (30)(31)(32). These differences in gene expression may underlie the decreased virulence observed in the LSO strain.

DISCUSSION
This study establishes that A. baumannii AB5075 produces two distinct subpopulations of opaque variants, designated VIR-O and LSO. The two subpopulations differ not only in their rates of switching to AV-T but also in their gene expression and virulence. The switching frequencies of these opaque variants were positively correlated with copy number of a locus in RI-2 on plasmid p1AB5075, with VIR-O variants exhibiting two or more copies of this locus. Copy number of this locus appears to vary through homologous recombination between two alleles of intI, leading to the interconversion of the LSO and VIR-O subpopulations. The VIR-O variant gives rise to the LSO variant at a higher frequency than the reverse; this is likely due to the inherent instability of tandem duplications, which are easily lost through recombination between their large regions of homology (33). High-level amplification of this locus can also occur, leading to hyperswitching to AV-T and increased resistance to tobramycin (11). The element controlling switching is encoded toward the 5= end and upstream of the aadB gene in RI-2, and our evidence suggests that this element is likely an sRNA. Switching stimulated by overexpression of aadB required transcription from the integron cassette  promoter P c but did not appear to require any of the small ORFs predicted upstream of aadB. Northern blotting revealed multiple small transcripts produced downstream of P c , providing further evidence that one or more sRNAs encoded in this region are responsible for controlling switching.
While our results suggest that an sRNA produced from the 5= end of the aadB transcript is responsible for stimulating switching, there are many open questions about this model that remain to be answered. Our Northern blotting results have specifically implicated a transcript of ϳ300 nt in switching, as an sRNA of this length was found to be produced from overexpression plasmids that stimulated switching but was reduced or absent from a plasmid lacking stimulatory activity (Fig. 6). Future studies should determine the precise sequence of this sRNA as well as the mechanism by which it is produced. Disruption of nonessential ribonucleases did not reveal any differences in production of the ϳ300-nt sRNA ( Table 1), suggesting that this sRNA is either generated by transcriptional termination or via processing by an essential RNase not examined here. Finally, the target(s) of this sRNA, as well as any chaperones involved in target binding, remains to be identified. Our data indicate that the target(s) involved in switching is not encoded on p1AB5075, as LSO Δp1AB5075 transformed with pWHaadB still exhibits increased switching (data not shown). Genes previously implicated in switching such as ompR, envZ, arpAB, ABUW_1645, and relA were not differentially expressed between LSO and VIR-O variants based on the RNA-seq analysis. However, since sRNAs act at the posttranscriptional level, regulation of these genes by the sRNA may be missed in this analysis. Identification of the chromosomal target(s) of this sRNA will be extremely valuable for understanding the regulation of switching in A. baumannii.
The putative sRNA described here exhibits a number of unexpected genetic and regulatory features. First, it is interesting that the putative sRNA is encoded on a transmissible plasmid. Although this is unusual, there is precedence for horizontally acquired sRNAs being produced in other bacterial species. For example, Salmonella enterica serovar Typhimurium has been shown to produce sRNAs from horizontally acquired genetic elements (34,35). One of these, TnpA, is produced from the 5= UTR of a transposase gene and functions to regulate expression of virulence genes by directly binding to mRNA of the pathogenicity regulator invF (35). A plasmid-derived sRNA has also been shown to regulate genetic competence in Legionella pneumophila (36). sRNAs associated with mobile elements have also been identified in Staphylococcus aureus (37), Xanthomonas campestris (38), and Coxiella burnetii (39). Taken together, these studies and ours suggest that horizontally transferred elements may be an underappreciated source of sRNAs.
It is also unusual that two or more copies of the sRNA-encoding locus are required to stimulate significant levels of opacity switching. Figure 3 demonstrates that the sRNA is not required for baseline levels of switching, as strain LSO Δp1AB5075, which does not harbor the sRNA, exhibits the same level of opacity switching as the wild-type LSO. Changes in switching to AV-T are only observed when copy number of the sRNA-encoding locus is increased, either through spontaneous amplification (Fig. 2 and 3) or overexpression on a multicopy plasmid (Fig. 4). This suggests that the sRNA must be expressed past a certain threshold before switching can be activated. This could potentially occur if its mRNA target(s) involved in switching is produced in great excess to the sRNA; perhaps when the sRNA is encoded in single copy, it is produced to insufficient levels to affect switching.
It is also interesting that the truncated plasmids pWHaadB-T, pWHaadB-TA, and pWHaadB-TB all caused much more variable switching phenotypes than the full-length plasmid pWHaadB. Although LSO colonies overexpressing the truncated plasmids exhibited increased average switching compared to that of an empty vector control, the differences were not found to be statistically significant due to the wide variability. The reason for this variability is unclear but could be due to variable levels of sRNA production from these plasmids. Even if the entire sequence for the active sRNA is harbored by each of these plasmids, it is possible that truncation of the inserts could affect the efficiency of posttranscriptional processing or transcriptional termination, which could potentially result in variable levels of transcripts between cells. If the sRNA only causes an observable phenotype when expressed past a certain threshold, then slight variations in transcript levels between colonies could result in dramatic differences in switching phenotypes. However, more research is needed to confirm this hypothesis.
This work is not the first report of a regulatory RNA element encoded in the promoter region of class 1 integrons. The promoter regions of aminoglycoside resistance integrons have also been reported to encode an aminoglycoside-responsive riboswitch, although the presence of this element is controversial (40)(41)(42). In 2013, Jia et al. characterized a putative 76-nt riboswitch that exhibits 97% sequence identity across 61 bp to the 5= UTR of aadB (40). Although the putative sRNA described here extends over a larger sequence than this riboswitch, it is interesting that the RI-2 promoter region may encode both a riboswitch and an sRNA. A riboswitch was recently reported to affect stability of an upstream sRNA harbored by Vibrio cholerae (43). In Listeria monocytogenes, a riboswitch has also been shown to act as an sRNA in trans (23), and a riboswitch-containing sRNA has also been identified in Enterococcus faecalis (44). More work will be needed to determine the interplay in regulation between the putative integron riboswitch and the sRNA described here.
It remains to be seen whether the mechanism controlling switching frequency described here is widespread in A. baumannii. A BLAST search with the first 300 nt of the aadB transcript yields hits to Ͼ100 A. baumannii genomes, with coverage ranging from 73% to 100%. Integrons encoding aminoglycoside modifying enzymes, including aadB, are also widely reported in the A. baumannii literature (45)(46)(47)(48)(49)(50). As the ϳ300-nt sRNA appears to be encoded mostly within the integron promoter region, it is possible that other integrons, even those not harboring aadB, still contain this transcript. Together, these pieces of evidence suggest that the active sRNA is likely to be widespread in this species. However, the data presented here demonstrate that in AB5075, the presence of this sRNA in single copy is not enough to stimulate switching, as duplications of the sRNA-encoding region are also required. Our results suggest that in AB5075, amplification likely occurs through homologous recombination between flanking intI alleles within the RI-2 integron (11). As of this writing, RI-2 has only been identified in two other A. baumannii strains besides AB5075 (51,52), suggesting that this mechanism of copy number variation may be restricted to certain A. baumannii strains. However, in other strains, amplifications of this locus could occur by other mechanisms, including transposition or increases in plasmid copy number. Indeed, amplification of a different aminoglycoside resistance-encoding integron has been reported in a clinical isolate of A. baumannii, even though this strain lacks the flanking intI alleles (53). Any A. baumannii strains that do not harbor the active sRNA (or are incapable of varying its copy number or expression) would be expected to behave as LSO variants. However, these strains could still have the ability to acquire the active sRNA, since RI-2 is carried on a transmissible plasmid (10,51,54). As VIR-O variants harboring duplications within RI-2 exhibit increased virulence, dissemination of this plasmid could be of public health concern.
It was unexpected that the LSO variant exhibited decreased levels of virulence compared to that of the VIR-O variant. Previous studies comparing virulence of VIR-O and AV-T isolates suggested that differences in virulence between the two variants are due to levels of capsule, which is a virulence factor that contributes to A. baumannii resistance to host lysozyme (4,55). Increased capsule levels partly underlie the appearance of opaque A. baumannii colonies, as AB5075 VIR-O mutants lacking capsule appear more translucent than the wild type (55). However, the in vivo results observed for the LSO variant demonstrate that an opaque colony appearance does not necessarily translate to virulence. While one could hypothesize that the decreased virulence of the LSO variant indicates a role for the AV-T form in infection, preliminary infection experiments with mixtures of LSO and AV-T variants indicate that this is not the case (data not shown). This suggests that other intrinsic factors in addition to capsule contribute to the increased virulence of the VIR-O variant compared to LSO. We hypothesize that the LSO variant is less virulent due to its decreased expression of multiple virulence factors, including ones involved in metabolism and pili production (see Data Set S1 in the supplemental material).
Our data suggest that AB5075 should maintain a low number of duplications of the sRNA-encoding locus to ensure maximal virulence. While Fig. 7 demonstrates that a single duplication of this locus results in a dramatic increase in virulence, Fig. 2D demonstrates that high-level amplification of this locus results in hyperswitching to the AV-T state, which should result in decreased virulence in vivo. Variant hetR-O2 characterized in Fig. 2D was first isolated using selection for increased tobramycin resistance (11), demonstrating that exposure to high levels of tobramycin can select for high-copy amplifications of RI-2 and hyperswitching to AV-T. This suggests that tobramycin treatment may lead to decreased virulence of AB5075 by increasing switching to AV-T; however, this hypothesis remains to be experimentally verified. A recent study demonstrated that pretreatment with the aminoglycoside kanamycin results in decreased virulence in AB5075, although the mechanism behind this appears to involve aminoglycoside-induced changes and not changes selected for by aminoglycosides (56).
Finally, the discovery of LSO variants has important implications for future studies using A. baumannii AB5075. As LSO, VIR-O, and AV-T variants all exhibit distinct phenotypes in terms of colony morphology, virulence, and global gene expression (4,5), it is critical that future studies of AB5075 take opacity variants into account when comparing phenotypes between mutant and control strains. This study demonstrates that opaque isolates in particular should be assessed for their degree of sectoring to AV-T prior to further experimentation to avoid spurious results in virulence and other studies.

MATERIALS AND METHODS
Bacterial strains and growth conditions. Stocks of bacteria were maintained at Ϫ80°C in 15% glycerol. Pure stocks of A. baumannii VIR-O isolates were generated as described previously (57). Liquid cultures were prepared in sterile LB broth, supplemented with tetracycline (Sigma-Aldrich, St. Louis, MO), hygromycin (Invitrogen, Carlsbad, CA), tobramycin (Sigma-Aldrich), or kanamycin (Sigma-Aldrich) as appropriate. To prevent opacity switching, which occurs at high culture density, cultures were grown overnight at room temperature or 37°C without shaking followed by growth at 37°C with shaking to the desired optical density at 600 nm (OD 600 ). MIC and cloning experiments were performed on LB medium supplemented with 1.5% agar. For switching assays and qualitative examination of switching, 0.5ϫ LB supplemented with 0.8% agar was used. Plates were supplemented with tetracycline, hygromycin, tobramycin, or kanamycin as indicated in the text and legends.
Wild-type LSO variants were isolated by plating serial dilutions of pure wild-type VIR-O stocks on 0.5ϫ LB with 0.8% agar. Rare colonies that lacked AV-T sectors at 24 h were passaged onto 0.5ϫ LB with 0.8% agar to confirm phenotypic stability; single colonies from restreaks were cultured and stocked as described above. The wild-type isolate designated set B VIR-O2 was derived by plating serial dilutions of an isolated LSO colony onto 0.5ϫ LB with 0.8% agar. Colonies were screened for sectoring at approximately 24 h of growth; a sectoring colony was passaged onto 0.5ϫ LB with 0.8% agar to confirm the phenotype. A single colony was then cultured and stocked as described above.
AV-T isolates carrying a single copy of the duplicated region were screened for by plating serial dilutions of an AV-T strain harboring two copies on 0.5ϫ LB with 0.8% agar. Colonies were then patched onto 0.5ϫ LB with 0.8% agar with and without 32 g/ml of tobramycin. Patches that grew poorly on tobramycin were struck from the plate without antibiotic onto 0.5ϫ LB with 0.8% agar, and single colonies were cultured and stocked as described above. Copy numbers of the duplicated region for the resulting isolates AV-T-56 and AV-T-80, as well as the parental AV-T isolate, were determined by qPCR as described below.
An LSO variant cured of plasmid p1AB5075 was serendipitously generated in an attempt to knockout aadB. A strain containing a T26 insertion in aadB was obtained from the AB5075 transposon library established at the University of Washington (10). Genomic DNA (gDNA) was prepared from an overnight culture of this strain grown at 37°C with shaking as described below. Diluted gDNA was used as a template for PCR using Phusion Hot Start II DNA polymerase (Thermo Scientific, Waltham, MA) with primers oSA65 and oSA66 (see Table S2 in the supplemental material) to amplify the aadB gene and the transposon disruption. The resulting fragment was ligated into the SmaI site of the counterselectable suicide plasmid pEX100 (58); the ligation was transformed into TransforMax EC100 electrocompetent E. coli (Lucigen, Middleton, WI), and transformants were selected by plating on 10 g/ml tetracycline. The resulting plasmid was designated pEX100-aadBTnS. A concentrated preparation of pEX100-aadBTnS was Regulation of Opacity Variation in A. baumannii ® September/October 2020 Volume 11 Issue 5 e02338-20 mbio.asm.org 13 made by purifying plasmid from 4 ml of E. coli culture using the Qiaprep Miniprep kit (Qiagen, Germantown, MD) followed by concentration through ethanol precipitation. This concentrated DNA was electroporated into an LSO culture of AB5075 as described below. Transformants were selected using 5 g/ml tetracycline. Colonies were then screened for sensitivity to 10% sucrose to identify transformants that had incorporated the suicide plasmid by homologous recombination. Cells that had undergone a second crossover to lose the plasmid backbone were selected for by culturing for several hours without antibiotic at 37°C with shaking followed by plating serial dilutions on no salt LB with 10% sucrose. Double-crossover colonies were then screened for tetracycline resistance to determine whether they had retained the T26 marker. The colony designated LSO Δp1AB5075 was determined to be tetracycline sensitive and was originally assumed to have reverted to the wild-type genotype upon the second crossover. We determined that LSO Δp1AB5075 had actually lost the p1AB5075 plasmid by attempting to amplify fragments of DNA from this plasmid using primer sets oSA69/70, oSA67/68, oSA86/87, and oSA104/105. All of these PCRs were unsuccessful, while control PCRs using primers in the chromosomal mutS gene (osA7 and oSA8) still amplified the expected band. Strains in which p1AB5075 had been reintroduced into LSO Δp1AB5075 were generated by electroporation. gDNA was prepared from VIR-O and LSO cultures grown to an OD of 0.5, as described below. gDNA from each strain was transformed into LSO Δp1AB5075 as described below. Transformants that had recovered the p1AB5075 plasmid were selected for by plating on 0.5ϫ LB with 0.8% agar supplemented with 2 g/ml tobramycin. Switching phenotypes were assessed qualitatively by observing transformants under a dissecting microscope with oblique illumination; colonies were passaged onto 0.5ϫ LB with 0.8% agar with 2 g/ml tobramycin and then cultured in LB to generate stocks as described above.
T26 insertions in RNase genes were obtained from the transposon mutant library prepared by Gallagher et al. (10). Initial screening was performed by electroporating the indicated plasmids directly into VIR-O colonies obtained from the library, using the method described below. Strains for Northern blotting were reconstructed in either the LSO Δp1AB5075 strain (rph::Tc, rnc::Tc) or a VIR-O isolate (remaining mutants) by transduction. Briefly, supernatants from stationary-phase cultures containing the desired library mutation were filtered and mixed with the appropriate AB5075 isolate grown to early-log phase. Mixtures were spotted onto LB plates and incubated for 4 h at 37°C. Spots were resuspended in LB, and transductants were selected for using 10 g/ml tetracycline. Transductants were stocked as described above, and insertions were confirmed by PCR with Taq DNA polymerase (New England Biolabs, Ipswich, MA) using primers given in Table S2.
Growth curves. Cultures were grown overnight without shaking to exponential phase (0.05 Ͻ OD Ͻ 0.5). Cultures were normalized to an OD of 0.05 and grown at 37°C with shaking to early stationary phase (OD Ͼ 1.0).
Electroporation of AB5075. Electroporations were conducted with cultures of AB5075 isolates grown at 37°C with shaking to late log phase. Cells were pelleted by centrifugation and washed three times with 10% glycerol to induce electrocompetence. Cells were mixed with either gDNA or plasmid DNA in 2-mm cuvettes and electroporated at 2.50 kV. Transformants were recovered in 1 ml LB for 30 min at 37°C without shaking followed by 1 h at 37°C with shaking. Transformants were then selected for by plating on the appropriate antibiotic.
Opacity switching assays. Switching frequencies were calculated as previously described (57). Switching assays were conducted on plates containing 20 ml of 0.5ϫ LB with 0.8% agar supplemented with 100 g/ml hygromycin or 5 g/ml tetracycline, as appropriate to ensure maintenance of plasmids. Assays were performed using six 24-h-old colonies from at least two independent experiments, unless otherwise noted.

VIR-O and LSO interconversion assays.
To estimate the rate at which VIR-O and LSO interconvert, suspensions of 24-h-old colonies (both variants) or cultures (VIR-O only) were serially diluted on 0.5ϫ LB with 0.8% agar to determine CFU; colony suspensions and cultures were then stored at 4°C. Following enumeration of CFU, suspensions and cultures were plated to obtain approximately 50 colonies per plate across several plates in order to screen several hundred colonies. After 48 h of growth, colonies were screened for sectoring using a dissecting microscope with oblique illumination. Colonies that appeared to exhibit the phenotype of the other variant were passaged onto 0.5ϫ LB with 0.8% agar alongside a VIR-O positive control to confirm the phenotype. The frequency of conversion was calculated by dividing the number of confirmed colonies of the other phenotype by the total number of colonies screened. gDNA isolation. Unless otherwise noted, cultures used for gDNA preparations were grown with shaking at 37°C to an OD 600 of 0.5. One milliliter of cells was pelleted through centrifugation; pellets were then resuspended in Tris-EDTA. Cells were lysed by incubation for 1 h at 37°C with 0.5% SDS and 400 g/ml proteinase K. Sodium chloride was added to a final concentration of 0.7 M, and DNA was extracted twice using equal volumes of phenol-chloroform and isoamyl alcohol. DNA was isolated by mixing with equal volumes of 95% ethanol until a precipitate was visible. DNA pellets were collected by centrifugation, washed twice with 75% ethanol, dried, and resuspended in molecular-grade water.
Whole-genome sequencing. PacBio WGS was performed at the University of Maryland Genomics Resource Center.
MIC assays. MICs were determined using a modified Etest assay as previously described (11). Briefly, two strains at an OD 600 of 0.1 were inoculated onto either side of an Etest strip (bioMérieux, Marcyl'Étoile, France) by pipetting 10 l of culture next to the bottom of the strip, tilting the plate to spread the culture up the side of the strip, and pipetting away excess culture at the top of the strip. MICs were assessed after 6 or 16 h, as noted in the text and legends. Experiments were performed two independent times to confirm the reproducibility of trends. soaked in 20ϫ SSC for 5 min. The membrane was baked for 2 h in a vacuum oven at 80°C with 330.2 mm Hg and subsequently treated according to the Southern blotting protocol. The membrane was cut in half to separate duplicate spots; one set was probed with oSA197, while the other set was probed with oSA191 as a control (Table S2). The blot was photographed following development overnight.
Mice. Wild-type (WT) C57BL/6J female mice were purchased from Jackson Laboratories and used at age 8 to 10 weeks. Mice were housed under specific-pathogen-free conditions at Yerkes National Primate Center, Emory University. Experimental studies were performed in accordance with the Institutional Animal Care and Use Committee guidelines.
Mouse pulmonary infection models. Approximately 5 ϫ 10 7 CFU were administered per mouse for infections to quantify the bacterial load. Overnight standing bacteria at room temperature were subcultured in LB broth and grown at 37°C with shaking to an OD 600 of ϳ0.15, washed, and resuspended in phosphate-buffered saline (PBS). Fifty microliters of bacterial inocula was inoculated intranasally (i.n.) to each mouse. Mice were anesthetized with isoflurane immediately prior to i.n. inoculation. At 24 h postinfection, the mice were sacrificed, and the lungs, spleen, and liver were harvested, homogenized, and plated for CFU on 0.5ϫ LB plates.
RNA sequencing. RNA-seq was performed at the Yerkes Genomics Core at Emory University. RNA-seq was performed in triplicates using samples generated from three independent cultures of the LSO and VIR-O variants. Upon sample collection, cultures were struck onto a 0.5ϫ LB and 0.8% agar plate and checked for the presence of contaminating AV-T variants; only samples with greater than ϳ95% opaque variants were used for RNA-seq. RNA purification was performed as outlined above, after which, samples were treated twice with Turbo DNA-free according to the manufacturer's instructions. To confirm that samples were not contaminated with DNA following DNase treatment, PCR of RNA samples was performed using primers abaR-qPCR-for and abaR-qPCR-rev (Table S2). Samples were then processed for RNA-sequencing at the Yerkes National Primate Research Center Genomic core.
Data availability. RNA sequencing information and files have been deposited to the GEO database under the accession numbers GSE156206, GSM4726730, GSM4726731, GSM4726732, GSM4726733, GSM4726734, and GSM4726735.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.