Colistin Dependence in Extensively Drug-Resistant Acinetobacter baumannii Strain Is Associated with ISAjo2 and ISAba13 Insertions and Multiple Cellular Responses

The nosocomial opportunistic Gram-negative bacterial pathogen Acinetobacter baumannii is resistant to multiple antimicrobial agents and an emerging global health problem. The polymyxin antibiotic colistin, targeting the negatively charged lipid A component of the lipopolysaccharide on the bacterial cell surface, is often considered as the last-resort treatment, but resistance to colistin is unfortunately increasing worldwide. Notably, colistin-susceptible A. baumannii can also develop a colistin dependence after exposure to this drug in vitro. Colistin dependence might represent a stepping stone to resistance also in vivo. However, the mechanisms are far from clear. To address this issue, we combined proteogenomics, high-resolution microscopy, and lipid profiling to characterize and compare A. baumannii colistin-susceptible clinical isolate (Ab-S) of to its colistin-dependent subpopulation (Ab-D) obtained after subsequent passages in moderate colistin concentrations. Incidentally, in the colistin-dependent subpopulation the lpxA gene was disrupted by insertion of ISAjo2, the lipid A biosynthesis terminated, and Ab-D cells displayed a lipooligosaccharide (LOS)-deficient phenotype. Moreover, both mlaD and pldA genes were perturbed by insertions of ISAjo2 and ISAba13, and LOS-deficient bacteria displayed a capsule with decreased thickness as well as other surface imperfections. The major changes in relative protein abundance levels were detected in type 6 secretion system (T6SS) components, the resistance-nodulation-division (RND)-type efflux pumps, and in proteins involved in maintenance of outer membrane asymmetry. These findings suggest that colistin dependence in A. baumannii involves an ensemble of mechanisms seen in resistance development and accompanied by complex cellular events related to insertional sequences (ISs)-triggered LOS-deficiency. To our knowledge, this is the first study demonstrating the involvement of ISAjo2 and ISAba13 IS elements in the modulation of the lipid A biosynthesis and associated development of dependence on colistin.


Introduction
Acinetobacter baumannii is a nosocomial opportunistic Gram-negative pathogen displaying extensive resistance to many classes of antimicrobial agents. An arsenal of virulence factors, such as biofilm production, capsular polysaccharides, drug efflux pumps, release Figure 1. Diversity of the growth pattern of A. baumannii clinical isolate. Initially, this isolate was characterized as susceptible to colistin by Etest on MHA, S ≤ 2 µg/mL (Ab-S), on the left. A bacterial subpopulation displaying colistin-dependent growth was selected after 8-10 passages on agar containing 16 µg/mL colistin sulphate. This colistin-dependent population (Ab-D) grows only along the colistin Etest strip on a pear-like shape area between 0.125 and 256 µg/mL, (middle panel). When the Ab-D subpopulation was passed 6 times on agar base medium and further applied for colistin Etest, a phenotypic heterogeneity within this bacterial population was observed (Ab-D after 6 passages on colistin-free medium). Most cells were growing in a colistin-susceptible manner and a certain amount retained a colistin-dependent growth pattern appearing along the colistin Etest strip (right panel).

Ultrastructural Traits of Colistin-Dependent A. baumannii
Development of resistance or dependence to colistin may occur via alterations in bacterial structures that are located outside or within the cell envelope, thereby affecting the bacterial fitness. To address this issue, we employed super resolution imaging by transmission electron microscopy (TEM) of the two A. baumannii subpopulations, Ab-S and Ab- Figure 1. Diversity of the growth pattern of A. baumannii clinical isolate. Initially, this isolate was characterized as susceptible to colistin by Etest on MHA, S ≤ 2 µg/mL (Ab-S), on the left. A bacterial subpopulation displaying colistin-dependent growth was selected after 8-10 passages on agar containing 16 µg/mL colistin sulphate. This colistin-dependent population (Ab-D) grows only along the colistin Etest strip on a pear-like shape area between 0.125 and 256 µg/mL, (middle panel). When the Ab-D subpopulation was passed 6 times on agar base medium and further applied for colistin Etest, a phenotypic heterogeneity within this bacterial population was observed (Ab-D after 6 passages on colistin-free medium). Most cells were growing in a colistin-susceptible manner and a certain amount retained a colistin-dependent growth pattern appearing along the colistin Etest strip (right panel).
When the MICs of 17 antimicrobial agents, including colistin, were evaluated using the broth microdilution method, the colistin MIC for the dependent Ab-D phenotype exceeded 8 µg/mL in comparison to MIC ≤ 2 µg/mL for Ab-S (Table S1). The latter did, however, displayed a high degree of drug resistance, for instance, to carbapenems, and was accordingly classified as extensively drug-resistant (XDR) [18]. Interestingly, the Ab-D subpopulation displayed lower MIC-values for meropenem, imipenem, amikacin, tigecycline, piperacillin-tazobactam, and amoxicillin-clavulanic acid compared to Ab-S (Table S1). The stability of the dependent phenotype was limited, but the characteristics were maintained during a few passages in colistin-free medium. They were, however, only partially reversible, as subsequent growth on antibiotic-free medium after six subculture passages yielded phenotypic heterogeneity within this bacterial population. Most cells displayed a colistin-susceptible manner but some retained a colistin-dependent growth pattern appearing along the colistin Etest strip (Figure 1).

Ultrastructural Traits of Colistin-Dependent A. baumannii
Development of resistance or dependence to colistin may occur via alterations in bacterial structures that are located outside or within the cell envelope, thereby affecting the bacterial fitness. To address this issue, we employed super resolution imaging by transmission electron microscopy (TEM) of the two A. baumannii subpopulations, Ab-S and Ab-D (Figure 2A,B). We observed that, under both conditions, Ab-S cells displayed a regular shape of the cell envelope with appendage-like structures, such as pili or fimbriae and a thick exopolysaccharide capsule layer. By contrast, Ab-D bacteria showed a cell envelope with increased membrane curvature, few appendages, and a thin capsule, especially when grown on MHA plates in close proximity of the colistin Etest strip. The Ab-D cells were  Black arrows point on appendage-like structures such as pili and fimbriae. White arrows specify outer membrane vesicles. Colored blue and green arrows point on exopolysaccharide capsule layers. (C) Quantification of the bacterial capsule thickness, nm. Columns represent the mean ± SE. Significant differences are indicated with *** when p > 0.001, as analyzed by two-tailed Student's t-test. Data from three different experiments, and 100 different capsule areas in at least 50 different cells per condition were analyzed (same color code as in insert images).

Variations in the Lipid A Content
A possible mechanism for colistin dependence is attributed to defects in or modifications of the lipid A moiety of the LPS, which may result in an impaired cell membrane [13]. We utilized Matrix-Assisted Laser Desorption Ionization-Time of Flight (MALDI-TOF) MS to investigate the lipid A structure in Ab-S and Ab-D. This method allows a rapid and accurate detection of lipid A possible modifications, including a potential pETN addition [19,20]. In Ab-S, the mass spectrum of lipid A was attributed by peaks at m/z values of 1910.5, 1728.3, and 1529.9 ( Figure 3). The two most intensive signals at m/z 1910.5 and m/z 1728.3 were produced by singly deprotonated bis-phosphorylated hepta-acyl and bis-phosphorylated hexa-acyl lipid A, which is consistent with previous studies [21,22]. The major peaks 1910.5, 1728.3, 1529.9, were missing in the mass spectrum for lipid A from the Ab-D cells (Figure 3). This likely reflects a lipid A deficiency in the Ab-D subpopulation. It is worth noting that no peaks were detected corresponding to previously described pETN-modified mono-and bis-phosphorylated hepta-acyl lipid A [19][20][21]23,24], indicating that colistin dependence in Ab-D is not associated with the addition of pETN. Taken together, MALDI-TOF MS analyses demonstrated structural deficiencies in lipid A, that most likely lead to structural defects also in the LOS and outer membrane. 186 * Complete genome sequences were searched against the NCBI bacteria database in order to determine the species. The reference genome with the best match is gathered and mapped reads refers to the percentage of the genome that have been matched against the reference genome. ** Unmapped percentage are reads that did not map against any previous reference genomes in NCBI.

Whole Genome Sequencing Reveals Genes Associated with Antibiotic Resistance
To investigate potential genetic differences, we performed whole genome sequencing (WGS) of Ab-S and Ab-D with downstream bioinformatics analyses. The whole genome size of A. baumannii was 4,111,821 and 4,125,623 base pairs (bp) for the Ab-S and Ab-D populations (Table 1, Supplementary files S1 and S2), whose genome data could be assembled into 121 and 186 contigs for Ab-S and Ab-D, respectively. As many as 109 contigs (59%) in Ab-D included less than 1000 bp, which is significantly more than the 35% in Ab-S. On the other hand, the number of sequences longer than 1000 bp was nearly the same for Ab-S and Ab-D (79 and 77 respectively). When the complete genome sequences were searched against the NCBI bacteria database, the reference genome with NCBI accession NC_011586 corresponding to A. baumannii MDR strain AB0057 was identified as closest in sequence with 95.1% and 90.9% sequence match for Ab-S and Ab-D, respectively. This indicates that the strain used here might have evolved from the AB0057, first isolated from a blood-stream infection patient at Walter Reed Army Medical Center [25][26][27]. Upon multi-locus sequence typing (MLST), Ab-S and Ab-D had sequence type 1 (ST1). The assembled data were also searched for antimicrobial resistance genes using the ResFinder database [28]. This analysis of the resistome revealed presence of five groups of genes associated with antibiotic resistance identical in both Ab-S and Ab-D subpopulations: aminoglycoside-modifying enzymes (aac(3)-IId, aph(3 )-VIa, aph(3 )-Ic, aac(3)-Ia, aadA1); beta-lactamases (bla OXA-69 , bla TEM-1D , bla OXA-23 , bla ADC-25 , bla NDM-1 ); phenicols (floR, catA1); sulphonamide (sul1) and tetracycline (tetA) ( Table 2). This list of the identified resistance genes was rather similar to that for AB0057 [25,29]. No mobile plasmid-mediated resistance mcr genes were found even after manual verification and no differences between Ab-S and Ab-D in polymyxin resistance-associated genes pmrA, pmrB and pmrC were identified. We further analyzed the plasmid content of susceptible and dependent A. baumannii using PLSDB, a resource of complete bacterial plasmids retrieved from the NCBI database [30]. PLSDB analysis identified 4 plasmids and no difference between Ab-S and Ab-D was observed (Table S2). SNPs analysis was performed, but did not find any differences between the genomes. However, as the variant call is based on read mapping, the SNP analysis cannot reveal differences in insertion sequences (ISs) elements, as these can be present in multiple copies and map to multiple different positions.

In Silico Genome Analysis Displays Multiple ISs
ISs mediate genome rearrangement, disturb coding sequences by insertional inactivation, and modulate gene expression enabling bacterial adaptability and resistance to antibacterial agents [31]. Some ISs are known to be involved in both colistin resistance [32,33] and dependence development [34]. Therefore, we further investigated the ISs content in genome sequences of Ab-S and Ab-D. Multiple ISs elements, both truncated and full-length, were identified using ISfinder [35]. ISfinder recognized 411 copies of IS-related events present in 73 contigs of the Ab-S dataset among all sequences and 652 IS-related events in 140 contigs of the Ab-D dataset (Table S2). Both subpopulations shared similar ISs composition from 13 families (IS5, IS6, IS4, IS3, ISNCY (IS Not Classified Yet), Tn3, IS30, ISL3, IS1, IS66, IS256, IS1182 and IS481) with ISs elements located predominantly in the beginning and at the very end of contigs (Table 3). ISAba1, ISAjo2, and ISAba13 had more than 10 ISs copies per genome detected in both Ab-S and Ab-D (Table 4), but the number of full-length ISs was low. WGS identified two copies of ISAjo2 full-length gene and three copies of ISAba13 full-length gene. ISAba1 and ISAba123 were encoded by single full-length gene copy in either of them. WGS identified two copies of ISAjo2 full-length gene and three copies of ISAba13 full-length gene in both Ab-S and Ab-D. ISAba1 and ISAba123 were encoded by single full-length gene copy each in both Ab-S and Ab-D. No additional novel ISs were detected in Ab-D in comparison to Ab-S. In Ab-S, the IS5 family showed the largest number of ISs and the most frequently detected IS element was ISAba1 (n = 26) belonging to the IS4 family, and with a similar number for ISAba33 (Table  S3). It should be noted, however, that ISAba33 shares 87% identity with ISAba1, which is typical for ISs belonging to the same family; thus, many identified ISs are difficult to distinguish due to sequences overlap. The most substantial difference between Ab-S and Ab-D was a dramatic increase in the number of ISAjo2 from 18 in Ab-S to 84 in Ab-D (Table 4). ISAjo2 was initially described for the Acinetobacter johnsonii genome and belongs to the ISNCY family [36]. Interestingly, less than 1000 bp short ISAjo2-borne contigs were overrepresented (77%) in Ab-D compared to Ab-S (22%). This difference can likely be attributed to the high levels of DNA repetitiveness, i.e., the presence of multiple identical or highly homologous ISs flanked by short terminal inverted repeats creating ambiguities in alignment and assembly, resulting in the presence of multiple short contigs [37].

Proteome Alterations
Genomics alone is not enough for understanding the biology and phenotypic capabilities of the organisms, such as responses to environmental perturbations. Thus, to provide a more complete characterization we utilized a comprehensive strain-specific combination of genomics and proteomics. First, to generate a complete set of possible protein-coding sequences, the acquired WGSs of Ab-D and Ab-S were translated in silico in six reading frames. Based on detected open reading frames (ORFs), phenotype-specific Ab-D and Ab-S databases, containing 60,560 and 60,210 potential protein sequences, respectively, were built. (Supplementary files S3 and S4). Second, we used the resulting protein databases in FASTA-format for analyses of high-resolution mass spectrometry data. This approach allowed the identification and relative quantification of around 2000 proteins and protein clusters; 196 proteins were expressed differentially, out of which 67 proteins were upregulated (Table S4) and 129 proteins were downregulated (Table S5). The major groups of differentially expressed proteins were secretion systems proteins, i.e., efflux pumps and proteins, ABC transporters and porins, but also other outer membrane components and proteins associated with antibiotic resistance (Table 5). We also identified 30 hypothetical differentially expressed proteins.

Upregulation of the Type VI Secretion System and RND-Type Efflux Proteins
The structural components of the type VI secretion system (T6SS), i.e., valine-glycine repeat protein G (VgrG or TssI), hemolysin-coregulated proteins (Hcp or TssD) and type VI secretion system ATPase TssH (ClpV) were dramatically upregulated in the Ab-D subpopulation (Tables S4 and S5). Interestingly, all the three out of a possible three (according to our WGS results) isoforms of the tip protein VgrG were upregulated. Likewise, the relative abundance of proteins belonging to type IV secretion system (T4SS) and Sec protein translocase complex were increased significantly in Ab-D. Additionally, the levels of proteins belonging to AdeABC and AdeIJK bacterial efflux pumps of the Resistancenodulation-cell division (RND) superfamily transporters increased two-fold in Ab-D, while the nucleotide sequences of the corresponding contigs encoding AdeABC, AdeIJK, and MacA were identical in Ab-S and Ab-D.

Downregulation of Proteins Involved in Maintenance of Outer Membrane Asymmetry
All proteins identified as involved in the maintenance of the asymmetric lipid distribution, with lipopolysaccharides at the outer leaflet and phospholipids at the inner leaflet of the outer membrane, were downregulated or missing in Ab-D. These included MlaA, MlaD, and PldA involved in retrograde glycerophospholipid transport and lipid degradation [38,39], acyl-ACP-UDP-N-acetylglucosamine O-acyltransferase LpxA catalyzing the first step of lipid A biosynthesis [6], LptA and LptD lipopolysaccharide transport (Lpt) [40,41] and periplasmic carrier protein LolA [42,43]. Almost all differentially expressed proteins identified as transporters (see "Other transporters" in Table 5, Tables S4 and S5) were downregulated as well as many other outer membrane proteins (Table S5). The only exception was the C4-dicarboxylate transporter (DctA) involved in carbon metabolism, which was upregulated more than seven times in the Ab-D population. Interestingly, the ATP-dependent tetradecameric serine protease ATP-binding subunit ClpX and Lon protease (ATP-dependent protease La) were upregulated in the Ab-D subpopulation (Table 5). Table 5. Selected groups of proteins differentially expressed in Ab-S and Ab-D. Fold changes were calculated with Ab-S as the reference category. Fold change value 0 means that protein was detected in Ab-S, but not in Ab-D.

Proteins
Accession

Alterations of Other Proteins Associated with Antibiotic Resistance
We identified changes in the levels of several other proteins, earlier known to be involved in antibiotic resistance mechanisms (Table 5). We observed a significant increase in CarO (the carbapenem resistance-associated outer membrane protein) and di-hydropteroate synthase DHPS (the sulfonamides target [44]) expression levels in Ab-D. Three beta-lactamases, encoded by bla NDM-1 , bla ADC-25 , and bla OXA-23 genes (and detected by Res-Finder in both Ab-S and Ab-D (Table 2)), were downregulated in the Ab-D subpopulation (Table 5). In summary, we found significant proteome alterations in Ab-D, indicating that colistin dependence is a result of a complex cellular response. Next, we evaluated if the observed protein expression alterations were consistent with the genetic differences between Ab-S and Ab-D. Out of all significantly downregulated proteins, sequence analyses of the lpxA, mlaD and pldA genes in Ab-D revealed ISs elements. Thus, the ISAjo2, belonging to the IS1202 group of the ISNCY family, was inserted into both lpxA and mlaD genes, and ISAba13 from IS5 family group IS903 into the pldA gene, acting as gene knockouts. Figure 4 illustrates the organization of the three gene clusters containing lpxA, mlaD and pldA in Ab-S and Ab-D. Interestingly, that two phenotypically distinct populations, obtained from Ab-D after six subculture passages on antibiotic-free medium, contained the same ISs inserted in mlaD and pldA, but cells that returned to colistin-sensitive growth pattern (Figure 1, right)  elements. Thus, the ISAjo2, belonging to the IS1202 group of the ISNCY family, was inserted into both lpxA and mlaD genes, and ISAba13 from IS5 family group IS903 into the pldA gene, acting as gene knockouts. Figure 4 illustrates the organization of the three gene clusters containing lpxA, mlaD and pldA in Ab-S and Ab-D. Interestingly, that two phenotypically distinct populations, obtained from Ab-D after six subculture passages on antibiotic-free medium, contained the same ISs inserted in mlaD and pldA, but cells that returned to colistin-sensitive growth pattern (Figure 1, right) have lost the ISAjo2 interrupting lpxA. Is elements may appear both within coding and intergenic regions of genes, where they can control neighboring genes and affecting their expression. Still, no ISs were detected in close proximity of genes encoding the three most upregulated proteins, i.e., the isoforms of the type VI secretion system tip proteins VgrG and Hcp. Pairwise sequence alignment with Stretcher (EMBL-EBI) [45] disclosed 100% identity of the VgrG-and Hcpencoding contigs from Ab-S and Ab-D. According to Zhu et al., mutations in katG, encoding a catalase involved in reactive oxygen species scavenging, affected polymyxin dependence in MRD A. baumannii AB5075 [15] and inactivation of mrcA, encoding the penicillin-binding protein A1, has been suggested to promote LOS loss [7]. Nevertheless, no mutations/substitutions/insertions in katG or mrcA were present in Ab-S and Ab-D.  Is elements may appear both within coding and intergenic regions of genes, where they can control neighboring genes and affecting their expression. Still, no ISs were detected in close proximity of genes encoding the three most upregulated proteins, i.e., the isoforms of the type VI secretion system tip proteins VgrG and Hcp. Pairwise sequence alignment with Stretcher (EMBL-EBI) [45] disclosed 100% identity of the VgrG-and Hcp-encoding contigs from Ab-S and Ab-D. According to Zhu et al., mutations in katG, encoding a catalase involved in reactive oxygen species scavenging, affected polymyxin dependence in MRD A. baumannii AB5075 [15] and inactivation of mrcA, encoding the penicillin-binding protein A1, has been suggested to promote LOS loss [7]. Nevertheless, no mutations/substitutions/insertions in katG or mrcA were present in Ab-S and Ab-D.

Lipid A Modifications and LOS Deficiency in Colistin-Dependent A. baumannii
The present study showed insertion of ISAjo2, which belongs to the IS1202 group of the ISNCY family, into lpxA gene in the colistin-dependent subpopulation Ab-D ( Figure 4A). Insertional disruption of the lpxA gene knocked it out and abolished acyl-ACP-UDP-Nacetylglucosamine O-acyltransferase expression in Ab-D (Table 5 and Table S5), which catalyzes the first step of lipid A biosynthesis. This event entirely terminates lipid A biosynthesis and results in a LOS-deficient A. baumannii phenotype [6] (Figure 5), which we confirmed here (Figures 2 and 3). Indeed, ISs and other defects in the lpxACD gene system are known to be associated with both colistin resistance [32,33] and dependence [34]. In the latter case, ISAba1 was disrupting the lpxC gene encoding UDP-3-O-acyl-N-acetylglucosamine deacetylase [34]. ISAjo2 was earlier identified in MDR A. johnsonii, where two copies of ISAjo2 flanked the region containing the carbapenemase gene bla OXA-58 [36]. We also noted that ISAjo2 was much more abundant in the Ab-D than in Ab-S (Table 4). This difference might be explained either by a redundancy of the genome assembly, or by ISAjo2 transposition triggered by subinhibitory concentrations of colistin [31]. In addition, both Ab-S and Ab-D displayed a high number of ISs elements accumulated in their genomes (Table 3), which likely reflects the evolutionary history of the parent strain.
We identified two other genes inactivated by insertions, i.e., mlaD and pldA, encoding proteins involved in maintenance of OM asymmetry [38] (Figure 5). In Ab-D, the ISAjo2 insertion interrupted the mlaD gene ( Figure 4B), encoding a membrane-anchored periplasmic protein MlaD of the Mla pathway, mediating retrograde transport of glycerophospholipids [46]. ISAba13 from the IS5 family group IS903 was located in the pldA gene ( Figure 4C). ISAba13 has previously been found in carbapenem-resistant A. baumannii [47,48] upstream from bla OXA-94 [48]. PldA is an outer membrane phospholipase catalyzing the hydrolysis of acyl ester bonds in phospholipids and degradation of mislocalized phospholipids from the surface-exposed outer leaflet. It was hypothesized that Mla and PldA systems function independently to prevent accumulation of phospholipids at the cell surface through distinct, yet related mechanisms [49] (Figure 5). We also verified mlaD and pldA inactivation and downregulation of MlaA in Ab-D by comparative proteomic analysis (Table 5 and Table S5). Our results are in agreement with the findings by Powers and Trent [38] on how LOS-deficient and highly polymyxin B-resistant A. baumannii can improve their fitness by elimination of the two OM asymmetry-maintenance systems, Mla and PldA. This was discovered, however, in colistin-resistant A. baumannii obtained after 120 generations of passaging on 10 µg/mL polymyxin B, while we observed the emergence of phenotypically colistin-dependent subpopulation already after 8 passages on colistin [38]. Recently, a transcriptomic analysis of a colistin-dependent isolate in relation to its parent colistin-susceptible MRD A. baumannii AB5075 revealed 1.7-fold increased expression of mlaD and 2.0-fold decreased expression of pldA in colistin-dependent bacteria [15]. We believe that this discrepancy can be attributed to the difference between strains.

Upregulation of the Type VI Secretion System, Efflux Proteins and Proteases in Colistin-Dependent A. baumannii
The most dramatic protein upregulation in LOS-deficient Ab-D population of A. baumannii was attributed to secretion system and efflux proteins ( Figure 5). These proteins help bacteria to export specific bacterial products to the cell surface, the extracellular environment and to other bacteria or eukaryotic cells, thereby promoting bacterial virulence [53]. Gram-negative bacteria harbor multiple types of secretion systems and T6SS is the most prevalent. T6SS is a macromolecular envelope-spanning apparatus that injects toxic bacterial effector proteins directly into eukaryotic target host or other prokaryotic cells, which is important for virulence and inter-bacterial competition. It mimics inverted phage tail and tube [54].
The inner tube of this structure is comprised of hemolysin-coregulated protein (Hcp or TssD) assembled into stacked hexameric rings, permitting effectors to pass through the tube center. This inner tube is capped by the valine-glycine repeat protein G (VgrG or TssI) trimer, forming the needle-like structure on the top of the inner tube allowing it to penetrate the host membrane. It is surrounded by a contractile sheath, which is disassembled and recycled by an AAA+-type ATPase TssH (ClpV) after the delivery of toxic effectors into a recipient cell. The perturbation of the cell envelope caused by membrane-tar- Various studies have indicated that, beside downregulation of glycerophospholipid transport and lipid degradation, lack of LOS can also lead to altered expression of critical transport and biosynthesis systems, modulating the composition and structure of the bacterial envelope [50], which is in agreement with our observations (Figure 2), and reducing the capacity for adhesion and formation of biofilms [51,52]. We found that in LOS-deficient Ab-D, many other proteins involved in maintenance of the asymmetric lipid distribution were downregulated or missing. This included proteins LolA from the Lol lipoprotein transport system, LptA and LptD lipopolysaccharide transport proteins responsible for the transport and assembly of LPS [40,41] and many other transporters and outer membrane proteins (Table 5 and Table S5).

Upregulation of the Type VI Secretion System, Efflux Proteins and Proteases in Colistin-Dependent A. baumannii
The most dramatic protein upregulation in LOS-deficient Ab-D population of A. baumannii was attributed to secretion system and efflux proteins ( Figure 5). These proteins help bacteria to export specific bacterial products to the cell surface, the extracellular environment and to other bacteria or eukaryotic cells, thereby promoting bacterial virulence [53]. Gram-negative bacteria harbor multiple types of secretion systems and T6SS is the most prevalent. T6SS is a macromolecular envelope-spanning apparatus that injects toxic bacterial effector proteins directly into eukaryotic target host or other prokaryotic cells, which is important for virulence and inter-bacterial competition. It mimics inverted phage tail and tube [54].
The inner tube of this structure is comprised of hemolysin-coregulated protein (Hcp or TssD) assembled into stacked hexameric rings, permitting effectors to pass through the tube center. This inner tube is capped by the valine-glycine repeat protein G (VgrG or TssI) trimer, forming the needle-like structure on the top of the inner tube allowing it to penetrate the host membrane. It is surrounded by a contractile sheath, which is disassembled and recycled by an AAA+-type ATPase TssH (ClpV) after the delivery of toxic effectors into a recipient cell. The perturbation of the cell envelope caused by membrane-targeting antibiotics, such as polymyxin B, was earlier reported to serve as the signal triggering the T6SS activation in Gram-negative bacteria [55]. Nevertheless, a repressed expression of T6SS genes observed for LOS-deficient colistin-resistant A. baumannii [50], which is opposite to our findings. We suggest that the upregulation of T6SS in the LOS-deficient colistin-dependent population of A. baumannii may be a part of a cellular response to the LOS-deficiency caused by exposure to colistin.
In addition, we observed an upregulation of RND superfamily transporters: AdeABC and AdeIJK bacterial efflux pumps ( Figure 5). Overproduction of these in clinical isolates is associated to MDR due to their very broad substrate profiles [56][57][58][59][60][61]. In particular, they play a major role in tigecycline resistance in A. baumannii, where AdeIJK is able to efflux β-lactams. Nevertheless, we observed that the Ab-D subpopulation displayed increased sensitivity to carbapenems, tigecycline, piperacillin-tazobactam, and amoxicillin-clavulanic acid compared to the Ab-S, which could be due to increased OM permeability.
Overexpression of AdeABC or AdeIJK may also affect the expression of various proteins involved in adhesion and biofilm formation (72). In LOS-deficient A. baumannii increased expression of adeIJK and macAB-tolC has been attributed to an intracellular accumulation of toxic substances [50]. Moreover, in E. coli and Salmonella enterica cells with higher acrAB expression (AcrA corresponds to AdeI, in A. baumannii), higher spontaneous mutation frequencies were observed in response to ciprofloxacin, tetracycline, and chloramphenicol, indicating involvement of AcrAB efflux pump in initial stages of permanent antibiotic resistance [62].
Interestingly, the ATP-dependent tetradecameric serine protease ATP-binding subunit ClpX and the Lon protease (ATP-dependent protease La) were found among proteins upregulated in the colistin-dependent subpopulation (Table S1). These energy-dependent proteases, also functioning as chaperones, are involved in degradation of misfolded, damaged and short-lived regulatory proteins playing a significant role in bacterial stress response [63] and virulence regulation [64,65].

Bacterial Cultivation and Antibiotic Susceptibility Testing
The A. baumannii clinical isolate was obtained from the abdominal cavity of a patient during 2013 at the Department of Clinical Microbiology, Linköping University Hospital, Sweden and the susceptibility profile of this parental strain named AB1 was earlier characterized by Nordqvist et al. [66]. To test for colistin resistance, bacterial suspensions with 5 × 10 5 CFU/mL (colony forming units/mL), or 0.5 McFarland units in 0.9% NaCl were grown for 24 h at 37 • C on Mueller-Hinton agar (MHA) (Becton Dickinsson, Franklin Lakes, NJ, USA) with Etest strips (BioMérieux, Marcy l'Etoile, France) covering colistin concentrations between 0.016 and 256 µg/mL. The A. baumannii CCUG 19,096 strain (Culture Collection, Göteborg University, Sweden) was used as a control for the susceptibility test [67]. For selection of a colistin-dependent subpopulation, the isolate was cultured for 8-10 passages on agar containing 16 µg/mL colistin sulphate according to [66]. (Merk Sigma Aldrich, St. Louis, MO, USA) and parallel colistin-Etests to follow the growth pattern. Two A. baumannii subpopulations, susceptible to colistin (Ab-S) and colistin-dependent (Ab-D), were grown on MHA plates or MHA plates with a colistin Etest strip, collected and further processed for analyses on transmission electron microscopy (TEM). For lipid analyses with matrix-assisted matrix desorption-ionization (MALDI-TOF) MS, WGS and proteomics, Ab-S and Ab-D subpopulation were grown on MHA plates with colistin Etest strips. In addition, Ab-D subpopulation was cultured for 6 passages on Columbia agar base medium and further applied for colistin Etest to monitor the growth pattern.

Transmission Electron Microscopy
Bacteria were collected, washed, fixed in 3% glutaraldehyde (Polysciences, Warrington, PA, USA) in 0.1 M Na cacodylate buffer, pH 7.4 for 30 min at room temperature (RT), centrifuged and embedded in 4% gelatin. Then followed 3% glutaraldehyde fixation in sodium cacodylate buffer pH 7.4 for 2 h at RT, washing with the same buffer and postfixation in 1% osmium tetroxide (Polysciences) for 1 h at 4 • C, rinsing and staining with 2% uranyl acetate (Polysciences in 50% ethanol and dehydration in a series of ascending concentrations of ethanol and acetone. Prior to embedding in Durcupan ACM epoxy embedding medium kit (Merk Sigma Aldrich) two-step infiltration was done. Ultrathin sections of 70-nm thickness were prepared using a Leica EM UC7 ultramicrotome (Leica Microsystems GmbH, Wetzlar, Germany), collected onto formvar-coated slot grids, and counter-stained with uranyl acetate and lead citrate. The specimens were examined in a JEM 1230 transmission electron microscope operated at 100 kV (JEOL Ltd., Tokyo, Japan), and the images were taken with a Orius SC1000 CCD camera using Digital Micrograph software (Gatan, Pleasanton, CA, USA).
For measuring the bacterial capsule thickness, images of the cells were analyzed with the ImageJ software (NIH, Bethesda, MD, USA), yielding graphs with means (±SE) and statistical analyses based on paired two-tailed Student's t-tests. p-values < 0.05 (*), < 0.01 (**), and < 0.001 (***) were considered significant. At least 3 independent experiments were performed on separate days on different cell passages.

Lipid A Extraction
Bacteria were collected and lipid A was extracted by an improved rapid microextraction method previously described [34,68], with some modifications. Briefly, 10 mg of cells suspended in 100 µL of isobutyric acid −1 M ammonium hydroxide mixture (5:3, v/v) were incubated in a microwave oven for 60 s (400 W, 2450 MHz) and then centrifuged at 8000 g for 15 min. Supernatants were transferred to new tubes, mixed with an equal volume of water and vacuum dried. The obtained pellets were washed twice in 400 µL of methanol, and centrifuged at 5000 rpm for 15 min. Finally, the insoluble lipid A was solubilized in 100 µL chloroform-methanol-water mixture (3:1.5:0.25, v/v/v) and subjected to a mass spectrometry analysis.

Whole-Genome Sequencing and Data Analysis
DNA was extracted from the isolates using EZ1 DNA Tissue Kit (Qiagen, Hilden, Germany). After quantification with Qubit dsDNA High Sensitivity kit (Thermo Fisher Scientific), 20 ng of DNA was used for library preparation using QIAseq FX DNA Library Kit (Qiagen). DNA libraries were sequenced on the MiSeq platform (Illumina, San Diego, CA, USA) with 2 × 300 bp paired-end reads, and the samples obtained an average sequencing depth of 103×. Data analysis was performed in CLC Genomics Workbench v. 9.5.4 with the Microbial Genomics Module v. 1.6.2 (Qiagen). The analysis included determination of and mapping to the closest NCBI reference genome (NC_011586), multi locus sequence typing (MLST) with the Pasteur scheme [70] and de novo assembly. Assembled data were searched for antimicrobial resistance genes against the ResFinder database (Centre for Genomic Epidemiology, Technical University of Denmark) with a threshold of 98% for identity and 60% for length [28]. A bacterial plasmid database PLSDB (https://ccb-microbe.cs.uni-saarland.de/plsdb/, accessed on 23 June 2020) version v0.4.1-2-g1b893f22b9 [30] with maximal p-value set to 0.1 and minimal identity set to 0.99 was used for identification of plasmids. Winner-takes-all strategy that removes redundancy from the output was employed (hashes found in multiple queries were removed except for the query with highest identity). ISs were annotated using ISFinder (https://isfinder.biotoul.fr/, accessed on 4 May 2020) [35] with E-value threshold below 1 × 10 −4 for Blast hits. The global alignment tools Needle and Stretcher (EMBL-EBI, https://www.ebi.ac.uk/Tools/psa/, accessed on 8 March 2018) [45] were used for pairwise sequence alignment. Variant calling was first performed against the closest reference genome, and single nucleotide polymorphisms (SNPs) with a depth of coverage ≥ 20× and a frequency ≥90% were compared between the two subpopulations.

Database
The WGS nucleotide sequences of A. baumannii were translated in silico. The open reading frames (ORFs) prediction for potential protein encoding segments has been carried out using ORF Finder [71] from the National Center for Biotechnology Information (NCBI) with the following settings: bacterial, archaeal & plant plastid code; ATG and alternative start codons, minimal length 75 nucleotides. The generated FASTA-style databases had genomic position and frame information embedded into each header tag for a given sequence.

Protein Extraction
Bacterial cells collected from the agar plates were suspended in 50 mL phosphate buffered saline (PBS) pH 7.4, and washed twice by centrifugation at 3500× g for 20 minat 4 • C. The obtained pellets were resuspended with 0.5 mL lysis buffer (1 mM EDTA, pH 7.4, 2% SDS, 40 mM DTT in PBS) with 0.5 µL Pierce Universal Nuclease (Thermo Fisher Scientific, Waltham, MA, USA) and 5 µL Halt Protease Inhibitor Cocktail (Thermo Fisher Scientific) added. The homogenates were sonicated in an Ultrasonic disintegrator, Soniprep 150 (MSE, London, UK), operating at 20% duty cycle and 3−4 output for 2 min. Lysates were clarified by centrifugation at 15,000× g for 30 min at 4 • C. Protein samples were processed by a filter-aided sample preparation (FASP) method according to Wisniewski [72]. The digested peptide samples were vacuum dried, dissolved in 0.1% formic acid and the peptide concentration was estimated by absorbance measurements at 280 nm by NanoDrop (Thermo Fisher Scientific) prior to liquid chromatography-tandem mass spectrometry (LC-MS/MS) analyses. The proteome assessments were repeated at least 10 times.

Proteomic Analysis by LC-MS/MS
Peptides were separated by reverse phase chromatography on a 20 mm × 100 µm C18 pre-column followed by a 100 mm × 75 µm C18 column with particle size 5 µm (NanoSeparations, Nieuwkoop, Netherlands) at a flow rate of 300 nL/min on EASY-nLC II (Thermo Fisher Scientific) by a gradient of 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B) as follows: from 2% B to 30% B in 60 min; from 30% B to 100% B in 60 min. Automated online analyses were performed in positive mode by LTQ Orbitrap Velos Pro hybrid mass spectrometer (Thermo Fisher Scientific) equipped with a nano-electrospray source with Xcalibur software (v.2.6, Thermo Fisher Scientific). Full MS scans were collected with a range of 350-1800 m/z, a resolution of 30,000 (m/z 200), the top 20 most intense multiple charged ions were selected with an isolation window of 2.0 and fragmented in the linear ion trap by collision-induced dissociation with normalized collision energy of 30%. Dynamic exclusion was enabled ensuring peaks selected for fragmentation were excluded for 60 s.

Database Searching
The generated raw files were analyzed using Sequest HT in Proteome Discoverer (Thermo Fisher Scientific, San Jose, CA, USA, CS version 1.4.0.288) and the translated genome sequences based on WGS. The following search parameters were used: trypsin as a digestion enzyme; maximum number of missed cleavages 2; fragment ion mass tolerance 0.50 Da; parent ion mass tolerance 10.0 ppm; carbamidomethylation of cysteine as fixed modification and methionine oxidation as variable modifications.

Data Evaluation and Label-Free Quantification
Identified proteins were validated using SCAFFOLD software (Version 4.4.8; Proteome Software Inc., Portland, OR, USA). Identifications were based on a minimum of 2 peptides, minimum 95% peptide identification probability (using the Scaffold Local FDR algorithm), and minimum 99% protein identification probability using the Protein Prophet algorithm [73]. Proteins, which contained similar peptides, and which could not be differentiated based on LC-MS/MS analysis alone were grouped to satisfy the principles of parsimony. The label-free quantitative analysis was performed using total number of spectral counts; normalization was performed to account for variations between samples. Quantitative differences were statistically analyzed by Student's t-test with the Benjamini-Hochberg correction. Differences with p-values ≤ 0.05 were considered statistically significant.

Protein Homology Search
Protein sequences were annotated with search in the NCBI database using the Basic Local Alignment Search Tool (BLAST, National Center for Biotechnology Information, Bethesda, MD, USA) blastp algorithm and Uniprot BLAST (EMBI-EBI, Cambridgeshire, UK) software (https://www.uniprot.org/, accessed on 20 May 2020).

Availability of Data and Materials
The raw WGS data has been deposited in the NCBI Sequence Read Archive (SRA) (http: //www.ncbi.nlm.nih.gov/sra) and can be accessed through the accession PRJNA657148.
The mass spectrometry data has been deposited to the ProteomeXchange Consortium (www.proteomexchange.org) via the PRIDE [74] partner repository with the dataset identifier PXD020218.

Conclusions
We here performed molecular and cellular profiling of colistin-susceptible A. baumannii isolate, harboring a variety of resistance genes, and its colistin-dependent subpopulation, providing new insights into colistin dependence. We also demonstrated, for the first time, the potential involvement of ISAjo2 and ISAba13 insertion sequences in colistin dependence, confirming a role for ISs in colistin treatment. Our results suggest that colistin dependence in A. baumannii is likely the result of complex cellular events that occur concurrently with ISs-triggered LOS-deficiency.

Supplementary Materials:
The following are available online at https://www.mdpi.com/1422-006 7/22/2/576/s1, Table S1. Minimum inhibitory concentrations (µg/mL) of Ab-S and Ab-D subpopulations. Table S2. PLSDB analysis for plasmids in Ab-S and Ab-D subpopulations of A. baumannii. Table  S3. ISs annotation by ISFinder Ab-S and Ab-D subpopulations of A. baumannii. Table S4. Proteins upregulated in Ab-D subpopulation in comparison to Ab-S. Fold changes were calculated with Ab-S as the reference category, which implies that values over 1 mean that a protein is upregulated in Ab-D. Table S5

Data Availability Statement:
The raw WGS data has been deposited in the NCBI Sequence Read Archive (SRA) (http://www.ncbi.nlm.nih.gov/sra) and can be accessed through the accession PR-JNA657148. The mass spectrometry data has been deposited to the ProteomeXchange Consortium (www.proteomexchange.org) via the PRIDE [74] partner repository with the dataset identifier PXD020218.

Acknowledgments:
We thank Professor Karl-Eric Magnusson for his careful and critical reading of the manuscript. We are very grateful to the Core Facility at the Faculty of Medicine and Health Sciences, Linköping University for support with equipment for mass spectrometry and electron microscopy. The Medical Research Council of Southeast Sweden supported this work.

Conflicts of Interest:
The authors declare no conflict of interest.