Colistin Resistance Onset Strategies and Genomic Mosaicism in Clinical Acinetobacter baumannii Lineages

The treatment of multidrug-resistant Gram-negative infections is based on colistin. As result, COL-resistance (COL-R) can develop and spread. In Acinetobacter baumannii, a crucial step is to understand COL-R onset and stability, still far to be elucidated. COL-R phenotypic stability, onset modalities, and phylogenomics were investigated in a clinical A. baumannii sample showing a COL resistant (COLR) phenotype at first isolation. COL-R was confirmed by Minimum-Inhibitory-Concentrations as well as investigated by Resistance-Induction assays and Population-Analysis-Profiles (PAPs) to determine: (i) stability; (ii) inducibility; (iii) heteroresistance. Genomics was performed by Mi-Seq Whole-Genome-Sequencing, Phylogenesis, and Genomic Epidemiology by bioinformatics. COLR A. baumannii were subdivided as follows: (i) 3 A. baumannii with stable and high COL MICs defining the “homogeneous-resistant” onset phenotype; (ii) 6 A. baumannii with variable and lower COL MICs displaying a “COL-inducible” onset phenotype responsible for adaptive-resistance or a “subpopulation” onset phenotype responsible for COL-heteroresistance. COL-R stability and onset strategies were not uniquely linked to the amount of LPS and cell envelope charge. Phylogenomics categorized 3 lineages clustering stable and/or unstable COL-R phenotypes with increasing genomic complexity. Likewise, different nsSNP profiling in genes already associated with COL-R marked the stable and/or unstable COL-R phenotypes. Our investigation finds out that A. baumannii can range through unstable or stable COLR phenotypes emerging via different “onset strategies” within phylogenetic lineages displaying increasing genomic mosaicism.


Introduction
In the last twenty years, Acinetobacter baumannii has become a major threat not only as an important pathogen associated with nosocomial and community infections at various body sites, including the bloodstream, respiratory tract, urinary tract, surgical sites, and wounds [1], but also because it poses a huge challenge in clinical settings due to its intrinsic and acquired antimicrobial resistance. The development of multidrug resistance (MDR) in A. baumannii is due to genomic mutations, acquired antimicrobial resistance genes, or modifications in the expression of some genes [2]. Because of the significant increase in the isolation of carbapenem-resistant A. baumannii (CR-A. baumannii) and, until recently, the lack of active drugs against this microorganism, colistin (COL) has been used to treat CR-A. baumannii infections despite its nephrotoxicity, representing, in combination, a last-line therapy. Eventually and not unexpectedly, the emergence of colistin-resistant (COL R ) A. baumannii was reported worldwide [3] and it is caused by mutations or altered expression of diverse genes-mainly pmr and lpx genes [4] as well as galU [5]-or the acquisition of the plasmidic gene mcr [6]. Nonetheless, colistin resistance (COL-R) is still low or variable across countries, while, as reported by Karokostantis (2020), the prevalence of COL heteroresistance seems to be more common [3,7,8]. Heteroresistance is usually defined as the presence of subpopulations with MIC values higher (variably defined as equal to or more than two-to eight-fold) than the MIC in the main population, is often associated with the previous use of the drug, and represents major trouble in clinical settings as resistant subpopulations can emerge during treatment, resulting in treatment failure. Colistin heteroresistance in A. baumannii is even more alarming as it was detected even in isolates without prior exposure to colistin, suggesting that it could be an inherent characteristic of this species [8,9]. Moreover, a concentration of antibiotic >128 mg/L was shown to be necessary to prevent the onset of colistin heteroresistant A. baumannii mutants, as previously reported [10]; this is much higher than colistin serum concentrations (median 2.36 mg/L) associated with approximately 50% toxicity rates [10]. Adding to the complexity in fully understanding the COL R A. baumannii threat, there are cases where COL minimum inhibitory concentration (MIC) values vary over time for the same A. baumannii strain, ranging from values lower than the susceptibility cut-off to higher values [11] even in the absence of heteroresistant subpopulations.
As already stated, different studies showed that COL heteroresistance in A. baumannii is common, but the real frequency rate is not determined [12], probably due to the absence of easy-to-perform tests, ranging from 18% to 100% of COL heteroresistant clinical strains. Furthermore, studies do not usually perform either resistance induction assays or MIC tests on strains cultured for different days [13,14]. These lack of analyses could lead to a misunderstanding in the detection of the COL heteroresistance, as it may result in underestimated or overestimated since the induction assays could promote the raise of COL R subpopulation, whereas culturing resistant strains for a long time could result in the loss of COL-R.
Although some studies have already investigated the presence of well-known mutations in the genes responsible for COL-R in resistant and heteroresistant A. baumannii strains [3,[15][16][17], to the best of our knowledge no one has ever analyzed their whole genomes to group them based on their similarities and differences. Herein, we conducted a phenotypic and genomic characterization as well as investigated the phylogenetic relationship in A. baumannii samples displaying different levels of colistin resistance and heteroresistance, suggesting three different strategies that A. baumannii can use to become resistant.

COL MIC
In 10 repeated MIC assays, the 1R, 2R, and 3R strains showed a stable and high COL MIC of 128 mg/L defining the "homogeneous-resistant" onset phenotype, and hence Full-Resistance. Contrarily, in the same 10 times repeated MIC tests, the second group of A. baumannii strains showed unstable COL MICs, as follows: COL MIC values ranging from 1 to 64 mg/L, with the most frequent value being 8 mg/L in 4R; from 2 to 16 mg/L, with the most frequent value being 16 mg/L in 5R; from 2 to 128 mg/L, with most frequent values being 64 and 2 mg/L in 6R and 7R, respectively; from 2 to 32 mg/L, with the most frequent value being 32 mg/L in 8R; and from 1 to 128 mg/L with 4 mg/L as the most frequent value in 9R (Table 1).

COL Resistance Induction
Colistin resistance induction assays showed a drug-inducible COL-R phenotype reported as "COL-inducible" onset phenotype in 6R and 8R. The induced 6R and 8R variants, having a stable COL MIC > 256 mg/L, were considered as having Adaptive-Resistance. Strains 4R, 5R, 7R, and 9R showed no growth at colistin concentrations ≥ 2 mg/L (Table 1).

Hetero-Resistance
COL Population Analysis Profiles (PAPs) revealed the presence of subpopulations able to grow up to COL 64 mg/L in 4R, 5R, and 7R, and up to 128 mg/L in 9R A. baumannii, as shown in Figure 1. Furthermore, the presence of two morphologically different subpopulations, named COL R variant-1 and COL R variant-2, in COL 32 mg/L agar plates in 4R, 5R, and 7R as well as in COL 128 mg/L agar plates in 9R was also detected (Table 1), (Figure 2). This phenomenon was referred to as "heteroresistance" and identified the "subpopulation" onset phenotype. The COL R variant-1 and COL R variant-2 showed a 2 to 6-fold increase in COL MIC in 4R, 5R, 7R, and a 1-fold COL MIC increase in 9R (Table 1).

LPS Quantification and Cell-Envelope Charge
Lipopolysaccharide (LPS) and cell envelope net positive charge quantification tested in all strains on a normalized total number of bacterial cells are shown in Table 1.

LPS Quantification and Cell-Envelope Charge
Lipopolysaccharide (LPS) and cell envelope net positive charge quantification tested in all strains on a normalized total number of bacterial cells are shown in Table 1.

Phylogenetic Tree and Genomic Typing
Genomic Phylogeny (gPhyl) recognized three different lineages, as shown by the Whole Genome SNP-based phylogenetic tree ( Figure 2). 5R, 6R, 7R, 8R, and 9R A. baumannii were grouped in gPhyl lineage-I, 3R, and 4R in gPhyl lineage-II and 1R and 2R in R Figure 2. Representative Morphologies of the COL R A. baumannii variants detected in 4R, 5R, and 7R detected on 32 mg/L COL-agar plates and 128 mg/L COL-agar plates in 9R strains.

LPS Quantification and Cell-Envelope Charge
Lipopolysaccharide (LPS) and cell envelope net positive charge quantification tested in all strains on a normalized total number of bacterial cells are shown in Table 1.

Phylogenetic Tree and Genomic Typing
Genomic Phylogeny (gPhyl) recognized three different lineages, as shown by the Whole Genome SNP-based phylogenetic tree ( Figure 2). 5R, 6R, 7R, 8R, and 9R A. baumannii were grouped in gPhyl lineage-I, 3R, and 4R in gPhyl lineage-II and 1R and 2R in gPhyl lineage-III ( Figure 3). gPhyl lineage-I only showed unstable COL R A. baumannii, gPhyl lineage-II had one stable and one unstable COL R A. baumannii, whereas gPhyl lineage-III only included stable COL R A. baumannii.
10, x FOR PEER REVIEW   Table 2).   All A. baumannii lacked the CRISPR/Cas systems, with the exception of 3R carrying 3 questionable (not confirmed) CRISPR regions with 1 spacer, and 6R having 2 questionable (not confirmed) CRISPR regions with 1 spacer, with both strains lacking Cas genes. Abundant mobile genetic elements including Insertion Sequences and Transposons were found in all A. baumannii, whereas the vB_AbaS_TRS1 and the Bϕ_B1251 prophages were only found in 1R, 2R, 3R, and 4R. All strains were assigned to the genomic macrorestriction profile "A", being indistinguishable by PFGE (Table 2).
Genomic Single Nucleotide Polymorphisms (SNPs) mapping on A. baumannii ACICU RefGen returned different nsSNPs in genes previously associated with COL-R mechanisms in A. baumannii and other Gram-negative species.
In detail, all unstable members had a characterizing moderate impact effect (MI) galU nsSNP determining the I245T AminoAcid (AA) change, whilst only 3 out of 6 showed pmrB nsSNPs. All stable COL full resistant members had diverse MI nsSNPs in both galU and pmrB. Furthermore, all the stable COL full resistant members had the galU nsSNP determining the I273V associated, in two strains, to the Q140L, along with two AA changes in PmrB (P170L and A138T) and one in LpxA (N5V) ( Table 1).
No nsSNPs were detected in pmrAC, lpxDCB, and proB in all A. baumannii on A. baumannnii ACICU RefGen mapping.

Discussion
The outbreak of antibiotic resistance and the emergence of MDR A. baumannii highlight the need to understand how resistant strains emerge and spread. The overuse and inappropriate consumption of colistin strongly correlate with the emergence of COL R A. baumannii [18].
Colistin resistance in A. baumannii has been reported worldwide, with the highest resistance rates in Asia and then in Europe [19]; however, the rate of colistin heteroresistance in A. baumannii turned out to be higher than that of resistance [9,[20][21][22][23].
In this context, we investigated the stability of COL-R phenotypes in A. baumannii phenotypically outlining the different onset strategies and genomically characterizing the phylogenetic lineages having different "stability and onset phenotypes".
Our data, for the first time, described the phenomenon of the variable behavior of the COL-R phenotype in a group of clinical A. baumannii. Our observations of the MIC values, repeated several times, lead us to conclude that a stable COL-R phenotype was exhibited only in 3 out of 9 A. baumannii. In A. baumannii, high levels of COL-R (MIC 128 mg/L) were detected within the entire bacterial population emerging, thus, with a "homogeneousresistant" onset phenotype, and hence considered as "Full-Resistance". The Full-Resistant A. baumannii were already described and our data was in agreement with previously published data and well-characterized strains [2,24]. Contrarily, the COL-R phenotype instability was described by our group, for the first time, in 6 A. baumannii showing mainly variable and lower COL MICs ranging from 1 to 128 mg/L. This original observation opens a new perspective in the understanding of the COL-R emergence strategies under colistin pressure and could be crucial for the clinical implication in the treatment of severe MDR Gram-negative infections.
The unstable COL-R phenotype was related by our group to two different onset strategies, i.e., a "COL-inducible" onset phenotype, considered as an "Adaptive Resistance" -in agreement to prior observations [25,26]-and a "Subpopulation" onset phenotype considered as "Heteroresistance" previously described by another group [27]. In our A. baumannii, the "COL-inducible" onset-phenotype emerged via an induction mechanism related to a constant long term colistin exposure, whilst the "Subpopulation" onset-phenotype emerged spontaneously in a subset of populations with higher COL-R levels, detectable as morphological variants with different COL MICs, exclusively in the presence of high COL concentrations (i.e., ≥COL 32/128 mg/L agar plates).
No correlation was found between the COL-R stability and the different onset strategies with LPS amount and cell-envelope charge. Therefore, we hypothesized that LPS alterations and perturbations in the negative envelope charge can exist as single or coexisting features, although in a strain-dependent manner and independently to the stability of the COL-R phenotype. In our opinion, the co-occurrence of a pool of the two factors might affect the envelope structure and charge both directly and/or indirectly.
Genomic Phylogeny arises firstly new consideration about the dynamics of COL R stable phenotype emergence, leading us to evidence that the progression from an unstable and stable phenotype appeared related to specific phylogenetic lineages characterized by high genomic mosaicism due to acquisition or loss of MGEs or Phages. The Genomic Phylogeny clustered our clinical A. baumannii sample in 3 phylogenetic lineages and the phylogram indicated gPhyl lineage-I as the common ancestor of gPhyl lineage-II and gPhyl lineage-III. Interestingly, gPhyl lineage-I only included unstable COL R A. baumannii with hetero-or adaptive COL-R, gPhyl lineage-II clustered one stable, homogeneous, COL Full-Resistant A. baumannii together with one unstable, heteroresistant COL R A. baumannii, whereas gPhyl lineage-III only included stable, homogeneous Full-Resistant COL R A. baumannii. Notably, the analysis of the evolutionary relationship of the gPhyl lineages clearly showed that gPhyl lineage-II, including unstable and stable COL-R phenotypes, may be considered as a "hybrid lineage" branching out from gPhyl lineage-I as a midpoint in the evolution of these strains, whereas gPhyl lineage-III, last differentiated, with only stable Full-Resistant COL R A. baumannii, may be considered as an endpoint in the COL R A. baumannii strain evolution, in agreement with the progression of COL-R stability acquisition.
This described phylogenetic evolution for the first time supports the hypothesis that the acquisition of a stable, homogeneous, full COL-R moves its first steps in the subpopulation or the adaptive onset strategy that progresses until reaching stability and homogeneity.
Furthermore, the gPhyl lineage-I A. baumannii strains belonged to the ST-1808 and ST-348, KL-9, OCL-1, and resistome-IV type; the gPhyl lineage-II A. baumannii strains belonged to diverse MLSTs (ST-1816 and ST-195 for 3R and ST-218 for 4R), KL-3 or KL-28, OCL-1, and resistome-II or III type; the gPhyl lineage-III A. baumannii strains belonged both to the ST-1839, KL-22, OCL-3, and resistome-I type, showing that belonging to a specific gPhyl lineage is related to intrinsic clonality of its members. By contrast, the mobilome and prophage profile showed a mosaic of mobile genetic elements (MGEs) and phages ranging from a higher number of transposable elements, such as 6-8 different insertion sequence elements (ISs) and one transposon (Tn6207) in gPhyl lineage-I unstable COLR A. baumannii, to a lower IS set (3-6) and one transposon (Tn6080 or Tn2006) associated to one and/or two different phages (vB_AbaS_TRS1 and the Bϕ_B1251) as in gPhyl lineage-II and III. The MGE and phage toolbox showed dynamism inside the highly clonal conserved A. baumannii genomes, resulting in increased variability and complexity related to the transition from an unstable to a stable COL-R phenotype, besides being responsible for genome evolution and high antimicrobial adaptation ability. Furthermore, resistomics highlighted that the gPhyl lineage-III-with its stable, homogeneous and full resistancesurprisingly appeared short of acquired resistance determinants, such as the genes involved in tetracycline and macrolide resistance, with respect to the other two mainly unstable phylogenomic lineages.
Furthermore, we can speculate that genetic variations in genes already related to polymyxin B (PB) and COL resistance seem, for the first time, related to the progression towards a stable COL R phenotype. New mutations in galUencoding an enzyme related to polymyxin B (PB) resistance in Proteus mirabilis [28]-and in pmrB, involved in the COL-R in A. baumannii [4] and Enterobacterales [29], were found out in the sample. New nsSNPs, computationally predicted affecting protein functionality, were mapped in GalU ligand binding site I245T in the unstable COL R and I273V in the stable one. GalU-UTPglucose-1-phosphate uridylyltransferase is a key enzyme involved in the pathway of the lipopolysaccharide biosynthesis, which is part of outer membrane biogenesis [30]. P. mirabilis knockout mutants in GalU were found to be extremely sensitive to PB, presumably because of changes in LPS [29]. Likewise, no mcr-mediated resistance mechanisms are implicated in the stability of COL-R phenotypes as mcr-genes were not found [6].
Of note, the genomic snapshot of the three different "COL-R onset phenotypes showed that the progression from COL-R instability to stability seems to be associated with a gradual genome evolution from "low-complexity" to "high-complexity" lineages through a pool of complex rearranged genomes.
In conclusion, our investigation finds out a new correlation between the in vitro COL-R stability and onset strategies as well as the membership to specific phylogenetic lineages evolving through complex genomic rearrangements involving whole-genome SNPomes, MGEs, Resistomes, Phage Profiling.
Our data demonstrated, for the first time, the association between the shift to unstable to stable COL-R phenotypes and increasing genomic mosaicism within the intrinsic high clonality of COL R A. baumannii. Likewise, new mutations putatively affecting the stability and functionality of GalU and PmrB can speculatively affect the dynamics of the COL-R stability acquisition.

Bacterial Strains
Nine clinical Extensively Drug-Resistant (XDR) A. baumannii, named 1-9R, were isolated from patients treated with sodium colistithemate for a long time, 7-14 days, at the Catania and Pisa University Hospitals (Table 3)  [31] mass spectrometry (Bruker Daltonics, Billerica, MA, USA). Antibiotypes were determined by the Phoenix system (Becton, Dickinson and Company, MD, USA) [32] and the results were confirmed by the Sensititre system (Thermo Fisher Scientific, Waltham, MA, USA) [33]. The strains were stored in glycerol at −80 • C until further analysis.

COL Minimum Inhibitory Concentrations (MICs)
Colistin MICs were determined by microdilution assay using sulfate colistin ≥ 15,000 U/mg (Sigma-Aldrich, Castle Hill, New Galles, Australia) in cation-adjusted Muller-Hinton broth (BBLTM MHBCA II; Becton, Dickinson and Company, MD, USA) as recommended by the Clinical and Laboratory Standards Institute (CLSI) [34] and the EUCAST guidelines. The results were interpreted according to the current colistin breakpoints for A. baumannii reported in EUCAST v9.0 [35] (≤2 mg/L for COL-susceptibility (COL-S) and >2 mg/L for COL-resistance (COL-R)). Pseudomonas aeruginosa ATCC27853 and Escherichia coli ATCC25922 were used as control strains for these assays. MIC assays were repeated 10 times almost every week for three months.

COL Resistance Induction Assay
COL-R induction assays were performed by growing daily A. baumannii strains at COL concentrations gradually doubled from COL 0.5 mg/L to COL 32 mg/L in Brain-Heart Infusion Broth (BHIB-Oxoid, Basingstoke, UK). Then, colistin resistance was re-evaluated by plating cultures in COL 16 mg/L agar-plates and the COL MICs were re-evaluated as described above. Each COL-R induction assay was performed in three biological replicates.

COL Hetero-Resistance Detection
Population Analysis Profiles (PAPs) were defined to evaluate the presence of A. baumannii subpopulations with a diverse rate of COL-R [9,36]. PAPs were performed by plating 50 µL of a 10 8 CFU/mL bacterial suspension in Mueller-Hinton (MH) agar containing serial dilutions (2-128 mg/L) of sulfate colistin. After 48 h of incubation at 37 • C, the presence of "differential rate" COL-R subpopulations were evaluated. A. baumannii ATCC19606 was used as the control strain. For each subpopulation, "differential rate" COL-R detection was performed in triplicate.

Surface Charge Determination
The cytochrome c binding assay was performed as a surrogate measure of the relative net positive surface charge of the strain as previously described [37,38]. Briefly, cells were grown overnight in TSB media to an OD 575 = 10 (2 × 10 10 CFU/mL), washed with 20 mM MOPS buffer (pH 7.0) three times, and re-suspended in the same buffer at OD 575 = 10 (2 × 10 10 CFU/mL). Cells were incubated with 0.5 mg/mL cytochrome c for 10 min and the amount of cytochrome c remaining in the supernatant was determined spectrophotometrically at OD 530 nm. The greater the amount of unbound cytochrome c detected in the supernatant, the higher the net positive charge of the bacterial surface. Data were converted and expressed as the mean (±SD) amount of bound cytochrome c. At least three independent runs were performed on separate days.

Direct LPS Quantification
LPS was extracted by the hot phenol-water method as previously described [39,40]. Briefly, each A. baumannii culture was grown up to OD 650 = 10 (2 × 10 10 CFU/mL) in Cation adjusted Mueller Hinton broth, and cells were harvested by centrifuging the tubes at 10,000× g for 5 min. The pellets were washed twice in PBS (pH = 7.2, 0.15 M) containing 0.15 mM CaCl 2 and 0.5 mM MgCl 2 , then they were resuspended in 10 mL PBS and sonicated for 10 min on ice. Treatments with proteinase K (100 µg/mL) at 65 • C for 1 h, and with DNase (20 µg/mL) and RNase (40 µg/mL) incubating the pellet at 37 • C overnight in the presence of 1 µL/mL 20% MgSO 4 and 4 µL/mL chloroform, were performed prior to extraction step to eliminate contaminating protein and nucleic acids. Subsequently, an equal volume of hot (65-70 • C) 90% phenol was added to the mixtures followed by vigorous shaking at 65-70 • C for 15 min. Suspensions were then cooled on ice, transferred to 1.5 mL polypropylene tubes, and centrifuged at 8500× g for an additional 15 min. Supernatants were transferred to 15 mL conical centrifuge tubes and phenol phases were re-extracted using 300 µL dH 2 O. Sodium acetate at 0.5 M final concentration and 10 volumes of 95% ethanol were added to the extracts and samples were stored at −20 • C overnight to allow LPS precipitation. On the last day, tubes were centrifuged at 2000× g 4 • C for 10 min and the pellets were resuspended in 1 mL dH 2 O. LPS absorbance at OD 259 was measured on the same day in a spectrophotometer (Thermo Scientific™ GENESYS 10S UV-Vis, Waltham, MA, USA), and the quantification was obtained by comparison with a standard curve made with scalar dilutions of extra pure E. coli LPS (Sigma, St. Louis, MO, USA). or capsule), was used to investigate the bacterial surface polysaccharide locus typing and variant evaluation [49,50].

Single Nucleotide Polymorphisms (SNPs)
Single Nucleotide Polymorphism (SNP) calls were carried out from the PE library raw reads as already published [2]. A. baumannii ACICU (CP000863.1) genome was used as the reference genome sequence for SNP mapping.

Genomic Single-Nucleotide Polymorphism Effect Prediction
The prediction of the whole-genome SNPs (SNPome) effects was performed by SnpEff (v.4.3T), a genomic variant annotation and functional effect prediction toolbox as previously published [51,52].