key: cord-0292040-k10dzhxi
authors: Huo, Wenwen; Busch, Lindsay M.; Hernandez-Bird, Juan; Hamami, Efrat; Marshall, Christopher W.; Geisinger, Edward; Cooper, Vaughn S.; van Opijnen, Tim; Rosch, Jason W.; Isberg, Ralph R.
title: Immunosuppression broadens evolutionary pathways to treatment failure during Acinetobacter baumannii pneumonia
date: 2022-01-03
journal: bioRxiv
DOI: 10.1101/2021.04.07.438861
sha: d0ccbf4d56c334128347cbaa0ce069a2dc24895d
doc_id: 292040
cord_uid: k10dzhxi

Acinetobacter baumannii is increasingly refractory to antibiotic treatment in healthcare settings. As is true of most human pathogens, the genetic path to antimicrobial resistance (AMR) and the role that the immune system plays in modulating AMR during disease are poorly understood. Here we reproduced several routes to fluoroquinolone resistance, performing evolution experiments using sequential lung infections in mice that are replete or depleted of neutrophils, providing two key insights into the evolution of drug resistance. First, neutropenic hosts acted as reservoirs for the accumulation of drug resistance during drug treatment. Selection for variants with altered drug sensitivity profiles arose readily in the absence of neutrophils, while immunocompetent animals restricted the appearance of these variants. Secondly, antibiotic treatment failure in the immunocompromised host was shown to occur without clinically defined resistance, an unexpected result that provides a model for how antibiotic failure occurs clinically in the absence of AMR. The genetic mechanism underlying both these results is initiated by mutations activating the drug egress pump regulator AdeL, which drives persistence in the presence of antibiotic. Therefore, antibiotic persistence mutations present a two-pronged risk during disease, causing drug treatment failure in the immunocompromised host while simultaneously increasing the emergence of high-level AMR.

Acinetobacter baumannii is a Gram-negative opportunistic pathogen, one of the high-53 priority ESKAPE organisms that are increasingly difficult to treat due to multiple antibiotic 54 resistance 1 . A significant proportion of healthcare-associated diseases caused by this group of 55 pathogens, such as ventilator associated pneumonia, is linked to their documented multi-drug 56 resistance (MDR) 2-5 . Of particular importance are patients in intensive care units (ICU) who are 57 critically ill and have depressed immunological clearance mechanisms that increase the risk of 58 infection by MDR pathogens 6,7 . As a consequence, the healthcare environment with its 59 immunologically compromised patients could provide a unique niche for selection of MDR isolates 60 8 . Overlaying these issues is the fact that for many patients in healthcare settings, antibiotic 61 treatment failure is common but is often unexplained, as resistant organisms cannot be identified 9 . 62

A. baumannii clinical isolates have demonstrated a remarkable ability to successfully battle 63 antibiotic treatment in the clinic, resulting from high intrinsic resistance to antimicrobials and the 64 acquisition of drug resistance elements by the organism 10,11,12, 13 . A critical missing link is a detailed 65 roadmap for the stepwise evolution of antibiotic resistance in the clinic, particularly in identifying 66

A. baumannii subpopulations most likely to give rise to drug treatment failure. Furthermore, it is 67 largely unknown if there exists a patient group that provides the reservoir for antimicrobial 68 resistance (AMR) acquisition. Particularly for healthcare-associated diseases, patient groups 69 susceptible to A. baumannii are by their nature compromised in a number of fashions, with the 70 potential for providing reservoirs for AMR evolution. The range of individuals with altered immune 71 function in these settings may allow for a diversity of host targets that can act as primary amplifiers 72 of resistance, with eventual spread to individuals with different sets of susceptibilities. Therefore, 73 as a model for healthcare-associated pneumonia we aimed to explore whether depletion of a single 74 arm of innate immunity in mice could help shape the antibiotic treatment outcome and support the 75 evolution of resistant organisms. 76

Fluoroquinolones (FQ) such as ciprofloxacin initially showed excellent activity against A. in addition to the patient's underlying condition and hospitalization status, prior exposure to 85 fluoroquinolones is also a risk factor for A. baumannii colonization and infection 17 , indicating that 86 resistance to this antibiotic class is linked to either increased pathogenic potential of the isolate or 87 is highly associated with acquisition of MDR. Unknown is whether there exist early adaptive 88 mutations that enable precursor populations of A. baumannii to act as ancestors to drug resistance. 89

There has been little study of whether AMR can be suppressed by the immune response. 90

Landmark mathematical modeling work argues that the immune response can largely limit the 91 outgrowth of persisters or other bacterial variants that exhibit intermediate resistance levels 18 . 92

Another study indicates that the cytokine response may control waves of AMR variants 19 . Given the 93 limited analysis of how antibiotic resistance evolves in the clinic, and the lack of a detailed 94 interrogation of the role played by the immune response in controlling selection of AMR, we sought 95 to identify the steps that lead to resistance in the presence or absence of a single arm of innate 96 immunity. The rationale behind this approach is that clinical antibiotic resistance is associated 97 with mutations located outside well-characterized drug targets, and these are difficult to identify 98 unrelated to resistance, as single colony isolates showed no increased survival in the presence of 170 CIP (Fig. 2E ). All the other mutations identified by sequencing the pools appeared unrelated to 171 resistance, as none survived incubation on solid medium containing 1g/ml CIP (Figs. 2B, 2E). 172

Furthermore, the analysis of individual colonies isolated after plating bacterial populations on 173 antibiotic-free medium after the 15 th mouse passage identified isolates with CIP MICs that were 174 only slightly elevated relative to the WT (Fig. 2E , left panel; WT MIC = 0.25-0.5). 175

Variants with altered CIP sensitivity arose in the immunocompetent mice, but they 176 remained at low levels in these populations, as predicted by the previous mathematical modeling 177 study ( Fig. 2A ) 18 . Mutations causing decreased drug sensitivity were identified by isolating single 178 colonies on solid medium containing 1g/ml CIP, and all twelve of the single colonies selected from 179 the 10 th passage had the identical mutation in adeL, the regulator of the AdeFGH egress pump 25, 26 . 180

The presence of the mutation in multiple Lineages is consistent with it existing in the initial 181 bacterial culture prior to the first inoculation of all three Lineages. Interestingly, 8 distinguishable 182

Lineage-specific strains were identified that were derived from the parental AdeL C312* mutation, 183 consistent with downstream mutations arising during mouse passage (Fig. 2E, right panel) . These 184 strains appeared to be largely lost by passage 15 in the immunocompetent mouse ( Fig. 2A, E) . We 185 conclude that although strains with mutations that activate the AdeFGH egress pump are detected, 186 they are unable to overgrow the WT and are eventually depleted during later passages in 187 immunocompetent mice (Fig. 2B) . 188

Consistent with the dynamic nature of resistance acquisition in neutrophil-depleted mice 189 ( Fig. 2B) , pool sequencing uncovered mutations that drove the stepwise trajectory to CIP R in later 190 passages (Fig. 2D ). Unlike the immunocompetent lineages, first step mutations that occurred within 191 adeL after 6 passages in neutrophil-depleted mice were easily detected by whole pool sequencing. 192

Each Lineage harbored different adeL mutations in these early passages that disrupted the 3' end of 193 the gene, consistent with previous work arguing that alterations in the C-terminus of the AdeL 194 protein result in AdeFGH pump activation 25, 26 . Mutations in various other genomic sites arose 195 between passage 9 and 12, with many predicted to contribute to decrease susceptibility to CIP 196 treatment. For instance, the adeL312* mutant acquired a second mutation in gyrB that overgrew 197 Lineage 1. In addition, a mutant harboring a noncononical mutation in gyrA arose in Lineage 2 that 198 was outcompeted by a strain with multiple genetic alterations, while a quadruple mutant appeared 199 to overgrow the single adeL mutant in Lineage 3. 200

To identify the various genotypes linked to increased CIP resistance, we isolated single 201 colonies after 9 passages to verify that two transiently predominant variants had the predicted 202 genotypes, and also sequenced isolates from passage 15 after plating in the absence or presence of 203 CIP (1µg/mL). Isolated colonies in absence of drug selection from Lineages 2 and 3 from passage 9 204 showed the predicted adeL alleles (compare Fig. 2D and 2F ). In addition, after plating the passage 205 15 pools in the absence of drug, the two predicted gyrase alleles gyrA(A117E) and gyrB(Q447H) 206 that are observed infrequently in the clinic were also identified (Fig. 2F ). Having gyrA(A117E) 207 alone was able to raise the CIP MIC by 16-fold ( Fig 2F) . Clones having the gyrB(Q447H) allele were 208 only observed linked to adeL(C312*) at passage 15, resulting in a CIP MIC that was 8-fold higher 209 than that of the parent strain ( Fig 2E) . Besides the mutations in target proteins, other single 210 colonies that resulted in a similar MIC increase were found to have a combination of four mutations 211 (Fig. 2F) . When comparing the single colonies isolated that had a CIP of 1 g/mL to those isolated 212 without antibiotic, the single colony isolates had genomic changes that were predicted by the 213 sequencing of the pools, indicating that the mutants described here became predominant without 214 requiring CIP selection ex vivo. 215

Mutations in adeL allow for persistence in the presence of high levels of ciprofloxacin. 216

To deconvolve the function of the various mutations observed, we backcrossed mutations into the 217 parent strain and assessed their relative contributions to CIP resistance. Of particular interest were 218 adeL nucleotide changes identified in populations after the 6 th passage in neutrophil-depleted 219 mouse lineages, as CIP failed to efficiently restrict growth in the lung in these passages (Figs. 1A, 220 2D). The adeL gene (ACX60_06025) 26 has two predicted domains often associated with LysR type 221 transcriptional regulators: a helix-turn-helix domain and a substrate binding domain responsible 222 for regulation of the AdeFGH pump (Fig. 3A) . Of the adeL mutations that arose during the in vivo 223 passages, one is located within the predicted substrate binding domain while the others are at the C 224 terminal end comprising in-frame deletions or early termination codons ( Fig. 3A and 3B ). We 225 constructed strains with each of these mutations and observed 100-1000X increased adeG 226 transcription levels compared to the parental strain, consistent with AdeFGH efflux pump 227 overexpression (Fig. 3C) . 228

While these single adeL mutations resulted in an increase in MIC to ~1 g/mL CIP that 229 barely rose to the level of significance (Fig. 4C) , this level was below the recognized clinical 230 breakpoint 27 , and such strains would be indicated as susceptible. As these are first step mutations, 231 we hypothesize that variants harboring these changes may facilitate the outgrowth of more 232 resistant isolates by promoting tolerance or persistence in the presence of antibiotic. To test this 233 model, we evaluated the survival of adeL mutants during exposure to high levels of CIP 28-32 . 234 Exposure to 10 g/mL CIP, roughly 20X MIC of the parental strain, led to the majority of the adeL 235 mutants (> 99%; MDK99) being rapidly killed (within 1 hour), mimicking what was observed for 236 the parental strain (Fig. 3D) . However, unlike the parental strain, a subpopulation of the adeL 237 mutant was able to persist through 24 hours of drug exposure (Fig. 3D ). This phenotype was 238 dependent on pump overproduction, as an adeL mutant strain deleted for the adeFGH operon was 239 indistinguishable from the parental strain in regards to persistence (Fig. 3D) . These data argue that showing increased MICs above the parental strain (Table S1 ; Fig. 4C ). The mfsA(-9) mutation is 254 located 9 bp upstream of a predicted MFS efflux pump, within 2 bases of a previously reported 255 mutation selected during evolution of CIP resistance in A. baumannii in planktonic conditions 34 . 256 Consistent with MFS transporter over-production, we found that the mfsA(-9) mutation increased 257 the mfsA mRNA levels by ~9-fold relative to the parental strain, based on q-rtPCR analysis (Fig. 4B) . 258

Significantly, when the mfsA(-9) mutation was combined with the adeLP131H mutation, the CIP MIC 259 increased to ~4ug/ml approaching that of the quadruple mutant strain isolated from mouse passage 260 15 ( Fig. 4C ; Table S1 ). A single nucleotide change located nearby the mfsA(-9) mutation that arose 261 transiently in the immunodepleted Lineage 3 ( Fig. 2D ; mfsA(-19)) behaved nearly identically (Figs. 262 4B,C). Therefore, combining two mutations that resulted in upregulation of both the AdeFGH pump 263 and a putative MFS transporter was sufficient to provide CIP resistance above the clinical 264 breakpoint, approaching that seen for the quadruple mutant. 265 A hotspot for mutations within lpxD is tightly linked to fluoroquinolone resistance in 266 clinical isolates. The mutation lpxD T118_A119insA (locus tag: ACX60_07955; GenBank: 267 AKQ26661.1) was consistently found in CIP R isolates from mice. This alteration within an 268 acyltransferase involved in lipooligosaccharide (LOS) biosynthesis 35-37 showed no effect on MIC 269 ( Interestingly, the alteration at residue E117 (z-score: 8.98; p-value, < 0.00001) directly 289 proceeds the T118 allele identified in our mouse experiments, with both variants predicted to be in 290 a turn between two beta sheets (Fig. 5D) 38 . Although the significance of this linkage is unclear, the 291 fact that the lpxD T118_A119insA has a fitness defect in broth (Fig. 5E ), raises the possibility that 292 mutations in this turn could contribute to a persistence phenomenon in tissues leading to evolution 293 of drug resistance. 294

Evolutionary replay experiments reconstruct the pathway to drug resistance. The 295 observation that resistant organisms outcompeted a persistent strain ( Fig. 2D ) could be due to a 296 coincidence that occurred during serial passage or due to antibiotic-driven selection for strains 297 with increased MIC over persistent strains. We performed three evolutionary replay experiments 40 298 with neutrophil-depleted Lineage 3 variants to distinguish between these possibilities. In the first 299 ( Fig. 6A ), bacteria were harvested after passage 9 and cycled 3 times in duplicate in neutrophil-300 depleted animals to test if the AdeL I335A336 allele, a 6 nucleotide in-frame deletion identified in 301 cyclophosphamide treated animals, would again be outcompeted by low abundance drug-resistant 302 mutants (Fig. 2D ). The second approach was to passage three times a mix of a colony-purified drug 303 persistent AdeL I335A336 single mutant (P9Cy3.1; Table S1 ) with a colony-purified drug resistant 304 quadruple mutant (Cy31; Table S1) in the approximate ratio present at passage 9 (95:5; Fig. 6B ). 305

The third was to test the model that neutrophils restrict the outgrowth of strains having activating 306 mutations in adeL (Fig. 6C) . 307

When the adeL mutant was analyzed in competition within the pool (Fig. 6A ) or when 308 mixed 95:5 with colony purified quadruple mutant (Fig. 6B) , we observed an increase in bacterial 309 burden and fraction of bacteria showing CIP MIC >2 after passaging the pools in neutrophil-310 depleted CIP-treated mice (Supp. Fig. 4 ). Whole genome sequencing of the bacterial populations 311 harvested after each passage revealed a near identical evolutionary trajectory as seen from the 312 original experiment, with the single mutant associated with drug persistence eventually 313 outcompeted by drug resistant mutants (Fig. 6) . Furthermore, in one of our replicate Lineages (Fig.  314 6A), another mutation located in the intergenic region between rpmL and mfsA (mfsA(-19)) arose 315 that had appeared transiently in immunodepleted Lineage 3 (Fig. 2D ), further arguing for the 316 contribution of this regulatory region in resistance evolution. Therefore, these two reconstructions 317 (Figs. 6A,B) demonstrate that the evolutionary trajectory is reproducible, and can be regenerated 318 by phenotypically similar mutations that arise spontaneously. 319

In the third experiment, single adeL mutants were unable to overgrow the WT in the 320 presence of CIP in immunocompetent animals, in contrast to what was observed in passage 9 of the 321 neutrophil-depleted animals. When a 5:95 (adeL:WT) mixture was passaged in immunocompetent 322 mice, the mutant was unable to increase its population share (Fig. 6C) , mimicking the failure of adeL 323 activating mutations to overgrow the pool in any of the immunocompetent Lineages ( Fig. 2A) . In 324 contrast, CIP R mutants having canonical resistance alleles in gyrA or gyrAparC were able to compete 325 efficiently with WT in the presence of neutrophils, with the GyrA(S81L) ParC(S84L) double mutant 326 growing from 5% to 70% of the population within 3 passages (Fig. 6C ). This argues that depleting 327 the immune response allows the outgrowth of persistence mutants, facilitating complex pathways 328 to drug resistance that are blocked in immunocompetent hosts. 329

Ciprofloxacin treatment within the host simultaneously selects for increased 330 resistance and enhanced fitness. Antibiotic resistance is often associated with a fitness cost 41 that 331 may be compensated over time by continued selection of secondary genotypes that overcome these 332 costs. The inability of the adeL mutant to overgrow WT during infection in the presence of CIP in 333 immunocompetent animals may reflect this point. Mutants with high resistance, however, could 334 maintain their selective advantage over non or low-resistance mutants even in the absence of drug 335 treatment. To test this model, the relative fitness of the drug resistant mutants derived from 336 neutrophil-depleted Lineages was compared to the parent strain AB17978. In the absence of CIP, 337 variants isolated after passage in neutrophil-depleted mice showed varying degrees of subtle 338 growth impairment in broth relative to the parental strain ( Fig. 7A ; Cy11, p< 0.01). During animal 339 infections in the absence of antibiotic, these growth defects were amplified. When 340 immunocompetent mice were challenged with the high fitness gyrAparC mutant in competition 341 with the WT, the two strains showed similar levels of colonization (Fig. 7B ). In contrast, each of the 342 drug resistant mutants derived from neutrophil-depleted mice competed poorly in 343 immunocompetent animals ( Fig. 7B ; p <0.0001). 344

Although the lowered fitness of mutants in immunocompetent animals further supports 345 the model that outgrowth of unique resistance alleles occurs in the absence of neutrophils, it did 346 not directly test the model that selection in neutropenic mice increased fitness relative to first-step 347 CIP R variants. In the absence of CIP treatment, the parent strain showed a subtle advantage over the 348 Lineage 3 quadruple mutant isolate Cy31 (Table S1) remained low in immunocompetent mice, and this did not lead to treatment failure (Fig. 1B) . 370

The absence of neutrophils resulted in the identification of rare alleles associated with drug 371 resistance. In contrast to the canonical target site GyrAS81LparCS84L mutation which shows high 372 fitness under conditions tested, the gyrAA117E and gyrBQ447H alleles rarely observed in clinical 373 strains predominated in two of the lineages from neutrophil-depleted animals ( Fig 2E) . These 374 unusual isolates showed fitness defects during disease and/or in culture (Fig. 7) that make them 375 unlikely to arise in the presence of intact immune functions, consistent with evolutionary 376 trajectories being determined by the immunological state of the host. Therefore, we propose that 377 immunocompromised hosts are incubators for the generation of unique drug resistant variants, 378 with uncertain outcomes after pathogen exposure to the community at large. Although the variants 379 we described have reduced fitness relative to WT, continued passage demonstrated that stepwise 380 increase in drug resistance was associated with stepwise increase in fitness (Fig. 7) . Similar fitness 381 effects have been obtained during continued passage in culture 34 . The recent demonstration of 382 enhanced viral evolution of SARS-CoV-2 in an immunocompromised patient undergoing 383 therapeutic antibody treatment is a graphic example of the potential interplay between immunity, 384 antimicrobials and evolution in the clinic, with potential largescale community effects 45 Persistence/tolerance mechanisms may play an important role in the relapse of bacterial infections 395 57-59 . In our study, we found that persistence in A. baumannii can be promoted by upregulation of a 396 drug efflux pump, AdeFGH 26 . It is known that clinical isolates often overproduce efflux pumps that 397 remove multiple antibiotics from the bacterial cytoplasm 26,60 , and that antibiotic persistence is 398 associated with such pump upregulation 61,62 . Of note, mutations upregulating this pump are rarely 399 isolated in the laboratory through simple antibiotic single-step selections on solid agar, indicating 400 there may be a special environment in the lung that allows outgrowth of these mutants. 401

There are three important repercussions from the analysis of the AdeL efflux overproducer 402 mutations. First, overproduction results in drug treatment failure during pneumonic disease in the 403 mouse, in spite of the fact that these bacterial strains are CIP-sensitive based on the international 404 standard for clinical breakpoints 27 . Therefore, clinical drug treatment failure in the absence of 405 identified antibiotic resistance 9 could be explained by the outgrowth of drug persistent mutants, 406 particularly in the immunocompromised host. Second, the adeL mutations provide a molecular 407 basis for A. baumannii to develop second-step variants that lead to CIP resistance above the clinical 408 breakpoint. Of interest, inhibitors of efflux pumps have been suggested to re-sensitize bacteria to 409 multiple antibiotics 63 , but perhaps such a regimen may also delay progression to resistance. Finally, 410

isolates from drug treatment failure occurring in immunocompromised patients could generate a 411 pool of precursor mutants giving rise to resistant isolates that eventually infect a broad range of 412 patients. A number of studies have similarly argued that efflux pump mutations provide a critical 413 pool for the multi-step evolution of antibiotic resistance 33,64,65 414

We propose a model of bacterial resistance progression in immunodepleted hosts (Fig. 8) . Genomic analyses can identify allelic variants linked to drug resistance, but verifying the 424 functional importance of these alleles is hindered by founder effects that are often difficult to 425 discount. The lpxD E117K mutation is one such allele associated with resistant gyrA S81L-426 containing genomes. The fact that the adjacent lpxD T118 insA allele was isolated during multiple 427 passages in immune-depleted mice argues for a role of these altered residues in supporting the 428 outgrowth of drug resistant organisms. The lpxD T118 insA mutation likely alters envelope 429 function. This raises the possibility that the mutation causes subtle changes in permeability that 430 slow drug access to target, or else the slowed growth of strains harboring this allele increases drug 431 tolerance during growth in tissues 52 . Our inability to demonstrate increased tolerance to CIP and 432 the fact that the clinical lpxD E117K did not show a fitness defect when grown in LB broth (Fig. 5E ) 433

argues that the critical phenotypes may be observed exclusively during growth in tissues, making 434 them difficult to evaluate. 435

In summary, we hypothesize that resistance progression in clinical isolates follows a similar 436 trend witnessed in our experiments, albeit with greater complexity in the human. Drug sensitive 437 bacteria may colonize both healthy and immune system-compromised patients, but resistance 438 evolution occurs more rapidly within hosts having impaired neutrophil function. The canonical 439 resistance mutations that arise within healthy individuals may compete efficiently with host-440 evolved mutations in patients having impaired neutrophil function. Hence, the transmission from 441 individuals having intact immune function may pose risks to immunocompromised patients. 442

Adding to the risks subjected to vulnerable patients, we found that A. baumannii persistence 443 mutants with MIC levels below the clinical breakpoint were associated with treatment failure, 444 emphasizing the difficulty in treating these patients (Fig. 1B) fresh LB broth and grown to exponential phase. A total of ~10 7 CFU were used for serial dilutions in 509 1x PBS and 10 l from each diluted culture was spotted on LB agar plates containing the following 510 concentrations of ciprofloxacin: 0, 0.25, 0.5, 1, 2, 4 and 8 g/ml. After an overnight incubation at 511 37C, colonies were counted, and CFU/ml was calculated. A detection limit of 100 CFU/ml was used. 512

The fraction bacteria resistant to a certain concentration (C 0 ; C 0 > 0) of ciprofloxacin is calculated as 513 (CFU/ml at C 0 -CFU/ml at all concentrations above C 0 ) / Total CFU. A stacked bar plot of the data 514 was generated using Prism GraphPad (Fig. 2) . 515 516 Whole Genome Sequencing. Genomic DNA (gDNA) was extracted from bacteria using the 517 DNeasy Blood and Tissue kit (Qiagen). Library preparation was performed using Illumina Nextera 518 with previously described modifications 73 . Libraries were enriched with 8 cycles of PCR and 519 sequenced using HiSeq2500 at Tufts University Core Facility using single-end 100 bp reads. Reads 520 were aligned to the A. baumannii ATCC 17978 genome and its plasmids (GenBank Accession: 521 CP012004; pAB1: CP000522; pAB2: CP00523; pAB3: CP012005), and variants were identified at 522 5% cutoff using breseq 0.32 74 . Reads were aligned to A. baumannii ATCC 17978 reference genome 523 (https://genomes.atcc.org/genomes/e1d18ea4273549a0) to include an additional 44kb gene 524 cluster 75 (Table S2) . 545

Triple technical replicates were examined per biological sample and at least three biological 546 replicates per strain were tested, with controls lacking reverse-transcriptase included to verify a 547 lack of contaminant genomic DNA. Transcript levels of specific targets from each strain were 548 evaluated by the comparative 2 -ΔΔCt method to the parental strain, normalizing to that of the 549 endogenous control 16s. The transcript level for each target across biological replicates was plotted 550 as mean + Standard Error of Mean (Fig. 3&4) . a Biotek plate reader with rotation. Growth was monitored by measuring A 600 at 15 min. intervals 558 for 16 hr, and the MIC was determined as the lowest concentration of drug that prevented growth, 559 using at least three biological replicates for each strain. The MICs across biological replicates were 560 plotted as mean + Standard Error of Mean (Fig. 4) . 561 562 Molecular cloning and mutant construction. Plasmids and primers used in this study are 563 listed in Table S2 . The mutant strains were constructed through sequential cloning into pUC18 77 564 then pJB4648 78 as previously described 13 . Briefly, the mutant allele from each strain of interest was 565 amplified alongside upstream and downstream segments to generate a PCR product ~1500 bp in 566 length flanked by appropriate restriction sites. The PCR product was then ligated into pUC18 and 567

propagated in E. coli DH5. The PCR product was then subcloned from pUC18 into pJB4648 and 568

propagated in E. coli DH5 pir. After sequence confirmation, the pJB4648 plasmid construct 569 containing the desired mutation was introduced into A. baumannii via electroporation. Markerless, 570 in-frame mutations were isolated via homologous recombination as described 13 . 571 572 Comparative sequence analysis. Genome sequences from 8666 A. baumannii clinical 573 isolates and ciprofloxacin resistance profiles for 2618 isolates were downloaded from PATRIC 574 database in Nov 2021 as described in results. MLST analysis was performed using Pasteur scheme 39 575 with the publicly available tool "mlst" (https://github.com/tseemann/mlst). The LpxD protein 576 sequence from ATCC 17978 was compared to these 2045 isolates using tblastn to identify the 577 amino acid locations of mismatches and gaps. The nucleotide sequences with the best matches to 578 the LpxD protein were extracted and translated into protein sequences. For each isolate, the 579 translated LpxD sequence was compared to the reference LpxD sequence from AB17978. The 580 nonsynonymous changes were recorded for each clinical isolate. The nonsynonymous changes at 581 each residue were then summarized to reflect the total number of clinical isolates showing those 582 changes. The raw numbers were further transformed into z-scores to determine if a particular 583 residue was significantly overrepresented by nonsynonymous changes linked to CIP R or CIP S 584 phenotypes (Supp. Fig. 1) . In order to reduce clonal effects, one clinical isolate per ST group was 585 selected randomly. Multiple aligned LpxD protein sequences are shown in Supp. Dataset 3. One 586 isolate had no hit for LpxD. The MLST profiles from 7 housekeeping genes were used to calculate 587 the Euclidean distance, and hierarchical clustering was determined using R (Fig. 5A) . A 588 phylogenetic tree from these 139 genomes was built using mashtree 79 and visualized in Geneious 589 Prime (https://www.geneious.com/) (Supp. Fig. 3) . The presence and absence of isolates was 590 summarized and visualized in a heatmap using Prism GraphPad v9 (Fig. 5B; Supp. Fig. 2) . Within 591 each group, the mismatches/gaps at each amino acid location within LpxD were displayed as 592 number of genomes having mismatches/gaps at that location. A table of the data was constructed  593 showing the specific amino acid in the rows, the genome groups harboring these mutations in the 594 columns, and the total number of mismatches/gaps in the cells. The results in each cell were then 595 normalized using z-scores with the mean and standard deviation from individual genome groups. 596

The z-score was plotted as function of each residue. the R 2 value was calculated for each growth curve assay (the R 2 ranges from 0 to 1, representing the 608 worst to the best fit). At least three biological replicates were performed for each test strain (Table  609 S3, Table S4 ) and the average doubling time and standard deviation was calculated. Statistical 610 significance was performed using one-way ANOVA followed by Dunnett's multiple comparisons. 611 612 613 Accession numbers. 614 The datasets generated during the current study are available in the Sequence Read Archive 615 (PRJNA485355). Detailed accession numbers for each sample are listed in Table 1 . to identify the genomic mutations acquired throughout passaging (after passage 3, 6, 9, 12 and 15). 652

The mutations were detected using a 5% abundance cutoff and filtered against the parent strain 653 Sequencing reads that support the findings of this study (Fig 2) 

No Drugs: No ESKAPE! An Update from the Infectious 789

Diseases Society of America

Phenotypic and Molecular Characterization of Acinetobacter baumannii 791

Clinical Isolates from Nosocomial Outbreaks in Los Angeles County

Antimicrobial-Resistant Pathogens Associated with Healthcare-794

Associated Infections Summary of Data Reported to the National Healthcare Safety Network at the 795 Centers for Disease Control and Prevention

Resistant Urinary Isolates of Escherichia coli from Outpatients Are Frequently Multidrug Resistant: 799 Results from the North American Urinary Tract Infection Collaborative Alliance-Quinolone 800

Resistance Study

Epidemiology of bacterial colonization at intensive care unit admission with 803 emphasis on extended-spectrum β-lactamase-and metallo-β-lactamase-producing Gram-negative 804 bacteria -an Indian experience

Acinetobacter baumannii: an emerging multidrug-resistant pathogen in 807 critical care

Virulence characteristics of 809 multidrug resistant biofilm forming Acinetobacter baumannii isolated from intensive care unit 810 patients

An increasing threat in hospitals: multidrug-resistant 812

Acinetobacter baumannii

The impact of initial antibiotic treatment failure: Real-world insights in 815 healthcare-associated or nosocomial pneumonia

Clinical and Pathophysiological Overview of Acinetobacter Infections: a 818 Century of Challenges

Relationship between beta-lactamase production, outer membrane 820 protein and penicillin-binding protein profiles on the activity of carbapenems against clinical 821 isolates of Acinetobacter baumannii

Multicity outbreak of carbapenem-824 resistant Acinetobacter baumannii isolates producing the carbapenemase OXA-40

Antibiotic modulation of capsular exopolysaccharide and 827 virulence in Acinetobacter baumannii

Mechanism of quinolone action and resistance

Fluoroquinolone resistance: 832 mechanisms, impact on bacteria, and role in evolutionary success

Rapid Determination of Quinolone Resistance in Acinetobacter spp

Acinetobacter infection

Exploring the collaboration between antibiotics and the immune 839 response in the treatment of acute, self-limiting infections

Rapid evolution and host immunity drive the rise and fall of carbapenem 842 resistance during an acute Pseudomonas aeruginosa infection

Fluoroquinolone Resistance in Acinetobacter baumannii

Prevalence of plasmid-mediated multidrug 848 resistance determinants in fluoroquinolone-resistant bacteria isolated from sewage and surface 849 water

Comparative proteomics analyses of 852

Acinetobacter baumannii strains ATCC 17978 and AB5075 reveal the differential role of type II

Lipid A Modification Systems in 888 Gram-Negative Bacteria

The firA gene, a locus involved in the expression of rifampicin 891 resistance in Escherichia coli

The firA gene of Escherichia coli 894 encodes UDP-3-O-(R-3-hydroxymyristoyl)-glucosamine N-acyltransferase. The third step of 895 endotoxin biosynthesis

Open-access bacterial population genomics: BIGSdb 900 software, the PubMLST.org website and their applications

Contingency and determinism in evolution: 903 Replaying life's tape

The fitness costs of antibiotic resistance mutations

Comparative genomics of clinical 907 strains of Pseudomonas aeruginosa strains isolated from different geographic sites

Genetic variation of a bacterial pathogen within individuals with 910 cystic fibrosis provides a record of selective pressures

Parallel bacterial evolution within multiple patients identifies 913 candidate pathogenicity genes

Persistence and Evolution of SARS-CoV-2 in an Immunocompromised Host

RelA Mutant Enterococcus faecium with Multiantibiotic Tolerance Arising 917 in an Immunocompromised Host

Bacterial Transcription as a Target for Antibacterial Drug 919

Acquired Antibiotic Resistance Genes: An Overview

hipA, a newly recognized gene of Escherichia coli K-12 that 924 affects frequency of persistence after inhibition of murein synthesis

HipBA-promoter structures reveal the basis of heritable multidrug 927 tolerance

A Physiological Basis for Nonheritable 929 Antibiotic Resistance

Mechanisms of Bacterial Tolerance and Persistence in the Gastrointestinal 931 and Respiratory Environments

Bacterial persistence as a 934 phenotypic switch

Persister cells, dormancy and infectious disease

Persisters: a distinct physiological state of E. coli

Isocitrate lyase mediates broad antibiotic 940 tolerance in Mycobacterium tuberculosis

Molecular mechanisms and 943 clinical implications of bacterial persistence

Persistent bacterial infections and persister cells

Formation, physiology, ecology, evolution and 948 clinical importance of bacterial persisters

RND-type efflux pumps in multidrug-resistant 951 clinical isolates of Acinetobacter baumannii: major role for AdeABC overexpression and AdeRS 952 mutations

Enhanced Efflux Activity Facilitates Drug Tolerance in Dormant Bacterial Cells

Mechanisms of bacterial persistence during stress 956 and antibiotic exposure

Efflux-Pump Upregulation: From Tolerance to High-958 level Antibiotic Resistance?

Heterogeneity in efflux pump expression predisposes 961 antibiotic-resistant cells to mutation

Role of AcrAB-TolC multidrug efflux pump in drug-resistance acquisition by 963 plasmid transfer

The impact of initial antibiotic treatment failure: real-world insights in 965 patients with complicated, health care-associated intra-abdominal infection

When antibiotics fail: a clinical and 968 microbiological perspective on antibiotic tolerance and persistence of Staphylococcus aureus

Targeting Antibiotic Tolerance, Pathogen by 971 Pathogen

The impact of initial antibiotic treatment failure: Real-world insights in 973 patients with complicated urinary tract infection

Murine Oropharyngeal Aspiration Model of 979 Ventilator-associated and Hospital-acquired Bacterial Pneumonia

Long-term experimental evolution in Escherichia coli. IV. 982 Targets of selection and the specificity of adaptation

A global regulatory 985 system links virulence and antibiotic resistance to envelope homeostasis in Acinetobacter 986 baumannii

Identification of mutations in laboratory-evolved microbes 988 from next-generation sequencing data using breseq

Identification of Two Variants of Acinetobacter baumannii Strain ATCC 991 17978 with Distinct Genotypes and Phenotypes

Predicting the functional effect of 994 amino acid substitutions and indels

Improved M13 phage cloning vectors and host 997 strains: nucleotide sequences of the M13mpl8 and pUC19 vectors

Identification of linked Legionella pneumophila 1000 genes essential for intracellular growth and evasion of the endocytic pathway

Mashtree: a rapid comparison of whole genome sequence files

The PATRIC Bioinformatics Resource Center: expanding data and analysis 1005 capabilities