Co‐evolutionary adaptations of Acinetobacter baumannii and a clinical carbapenemase‐encoding plasmid during carbapenem exposure

Abstract OXA‐23 is the predominant carbapenemase in carbapenem‐resistant Acinetobacter baumannii. The co‐evolutionary dynamics of A. baumannii and OXA‐23‐encoding plasmids are poorly understood. Here, we transformed A. baumannii ATCC 17978 with pAZJ221, a bla OXA−23‐containing plasmid from clinical A. baumannii isolate A221, and subjected the transformant to experimental evolution in the presence of a sub‐inhibitory concentration of imipenem for nearly 400 generations. We used population sequencing to track genetic changes at six time points and evaluated phenotypic changes. Increased fitness of evolving populations, temporary duplication of bla OXA−23 in pAZJ221, interfering allele dynamics, and chromosomal locus‐level parallelism were observed. To characterize genotype‐to‐phenotype associations, we focused on six mutations in parallel targets predicted to affect small RNAs and a cyclic dimeric (3′ → 5′) GMP‐metabolizing protein. Six isogenic mutants with or without pAZJ221 were engineered to test for the effects of these mutations on fitness costs and plasmid kinetics, and the evolved plasmid containing two copies of bla OXA−23 was transferred to ancestral ATCC 17978. Five of the six mutations contributed to improved fitness in the presence of pAZJ221 under imipenem pressure, and all but one of them impaired plasmid conjugation ability. The duplication of bla OXA−23 increased host fitness under carbapenem pressure but imposed a burden on the host in antibiotic‐free media relative to the ancestral pAZJ221. Overall, our study provides a framework for the co‐evolution of A. baumannii and a clinical bla OXA−23‐containing plasmid in the presence of imipenem, involving early bla OXA−23 duplication followed by chromosomal adaptations that improved the fitness of plasmid‐carrying cells.


| INTRODUC TI ON
Acinetobacter baumannii is a Gram-negative nosocomial pathogen that has acquired resistance to multiple antibiotics, making clinical therapy challenging (Wong et al., 2017). Carbapenem-resistant A. baumannii (CRAB) mostly affects debilitated patients, resulting in increased morbidity, mortality, and healthcare costs (Hsu et al., 2017;Partridge et al., 2018). The emergence and prevalence of CRAB is a global public health threat because carbapenems are powerful, broad-spectrum β-lactams widely regarded as antibiotics of last resort. Overexpression of genes encoding oxacillinases (OXAs) is the most common carbapenem resistance mechanism in A. baumannii (Hamidian & Nigro, 2019;Wong et al., 2017).

Enzymatic hydrolysis of carbapenems mediated by OXAs is slow.
However, strong promotors provided by insertion sequence (IS) elements inserted immediately upstream of bla OXA genes can lead to overexpression resulting in clinical resistance (Tietgen et al., 2019).
Mobile bla OXA genes bracketed by IS elements in composite transposons such as Tn2006, Tn2007, Tn2008, and Tn2009 can be transferred horizontally and play an important role in the dissemination of carbapenem resistance in A. baumannii (Hamidian & Nigro, 2019;Yoon et al., 2017).
Plasmids harboring carbapenem resistance genes are important contributors to the spread of carbapenem resistance in A. baumannii Kuo et al., 2018;Liu et al., 2021). The carbapenemase gene bla OXA−23 is predominant in Chinese CRAB isolates and the 77,530-bp conjugative plasmid pAZJ221 is an important vector for horizontal transfer of bla OXA−23 (Liu et al., 2015). The 1164 bp replication initiation (rep) gene of pAZJ221 is identical to that of A. baumannii plasmid pABTJ1 (Huang et al., 2012), and pABTJ1 has been designated the reference plasmid for group GR25 under a typing scheme for A.
baumannii plasmids based on the nucleotide identity of rep genes (Bertini et al., 2010;Salgado-Camargo et al., 2020). pAZJ221 carries a complete set of genes associated with conjugation and the composite transposon Tn2009, which contains bla OXA−23 bracketed by ISAba1 (Liu et al., 2015). While many conjugative plasmids carry beneficial traits, they can exert physiological fitness costs-of-carriage resulting in reduced growth rates relative to plasmid-free strains (Harrison & Brockhurst, 2012). A number of studies have demonstrated that these fitness costs can be compensated for through adaptive evolution, and compensation may be a key mechanism driving the spread of successful antibiotic-resistant strains (Hall et al., 2021;MacLean & San Millan, 2019;Porse et al., 2016).
Extraordinarily high levels of antibiotic consumption in hospitals impose a strong selective pressure that drives the evolution and spread of antibiotic resistance determinants in nosocomial pathogens (San Millan, 2018). Thus, we hypothesized that plasmid-host adaptations between the clinically relevant plasmid pAZJ221 and its A. baumannii host would occur in the presence of a carbapenem.
To elucidate the mechanisms of compensatory adaptations, the widely available and well-characterized laboratory strain A. baumannii ATCC 17978 was used as a host (Ridenhour et al., 2017;Santos-Lopez et al., 2019;Zlamal et al., 2021). We subjected A. baumannii ATCC 17978 containing pAZJ221 (ATCC 17978/pAZJ221 hereafter) to experimental evolution at a sub-inhibitory concentration of imipenem for nearly 400 generations and applied population-wide whole-genome sequencing (WGS) to elucidate the dynamics of genomic adaptations. We constructed mutants to determine the role of specific mutations or an evolved variant of the plasmid in the coevolutionary process. Our work sheds light on the dynamics of the establishment of pAZJ221 in a new A. baumannii host and verified the effects of parallel chromosomal mutations and plasmid evolution on host fitness and plasmid characteristics.

| Bacterial strains and media
To construct the ancestral strain for the subsequent adaptive evolution experiment, we introduced the plasmid pAZJ221 to A. baumannii ATCC 17978 through electroporation. The transformed strain was called ATCC 17978/pAZJ221. Mueller Hinton (MH) Broth (Oxoid, UK) or Mueller Hinton (MH) Agar (Oxoid, UK) were used throughout this study.

| Minimum inhibitory concentration (MIC) measurements
MIC assays were performed by the agar dilution method. Results of susceptibility testing were interpreted according to the Clinical and fitness under carbapenem pressure but imposed a burden on the host in antibioticfree media relative to the ancestral pAZJ221. Overall, our study provides a framework for the co-evolution of A. baumannii and a clinical bla OXA−23 -containing plasmid in the presence of imipenem, involving early bla OXA−23 duplication followed by chromosomal adaptations that improved the fitness of plasmid-carrying cells.

K E Y W O R D S
Acinetobacter baumannii, conjugative plasmid, experimental evolution, fitness, horizontal gene transfer, OXA-23, plasmid-host adaption Laboratory Standards Institute guidelines (CLSI, 2021). Escherichia coli ATCC 25922 was included in all MIC assays as a control strain.

| Preparation of electrocompetent cells
A fresh overnight culture was diluted 1:100 in 20 ml MH broth. Cells were grown at 37°C shaking at 200 rpm to logarithmic phase. Cells were harvested by centrifugation, suspended in 20 ml sterile H 2 O, and placed on ice for 30 min. Cells were collected by spinning for 3 min at 5000 rpm at 4 °C and washed with 20 ml 10% glycerol. This spinning and washing procedure was repeated three times. Cells were harvested by centrifugation and the supernatant was discarded, after which cells were suspended in 2 ml 10% glycerol, transferred to microfuge tubes (100 μl/tube), and stored at −70°C until use.
Electrocompetent cells stored at −70°C were thawed on ice. 200 ng plasmid DNA was added, and cells were transferred to the cuvette.
Electroporation conditions were 1800 V, 25 μF, and 200 Ω. After electroporation, 1 ml MH broth was immediately added to the cuvette and the contents were transferred to sterile culture tubes for 1-h growth, then the culture was spread on MH agar plates containing selective antibiotics, and the plates were cultured at 37 °C overnight. Transformants were verified by PCR and Sanger sequencing.

| Evolutionary experiment
Three independent colonies of ATCC 17978/pAZJ221 were incubated in MH broth containing 16 mg/L imipenem and shaking (200 rpm) at 37°C for approximately 20 h. Every day, 20 μl overnight culture containing approximately 10 7 cells was taken and passaged into 2 ml fresh media at a 1:100 dilution ratio. Three control groups were set for the mutation analysis, each containing four biological replicates: (i) ATCC 17978 without pAZJ221 under sub-MIC imipenem pressure (0.0625 mg/L); (ii) ATCC 17978 without pAZJ221 in antibiotic-free MH broth; (iii) ATCC 17978/pAZJ221 in antibiotic-free MH broth. All populations were passaged at a 1:100 dilution ratio daily (i.e., approximately 6.6 generations per day) for a total of 60 days (approximately 400 generations in total) (Cooper, 2018). Every 10 days, stocks were generated for each population and stored frozen at −70°C.

| Whole-genome sequencing and sequence analyses
To identify the mutations in the populations and their allele frequencies, total DNA was extracted using a QIAamp DNA Minikit (Qiagen, Valencia, CA, USA) from a fresh culture inoculated from frozen stocks (see above). Eighteen populations of the experimental group and 71 populations of control groups were sequenced. Quality and quantity of genomic DNA were determined by visualization on agarose gels and spectroanalysis using a NanoDrop ND-1000 (ThermoFisher, Waltham, MA, USA). A 300-bp library for Illumina paired-end sequencing was constructed using a Paired-End DNA Sample Prep Kit (Illumina Inc.). The reads coverage of evolving populations on each reference sequence is listed in Table S1.
Mutations and their frequencies were determined using the software Breseq 0.27.1 with the default parameters and the -p flag (Deatherage & Barrick, 2014). IS element rearrangements in population samples were predicted by ijump (https://github.com/sleyn/ ijump). Reference sequences were downloaded from NCBI (ATCC 17978-mff: CP012004.1, pAB1: CP000522.1, pAB2: CP000523.1, pAB3: CP012005.1, pAZJ221: KM922672.1). We corrected the published ATCC 17978-mff reference backbone for the variants that were detected in our ATCC 17978 strain (SRR12569044), which had been sequenced by Illumina paired-end sequencing. Mutations were manually curated and filtered under the following criteria: (i) mutations were filtered if the gene or the intergenic region was found to contain an identical nucleotide mutation in any control group or the ancestor strain; (ii) mutations that did not reach a cumulative frequency of at least 20% across all populations were removed; (iii) mutations that did not change in frequency over time, indicating mismapped reads or sequencing errors. The criteria were set in line with a published study (Scribner et al., 2020). Strains with parallel mutations were screened by isolating separate colonies from the corresponding evolving populations, and parallel mutations were verified by PCR and Sanger sequencing. Primers are listed in Table S2.
The locations of small RNAs (sRNAs) were predicted by "AcinetoCom" (Kröger et al., 2018), and RNA secondary structures were predicted by MXfold (Sato et al., 2021). Putative target mRNAs for sRNAs were predicted by TargetRNA2, and the COG categories of target genes were annotated by eggnog-mapper (Cantalapiedra et al., 2021;Huerta-Cepas et al., 2019;Kery et al., 2014). The conservation degree of a gene product among A. baumannii strains was determined using BLASTp to query the nonredundant protein sequences database on the NCBI server. Signal peptide and cleavage sites were predicted using SignalP 6.0 (Teufel et al., 2022). Conserved domains of proteins were analyzed by Conserved Domain Database (CDD) and SMART Marchler-Bauer et al., 2017;Teufel et al., 2022). HHpred tool was employed for protein remote homology detection (Zimmermann et al., 2018). The interpretation of mutation effects and the molecular modeling were performed by AlphaFold 2.0 (Jumper et al., 2021).
To confirm the existence of a bla OXA−23 duplication, two single colonies were isolated from lineage 2 on day 10 and day 60, respectively. Strains were sequenced using the Nanopore MinION platform (Oxford Nanopore Technologies, Oxford, United Kingdom). Longread library preparation for Nanopore sequencing was performed with a 1D sequencing kit (SQK-LSK109; Nanopore) without fragmentation. The libraries were then sequenced on a MinION device with a 1D flow cell (FIO-MIN106; Nanopore) and base called with Guppy v2.3.5 (Nanopore). The long-read data were used in a hybrid de novo assembly using Unicycler v0.4.8 (Wick et al., 2017).

| Fitness evaluation
The growth curve, the area under the curve (AUC), and the maximum optical density (max OD) were used as indicators of fitness (San Millan, 2018). We defined a statistically significant advantage in at least one of the two parameters, AUC and max OD, as indicative of a fitness advantage. Three independent cultures per tested strain were grown overnight, diluted 1:200 in MH broth, and added to a flat-bottom honeycomb 100-well plate in three replicates. The plate was incubated at 37°C under agitation. OD 600 of each culture was determined in 10-min intervals for a total duration of 20 h using Bioscreen C Analyser (Oy Growth Curves Ab. Ltd.). Blank wells, which only contained the medium, were used to adjust for changes in OD 600 not caused by microbial growth. Where required for the experiment, imipenem was added in MH broth to reach a final concentration of 16 mg/L. The max OD was regarded as the OD 600 value determined at the time point of 1200 min. Growth curves were constructed and used to calculate AUC using GraphPad Prism 8.

| Mutant construction
We modified the suicide vector pMo130-Tel R , which contains a tellurite-resistance marker and a sacB gene conferring sucrose sensitivity (Amin et al., 2013). We replaced the tellurite-resistance marker with a hygromycin-resistance marker and called the modified vector pMo130-Hyg R . We used the modified vector to construct mutants as described previously . A 2-kb fragment containing the mutation was generated by PCR and cloned into the BamHI and XbaI site of pMo130-Hyg R using the ClonExpress® II One Step Cloning Kit (Vazyme Biotech Co., Ltd.) as previously described . pMo130-Hyg R derivatives were introduced to A. baumannii ATCC 17978 by electroporation and would integrate into the chromosome by homologous recombination (first-crossover).
Transformants were selected on MH agar containing 100 mg/L hygromycin, where they turned yellow when sprayed with 0.45 M pyrocatechol. Transconjugants were cultured in LB broth containing 10% sucrose to select double recombinants, and serial dilutions were spread onto MH agar plates and sprayed with 0.45 M pyrocatechol to monitor excision of the suicide vector. The mutation was confirmed by PCR and Sanger sequencing using primers in Table S2.

| Quantification of plasmid copy numbers and bla OXA−23 copy numbers
The copy numbers of plasmids and bla OXA−23 were determined by quantitative polymerase chain reaction (qPCR) using a LightCycler ® 480 instrument (Roche Molecular Diagnostics). The experiment was performed by referring to San Millan et al. (2014). Three independent DNA extractions were performed per strain or population. DNA extractions were performed from 2 ml of MH broth cultures without antibiotics and using a QIAamp DNA Minikit (Qiagen). DNA was quantified using Qubit 3.0 (Invitrogen) following manufacturers' instructions. A previous study showed that accurate qPCR-based copy number results would be obtained with total DNA extracts treated with a restriction enzyme that linearizes the plasmid (Providenti et al., 2006). Therefore, 2 μl DNA from each sample was digested using 2.5 μl FastDigest EcoRI (Thermo Fisher Scientific) for 30 min at 37°C. EcoRI was inactivated at 80°C for 5 min. No EcoRI targets are present in the amplification product of any reaction. We used pAZJ221 putative origin of replication gene repA and the rpoD chromosomal monocopy gene to compare the ratio of plasmid and chromosomal DNA.
To calculate primer efficiency, serial dilutions of ATCC 17978/ pAZJ221 wild-type EcoRI-digested genomic DNA were used to create a standard curve for each pair of primers. The average cycle threshold (CT) was obtained for each point and plotted against the log 10 of each concentration. The slope of the regression was obtained, and the efficiency of the primers was calculated using the following formula: (10 1/slope − 1) × 100% (Dorado-Morales et al., 2021).
Primer sequences, the efficiency of the reactions, the targets, and the size of amplicons are presented in Table S3. To calculate the copy number of bla OXA−23 per plasmid, we used the formula: In this formula, bn is the bla OXA−23 copy number per plasmid, Sb is the sizes of bla OXA−23 amplicons (in bp), Eb is the efficiency of bla-OXA−23 qPCRs, and Ctb is the threshold cycles of bla OXA−23 reactions.

| Plasmid stability determination
We assessed the stability of pAZJ221, in the absence of selective pressure, in the constructed mutants and in ATCC 17978

| Conjugation rate and conjugation frequency
We measured the conjugation rates and conjugation frequencies (the number of transconjugants per recipient) to assess plasmid conjugation ability. The conjugation rate is expressed as mL cell −1 h −1 (Alderliesten et al., 2020). Conjugation between plasmid-containing donors and plasmid-free recipients was measured in antibiotic-free MH broth as previously described by Simonsen et al. (1990). In this study, the donor is wild-type ATCC 17978/pAZJ221 or mutant/pAZJ221 or wild-type ATCC 17978/ pAZJ221E, and the recipient is a rifampin-resistant ATCC 17978. as follows: ᵠ is the maximum growth rate of mixed donor and recipient cultures, which is calculated from OD 600 measured at two time point (hour), t a and t b , during the exponential phase: Another parameter conjugation frequency (T/R) was also evaluated using the same stationary phase cell culture.

| Statistical analysis
Statistical analyses were performed following the principles below. Therefore, we measured the effect of pAZJ221 on the fitness of ATCC 17978 by assessing three parameters of bacterial growth: the growth curve, AUC, and max OD. We confirmed that pAZJ221 imposed a fitness cost on ATCC 17978, and ATCC 17978 containing pAZJ221 exhibited significantly decreased fitness in the presence of sub-MIC imipenem relative to antibiotic-free conditions ( Figure 1). Thus, we designed an evolutionary experiment to investigate compensatory host-plasmid co-evolution that alleviates the cost of pAZJ221 in the presence of a carbapenem. The experimental setup is shown in Table 1.
Three experimental lineages were propagated. For each, a population was stored from each of six time points (day 10, 20, 30, 40, 50, and 60) such that each lineage was represented by six sampled  (Figure 2). In lineage 1, the max OD on day 50 was lower than that of the ancestral strain, possibly due to biofilm formation in the later stages of the experiment ( Figure S1).

| Dynamic allele frequency changes
To determine the genetic bases for the observed evolutionary adaptations, we conducted whole-population WGS of evolved populations, control populations, and the ancestral clone to track allele frequency dynamics. We found that adaptations might be the result of mutations in the form of single-nucleotide polymorphisms (SNPs) within genes or in intergenic regions. We defined a "genuine" mutation as one that only occurred in the experimental group. If the same SNP was observed in both the experimental group and the control groups, we did not regard it as a mutation that arose in response to the presence of pAZJ221 and imipenem. In total, we detected 26 genuine chromosomal point mutations: 65.4% (17/26) were nonsynonymous, and 34.6% (9/26) were in intergenic regions. Of nine SNPs in intergenic regions, five were predicted to occur in sRNAs.
We observed dynamic changes of allele frequencies in all lineages. The frequencies of some mutations initially increased before decreasing and disappearing, while other mutations appeared and gradually became dominant. Different mutations competed in the same lineage, supporting the idea of clonal interference (Gerrish & Lenski, 1998;Hughes et al., 2012;Lang et al., 2011).
A complete list of mutations detected in the experimental group and control groups is provided in Table S5, and the allele frequencies of genuine mutations in the experimental group are shown in Figure S2.

| Mutations on parallel targets
Parallelism at the level of nucleotides, amino acids, genes, or pathways in replicate populations is strong evidence of selection on that target and might suggest that these alterations provide an ideal adaptive solution (Cooper, 2018

| Mutations in sRNAs
The details and acronyms of mutations in parallel loci are provided in Abbreviation: IPM, imipenem. a "+" means that the corresponding variable exists. would increase the size of the hairpin loop in sRNA33 ( Figure S4).

| Carbapenem MICs of the ancestor and mutants were indistinguishable
To address whether mutations on sRNAs and within abDGC1 played a role in the adaptation of ATCC 17978 to pAZJ221, we introduced six mutations (listed in

F I G U R E 3
Growth curve (a), the area under the curve (b), and the maximum OD 600 (c) of ATCC 17978 and 6 constructed mutants without pAZJ221 in antibiotic-free MH broth. Box plots show median, lower, and upper quartiles, the minimum value, and the maximum value. Data were collected for 20 h at 37°C with shaking. Statistical analysis was performed through Dunnett's multiple comparisons test. When the data do not follow normal distribution, Dunn's multiple comparisons test was performed. *, adjusted p value < 0.05; **, adjusted p value < 0.01; ***, adjusted p value < 0.001;****, adjusted p value < 0.0001; ns, not statistically significant. Wild Type, ATCC 17978 improved fitness may have been more important than the extent of resistance in these experiments. Thus, we evaluated the fitness of constructed mutants.

| Diverse fitness effects of chromosomal parallel mutations
Our study involved two variables: the presence of pAZJ221 and the presence of imipenem pressure. Thus, we obtained the growth curves of strains under three conditions to elucidate whether, and under which conditions, these six mutations could confer advantages to the host.
For the first condition, we compared the growth curves, AUCs, and max ODs of the wild-type ATCC 17978 and six constructed isogenic mutants harboring mutations listed in Table 3. These strains do not carry pAZJ221 and were tested in antibiotic-free MH broth as they are carbapenem-susceptible. The results showed that two mutations upstream of tetR1 were beneficial, exhibiting significantly higher max OD values than the wild-type strain (Figure 3).
We transformed pAZJ221 into the wild-type ATCC 17978 and six constructed mutants and then obtained growth curves in antibiotic-free MH broth ( Figure 4) and 16 mg/L imipenem MH broth ( Figure 5), respectively. Mutations upstream of tetR1 still conferred fitness advantages in the presence of pAZJ221 with or without imipenem pressure. The mutation tetR2(C-378A) did not provide any fitness advantage under all three conditions at the resolution of the growth curve experiment, which means that it is a neutral mutant allele under the tested conditions. Mutations in abDGC1 had diverse effects on fitness, abDGC1(D589Y)-producing ATCC 17978 exhibited a slightly higher AUC value than wild-type in the presence of pAZJ221 in MH broth, but its max OD value was not significantly different from its wild-type counterpart. When growing in the presence of imipenem, the AUC and max OD values of abDGC1(D589Y)/ pAZJ221 were all higher than those of wild-type/pAZJ221, indicating a positive fitness effect under imipenem pressure. As for the mutation abDGC1 (D341N) and tetR2(G-399A), higher fitness was only observed when pAZJ221-containing mutants were subject to imipenem pressure.
The effects of mutations are summarized in Table 3. The pattern of fitness contribution indicated three strategies to mitigate the fitness cost imposed by pAZJ221 and imipenem exposure: was relieved as a collateral cost-mitigating effect of host adaptation; this strategy was also reported by a recent study (Kloos et al., 2021). Other mutations located upstream of tetR1 were also detected in the control groups, including SNPs and 1-bp deletion (Table S5) was impaired, as its conjugation frequency was lower than pAZJ221 ( Figure 9a).

| Changes in plasmid kinetics
Plasmid kinetics are widely assumed to influence the spread of antibiotic resistance genes in bacterial populations and ecosystems (Campos et al., 2020). The average copy number of pAZJ221 in the ancestral host was 1.5 copies per chromosome (1.50 ± 0.17). We determined the copy number of pAZJ221 in six constructed mutants and the copy number of pAZJ221E in wild-type ATCC 17978. All strains carried around 1.5 copies of pAZJ221 or pAZJ221E per cell (Table S8).
Conjugation is also known to be energetically expensive and en- The sequence that has been duplicated in pAZJ221E is indicated by a labeled red line significant reduced conjugation rates or conjugation frequencies of pAZJ221 were observed when pAZJ221 was electroporated into the constructed mutants except for abDGC1(D589Y) (Figure 9a,b).
The result indicated that a decreased conjugation ability was a common evolutionary trend for the evolution of ATCC 17978/pAZJ221.
As for the plasmid stability of mutant/pAZJ221combinations, after 10-day propagation (about 100 generations) only mutants harboring mutations upstream of tetR1 showed significantly lower plasmid percentages than that on the first day ( Figure 9c).

| DISCUSS ION
We observed that two mutations upstream of tetR1 could improve fitness in a new environment where neither pAZJ221 nor imipenem pressure was present, suggesting that these SNPs had pleiotropic fitness effects (Kinsler et al., 2020). Improved fitness in multiple conditions could favor the success of strains and prevent their loss within populations. Strains with the mutations tetR1(C-424T) and tetR1(G-339T) might be well-suited to respond to future evolutionary challenges and to serve as stable donors or recipients of plasmids such as pAZJ221.
Gene duplication is a major source of evolutionary innovation (Toll-Riera et al., 2016). Our results emphasized the importance of gene instability in rapidly tuning levels of gene expression, as noted in other studies (Girgis et al., 2021;Tomanek et al., 2020). Fitness advantages resulting from bla OXA−23 duplication were observed soon after pAZJ221 entered a new host exposed to a carbapenem, and this advantage might facilitate population expansion for the accumulation of chromosomal mutations that stabilize the host-plasmid pair.

F I G U R E 8
Growth curve (a), the area under the curve (b), and the maximum OD 600 (c) of ATCC 17978/pAZJ221 and ATCC 17978/ pAZJ221E in MH broth and 16 mg/L imipenem MH broth (IPM) are analyzed and compared. Statistical analyses were performed through unpaired t-tests. When the data did not follow normal distribution, Mann-Whitney test was performed. *p < 0.05; ****p < 0.0001; ns, not statistically significant. (d) bla OX A-23 copy number per plasmid in ATCC 17978/pAZJ221E on four time points during propagation in antibioticfree MH broth. The copy number is monitored using qPCR, and three biological replicates are included. Error bars depict the standard deviation of the mean, and the dots represent the data of each replicate We noted that plasmid fitness and host fitness are not necessarily aligned. Among five mutations that conferred fitness advantages in the presence of pAZJ221 under imipenem pressure, four impaired the transfer ability of pAZJ221 and two decreased plasmid stability (Table 3). A reduced conjugation rate is associated with plasmid fitness cost amelioration, representing a shift toward higher investment in vertical transmission (Dahlberg & Chao, 2003;Harrison & Brockhurst, 2012;Turner et al., 1998). Our results are in line with the work of Hall et al. whose model predicted that amelioration is a more successful strategy than high conjugation rates, and that established bacteria-plasmid relationships would ultimately trend toward low conjugation rates, because high conjugation rates are only transiently beneficial (Hall et al., 2017).
We propose a potentially general mechanism of A. baumannii adaptations to the OXA-23-producing plasmid pAZJ221. First, duplication of bla OXA−23 conferred an immediate advantage in the presence of a carbapenem. Next, chromosomal mutations with diverse effects appeared. Five out of six parallel mutations were verified to confer fitness advantages in the presence of pAZJ221 under imipenem pressure, and all but one of them impaired the plasmid's conjugation ability.
There are some limitations for translating our findings to realworld situations. First, our study was conducted under laboratory conditions while in the wild, pathogens exist in more complex environments. Additionally, the mechanisms responsible for beneficial effects on the fitness of the mutations require further investigation.
Finally, more clinically relevant A. baumannii strains should be considered for further evolutionary experiments.
Despite the aforementioned limitations, our study sheds new light on the co-evolution of an OXA-23-producing plasmid that co-evolves with a new A. baumannii host under sustained, sub-inhibitory carbapenem exposure. To combat the spread and persistence of CRAB, novel therapeutic approaches that target mechanisms contributing to the emergence and dissemination of successful CRAB strains are urgently required. It is also crucial that the ongoing evolution of CRAB and associated carbapenem resistance plasmids are monitored in clinical environments globally.

CO N FLI C T O F I NTE R E S T
None to declare.

DATA AVA I L A B I LIT Y S T S TE M E NT S
Sequencing data were deposited to NCBI as Bioproject PRJNA