Whole-Genome-Sequence-Based Characterization of Extensively Drug-Resistant Acinetobacter baumannii Hospital Outbreak.

Carbapenem-resistant Acinetobacter baumannii (CRAB) has been implicated in hospital outbreaks worldwide. Here, we present a whole-genome-based investigation of an extensively drug-resistant CRAB outbreak rapidly spreading and causing high incidences of mortality at numerous wards of a large tertiary hospital in Lebanon. This is the first study of its kind in the region. Two circulating clones were identified using a combination of molecular typing approaches, short- and long-read sequencing and Bayesian transmission network analysis. One clone carried blaOXA-23 on Tn2006 (ST-1305, ST-195, and ST-218), and another carried blaOXA-72 on a pMAL-1 plasmid (ST-502 and ST-2059, a new ST). A pMAL-2 plasmid was circulating between the two clones. The approaches implemented in this study and the obtained findings facilitate the tracking of outbreak scenarios in Lebanon and the region at large.

resistant to levofloxacin, and one isolate (ACM-3) showed intermediate resistance to levofloxacin. Additionally, 95% (n ϭ 39) were resistant to ceftazidime, imipenem, meropenem, and piperacillin-tazobactam, and 82.9% (n ϭ 34) were resistant to cefepime. All isolates were susceptible to colistin, as determined by the obtained MIC values. Based on the findings described above, 95% of the isolates were classified as being extensively drug resistant (XDR) (20) (Fig. 2).
The antibiotic susceptibility testing results were confirmed through in silico detection of resistant determinants. Four different aminoglycoside-inactivating enzymes and their variants were detected. Aminoglycoside acetyltransferase (AAC) aac(3=)-Ia and ant(3Љ)-Ia, belonging to the aminoglycoside nucleotidyltransferase (ANT) family, were found in 56% (n ϭ 23) and 53.6% (n ϭ 22) of the isolates, respectively. Five variants of the aminoglycoside phosphotransferase (APH) enzyme were also identified. The most common was aph(3=)-VIa, being detected in 73% (n ϭ 30) of the isolates. strA and strB variants of the aph enzyme were detected in 46.3% (n ϭ 19), aph(3=)-Ia in 34% (n ϭ 14), aph(6=)-Id in 2.4% (n ϭ 1), and armA in 46.3% (n ϭ 19) of the isolates. Acinetobacter-derived cephalosporinase genes bla ADC-73 and bla ADC-74 were identified in 62.5% (n ϭ 25) and 37.5% (n ϭ 15) of the isolates, respectively. ISAba1 was detected upstream of bla ADC-73 and bla ADC-74 in the opposite direction. All of the isolates carried bla OXA-66 , while only ACM-14 was positive for bla OXA-359. Class A ␤-lactamase (bla TEM-1D ), in addition to macrolide resistance determinants mph(E) and msr(E), were also found in 31.7% (n ϭ 13) and 44% (n ϭ 18) of the isolates, respectively. PFGE. As we suspected an outbreak scenario, we first typed the isolates using PFGE. Isolates showing a difference of more than three bands were classified as belonging to different pulsotypes (21). Accordingly, 21 different pulsotypes were identified and were aligned with the STs but not with the source of isolation. The isolates, based on the PFGE output, appeared to be circulating in the different wards of the hospital. The first clade included  MLST. In silico MLST analysis, based on seven housekeeping genes (cpn60, gdhB, gltA, gpi, gyrB, recA, and rpoD) using the Oxford scheme, revealed the presence of eight different STs compared to only five STs identified using the Pasteur MLST scheme. Given that the Oxford MLST scheme was more capable at differentiating between closely related isolates, we adopted it for all further analysis. ST-502, CC92  by numbers below branches as well as by branch coloring (green, high confidence levels; red, low confidence levels). Dashed lines connect labels with particular leaf nodes in order to avoid labels to overlap the branches. Classes of antibiotics are given the following markings: A, aminoglycosides; B, ␤-lactams; Q, quinolones; T, tetracyclines; GM, gentamicin; AK, amikacin; TZP, piperacillin-tazobactam; CAZ, ceftazidime; FEP, cefepime; IMP, imipenem; MEM, meropenem; CIP, ciprofloxacin; LEV, levofloxacin. Dark blue indicates resistant, light blue indicates intermediate susceptibility, and blank indicates sensitive. (B) Unrooted phylograms inferred from SNP data from core genomes of strains belonging to the Tn2006-mediated clone and cluster pMAL-1-mediated clone, respectively. Red circle in the pMAL-1-mediated clone highlights ST-2059 isolates. Bootstrap values are indicated by numbers below branches as well as by branch coloring (green, high confidence levels; red, low confidence levels). Dashed lines connect labels with particular leaf nodes in order to avoid labels to overlap the branches. the isolates, respectively. Two singletons, ST-1305, CC208 (2-3-1-96-12-2-3; ACM-7)  (ST-2-like, CC2 Pasteur scheme), and ST-425, CC92 (2-3-1-100-3-2-3; ACM-39) (ST-2-like,  CC2 Pasteur scheme), were also among the detected ST types. Mobile genetic elements. The A. baumannii PCR-based replicon typing (AB-PBRT) method was used to characterize the circulating plasmids. AB-PBRT categorizes A. baumannii plasmids into homology groups (GRs) based on the nucleotide homology of their respective replicase genes. AB-PBRT results showed that 21.9% (n ϭ 9) of the isolates carried homology group GR2, 2.4% (n ϭ 1) carried GR6, and 68.3% (n ϭ 28) carried both GR1 and GR6, related to a wide range of plasmids (22).
Tn2006-linked clone. Nineteen CRAB (46.3%) isolated from 14 patients carried the bla OXA-23 carbapenemase gene in addition to intrinsically carrying bla OXA-66 . These constituted the Tn2006-linked circulating clone in the outbreak. bla OXA-23 is usually associated with plasmids or integrated through transposons into the A. baumannii chromosome (10). These transposons are highly diverse and were designated Tn2006, Tn2007, Tn2008, and Tn2009 (23). bla OXA-23 was carried by a chromosomally integrated Tn2006. The mobilization of Tn2006, being detected in ST-195, ST-218, and ST-1305, was facilitated by the ISAba1 bracketing of bla OXA-23 . The remaining constituents of Tn2006 were yeeA (encoding the putative DNA methylase), DEAD (encoding the putative Asp-Glu-Ala-Asp helicase), and ATPase (encoding the putative AAA ATPase) genes.
The integration site of Tn2006 was the same among all the isolates of the Tn2006linked clone. Tn2006 was integrated inside a dienelactone hydrolase family protein, dividing it into two fragments, one 670 bp upstream of Tn2006 and a smaller 145-bp fragment downstream of Tn2006.
Tn2006 was associated with the AbaR25 resistance island, as has been previously described for Tn2006 (23,24). AbaR25 had the same genetic environment as AbaR25-I type (24).
In order to better characterize the higAB system found on pAB-MAL-2, we compared it with a recently described plasmid, pAB120, having the higBA2 Ab TA module (27). Alignment of the TA components revealed 60.3% and 58.8% similarity to higB2 and higA of pAB120, respectively. The higAB system was preceded by ISAba125 (Fig. S1). BLAST analysis showed 100% identity to higB2 of A. baumannii ACICU plasmid pACICU2 (GenBank accession number CP031382) and to the type II toxin-antitoxin system RelE/ParE in many other A. baumannii plasmids. higA was 100% identical to antitoxin higA1, linked to A. baumannii ACICU plasmid pACICU2 and to transcriptional regulators in other A. baumannii plasmids.
Whole-genome SNP phylogenetic analysis. In total, 71,864 SNP sites were identified. Isolates were separated into two major clusters, of which two representative genomes were chosen to be additionally sequenced using PacBio long-read sequencing technology, ACM-2 (representing the isolates carrying bla OXA-23 and forming the Tn2006-mediated clone) and ACM-17 (representing the isolates carrying bla OXA-72 and forming the pMAL-1-mediated clones). The results obtained confirmed the presence of two circulating clones. A total of 1,207 SNP sites were detected in the Tn2006-linked clone, while only 57 SNPs were detected in the pMAL-1-linked clone (Fig. 2).
Outbreak analysis. To reconstruct the suspected outbreak based on statistical analyses and core genome alignments, and taking into consideration collection dates, we used the R package Outbreaker (28). Outbreaker allows for a Bayesian reconstruction of disease outbreaks by combining both epidemiologic and genomic data. As a result, a bell-shaped distribution of the mutations per site and generation was observed in the case of the Tn2006-linked clone, along with a tight clustering of the isolates in an interconnected nodular network (Fig. S4). Clearly localized SNP-generating hot spots were favored to produce genomic diversity, as both recombination-and non- recombination-induced single-nucleotide variants almost always occurred in a few defined regions (SNP density ranging from 0.0003 to 0.0012). This included genes encoding diacylglycerol kinase, 3=,5=-cyclic-nucleotide phosphodiesterase, BapA prefixlike domain-containing protein, translation error-prone DNA polymerase V autoproteolytic subunit, and epoxyqueuosine reductase QueH. An exponential decrease of the infectivity of the cases based on the probability of infecting another patient up to 20 days after infection was observed (Fig. S5).
On the other hand, in the pMAL-1-linked clone, the pairwise genetic distances had a clear bimodal distribution showing two bell-shaped curves (Fig. 4A). Based on this distribution, we speculated that the pMAL-1-linked clone encompassed two subclones. The mean mutation rate was 10, ranging from 0 to 18 for subclone I, while for subclone II we found 35 mutations, ranging between 28 and 45 (Fig. 4B). A high density of SNPs (ranging between 2 ϫ 10 Ϫ4 and 10 ϫ 10 Ϫ4 ) was observed within specific loci in the core genomes. These were genes encoding BapA prefix-like domain-containing protein, DUF2750 domain-containing protein, stress-induced protein, and putative pilus assembly protein FilE.

DISCUSSION
In this study, a CRAB hospital outbreak caused by two distinct clones was tracked using PFGE, MLST, and WGS analysis over 9 months. Our analysis revealed a pMAL-1linked clone, including ST-502 and ST-2059 isolates carrying bla OXA-72 , along with two    . Interestingly, out of the many transposons carrying the bla OXA-23 gene, only Tn2006 has been shown to be highly mobile (29). bla OXA-23 was originally identified in Acinetobacter radioresistens, with bla OXA-23 -mediated resistance to carbapenem being detected only in the presence of a strong promoter, such as ISAba1 (29). ISAba1, which belongs to the IS4 family of insertion sequence elements, provides a promoter motif mediating the increased expression of bla OXA-23 and is involved in facilitating its own mobilization not only to A. baumannii but also to Proteus mirabilis (10). Upon its first description in 1995 in A. baumannii, bla OXA-23 carried on plasmids or transposons became widespread and was detected in isolates recovered from the United Kingdom, France, Romania, Brazil, South Korea, United Arab Emirates, Egypt, and Iraq (10,30,31). bla OXA-23 was detected previously in GC2 CRAB isolates in Lebanon and was recognized as being the most common carbapenemase in the country (17,30). The Tn2006-linked clone also carried a cryptic pA85-2 plasmid that appears to be commonly found in the GC1 lineage (32). The advantage conferred by this plasmid is still unknown.
The pMAL-1-linked clone included ST-502 and ST-2059 isolates and harbored the bla OXA-72 gene. This is the first report of ST-502 and ST-2059 A. baumannii in Lebanon. Carbapenem-resistant ST-502 isolates have been reported previously in Brazil (33), Bulgaria (34), and South Africa (35). bla OXA-72 was first identified in 2004 in A. baumannii isolated from Taiwan, and since then it was detected in a few other countries, including Brazil, France, and the United States (36). The pMAL-1 plasmid was very similar to p2ABST25, with the latter being negative for ISAba31 (26). The presence of ISAba31 upstream of bla OXA-72 has been reported previously in pMAL-1 (26), yet its effects on bla OXA-72 expression require further investigations. No previous reports showed the presence of p2ABST25-positive isolates in Lebanon or any of the neighboring countries, and as such pMAL-1 could have been introduced rather than evolved from another circulating plasmid.
We also revealed that the Tn2006-linked clone harbored a pA85-2 plasmid carrying persistence-and virulence-related genes, such as a BrnT/BrnA toxin-antitoxin system, a TonB-dependent receptor, and a septicolysin. The BrnT toxin-antitoxin system is required for RNA cleavage and control of bacteriostasis in Brucella abortus (37). TonBdependent transporters are outer membrane proteins that bind and transport siderophores in addition to vitamin B12, nickel complexes, and carbohydrates (38). Septicolysin, on the other hand, is a pore-forming toxin with cytolytic activity that mediates invasion (39).
The pAB-MAL-2 plasmid was found circulating in both the Tn2006-and the pMAL-1-linked outbreak clones. pAB-MAL-2 detected in this study was slightly different from that in previous reports (26). The plasmid was additionally positive for a TA system showing 60% similarity to a recently described higBA2 Ab TA module on pAB120 (27). The higAB system found on the pAB-MAL-2 plasmid in our isolates was preceded by ISAba125. ISAba125 was previously associated with the dissemination of bla NDM-1 (40). The presence of ISAba125, a strong promoter, upstream of the TA system (41) is an important finding. TA systems are genetic loci involved in plasmid maintenance and in regulating bacterial stress responses linked to pathogen virulence and formation of drug-resistant persister cells and biofilms (42). The higBA2 gene detected in this study is a reverse TA, where the toxin gene is the first in the operon. HigB2 functions as an RNase capable of conferring maintenance of unstable plasmid in Acinetobacter, and it is neutralized with HigA2 antitoxin (27).
A total of 1,207 SNP sites were detected within the Tn2006-linked clone harboring bla OXA-23 , compared to only 57 SNP sites identified within the pMAL-1-linked clone harboring bla  . Bayesian analysis revealed the transmission events with a high degree of certainty. During the outbreak, the isolates underwent genomic remodeling at clearly defined loci. Recombination events, as well as increased variance in the nucleotide sequences within a few genes, served as the primary drivers of diversity. In particular, and in both clones, a high density of SNPs was observed in BapA, a biofilm-associated protein mediating biofilm formation and adhesion to host cells (43).
In the Tn2006-linked clone, one of the genes that accumulated a large number of SNPs encoded a diacylglycerol kinase, a small integral membrane protein (44). Diacylglycerol kinase was previously shown to be upregulated in colistin-resistant A. baumannii and was hypothesized to increase the integrity of the cell membrane, favoring drug resistance (45).
In conclusion, by following the pattern of transmission and based on the molecular epidemiology and genomics data, we confirm the polyclonal nature of the outbreak. Using PFGE, MLST, and PCR typing methods in combination with WGS data and SNP-based Bayesian transmission network analysis, we were able to characterize at high-scale resolution a hospital CRAB outbreak caused by a Tn2006-linked clone carrying bla OXA-23 and a pMAL-1-linked clone carrying bla OXA-72 . An outbreak driven by chromosomally encoded resistance determinants had a more consistent evolution and was easier to retrace using core genome sequence data from short genome reads. Alternatively, a plasmid-driven outbreak followed less of a straight evolutionary path during the sampling. Both clones additionally harbored circulating plasmids, ensuring the survival of the fittest and contributing to virulence.

MATERIALS AND METHODS
Ethical approval. Ethical approval was not required, as the isolates were collected as part of routine clinical care and patient data collection followed patient discharge from the hospital and/or death. No additional isolates were collected beyond those obtained from routine clinical care, and no diagnostic or treatment decisions were affected by the outcomes of this study.
Bacterial isolates. A total of 41 A. baumannii isolates were collected between April and December 2016 from hospitalized patients at a 544-bed hospital in Lebanon and were designated ACM-1 to -37 based on the date of their isolation and ACM-38 to -41 for the isolates with no trackable records. The initial identification of Acinetobacter species was carried out using automated microbial identification systems (Vitek and BD Phoenix).
Antimicrobial susceptibility testing. Antimicrobial susceptibility testing was performed using the disk diffusion method on Mueller-Hinton agar to determine resistance patterns against nine different antibiotics: gentamicin, amikacin, piperacillin-tazobactam, ceftazidime, cefepime, imipenem, meropenem, ciprofloxacin, and levofloxacin. The results were interpreted according to Clinical and Laboratory Standards Institute (CLSI) guidelines (M100-S29) (46). Isolates were considered extensively drug resistant (XDR) if they were nonsusceptible to at least one agent in all but two or fewer antimicrobial classes (20).
MICs were determined using the Etest (bioMérieux, France) method for gentamicin, cefepime, ciprofloxacin, trimethoprim-sulfamethoxazole, amikacin, piperacillin-tazobactam, and colistin and the broth microdilution method for colistin, meropenem, and imipenem (Sigma-Aldrich, USA). For the Etest, bacterial suspensions with turbidity equivalent to 0.5 McFarland standard were prepared and spread onto Mueller-Hinton agar (MHA) (Bio-Rad Laboratories, Inc., USA), followed by the addition of Etest strips. The plates then were incubated at 37°C for 24 h. For the broth microdilution method, cation-adjusted Mueller-Hinton broth (Sigma-Aldrich, USA) was used according to CLSI recommendations. Serial dilutions of each of the three antibiotics (32 to 0.062 g/ml) were prepared. A. baumannii isolates were suspended to 0.5 McFarland turbidity and added to 96-well microdilution plates. Microplates were incubated at 35°C for 20 to 24 h. The lowest antimicrobial drug concentration at which there was no growth was considered the MIC.
PFGE. Genomic DNA plugs of A. baumannii were prepared according to the protocol developed by Seifert et al. (47). Briefly, plugs were digested using ApaI (Thermo Fisher Scientific, MA, USA) for 2 h at 37°C. DNA fragments were then separated on 1% Seakem Gold gel using a CHEF DR-III system (Bio-Rad Laboratories, Inc., CA, USA) for 16 h with initial and final switch time of 7 s and 20 s, respectively. To compare the patterns on different gels, Salmonella enterica subsp. enterica serovar Braenderup (ATCC BAA664TM) was used as the reference for band size using XbaI restriction digestion (Thermo Fisher Scientific, MA, USA). The gel then was stained with ethidium bromide and viewed under UV light. Fingerprints were analyzed using BioNumerics software, version 7.6.1 (Applied Maths, St-Martens-Latem, Belgium). Bands not detected automatically were manually assigned. Fingerprints were clustered according to the band-based coefficient, which measures similarity based upon common and different bands with 0.5% optimization and 0.5% tolerance of band matching.
Conjugation. Conjugation experiments were performed as described previously (48)  msphere.asm.org 9 to sodium azide, as a recipient. ACM-17 and E. coli J53 were mixed at a ratio of 4:1 (donor to recipient) in Luria-Bertani broth, and the mixture was incubated at 37°C for 1 h. To select for transconjugants, serial dilutions of the cultures were plated on UriSelect agar containing imipenem (4 g/ml) and sodium azide (100 g/ml). Whole-genome sequencing. DNA extraction was performed using the NucleoSpin tissue kit (Macherey-Nagel, Germany) according to the manufacturer's instructions. Library preparation was done using the Nextera XT DNA library preparation kit (Illumina). Sequencing of the library was done on an Illumina MiSeq using a paired-end 500-cycle protocol with a read length of 250 bp. FastQC, version 1.0.0 (49), was used for quality control. Reads were trimmed by the Trimmomatic tool, v0.36 (50). Paired-end reads were assembled using SPAdes, v3.12.0 (51), and annotated using the RAST server (http://rast.nmpdr .org) (52).
Two isolates (ACM-2 and ACM-17) were additionally sequenced using PacBio long-read sequencing technology on the Sequel platform (Pacific Biosciences, CA, USA). ACM-2 represented the Tn2006mediated clone, and ACM-17 represented the pMAL-1-mediated clone. Library preparation was done according to the manufacturer's instructions for microbial multiplexing. G-tubes (Covaris, USA) were used for DNA shearing, and no size selection was performed. Resulting chromosomal contigs of ACM-2 and ACM-17 were corrected by Illumina data with Pilon, v1.23 (53), and the overlapping ends of chromosomes were trimmed after manual inspection of reads mapped by BWA-MEM algorithm as implemented in BWA, v0.7.17 (54), and Bowtie, v2.3.4.2 (55). Genome assembly was done using HGAP4 (56) with minimum seed coverage of 30. PCR-based amplification was used to fill the sequence gaps.
Genome analysis. Resfinder (57), MLST 1.8 server, ISfinder database (58), BLAST (59), and the open reading frame (ORF) finder tool (59) were utilized to identify resistance genes, sequence types (STs), ISs, and plasmid identification and annotation, respectively. Mauve (version 2.3.1) was used for comparative genome alignments (60). The core genome alignments of each clone were used for outbreak analysis. Ape was used to infer basic phylogeny of the aligned core genomes (61). Adegenet was used to cluster genetic fragments that were found to be identical between the isolates from each outbreak and to perform a multivariate analysis of the genetic markers using the SeqTrack algorithm (62); this was used to visualize the genetic relatedness of the isolates spatially. plasmidSPAdes, version 3.10.1 (63), was employed to generate separate plasmid contigs, and the output was visualized through Bandage and an assembly graph viewer (64).
We used Outbreaker (28) to help elucidate the circulating clones that were directly linked to the studied outbreak. Outbreaker is based on collection dates and a reconstruction of the outbreak based on the evolution of core genome SNPs outside identified recombination spots in all studied genomes.
Phylogenetic analysis. For the purposes of phylogenetic analysis of all the recovered isolates, along with ACM-14, which was used as an outgroup, core genome sequences were extracted from the contigs using an in-house script. The script was used to extract all nonrepetitive homologous sequences that are longer than 399 bp using NUCmer, version 3.1 (66), output as generated by QUAST, version 5.0.0 (67). Core genome sequences were aligned by MAFFT, version 7.407 (68), using default gap penalties and the memsave parameter, FFT-NS-2 strategy, and iterative refinement with a maximum of two iterations. SNP sites then were extracted from aligned sequences using the SNP-sites tool, version 2.4.0 (69).
In order to infer the phylogeny of the two clones, quality-trimmed Illumina reads were mapped to the chromosomal sequences (ACM-2 or ACM-17) using Bowtie, version 2.3.4.2 (55). SNPs were consequently called using VarScan, version 2.4.3 (70), with minimum read depth set to 8, minimum base quality of 20, and variant allele frequency of Ն0.8. All sites where at least one isolate had a read depth of Ͻ8 were removed from the final data sets.
All three data sets were analyzed by jModelTest, version 2.1.10, to determine the most appropriate models of nucleotide substitution (71). Phylogenetic analyses were performed by RAxML, version 8.2.10 (72), and robustness of the inferred topologies was assessed by 500-bootstrap replicate analyses. Topologies of the trees were visualized using iTOL, version 4.3.2 (73), and edited by Inkscape, version 0.92 (www.inkscape.org), and Gimp, version 2.10.6 (www.gimp.org).
Plasmid typing. The PCR-based replicon typing method for A. baumannii developed by Bertini et al. (22) was used to determine the plasmid content of the 41 isolates. With this method, the 27 known A. baumannii replicase (rep) genes are grouped into 19 homology groups according to their nucleotide sequence similarities. Six multiplex PCRs were done using primers designed by Bertini et al. (22).
Accession number(s). The draft genomes were deposited in the NCBI databases under the accession numbers listed in Table S3 in the supplemental material.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.