Whole-Genome Sequencing for Investigating a Health Care-Associated Outbreak of Carbapenem-Resistant Acinetobacter baumannii

Carbapenem-resistant Acinetobacter baumannii (CRAB) outbreaks in hospital settings challenge the treatment of patients and infection control. Understanding the relatedness of clinical isolates is important in distinguishing outbreak isolates from sporadic cases. This study investigated 11 CRAB isolates from a hospital outbreak by whole-genome sequencing (WGS), utilizing various bioinformatics tools for outbreak analysis. The results of multilocus sequence typing (MLST), single nucleotide polymorphism (SNP) analysis, and phylogenetic tree analysis by WGS through web-based tools were compared, and repetitive element polymerase chain reaction (rep-PCR) typing was performed. Through the WGS of 11 A. baumannii isolates, three clonal lineages were identified from the outbreak. The coexistence of blaOXA-23, blaOXA-66, blaADC-25, and armA with additional aminoglycoside-inactivating enzymes, predicted to confer multidrug resistance, was identified in all isolates. The MLST Oxford scheme identified three types (ST191, ST369, and ST451), and, through whole-genome MLST and whole-genome SNP analyses, different clones were found to exist within the MLST types. wgSNP showed the highest discriminatory power with the lowest similarities among the isolates. Using the various bioinformatics tools for WGS, CRAB outbreak analysis was applicable and identified three discrete clusters differentiating the separate epidemiologic relationships among the isolates.


Introduction
Carbapenem-resistant Acinetobacter baumannii (CRAB) is an important pathogen in healthcare-associated infections and leads to high mortality, especially in intensive care units [1]. A. baumannii can cause various conditions, such as bacteremia, bloodstream and surgical wound infections, and ventilator-associated pneumonia [2]. The emergence of multidrug-resistant A. baumannii is a challenge in the treatment of patients and infection control since it has the ability to survive on the surface of plastics and can easily spread in the hospital environment [3]. CRAB is designated a critical-priority pathogen by the World Health Organization for the development of new antimicrobial drugs [4]. Currently, only a few therapeutic options, including colistin and tigecycline, are available for CRAB with increasing resistance rates, which is a concerning situation [5].

Materials and Methods
This study investigated 11 A. baumannii isolates collected in 2015 and 2016 at Seoul National University Bundang Hospital; 9 of the isolates were obtained from the intensive care unit (ICU) and Ward A during a hospital outbreak in 2016, and 2 of the isolates were from patients with CRAB from 2015 in the same institution (Table 1). This study was approved by the institutional review board of Seoul National University Bundang Hospital (B-2008-628-305) on 21 July 2020 with a waiver of informed consent. In 2016, five patients who stayed in the ICU were diagnosed with CRAB, one of whom was transferred to Ward A. Environmental screenings (bedrails, suction catheters, and central station keyboard) of the ICU and Ward A were included. Acinetobacter isolates were identified by Gram stain morphology, and antibiotic susceptibility (AST) was evaluated with the Vitek2 system (bioMerieux, Marcy l'Etoile, France). The AST results were interpreted based on the Clinical and Laboratory Standards Institute (CLSI) M100 guidelines [20].

Whole-Genome Sequencing
WGS was performed using an Illumina HiSeq4000 (Illumina, San Diego, CA, USA). The results were randomly labeled for this study (#1-#12, excluding #10). Clinical information was not given until the analysis was completed. The whole-genome sequences of eleven A. baumannii isolates were assembled and annotated using the Comprehensive Genome Analysis service with the default parameters on the Pathosystems Resource 127 Integration Center (PATRIC) 3.6.6 supported by the National Institutes of Health (https://www.patricbrc.org) [21]. The genome was annotated via Rapid annotation using subsystem technology tool kit (RASTtk) [22] and was assigned a unique genome identifier. Pathogenwatch, developed by the Center for Genomic Pathogen Surveillance (https://pathogen.watch), was additionally utilized for genome assembly [18]. FASTQ files were uploaded and assembled, and the organism was predicted. De novo assembly was additionally performed with Bionumerics 7.6 (Applied Maths, Sint-Martens-Latem, Belgium) using the default settings.
Core genome (cg) MLST clustering was performed on Pathogenwatch, where profiles are clustered by calculating distances between each assembly that shares a given cgMLST scheme based on 2390 targets for A. baumannii. The distance is calculated as the number of different loci for the scheme, ignoring any that are missing. These were then clustered using single linkage clustering based on the calculated pairwise distances in Pathogenwatch based on https://www.cgmlst.org/ncs/schema/3956907/.

Single Nucleotide Variation Analysis
SNV analysis was conducted using the variation analysis service of the PATRIC online server. Isolate #1, which was a non-outbreak isolate from 2015, was selected as the reference genome for the analysis. Burrows-Wheeler Aligner-mem was used for the aligner, and FreeBayes was used for the SNV caller. Only SNVs reported to have a "high" effect, nonsense and frameshift variants, were filtered. Synonymous or missense variants were neglected.
Whole-genome single nucleotide polymorphism (wgSNP) analysis was performed with Bionumerics 7.6 (Applied Maths) using Isolate #1 as the reference genome, and "Strict SNP filtering" was applied.

Cluster Analysis
A phylogenetic tree was created using the Phylogenetic Tree Building Service of the PATRIC online server with the codon tree method. The codon tree method selects single-copy PATRIC PGFams and analyses aligned proteins and coding DNA from singlecopy genes using Randomized axelerated maximum likelihood (RAxML) (v8.2.11). One hundred genes were utilized, and the allowed deletions and duplications were set as 0, the default value. Cluster analysis was performed on the wgMLST and wgSNP data with Bionumerics 7.6 using categorical differences with a scaling factor of 100. Clustering was performed with the unweighted pair group method with arithmetic mean.

Antimicrobial Resistance Prediction
AMR was predicted by ResFinder 4.0 [25] with the default settings, threshold for %ID as 90% and minimum length of 60%, and KmerResistance 2.2, using the species determination on maximum query coverage, select identity threshold as 70%, and threshold for depth corr as 10% [26,27].

Repetitive Element PCR
DNA was extracted with an UltraClean microbial DNA isolation kit (MoBio Laboratories, Carlsbad, CA, USA), and the concentration was assessed with a NanoDrop. DNA was amplified with a DiversiLab Acinetobacter kit (bioMerieux, France) following the manufacturer's instructions. Analysis was performed with web-based DiversiLab software using the Pearson correlation (PC) method, which weighs peak intensities; the Kullback-Leiber (KL) method, which weighs the absence of peaks; and the extended Jaccard (XJ) method, which is sensitive to the presence of peaks, for analysis. In this study, strains with greater than 95 % similarity were considered similar, and strains with less than 95% similarity were considered different. We analyzed the strain diversity of 11 isolates in two batches, with two isolates commonly included in both batches and eight other non-outbreak isolates included in the rep-PCR analysis.

Results
This study comprised of 11 A. baumannii isolates collected in 2015 and 2016; nine isolates were from a hospital outbreak in 2016 in the ICU, and two isolates (#1 and #7) were from patients with CRAB from 2015 in the same institution (Table 1). Specimens were from blood (n = 3), sputum (n = 2), transtracheal aspirate (n = 2), and surveillance culture/swab (n = 4). The AMR results from the specimens (n = 11) showed carbapenem resistance.

Whole-Genome Sequencing Results
From WGS, the isolates were assigned as A. baumannii. Due to the difference in assembly, the genome lengths and contigs varied according to different bioinformatics tools (Table S1).
Based on the cgMLST scheme with 2390 targets on Pathogenwatch, at the threshold of 10, isolates were clustered into three groups of #2, #6, #11, #12, and #5; #1 and #7; and #3 and #8. Isolates #4 and #9 were not clustered with any other isolates. At the threshold of 30, #4 was clustered with #3 and #8, but #9 remained separate from the other isolates. The de novo assembled genomes submitted to the calculation engine by BioNumerics produced a wgMLST profile with 5619 loci. The comparison of characteristics showed similarity values of 84.81-100% for the isolates (Figure 2). and #12 were assigned as ST451; #3, #4, and #8 as ST369; and #1, #7, and #9 as ST191 ( Figure  1a). Based on the cgMLST scheme with 2390 targets on Pathogenwatch, at the threshold of 10, isolates were clustered into three groups of #2, #6, #11, #12, and #5; #1 and #7; and #3 and #8. Isolates #4 and #9 were not clustered with any other isolates. At the threshold of 30, #4 was clustered with #3 and #8, but #9 remained separate from the other isolates. The de novo assembled genomes submitted to the calculation engine by BioNumerics produced a wgMLST profile with 5619 loci. The comparison of characteristics showed similarity values of 84.81-100% for the isolates (Figure 2).

Antimicrobial Resistance Prediction
The antibiotic susceptibility testing results were predicted through in silico analysis of resistance determination by ResFinder and KmerResistance ( Table 2). All of the isolates showed coexistence of the OXA carbapenemase genes blaOXA-23 and blaOXA-66 and the Acinetobacter-derived cephalosporinase gene blaADC-25, while six isolates had the additional gene blaTEM-1D, and these results were concordant with ResFinder and KmerResistance.

Antimicrobial Resistance Prediction
The antibiotic susceptibility testing results were predicted through in silico analysis of resistance determination by ResFinder and KmerResistance ( Table 2). All of the isolates showed coexistence of the OXA carbapenemase genes bla OXA-23 and bla OXA-66 and the Acinetobacter-derived cephalosporinase gene bla ADC-25 , while six isolates had the additional gene bla TEM-1D , and these results were concordant with ResFinder and KmerResistance. Six different aminoglycoside-inactivating enzymes and their variants were detected. Aminoglycoside acetyltransferase (AAC) aac(6 )-Ib-cr was found in six isolates assigned as MLST (Oxford) ST191 and ST369. Three variants of aminoglycoside phosphotransferase, aph(3 )Ia, aph(3")Ib, and aph(6)-Id, were identified in the other five isolates, and all 11 isolates had variants in armA, a 16S rRNA methylase gene. The aac(6 )-Ib-cr variant was identified only with ResFinder.
The sulfonamide resistance gene sul1 was present in six isolates, and sul2 was present in the other remaining isolates. The phenicol resistance gene catB8 was present in six isolates with sul1. ResFinder and Kmer resistance showed different results for the macrolide resistance gene mphE, where all isolates were predicted to have this gene in KmerResistance, whereas ResFinder showed that only nine isolates had this. All of the isolates were predicted to be susceptible to colistin.

Discussion
Outbreak analysis of 11 A. baumannii isolates was performed with WGS and rep-PCR, and three clonal lineages were identified from the outbreak. The coexistence of blaOXA-23, blaOXA-66, blaADC-25, and armA with additional aminoglycoside-inactivating enzymes, predicted to confer multidrug resistance, was identified in all the isolates.
Although both the MLST Pasteur and Oxford schemes were based on seven loci, the Pasteur scheme could not differentiate the isolates, but the Oxford scheme was capable of differentiating the closely related isolates due to the greater number of sequence types [28,29]. However, A. baumannii has both high gene content variation [30] and substantial levels of recombination [31], suggesting insufficient discrimination among isolates, requiring a higher resolution for WGS [28,29]. ST191, ST369, and ST451 were identified in the isolates; these ST types are frequently isolated in Korea and other Asian countries [32,33] and thus could not discriminate the clones within the outbreak.
Clustering and phylogenetic analysis from wgMLST and wgSNP showed three clusters, and SNV analysis compared the similarities among the strains with high-effect SNVs. The discriminatory power was highest with the SNV analysis, but since the number of isolates was small, clustering results did not vary among the different methods. Cluster 1, including Isolates #2, #5, #6, #11, and #12, showed 100% similarity within the isolates by clustering analysis of wgMLST and wgSNP/SNV, suggesting a single origin. Isolate #2 was extracted from transtracheal aspirate specimens of an ICU patient. Isolate #5 was obtained from the keyboard of the main station at the ICU, and Isolate #6 was obtained from the bedrail of Ward A, where the patient from the ICU was transferred to. Isolates #11 and #12 were each obtained from suction catheters at the bedside from the ICU and Ward A, respectively. These results support the spread of the same clone in the ICU and Ward A

Discussion
Outbreak analysis of 11 A. baumannii isolates was performed with WGS and rep-PCR, and three clonal lineages were identified from the outbreak. The coexistence of bla OXA-23 , bla OXA-66 , bla ADC-25 , and armA with additional aminoglycoside-inactivating enzymes, predicted to confer multidrug resistance, was identified in all the isolates.
Although both the MLST Pasteur and Oxford schemes were based on seven loci, the Pasteur scheme could not differentiate the isolates, but the Oxford scheme was capable of differentiating the closely related isolates due to the greater number of sequence types [28,29]. However, A. baumannii has both high gene content variation [30] and substantial levels of recombination [31], suggesting insufficient discrimination among isolates, requiring a higher resolution for WGS [28,29]. ST191, ST369, and ST451 were identified in the isolates; these ST types are frequently isolated in Korea and other Asian countries [32,33] and thus could not discriminate the clones within the outbreak.
Clustering and phylogenetic analysis from wgMLST and wgSNP showed three clusters, and SNV analysis compared the similarities among the strains with high-effect SNVs. The discriminatory power was highest with the SNV analysis, but since the number of isolates was small, clustering results did not vary among the different methods. Cluster 1, including Isolates #2, #5, #6, #11, and #12, showed 100% similarity within the isolates by clustering analysis of wgMLST and wgSNP/SNV, suggesting a single origin. Isolate #2 was extracted from transtracheal aspirate specimens of an ICU patient. Isolate #5 was obtained from the keyboard of the main station at the ICU, and Isolate #6 was obtained from the bedrail of Ward A, where the patient from the ICU was transferred to. Isolates #11 and #12 were each obtained from suction catheters at the bedside from the ICU and Ward A, respectively. These results support the spread of the same clone in the ICU and Ward A by the transfer of a patient. In Cluster 2, #3 and #8 showed 100% similarity, suggesting the same clonal origin. Two patients were in the ICU during the same time period. The patients from Cluster 1 also stayed in the ICU during the same period as the patients in Cluster 2, suggesting multiple clones during the outbreak. Although #4 was clustered with #3 and #8, it showed similarities of 97.5% by wgMLST and 96.6% by wgSNP, and SNV analysis with PATRIC showed a similarity of 72.7% to #3, suggesting a different clone. A previous study suggested a threshold of 2.5 SNPs to distinguish outbreak isolates from non-outbreak isolates in A. baumannii [34], and the isolate sets with less than three SNV differences and sharing more than 90% of SNVs were #1 and #7; #2, #5, #6, #11, and #12; and #3 and #8, where the 2016 outbreak isolates were mainly grouped into two clones.
Despite the complexity of WGS, we were able to analyze WGS, including MLST, SNV and phylogenetic tree analysis, all commonly utilized for outbreak analysis, with the webbased public tools Pathogenwatch and PATRIC [11,21,35]. We used the DiversiLab system for rep-PCR and showed similar results to the WGS phylogenetic tree and core genome clustering, as previously reported [36,37]. rep-PCR can be performed in a relatively short time without extensive post-experiment analyses, but the exact sequences are not available, and variabilities are reported for interlaboratory comparison of fingerprints [38,39]. WGS has the advantage of reanalyzing the results with different sequence types and can provide additional information on AMR, virulence genes and transmission scenarios [10,40]. WGS has sufficient resolution to determine transmission within clonal outbreaks [34,37]. In most cases of outbreak investigation, the source case is not identified, and genomic variability increases over time, making it hard to determine the threshold point [41]. However, it should be noted that the "significance" of the difference between isolates should be judged based on a comprehensive understanding of the genetics and epidemiology of the pathogen, the setting within which the issue is being studied, and the tools being used in the investigation [41].
AMR prediction with ResFinder and KmerResistance showed high concordance, with most of the resistance genes identified commonly. The isolates showed a common antibiotic resistance profile reported in multidrug-resistant A. baumannii in Korea [32,42,43]. A previous study showed high concordance with ResFinder, an assembly-based tool, and KmerResistance, a read-based tool, in high-quality sequencing results, as in our study [26]. False-positive predictions through sequencing-based methods may be possible, requiring caution for interpretation [18,44]. Excluding the discrepancy of one isolate for resistance of sulfonamide gene and the discrepancies for prediction of quinolone resistance for five isolates, CRABs from the outbreak showed resistance to all the above antimicrobial agents except for colistin and the results correlated with the in silico prediction results. Only few comprehensive studies have investigated the concordance between the prediction for resistance by WGS and the conventional phenotypic antimicrobial susceptibility testing [45]. Previous studies have shown various concordance rates varying by antimicrobial agents [46,47]. The discordance of prediction with fluroquinolone resistance has been reported regarding the aac(6 )-Ib-cr, where an non-wild type is linked to specific sequences only [25] and the fluoroquinolone resistance are predicted through other genes including gyrA, parC, not included in the prediction tool, ResFinder [25,47]. The high concordance of genotype-phenotype correlation for antimicrobial resistance excluding the fluoroquinolone in our study, is possibly due to the inclusion of isolates within a single outbreak, which showed multi-resistance to the antimicrobial agents, where multi-resistance of A. baumannii is common in Korea. If various CRABs from different clinical background had been included, then the prediction may have showed discrepancies among the antimicrobial agents.
Despite the wide applicability of the WGS in outbreak analysis, the cost of WGS is still expensive, considering the equipment set up, cost of sequencing and bioinformatics analysis [48]. There are few studies evaluating the cost-effectiveness of bacterial WGS surveillance compared to the standard of care in detecting hospital outbreaks. Kumar et al. suggests that preliminary studies show WGS surveillance could be a cost-effective strategy [49,50]. However, clinical settings vary and cost-effectiveness has only been studied in certain organisms such as Klebsiella pneumoniae and Pseudomonas aeruginosa, thus more economical assessment would be necessary.
Although our study was a retrospective study, when WGS is performed during an outbreak, it may inform appropriate patient isolation protocols that could aid in the control of an outbreak [10]. Through recent advances in bioinformatics tools for WGS, outbreak analysis can be performed in a relatively short time. We characterized the clonal clusters in a nosocomial outbreak of CRAB in a tertiary hospital with WGS, supporting the use of WGS in healthcare infection epidemiologic studies.