Subtypes, resistance and virulence platforms in extended-drug resistant Acinetobacter baumannii Romanian isolates

Acinetobacter baumannii has emerged worldwide as a dominant pathogen in a broad range of severe infections, raising an acute need for efficient antibacterials. This is the first report on the resistome and virulome of 33 extended drug-resistant and carbapenem-resistant A. baumannii (XDR CRAB) strains isolated from hospitalized and ambulatory patients in Bucharest, Romania. A total of 33 isolates were collected and analyzed using phenotypic antibiotic susceptibility and conjugation assays, PCR, whole-genome sequencing (WGS), pulsed-field gel electrophoresis (PFGE) and MultiLocus Sequence Typing (MLST). All isolates were extensively drug-resistant (XDR), being susceptible only to colistin. The carbapenem resistance was attributed by PCR mainly to blaOXA-24 and blaOXA-23 genes. PFGE followed by MLST analysis demonstrated the presence of nine pulsotypes and six sequence types. WGS of seven XDR CRAB isolates from healthcare-associated infections demonstrated the high diversity of resistance genes repertoire, as well as of mobile genetic elements, carrying ARGs for aminoglycosides, sulphonamides and macrolides. Our data will facilitate the understanding of resistance, virulence and transmission features of XDR AB isolates from Romanian patients and might be able to contribute to the implementation of appropriate infection control measures and to develop new molecules with innovative mechanisms of action, able to fight effectively against these bugs, for limiting the spread and decreasing the infection rate and mortality.

The PFGE results revealed that there was no outbreak or spread of one single genotype in the clinical strains that were analyzed. However, clones I, II, III, IV and V were associated with hospital A, clones II, VIII and IX with the ambulatory unit B and clones IV, VI and VII were found only in hospital C.
Genomic analyses. The genomes of 7 selected CRAB strains encoded 01s, 14s, 10s, 18s, 24s, A07 and A14 were fully sequenced and analyzed to have a complete picture of the antibiotic resistance and virulence genes repertoires. The analyzed strains were selected for WGS based on the isolation source, all being isolated from healthcare-associated infections, respectively catheter-associated bloodstream infections, ventilator-associated pneumonia and central nervous system infections. Draft-genome sequencing analysis revealed that the chromosome size varied, as expected, between 3.86 and 4.071 Mbp. Instead, the analyzed strains harbored diverse mobile genetic elements. More than half of the CDS were functionally annotated by the RAST program. The general features of the genomes are presented in Table 2.

msr(E), mph(E) msr(E), mph(E)
Tetracycline Quinolones resistance   www.nature.com/scientificreports/ phospholipids degradation (plc, plcC, plcD) and apoptosis induction ( Table 3). Two of the strains (14s and A14) contain the hemO (hemoxigenase) gene, which was associated with the hypervirulent phenotype of A. baumannii LAC 4 19 . The ability to produce various virulence factors of these A. baumannii strains could explain their ability to persist and colonize the human host leading to a serious threat for hospitalized patients. The predicted serological typing scheme for the A. baumannii strains analyzed in this study shows some aspects related to their virulence. It is known that the major immunogenic polysaccharide which is produced A. baumannii as in important virulence factor is the capsular one (K) and not the somatic antigen O, since the non-capsulated strains don't cause infections 20 . The Supplementary Fig. 1 shows the genetic variability in the K and O loci, the metrics related to the match confidence, coverage, identity and number of genes, as well the locus type for each of the 7 strains that were sequenced in this study, and their relationships with regards to the subtype.
Pangenome, phage and genomic islands analysis. Genomic Islands. Genomic islands (GI) represent proof of horizontal gene transfer in a population; these DNA segments may integrate into the chromosome of the host and undergo transformation, conjugation or transduction 21 . The GI predictions showed that only 4 of the 7 strains belonging to ST1, ST2 and ST636, harbored genomic islands: 3 with putative type IV secretion system (T4SS) ICEs and one with putative IME.
The strain 18s has a putative IME region of 31 kb (41%GC) containing 39 ORFs, some of them encoding for TrbI T4SS component, Rep_trans relaxase and rve integrase.
Further, considering the reference genomes used for phylogenetic analysis and the same GI prediction method, the GI were encountered in 19 out of 69 genomes, with sets of genes similar to the selected sequenced samples from the same phylogenetic cluster. While most of the genomes have only one GI region predicted, 3 of the reference genomes had 2 GI regions (Suppl. Table 1).
Most of the genomes with predicted GI belong to the ST1 cluster. Compared with the references from the same subtype, 18s GI exhibits: (i) the lowest number of ORFs, thus fewer known protein types, even though its length is similar with other Putative IME GI from different references; (ii) the highest GC percentage (41.19%) in the genomic island (Suppl. Table 1).
When compared with the reference genomes, the strains 24s and 01s belonging to ST636, and A14 from ST2 have the longest GI, with the highest number of ORFs and similar GC.
Beside this prediction, traces of TnAbaR were also found. TnAbaR is a core composite transposon bound by inverted repeats and 2 copies of direct repeats at its ends, forming an AbaR resistance island. Usually, AbaRs are inserted in A. baumannii genes, leading to the loss of their function-in this case, comM (an ATPase-encoding gene known as a hotspot for the integration of AbaR). These islands consist mainly of MGEs, such as transposons or integrons, and various genes that confer MDR 22 .
Although they fully cover this resistance island, 14s and 01s strains have the lowest overall identity (20% and 25%, respectively) with TnAbaR23. The contig from 01s has 100% identity to the partial sequence of comM gene, while 14s has the highest local identity (97.1%) only to TnpA IS15DI transposase from Module_I. The strains 24s has an overall 47.1% identity with TnAbaR23, though having 92%, 100% and 100% identity to Module_I,_J and _K, respectively, while A07 has an overall 37.2% identity, though having 100% and 83.6% identity to Module_A and _B, respectively.
The other strains that partially cover TnAbaR23 also have low identities: 10s has an overall 27.1% identity (with 97.1% identity to TnpA gene from Module_I). A14 has an overall 32.8% identity, of which modules A to first half of Module_F have no coverage, the only high similarity region comprising the end of Module_F and the start of Module_G.
On the other hand, the contigs of 18s cover almost entirely TnAbaR23, with 99.97-100% identity to Module_A,_B,_C,_D,_F and _K, 93.4% to Module_G and 36% to Module_I. Almost all of Module_J has no coverage (highlighted in yellow in Suppl. Figure 2): the end part of the CadR (transcriptional regulator of MerR family), whole CadA (heavy metal transport/detoxification protein), whole LspA (prolipoprotein signal peptidase) and almost all of the transposase in this module (protein ID: AFB76410).
The TnAbaR-like island of 18s had most similarities to other TnAbaR islands, mostly on modules A to half of G, and K, given by the backbone transposon Tn6019 (Suppl. Figure 2). The highest similarity is with TnAbaR23 (as it can be seen in the identity distance matrix from Suppl. Figure 2). The region in Module_G (highlighted in cyan in Suppl. Figure 2) Figure 2), where the 18s resistance island may acquire structural variation compared to the others.
Phages. The most phage-abundant regions were found in 2 strains that belong to the ST492 subtype. In these strains we found the highest number of intact phages, phage species and attachment sites for phages compared to the other strains with other subtypes (Suppl. Table 1).
Pangenomes. The pangenome analysis shows that the total number of genes found in each of the samples is approximately 3700. Around 43% of these are represented by hypothetical proteins (Fig. 1b). The genomes were subsequently compared one with each other as a matter of unique genes. Then, the lowest numbers taken by subtype belong to ST492 and ST636, while the highest numbers are found in ST2 (6 to 8 times higher than the previous two) and ST1 (almost 3 times higher than in ST2). The ST492 is of particular interest, considering the subtype is relatively new and not much is known about it in literature. Taking into account that the highest number of phages, but no genomic islands were predicted in two ST492 strains, we took a closer insight into their genome. When compared to the other genomes considered in our phylogenetic analysis, the 2 ST492 strains contain 19 unique genes that were not found in any other genome from the selected ones (including 3 other ST492 reference genomes). Of these 19 genes, 17 are hypothetical and the other 2 are represented by: dnaB2 (a replicative DNA helicase) and bfrD1 (a putative TonB-dependent receptor BfrD). These 2 ST492 strains have 62 proteins in common, of which 57 are hypothetical proteins while the other 5 are represented by: fpvA (ferripyoverdine receptor), pucD (putative xanthine dehydrogenase subunit D), folE (GTP cyclohydrolase 1), bfrD1 and dnaB2. When considering all 5 ST492 genomes from the phylogenetic analysis set, then there are 18 proteins unique only to ST492 of which 11 are hypothetical and the other 7 proteins are: ptk (tyrosine-protein kinase), ptp (low molecular weight protein tyrosine-phosphatase), glxR2 (2-hydroxy-3-oxopropionate reductase), wbpA (UDP-N-acetyl-D-glucosamine 6-dehydrogenase), pglH (GalNAc-alpha-(1-> 4)-GalNAc-alpha-(1-> 3)-diNAcBac-PP-undecaprenol alpha-1,4-N-acetyl-D-galactosaminyltransferase), mshA3 (D-inositol-3-phosphate glycosyltransferase) and pglA (N,N'-diacetylbacillosaminyldiphospho-undecaprenol alpha-1,3-N-acetylgalactosaminyltransferase).
All 76 samples that were used for the phylogenetic analysis have in common approximatively 2000 core genes and present other 8-9000 accessory genes. This suggests that the selected genomes for the phylogenetic and pangenome analysis have some active MGEs, which may contain various types of resistance genes, integrases, transposases-already predicted and mentioned within the body of this manuscript. www.nature.com/scientificreports/ Phylogenetic analysis. The phylogenetic relationships between the selected 76 genomes show a clear clustering between the subtypes of interest in which the studied strains fit in. It also shows the separate evolution of ST492 from ST2 (as previously mentioned, ST492 and ST636 are single and respectively triple locus variants of ST2) 23 . The other 8 randomly selected subtypes cluster separately (Fig. 1a). Few other studies on ST492 have been mentioned in literature 24,25 , but the genomes may have not been uploaded in NCBI database 26 , although not much is known about this subtype, therefore only 5 genomes for ST492 have been used for the phylogenetic analysis. The 2 ST636 strains from this study (01S and 24S) group together, separately from other ST636, while the ST492 ones (10S and 14S), even if they are a subbranch of ST2, are in the same cluster with the other two ST2 samples (A07 and A14). Similar to the phylogenetic relationship and MLST predictions, some of the genes are expressed only in the isolates from the same branch (e.g., bla ADC-74 predicted to be present only in 01S and 24S isolates (ST636); bla ADC-11 , bla PER-1 only for A07 and A14 isolates belonging to the ST2 clone; bla ADC-30 tet(B), sul2-only for 10S and 14S isolates (ST492 clone). Based on the variant results, approximately 1.25% average of the whole A. baumannii genome length in each of the 7 strains proved to be variants. At the same time, most of the variants predicted by ARIBA are found in the variant list from snippy 27 , in the same or very similar genes.

Discussions
A. baumannii is one of the most successful pathogens responsible for nosocomial infections, occurred especially in patients admitted to intensive care units (ICUs), but also for community-acquired infections, being able to acquire resistance to carbapenems, fluoroquinolones and aminoglycosides 28 . Due to the limited options for the antibiotic treatment of the produced infections, CRAB isolates became a significant health problem worldwide 29 . Although antibiotic resistance may not be considered a traditional virulence factor, in case of A. baumannii it is by far the biggest driver of the clinical outcome by precluding the clinician's ability to eliminate the infecting strain. Our experimental data have shown a high prevalence of imipenem resistance (96.96%) among A. baumannii strains isolated in 2017-2018 from hospitalized and community-acquired infections in patients from Bucharest. This percentage was close to reported rates in Egypt, Turkey, Spain and Italy and higher than from Saudi Arabia [30][31][32][33][34] .
To date, in Romania, there has been reported a high prevalence of CRAB strains in different parts of the country including Bucharest, the capital city 35 . Even though bla OXA-58 has been identified in A. baumannii (ST1 clone) recovered from patients in our country 36 , the ST2 associated with bla OXA-23 remains the most common among the CRAB isolates. A pilot study from 3 Romanian hospitals-from Iași and Târgu-Mureș, performed between 2014 and 2015-showed the presence of carbapenemases OXA-23 and OXA-24/72 in A. baumannii nosocomial isolates 37 .
Regarding the distribution of the carbapenemases genes among the identified clones, it has been observed that the ST2 clone encountered in hospital A was associated with the production of bla OXA-23 gene (10/33; 30.30%) carbapenemase and revealed, in most cases, the presence upstream the carbapenemase gene, of the ISAba1 element (9/33; 27.27%). The presence of ISAba1 upstream bla OXA-23 and bla OXA-51 is required to confer resistance to carbapenems 38 . Although the relationship between ISAba1 upstream bla OXA-51 and carbapenem resistance was confirmed, this might not be enough to confer resistance, as A. baumannii isolates susceptible to carbapenems with the association ISAba1/ bla OXA-51 have already been described 39 . The clone harboring 2 carbapenemases: OXA-23 and OXA-51 was identified in the ambulatory sector (clinical unit B) ( Table 1); the ST636 clone was related with the presence of OXA-24 carbapenemase, most of the strains being isolated from hospital C (7/33; 21.21%) and 2 strains from the ambulatory. The other isolates belonging to STs 492, 312, 642 and ST1 were associated with bla OXA-24 gene ( Table 1). The distribution of these clones per hospital unit revealed that ST492 and ST1 were found exclusively in hospital C; ST642 in hospital A, while ST312 was identified in both hospital units (A and C) ( Table 1).
Currently, worldwide carbapenem resistant strains are mostly associated with international clone II, with bla OXA-23 as the main carbapenem resistance mechanism 12 . In Greece, it has been observed that the ST2 was the most common clone circulating in Greek hospital settings 40 . With regards to MDR, several other authors have demonstrated the association of bla OXA-23 , bla OXA-58 , bla OXA-72 and ST2 24,[41][42][43][44] . Furthermore, the international clone ST2 was found to be broadly spread among our country 45  Similar to our results, the presence of multiple ARGs for β-lactams, aminoglycosides, sulfonamide and tetracyclines was evidenced in clinical MDR A. baumannii isolates from Spain and Switzerland 49,50 .
Some of the identified plasmids (e.g. pACICU-2 like ST2 carrying bla OXA-23 ; pMAL-1like ST492 and ST1 carrying bla OXA-72 ) were previously described in our country, in A. baumannii strains belonging to different clones 35 www.nature.com/scientificreports/ gene; pACICU-2 like and pMAL-1 like ST2 carrying bla OXA-23 ) are reported for the first time in communityacquired and intra-hospital A. baumannii strains isolated in Bucharest. In our study, due to sequencing limitations, the bla OXA-23 gene location was not identified. Our data revealed that community-acquired A. baumannii harbored pACICU-2 linked to ST636 producing OXA-72; pMAL-1linked to ST492 and ST1 producing OXA-72 and clinical CRAB pACICU-2 like and pMAL-1linked to ST2 OXA-23 producing. The OC locus prediction shows that the analyzed strains contain 3 locus types: OCL1, OCL2 and OCL4. OCL1, present in 4 out of 7 strains, is known to be the most abundant gene cluster in the major global clone groups GC1 and GC2, of which ST1 and ST2 belong to. In this case the ST1 strain belongs to OCL4 and the ST492 to OCL1 (supporting thus the hypothesis that ST492, although rare, resulted from ST2, as suggested by phylogeny, or vice versa). The major difference is found between the strains belonging to the ST636 subtype, where wecB gene, known to be involved in the biosynthesis of sialic acid, occurs between the glycosyltransferases gtrOC8 and gtrOC9 52 .
On the other hand, the diversity is increased in the K loci, the 7 strains containing 5 different gene clusters: KL1, KL3, KL30, KL40 and KL77.
The major differences appear between the capsular export region and the repeat unit of translocase (wzx), with the presence or absence of various types of genes: UDP-N-acetylglucosamine-2-epimerase (mnaA) and UDP-N-acetylmannosamine dehydrogenase (mnaB) in the ST636 strain; UDP-d-GlcpNAcA epimerase (gne2) in ST1; d-glucosaminate PTS permease components EIIA, EIIB, EIIC (dgaA, dgaB, dgaC) in A14 from ST2; Photosystem I P700 chlorophyll a apoprotein A1 & A2 (psaA, psaB), Photosystem I iron-sulfur center (psaC), Photosystem I reaction center subunit II & IV & III (psaD, psaE, psaF) in A07 from ST2. Actually, A07 strain has the most diverse K locus from the analyzed strains-there is also an acyl/acetyl transferase (e.g. atr20) which is inserted in the simple sugar synthesis region. More than that, the insertion of atr20 may suggest a novel K locus even though it matches the K77 reference, according to the Kaptive documentation. Similarly may happen to the K loci in ST492 strains, since there is no coverage in the respective region for wzy, gtr63 and itrA2; this finding, together with the low matching confidence and missing genes between gna and wzx compared with the other samples, is suggesting a possible novel variant. We also revealed that the p-ACICU-2like and pMAL-1like carrying bla OXA-23 , p-ACICU-2like carrying bla OXA-72 and pMAL-1like carrying bla OXA-72 presented virulence-related genes involved in iron acquisition (such as the TonB-dependent receptor), adherence (ompA), biofilm formation (pgaABCD locus), invasion (septicolysin) and haemolytic activity against human erythrocytes, aiding in iron acquisition) (e.g. plc, plcC and plcD) 53 . TonB-dependent transporters are outer membrane proteins (OMPs) that bind and transport siderophores in addition to vitamin B12, nickel complexes, and carbohydrates and may be involved in the survival of bacteria in the lungs central nervous system and blood 54 . Septicolysin, on the other hand, is a pore-forming toxin with a cytolytic activity that mediates invasion of tissues or cells 55 .
Pulsed-field gel electrophoresis (PFGE). The clonality of bla OXA-23 and bla OXA-24 carrying A. baumannii was determined with PFGE using the ApaI enzyme and CHEF-DRII apparatus (Bio-Rad). Pulsotypes were defined as isolates with PFGE band patterns of 85% similarity or above 62 . The isolates were classified according to the criteria described 63  www.nature.com/scientificreports/ Mating experiments. Transferability of bla OXA-23 by conjugation was tested in solid mating using a rifampicin (RIF)-resistant Acinetobacter baylyi ADP1 as a recipient. Briefly, equal amounts (100 µL) of overnight cultures of the donor (A14; A17; 8A) and recipient strains were mixed and incubated in Brain heart infusion agar plates, cells were resuspended in saline solution and selected in plates containing RIF (300 mg/L) and MEM (0.5 mg/L) 64 . Characterization of the transconjugants was conducted by PCR to confirm the conjugative transfer.
Multilocus sequence typing (MLST). MLST  Whole-genome sequencing and bioinformatic protocols. The genomes of 7 A. baumannii strains isolated from different infections were sequenced using HiSeq and MiSeq X Ten, Illumina. The paired-end raw reads have been submitted to the GenBank under accession numbers: SRX4094320, SRX4094321, SRX4094322, SRX4094323, SAMN14525745, SAMN14525746, SAMN14525747 and were subsequently assembled using shovill 66 with SPADES 3.12.0 67 already implemented, while the quality of assembling was checked with QUAST 68 . The samples underwent a reference mapping step in Geneious Prime 2020.1.1 69 , with trimming in a 5 iteration step and taking into account only the reads with a quality factor higher than Q30 resulting in coverage of approximatively 35x-to-38x on the whole genome. Annotations were performed with RAST 70 and Prokka 71 while the prediction of resistance and virulence profiles was performed by using the following programs and databases: Abricate (compared against VFDB database for virulence factor prediction), ARIBA (downloading and preparing at first the reference data from NCBI database), ResFinder (compared against implemented NCBI database), PlasmidFinder (compared against implemented plasmidfinder database), PathogenFinder, CARD, PubMLST, ISfinder, PATRIC [72][73][74][75][76][77][78][79][80][81][82][83] . The resistance genes and virulence factors were selected based on the highest identity and coverage. Plasmid replicons were detected using AB-PBRT 84 by an in silico PCR approach sequences 85 . The untypable plasmids were compared with the NCBI nucleotide database, while the plasmids integrated within the bacterial chromosome were detected using the BLAST tool 86 using a command-line interface. All the other predictions were run by using default parameters. Integron analysis was performed using command-line blastn 86 , with 5'CS and 3'CS sequences (Supplementary file 1) as queries against assembled genomes. Integron sequences annotations were performed manually.
Additional plasmids analyses were performed using the PLSDB database Search tool 87 , using the Mash strategy, with search parameter set as default (max. p-value at 0.1 and min. identity 0.99).
The phylogenetic analysis was performed using the following protocol: (i) taking into account the subtyping information from MLST prediction for the 7 strains discussed in our study, the assembly contigs from 69 reference sequences were selected from the NCBI database 26 . After repeating the MLST prediction on the majority of A. baumannii genomes from NCBI (that were uploaded as contigs) the selection of the 69 sequences used for phylogeny was done randomly. Approximatively 20 genomes were selected from each subtype corresponding to the subtypes found in our samples (except ST492-only 3 out of ~ 6000 genomes were found) while other 8 genomes belong to different subtypes. (ii) Whole genome annotation with Prokka (using genetic code 11 and Genbank/ENA/DDJB compliance options) was performed for each of the 76 genomes selected for phylogeny. (iii) the output from Prokka was used for pangenome analysis using Roary 88 . (iv) the phylogenetic tree was generated (on the alignment of core genome gene sequences in Roary with MAFFT 89 , using FastTree 90 with Jukes-Cantor model, Maximum-likelihood and minimum-evolution NNIs and SPRs algorithms, implemented in Geneious Prime 2020.1.1. The phylogenetic tree image was generated using FigTree software (http:// tree. bio. ed. ac. uk/ softw are/ figtr ee/). The Genomic Islands predictions were performed with ICEfinder online tool 91 , while the phage predictions were done with PHASTER online tool 92 . K and OC locus predictions were performed using the Kaptive software 20 . TnAbaR sequences have been aligned with MAFFT, implemented in Geneious Prime.

Conclusions
Our results highlight the presence of an impressive armamentarium of ARGs and of mobile elements required to form their mobilization and transmission, such as the plasmids, transposons and ISs in CRAB strains isolated from Romanian patients. These strains belong to widespread clones reported in intrahospital infections or community-acquired patients. The most frequently encountered clones identified in A. baumannii clinical strains were ST2, ST636 and ST492, while the most frequent carbapenemase genes were bla OXA-23 or bla OXA-24 . Among resistance determinants, several virulence genes, as well as factors that contribute to the persistence of these bacteria in the hospital environment, have been detected. All sequenced A. baumannii isolates have the genetic equipment conferring them the ability to grow in iron-depleted media and to survive in the presence of desiccation, antimicrobials and toxic compounds. Our data will facilitate the understanding of resistance, virulence and transmission features of XDR AB isolates from Romanian patients and might contribute to the implementation of appropriate infection control measures for limiting the spread and decreasing the infection rate and mortality.