Comparative genomic analysis of Acinetobacter baumannii clinical isolates reveals extensive genomic variation and diverse antibiotic resistance determinants

Background Acinetobacter baumannii is an important nosocomial pathogen that poses a serious health threat to immune-compromised patients. Due to its rapid ability to develop multidrug resistance (MDR), A. baumannii has increasingly become a focus of attention worldwide. To better understand the genetic variation and antibiotic resistance mechanisms of this bacterium at the genomic level, we reported high-quality draft genome sequences of 8 clinical isolates with various sequence types and drug susceptibility profiles. Results We sequenced 7 MDR and 1 drug-sensitive clinical A. baumannii isolates and performed comparative genomic analysis of these draft genomes with 16 A. baumannii complete genomes from GenBank. We found a high degree of variation in A. baumannii, including single nucleotide polymorphisms (SNPs) and large DNA fragment variations in the AbaR-like resistance island (RI) regions, the prophage and the type VI secretion system (T6SS). In addition, we found several new AbaR-like RI regions with highly variable structures in our MDR strains. Interestingly, we found a novel genomic island (designated as GIBJ4) in the drug-sensitive strain BJ4 carrying metal resistance genes instead of antibiotic resistance genes inserted into the position where AbaR-like RIs commonly reside in other A. baumannii strains. Furthermore, we showed that diverse antibiotic resistance determinants are present outside the RIs in A. baumannii, including antibiotic resistance-gene bearing integrons, the blaOXA-23-containing transposon Tn2009, and chromosomal intrinsic antibiotic resistance genes. Conclusions Our comparative genomic analysis revealed that extensive genomic variation exists in the A. baumannii genome. Transposons, genomic islands and point mutations are the main contributors to the plasticity of the A. baumannii genome and play critical roles in facilitating the development of antibiotic resistance in the clinical isolates. Electronic supplementary material The online version of this article (doi:10.1186/1471-2164-15-1163) contains supplementary material, which is available to authorized users.


Background
A. baumannii, an important nosocomial pathogen, is becoming an increasing threat to hospital patients due to its ability to develop multidrug resistance (MDR) [1][2][3]. Drug resistance in A. baumannii is due to a combination of mechanisms, including the expression of β-lactamases, alteration of cell membrane impermeability, and increased expression of efflux pumps [4]. The drug resistance genes of A. baumannii isolates are often clustered into antibiotic resistance islands (AbaRs) that interrupt the ATPase gene (comM) [5]. For example, a 86-kb resistance island (RI) was found in A. baumannii strain AYE (AbaR1) [6], and a shorter RI was identified in the A. baumannii strain ACICU (AbaR2) [7]. RIs are thought to emerge from the integration of plasmids or other mobile elements, and some drug-susceptible strains lack these RIs [6]. In addition, plasmid-borne resistance genes have also been reported, e.g., the bla OXA-23 gene, which is associated with carbapenem resistance, has been identified in clinical A. baumannii isolates around the world [8,9].
Compared with current knowledge regarding antibiotic resistance mechanisms in A. baumannii, less is known regarding the virulence factors in this bacterium [10]. Several studies have focused on characterizing the formation of biofilms, one of the determinants involved in the pathogenesis in A. baumannii. For example, a chaperoneusher pili assembly system (csu locus) has been shown to be involved in attachment and biofilm formation in A. baumannii [11]. Other virulence factors identified include a siderophore-mediated iron acquisition system, AbaI autoinducer synthase, the BfmRS two-component regulatory system, the type VI secretion system (T6SS) [12] and lipopolysaccharide (LPS) [13]. The LPS found in A. baumannii, which is composed of lipids, O-antigen, and an outer core (OC) and inner core, has been shown to be a major contributor to the pathogenesis of infection [13]. The OC gene locus contains many genes encoding glycosyltransferase enzymes that catalyze the bonds between sugars in the OC structure [14].
Whole-genome sequencing studies comparing distinct drug-susceptible and MDR strains [1,15] or isolates from a single patient [16] have improved our understanding of the evolution of A. baumannii. To better understand the genomic variation and the antibiotic resistance mechanisms in A. baumannii, here we sequenced eight clinical A. baumannii isolates with various sequence types and drug susceptibility profiles and performed comparative genomic analysis.

Results
Susceptibility profiles, multilocus sequence typing (MLST) and whole-genome sequencing The susceptibility profiles for all sequenced strains are shown in Table 1. All 7 MDR strains were resistant to the antibiotics gentamicin (CN), ciprofloxacin (CIP), ceftriaxone (CTR), ceftazidime (CAZ), cefepime (FEP), and tetracycline (TE) but susceptible to polymyxin B (PB). The drug-sensitive strain BJ4 was sensitive or intermediate to all tested antibiotics except CTR.
We found that all 7 MDR strains correspond to global clone II (GC II). The strains BJ2, BJ6, and BJ7 share the same sequence type (ST), namely, ST208, and strains BJ1 and BJ5 share a type (ST191). In addition, strains BJ3 and BJ8 belong to ST218 and ST368, respectively. However, the drug-sensitive strain BJ4 shows a novel sequence type.
The basic whole-genome sequencing statistics are shown in Table 2. Illumina 100 bp paired-end sequencing produced more than 900 Mb of data for each of the eight strains, and the sequencing depth ranged from 239× to 473×. The GC content of the genomes was approximately 38.9%, as expected for the species. The size of the genomes varied from 3.86 to 4.03 Mb.

Phylogenetic analysis of A. baumannii isolates
A maximum-likelihood tree of the 8 sequenced genomes and 16 reported A. baumannii complete genomes were created based on core SNPs from whole-genome alignment ( Figure 1). The phylogenetic tree showed that the previously sequenced strains and all of the 7 MDR clinical isolates belonging to GC II formed a clade, while strains AB307-0294, AYE, and AB0057, which belong to GC I, grouped together. The BJ1 and BJ5 strains are closely related, while strains BJ2, BJ6, and BJ7 form another closely related group. Interestingly, strain BJ4, the drug-sensitive strain, is distinct from all of the sequenced MDR strains, which may indicate that it has a unique origin compared with other drug-resistant strains. A. baumannii core and unique genes We compared the gene contents of the 8 genomes with other A. baumannii reference genomes using the PanOCT analysis software [17], which utilizes conserved gene neighborhood (CGN) and frameshift detection in a weighted scoring scheme and the BLAST score ratio to effectively generate non-paralogous gene clusters. We found that the pan-genome continued to expand after the compilation of 24 genomes, whereas the number of core genes remained relatively stable with the addition of new strains ( Figure 2A). The size of the pan-genome was 8245 genes, and there are 1902 genes (core) shared among the 24 isolates ( Figure 2B). The number of unique genes ranges from 7 in strain BJ1 to 552 in strain SDF ( Table 3). Many of these unique genes are hypothetical, transposon-related and phage-related genes. Detailed information regarding orthologous groups and singletons of the strains is provided in Additional file 1: Table S1. The large number of unique genes in these genomes likely indicates frequent horizontal gene transfer events in A. baumannii. Hierarchical clustering of these strains based on gene content yields a dendrogram ( Figure 2B) that is similar to the core SNP-based phylogenetic tree ( Figure 1) in which strains from GC I form one group and strains from GC II form another group. We further analyzed the core and unique genes according to the various classes of the Clusters of Orthologous Groups (COGs) ( Figure 2C). We found that core genes were significantly enriched in genes belonging to class J (Translation, ribosomal structure and biogenesis; P value = 1.53e-09) and class F (Nucleotide transport and metabolism; P value = 0.0008301). In contrast, unique genes were significantly enriched in class L (Replication, recombination and repair; P value < 2.2e-16), class V (Defense mechanisms; P value = 1.868e-09), and class M (Cell wall/ membrane/envelope biogenesis; P value = 0.0005701).
To compare protein sequence evolution rates between the MDR isolates and the drug-sensitive isolates, we measured the nonsynonymous substitution rate (Ka or dN) in 1902 orthologous genes. We previously showed that this rate is a relatively consistent parameter for defining fast-evolving and slow-evolving protein-coding genes [18]. The fast-evolving genes we identified among the MDR isolates include many outer membrane proteins and stress-related proteins; one of these proteins is a phenazine biosynthetic PhzF-like protein that serves as an enzyme essential for phenazine synthesis. Phenazines are pigments, and many exhibit broad-spectrum antibiotic activity against bacteria, fungi, and parasites and can contribute to the ecological competence of the  Figure 1 Phylogenetic tree of A. baumannii isolates. A maximum likelihood tree was constructed using dnaml from the PHYLIP package, based on the core SNP in each genome. Bar, 0.01 substitution per nucleotide. "R" = multidrug resistance; "S" = drugsensitive; "UN" = unknown.
strains [19]. In contrast, the identified slow-evolving genes include many conserved hypothetical proteins and metabolism-related proteins. For example, SbmA, which is involved in the prokaryotic internalization of antimicrobial peptides (AMPs), was identified as a slowevolving gene [20].

SNPs among A. baumannii strains
The number of SNPs among the 7 MDR strains with distinct STs ranged from 920 to 2675 (Additional file 2: Table S2). The strains with the same STs showed fewer SNPs, ranging from 74 to 196. Among the 74 putative SNPs identified between BJ6 and BJ7, only 12 (16%) were nonsynonymous mutations; these SNPs were located within genes coding for outer membrane receptor for monomeric catechols, dihydropteroate synthase, fatty acid desaturase, and a putative RND superfamily exporter. We also found similar nonsynonymous mutations within all of the genes mentioned above between BJ1 and BJ5.
To identify SNP regions clustered among the 7 MDR strains, SNP density was estimated throughout the genomes using a sliding window of 5 kb. The resulting SNP density map shows a non-random distribution, with many regions having elevated SNP density (Additional file 3: Figure S1). One large region of elevated SNP density is around the origin of replication of the genome and the K locus, as reported by Snitkin et al. [21]. We also found other SNP clusters containing genes involved in heme utilization, arginine and proline metabolism, the ABCtype transport system, etc.

Large genomic variants among A. baumannii strains
We compared the genomes of our 8 A. baumannii strains with the reference genome of A. baumannii MDR-TJ, a multidrug resistance strain belonging to GC II group [22]. We identified many highly variable regions ( Figure 3); specifically, the following regions on the MDR-TJ genome are missing or have low identity with our strains: from 982 to 1,034 kb, 1,343 to 1,363 kb, 1,364 to 1,400 kb, 1,575 to 1,617 kb, 2,460 to 2,500 kb, 3,672 to 3,710 kb and 3,798 to 3,894 kb.
The region from 982 to 1,034 kb was predicted to be the prophage locus. The sequence of strain BJ1 in this locus is highly similar to that of the reference genome, while other strains have variable sequences in this region (low protein identity or missing). Interestingly, an ISAba1associated deletion of approximately 20 kb in a region of adhesion genes (csuE) from 1,343 to 1,363 kb was absent from the strain BJ2 and from the previously reported reference strain MDR-ZJ06 [23]. The region from 1,364 to 1,400 kb encompasses a cluster of genes involved in iron acquisition. The region from 1,575 to 1,617 kb was predicted to be the second prophage locus. The approximately 40-bp region from 2,460 to 2,500 kb, which encodes the entire type VI secretion system (T6SS), was completely absent from strain BJ4. The region from 3,672 to 3,710 kb, which encodes the entire AbaR-like genomic island (RIs), was also completely absent from strain BJ4. A more detailed analysis of this island is shown in Figure 4. The region from 3,798 to 3,894 kb contains many highly divergent genes, including several membrane proteins, stress-related proteins, and efflux pumps. Specifically, the region from 3,869 to 3,894 kb encompasses a series of genes encoding the O-antigen component of LPS.
In addition, the above-mentioned variable regions are always accompanied by several insertion elements, which may assist the integration of resistant and pathogenesisrelated genes and facilitate the transfer of drug resistance and pathogenic genes among strains. In addition, IS elements may enhance drug-resistance and virulence by promoting gene expression [24,25]. Furthermore, the CRISPR (clustered regularly interspaced short palindromic repeats) systems, which were identified in the genomes of three GC I strains (AYE, AB0057 and AB307-0294), were not present in any of the 8 sequenced strains.

AbaR-like resistance islands (RIs)
We compared the sequences of AbaR-like RIs in each A. baumannii isolate and found a series of variation events at this locus ( Figure 4). AbaR-like RIs inserted in the comM gene were identified in all 7 MDR strains: AbaR BJ2 and AbaR BJ3 shared the same structure, AbaR BJ6 and AbaR BJ7 shared a structure, and AbaR BJ1 and AbaR BJ5 shared a structure. In addition, AbaR BJ8 , AbaR BJ2 and AbaR BJ3 were novel AbaR-like RIs, while the remaining regions have been previously reported [23,26]. AbaR-like RIs were not found in strain BJ4, which may partly explain its susceptibility to antibiotics. AbaR BJ8 shares the same backbone with AbaR MDR-TJ and consists of two Tn6022 transposons and 3 resistancerelated regions (RR) (Figure 4). RR1a inserted at the 3'end of the island contains ISAba1Δ (mobile element), sul2 (conferring sulfonamide resistance), glmM (phosphoglucosamine mutase) and tnsA (transposase protein A); RR2 located between the two copies of Tn6022 bears the antibiotic resistance gene sul2; RR3 at the 5' end of AbaR BJ8 contains four resistance genes: tetA and tetR (conferring tetracycline resistance), strA and strB (conferring streptomycin resistance). There was also a cluster of ORFs inserted between Tn6022 and RR2; this cluster is designated "ORFs" in Figure 4. Compared with AbaR BJ8 , the RR1a region and a part of the 3'end of Tn6022 in AbaR MDR-TJ were absent. AbaR BJ2 and AbaR BJ3 had a truncated RR1a (with a tnsAΔ) designated as RR1b; the rest of the structure was identical to that of AbaR BJ8 . AbaR BJ6 and AbaR BJ7 included Tn6022Δ1 and RR3 segments, while AbaR BJ1 and AbaR BJ5 only contained the Tn6022Δ1 segment without resistance genes.
Comparative analysis of AbaR-like RIs in the 7 MDR strains and the other 7 GC II isolates was also performed ( Figure 4). One major distinction among the RIs in the GC II strains was the presence or absence of a second Tn6022 copy. Tn6022 or Tn6022Δ were mutual segments that existed in all of these strains, while the three resistance regions (RR1, RR2 and RR3) and ORFs were deleted or truncated in certain strains. In AbaR BJAB0714 , AbaR TCDC-AB0715 and AbaR 1656-2 , large fragments of inserted segments were found: the vertical arrow in AbaR BJAB07104 indicates the insertion of Tn6206 and the tra system specific to this island [27]; the arrow in AbaR 1656-2 indicates an inserted RR3a segment containing the resistance genes per-1 and strA (tnpA-tnpA2-gst-per-1-tnpA1-insB-IS3-strA) but not the tetA(B) and tetR(B) regions of RR3 [28]; and the arrow in AbaR TCDC-AB0715 indicates a large segment specific to this strain containing six IS26 elements and multiple resistance genes [29].
Interestingly, the comM region of the drug-sensitive strain BJ4 was interrupted by a novel genome island (GI, designated GI BJ4 ). This island was 29.3 kb in length and contained no antibiotic resistance gene ( Figure 5). Rather, five metal-resistance genes, including cueR, zntA, arsR, czcD and xre were identified within this island. Furthermore, 12 orfs of unknown function were present in GI BJ4 , including a tnsA gene encoding an endonuclease domain protein, an int gene encoding an integrase core domain protein, the transposon-related tniB and tniQ genes and an inserted IS1236 segment.

Other mobile elements containing resistance genes
Class 1 integron is an important factor for the horizontal transfer of resistance genes in A. baumannii, especially the aminoglycoside resistance genes [30]. Six of our MDR strains contained a class 1 integron, and their gene cassette arrays were as follows: the integron of BJ1 harbored a gene cassette array of aacC1-orfP-orfP-orfQ-aadA1; the integrons of BJ2, BJ6 and BJ7 included gene a cassette array of aacC1-orfA-orfB-aadA1; and the integrons of BJ3 and BJ5 had a gene cassette array of aacA4-catB8-aadA1. Among these gene cassettes, aacC1, aadA1 and aacA4 are aminoglycoside resistance genes; catB8 is group B chloramphenicol acetyltransferase; and  Structure of genomic islands in the BJ4 strain. The structure was drawn to scale according to the sequences of the genomic island in the genome and showed that the comM gene was interrupted by a novel genome island (designated as GI BJ4 ). This island was 29.3 kb in length and did not contain any antibiotic resistance genes. orf P, orfQ, orfA and orfB encode proteins of unknown function. In addition, all of the found integrase protein sequences were 100% identical.
We also identified a bla OXA-23 -containing transposon Tn2009 in all sequenced strains, except for BJ4 and BJ8. This result was consistent with the antimicrobial susceptibility testing. Tn2009 has been previously described in three Chinese A. baumannii strains: MDR-ZJ06, AB16 and MDR-TJ (in the pABTJ1plasmid) [22,23,30]. Genomic analysis revealed that Tn2009 was flanked by two directed repeats of ISAba1 elements at both ends, and the class D β-lactamase gene bla OXA-23 existed at the 3'-end and was adjacent to the ISAba1 element. IS elements represent another source of variability among A. baumannii isolates, and the insertion of ISAba1 might play a significant role in the expression of bla OXA-23 [24,25].

Comparative analysis of antibiotic resistance genes
A comparative analysis of antibiotic resistance genes was performed on the 8 sequenced strains and 16 reference strains, among which the BJ4, D1279779, AB307-0294, ATCC17978 and SDF strains are antibiotic susceptible (Table 4). There are four types of β-lactamase in all of these strains, including class A β-lactamase, class C β-lactamase, class D β-lactamase and the metallo-βlactamase superfamily. The four types of β-lactamase are encoded by various types of genes and are responsible for much of the multidrug resistance of these strains. Among the 8 sequenced strains in this study, the ampC, metallo-β-lactamase superfamily gene, putative class A β-lactamase gene and bla OXA-66 existed in all of the 7 MDR strains. The tem-1 and putative class C β-lactamase gene were shared by six of the seven MDR strains. However, per-1 is unique to BJ8, and bla OXA-69 is unique to BJ4. The per-1 gene also exists in MDR strain 1656-2. Per-1 is an extended-spectrum β-lactamase, and its induction might be responsible for resistance to all cephalosporins and cause difficulties in treating infections [31].
Resistance to aminoglycosides is primarily mediated by aminoglycoside-modifying enzymes (AMEs), which include three types: aminoglycoside N-acetyltransferase, aminoglycoside O-phosphotransferase, and aminoglycoside O-adenylyltransferase. These three types of AME genes, especially the aacA4, aacC1, strA, strB, aphA1 and aadA1 genes, are commonly found in the 7 sequenced MDR BJ strains and the 12 reference MDR strains. In contrast, the sequenced non-MDR strain BJ4 and the 4 reference non-MDR strains contain fewer AMEs genes: BJ4, D1279779, AB307-0294 and SDF each contain one aphA1gene, and ATCC17978 does not contain an AME gene. Mutations in the gyrA (Ser83Leu) and parC (Ser80Ile) genes are responsible for quinolone resistance in A. baumannii. In this analysis, all of the MDR strains except ZW85-1 contained a mutation in gyrA gene, while the parC gene mutation was only present in nine of the MDR strains. This result might indicate that gyrA (Ser83Leu) plays a more important role than does parC (Ser80Ile) in fluoroquinolone resistance. Among the 19 MDR strains, only 10 include the group A or group B chloramphenicol acetyltransferase genes. There may be other resistance mechanisms, such as efflux pumps, that contribute to chloramphenicol resistance in these MDR strains. The sulfonamide-resistant dihydropteroate synthase genes were present in all of the analyzed MDR strains, while in the 5 non-MDR strains, only ATCC17978 contained a sul2 gene.
The RND (resistance-nodulation-division) family efflux pump, consisting of the adeA, adeB and adeC genes, was present in most of these strains. This efflux pump requires the coexistence of all three genes (adeA, adeB and adeC) to function properly. The antibiotic-susceptible strains D1279779, ATCC17978 and SDF do not contain a functional AdeABC efflux pump. All of the MDR strains except BJAB0715 contain intact adeA, adeB and adeC genes, which might play a role in their antibiotic resistance. Some efflux pump genes belonging to the MFS (major facilitator superfamily) were also identified in several of the MDR strains, including tetA, tetB and tetG, which encode tetracycline efflux proteins, and cmlA and cmlA5, which encode chloramphenicol efflux proteins.

Discussion
In this study, we used whole-genome sequencing methods to characterize genomic variations and antibiotic resistance mechanisms in clinical A. baumannii isolates with various sequence types and drug susceptibility profiles. Although the isolates are closely related, we identified significant genetic differences and a high degree of genomic plasticity in these strains. Pan-genomic analysis of the 8 A. baumannii isolates and the other 16 complete genomes revealed that A. baumannii genomes were highly heterogeneous with respect to gene content and possessed a series of unique genes; these results are similar to those of previous studies [15,32]. The unique genes are enriched in COG class L (Replication, recombination and repair), class V (Defense mechanisms), and class M (Cell wall/membrane/envelope biogenesis), which suggests that these genes are critical for A. baumannii survival or are closely associated with the ability of the bacteria to adapt to challenging niches.
Phylogenetic analysis showed that the drug-susceptible isolate BJ4 was distinct from the other MDR strains, and its closer relationship with the AB0057 and AYE MDR isolates offers another perspective on the origins and acquisition of antibiotic resistance determinants. In addition, the close relationship among strains BJ2, BJ6, and BJ7 indicated they these strains may come from a common ancestor. The csuE deletion in strain BJ2 suggested that this   Table 4 Antimicrobial resistance-associated genes in the 8 BJ strains and 16 reference strains. 0 = absent, 1 = present, 1R = present and resistant (bearing a point mutation), 1S = present but sensitive (no point mutation)  Fluoroquinolones  1R  1S  1S  1R  1R  1S  1R  1R  1R  1R  1R  1S   1S  1S  1S  1R  1S  1S  1S  1S  1S  1S  1S  loss may have occurred after the ancestral strain entered the hospital, followed by the mixing of strains with and without csuE. A comparison of the gene content-based dendrogram with the core SNP tree revealed a similar clustering relationship. The slight difference in tree topology is primarily driven by (i) lateral gene transfer and (ii) IS-mediated phage-and plasmid-associated gene gain and loss. The CRISPR repeat elements, which are involved in a complex mechanism that inhibits invasive phage and plasmid DNA, were not present in any of the eight strains, which may partly explain the widespread distribution of phageand plasmid-related genes and the extensive genomic plasticity among A. baumannii isolates.
Many regions associated with IS-mediated deletions, including a deletion of the entire T6SS region, have been shown to be involved in interbacterial interactions [12]. We found that the T6SS region is conserved in all 7 of the analyzed MDR A. baumannii isolates. As antibiotic therapy appears to reduce interbacterial competition, this result is consistent with the hypothesis that the MDR phenotype is conferred by antibiotic resistance genes, indicating that the T6SS regions are less important [33]. The surface polysaccharide loci are highly variable among the 8 strains, which is consistent with a previous report that these regions are significant sources of variability within A. baumannii strains [14]. We also found that the OC locus (from 558 to 566 kb in Figure 3) was less variable and was highly conserved among the MDR strains, but this region was almost completely missing from the drug-sensitive strain BJ4. In addition, virulence gene analysis showed that a total of 10 putative virulence genes were present in all A. baumannii genomes, suggesting that these genes may play significant roles in the pathogenesis of A. baumannii infection.
AbaR-like RIs inserted in the comM gene were identified in all 7 of the analyzed MDR strains but not in the drugsensitive strain BJ4. This isolate contained a 29.3-kb new GI with five metal-resistance genes but no antibiotic resistance genes ( Figure 5). Therefore, we suggested that GIs inserted into the comM gene are not always associated with antibiotic resistance, and their function might be related to the adaption of the strain to its survival niche. We also found that the RI is highly variable in composition and is not the only contributor to the MDR phenotype. Resistance genes in other mobile elements are found outside the RIs, and they are able to contribute to drug resistance in each strain examined. Among the seven MDR strains, only strain BJ8 did not contain the bla OXA-23 -containing transposon Tn2009, suggesting that Tn2009 is an important carrier of the bla OXA-23 gene in clinical isolates. Furthermore, the detection of Tn2009 in both chromosome and plasmid DNA suggested that this transposon can be transferred easily between clinical strains.
Antimicrobial susceptibility testing indicated that, compared with the 7 MDR strains, the drug-sensitive strain BJ4 shows low-level resistance to piperacillin-tazobactam (TZP), ceftriaxone (CTR) and ceftazidime (CAZ). We hypothesize that this type of low-level resistance is likely caused by the four β-lactamase genes carried by this strain (Table 4). In addition, the aphA1 gene identified in BJ4 encodes resistance to kanamycin but not to gentamicin, amikacin or netilmicin [34]. This result may explain why this strain is susceptible to gentamicin and amikacin. The class D β-lactamase gene bla OXA-69 is unique to strain BJ4, while the 7 MDR strains contain a bla OXA-66 gene at the same genomic location. The intrinsic bla OXA-69 gene encodes oxacillinase, which can hydrolyze imipenem and meropenem at a low level [35]. It is reported that the presence of IS elements such as ISAba1 in the upstream of bla OXA-69 can up-regulate the resistance gene's expression level [25,36]; however, few IS elements were found in the drug-sensitive strain BJ4, and no IS elements are present upstream of the intrinsic resistance genes. In contrast, the MDR strains contain a series of IS element copies, including ISAba1, ISAba125, and IS26; importantly, ISAba1 elements were identified upstream of bla OXA-23 and other RIs in all MDR strains. These results may partly explain the variations in antimicrobial susceptibility between the MDR strains and the sensitive strain.
SNPs are another important source of genetic variation and may contribute to drug resistance and pathogenesis in A. baumannii [37]. Both phylogenic and SNP analysis indicated that the drug-sensitive isolate BJ04 is genetically distinct from other MDR A. baumannii strains. In addition, only a small proportion of SNPs are nonsynonymous among closely related clinical MDR A. baumannii strains with the same STs, indicating that these strains may undergo purifying selection on a genome-wide scale. Furthermore, mutation hotspots between MDR strains were identified in several genes associated with drug resistance, e.g., genes encoding dihydropteroate synthase, a target for sulfonamide antibiotics, and the putative RND superfamily exporter genes, which encode multidrug efflux pumps [38].

Conclusions
In this study, we used whole-genome sequencing to identify genetic variants in A. baumannii isolates. We performed comparative genomic analysis of 8 clinical A. baumannii isolates with 16 available complete A. baumannii genomes in the NCBI database. Our results shed new light on the importance of genomic variations, especially transposon-related and/or phage-related gene variations, in the evolution of A. baumannii. Furthermore, we suggest that the MDR A. baumannii strains harbor diverse antibiotic resistance mechanisms. Future studies focused on a larger sample of A. baumannii isolates from various hospitals and lineages are necessary to better understand the rapid development of antibiotic resistance in A. baumannii.

Bacterial isolates and antimicrobial susceptibility testing
The A. baumannii strains BJ1 to BJ8 were isolated from the 306 th Hospital of People's Liberation Army in Beijing, China. Identification of the isolates and antimicrobial susceptibility testing were performed using the bioMérieux VITEK-2 AST-GN13 system following the manufacturer's instructions. The minimum inhibitory concentration (MIC) of 11 antimicrobial agents was determined according to the recommendations given by the Clinical and Laboratory Standards Institute (CLSI) (Clinical and Laboratory Standards Institute, 2011) [1]. The reference strains Escherichia coli ATCC 25922 and Pseudomonas aeruginosa ATCC 27853 were used as quality controls.

DNA extraction, whole-genome sequencing, and annotation
Genomic DNA was extracted using the TIANamp bacteria DNA kit (Tiangen Biotech (Beijing) Co., Ltd.) according to the manufacturer's instructions. The genomic DNA was fragmented by ultrasonication, and the DNA fragments were subjected to the whole-genome sequencing workflow of the Illumina HiSeq 2000 system. Genome assembly was carried out by SOAPdenovo (http:// soap.genomics.org.cn). The detailed methods for genome assembly and annotation were described in another study [39]. To close gaps within the AbaR-like RIs and integrons, primer pairs were designed at the end of each gap using the genomes as templates. The PCR products were sequenced with an ABI 3730 automated DNA sequencer, and the sequences of these products were used to fill the gaps. Random primers within the AbaR-like RIs and integron regions were subsequently designed to reconfirm the accuracy of these sequences.

Multiple locus sequence typing (MLST)
To identify sequence types, we aligned the assembled sequences against seven housekeeping gene sequences (gltA, gyrB, gdhB, recA, cpn60, gpi, and rpoD) using BLAST and then extracted the aligned sequences by comparing them to allele profiles in the A. baumannii MLST database (http://pubmlst.org/).

SNP detection and analysis
The short reads were first aligned onto the MDR-TJ genome reference using the SOAP2 program [40]. Then, SOAPsnp was used to score SNPs from aligned reads [41]. The SOAPsnp results were filtered as follows: 1) the read coverage of the SNP site was greater than five; 2) the Illumina quality score of either allele was greater than 30; and 3) the count of the all of the mapped best base was more than twice the count of all of the mapped second best base. From all of the SNPs identified in the sequenced genome sequences, the SNP density was calculated throughout the MDR-TJ genome using a slidingwindow size of 5 kb. This window was moved at steps of 1 kb at a time, and the SNP number within each window size was counted. The construction of an SNP clustering map was performed using Circos [42].

Comparative genomics analysis
Genomic data used in comparative analysis were downloaded from the NCBI ftp server, including complete genome sequences of A. baumannii Multiple sequence alignments of the A. baumannii genomes were performed with Mugsy [43]. The phylogenetic tree was constructed using dnaml from the PHYLIP package [44] based on SNPs from the whole-genome alignment, and the genome of Acinetobacter sp. ADP1 was used as the outgroup. An all-against-all BLASTP search between every pair of protein sequences from each strain was performed. Orthologs were identified using PanOCT [17] with the BLASTP output (Identity 80%; Aligned length 30%; E-value < 1e −5 ). The map for core and pan-genome calculations in A. baumannii isolates was performed using PanGP [45]. The heatmap figure was generated using the R package pheatmap [46]. The map of ORF comparisons among A. baumannii genomes was constructed using Circos [42]. As shown in the phylogenetic analysis, the strain SDF was genetically the most distant from our strains; this strain is therefore more suitable for analyzing the nonsynonymous substitution rate and thereby defining the rapidly and slowly evolving protein-coding genes. We aligned the amino acid sequences of SDF with our sequenced strains, estimating the nonsynonymous substitution rates of orthologs based on the NG method using KaKs_Calculator 2.0 [47].
COG annotation was performed using the BLAST software against the COG database. COG enrichment analysis was determined using Fisher's exact test by comparing the prevalence of a target group of genes assigned to a specific COG category to the prevalence of genes in the whole genome assigned to that COG category. To identify possible virulence factors, the Virulence Factors Database (VFDB) [48] was aligned to the ORF protein sequences and filtered with 60% identity and 90% aligned length. To search the antibiotic resistance genes, the protein-coding sequences were aligned against the Antibiotic Resistance Database (ARDB) [49,50] using the similarity thresholds recommended in ARDB. PHAST [51] was used to identify the putative prophages in Acinetobacter genomes. ISs were identified using the IS Finder database (www-is.biotoul.fr) [52]. The detection of CRISPR loci in 8 sequenced draft genome sequences was performed using CRISPRFinder [42].