Utility of Whole-Genome Sequencing in Characterizing Acinetobacter Epidemiology and Analyzing Hospital Outbreaks

Abstract

Acinetobacter baumannii frequently causes nosocomial infections and outbreaks. Whole-genome sequencing (WGS) is a promising technique for strain typing and outbreak investigations. We compared the performance of conventional methods with WGS for strain typing clinical Acinetobacter isolates and analyzing a carbapenem-resistant A. baumannii (CRAB) outbreak. We performed two band-based typing techniques (pulsed-field gel electrophoresis and repetitive extragenic palindromic-PCR), multilocus sequence type (MLST) analysis, and WGS on 148 Acinetobacter calcoaceticus-A. baumannii complex bloodstream isolates collected from a single hospital from 2005 to 2012. Phylogenetic trees inferred from core-genome single nucleotide polymorphisms (SNPs) confirmed three Acinetobacter species within this collection. Four major A. baumannii clonal lineages (as defined by MLST) circulated during the study, three of which are globally distributed and one of which is novel. WGS indicated that a threshold of 2,500 core SNPs accurately distinguished A. baumannii isolates from different clonal lineages. The band-based techniques performed poorly in assigning isolates to clonal lineages and exhibited little agreement with sequence-based techniques. After applying WGS to a CRAB outbreak that occurred during the study, we identified a threshold of 2.5 core SNPs that distinguished nonoutbreak from outbreak strains. WGS was more discriminatory than the band-based techniques and was used to construct a more accurate transmission map that resolved many of the plausible transmission routes suggested by epidemiologic links. Our study demonstrates that WGS is superior to conventional techniques for A. baumannii strain typing and outbreak analysis. These findings support the incorporation of WGS into health care infection prevention efforts.

INTRODUCTION

An important role of clinical microbiology is to identify relationships between bacterial isolates. At a broad level, phenotypic and genotypic tests are used to categorize bacterial isolates into the same or different species. Within a bacterial species, techniques are used to group isolates into clonal lineages, which are groups of closely related bacteria that share a recent common ancestor but have spread regionally or globally. At a more local level, infection control practitioners must determine whether a group of isolates constitutes a hospital outbreak by ascertaining whether the isolates have a degree of similarity consistent with a common source within the hospital. Once isolates belonging to a hospital outbreak have been identified, similarities and differences between these isolates can be exploited to generate a transmission map to aid in finding the source of the outbreak and in disrupting ongoing pathways of transmission. For some groups of bacteria, such as Acinetobacter, discernment at each level has medically important consequences and therefore must be accomplished by hospital-associated clinical microbiology laboratories.

Within the Acinetobacter genus, the Acinetobacter calcoaceticus-Acinetobacter baumannii (ACB) complex encompasses the phenotypically related pathogens A. baumannii, A. pittii (formerly genomospecies 3), and A. nosocomialis (formerly genomospecies 13TU), and one species, A. calcoaceticus, which is not known to cause human disease (1). Of these, A. baumannii is the most common cause of infection. This bacterium is frequently isolated from critically ill hospitalized patients and often causes outbreaks (2–4). Infections with A. baumannii have been associated with high attributable mortality and increased length of hospital stay, with multidrug resistance often being a predictor of poor clinical outcomes (5, 6). However, the frequencies of A. nosocomialis and A. pittii as human pathogens are increasingly recognized (7, 8). Several studies have shown that these different species within the ACB complex exhibit unique epidemiologic niches, drug resistance patterns, and virulence characteristics within the nosocomial environment (9, 10). In addition, recent studies have demonstrated that patients infected with non-baumannii ACB complex bacteria have fewer comorbidities and improved clinical outcomes than patients with A. baumannii infections (7, 11–13). Thus, determining the species within the ACB complex that is responsible for an infection has important medical implications. Unfortunately, commercially available platforms and phenotypic identification methods used in clinical microbiology laboratories are unable to differentiate related species within the ACB complex, and, as such, infections are often reported clinically as “A. baumannii complex” or simply grouped together as “A. baumannii” (14). Molecular techniques, such as rpoB gene sequence analysis, are necessary to accurately identify ACB complex isolates to the species level (15, 16).

In recent years, multidrug-resistant (MDR) strains of A. baumannii have become common, and substantial effort has been devoted to defining the epidemiology of these strains. Initial studies attributed the global spread of MDR A. baumannii to three major clonal lineages identified by amplified fragment length polymorphism (AFLP) analysis that are referred to as international clones (ICs)-I, -II, and -III (17, 18). More recently, multilocus sequence typing (MLST) has become the gold standard for investigating the population structure of A. baumannii (19) and has linked sequence types (STs), such as ST1, ST2, and ST3, with IC-I, -II, and -III, respectively (20). MLST analyses have identified at least six additional clonal lineages with geographically broad distributions (17–19, 21, 22). The study of the global epidemiology and population structure of A. baumannii remains an area of active interest.

A. baumannii is a frequent cause of intrahospital outbreaks, and the ability to distinguish clinical A. baumannii isolates as genetically unique is of particular importance for outbreak analysis. The goal is to distinguish outbreak isolates from nonoutbreak isolates and to use strain typing information to define routes of transmission within an outbreak. Numerous different bacterial strain typing techniques have been described for this purpose. Older band-based fingerprint techniques, such as pulsed-field gel electrophoresis (PFGE), AFLP analysis, and repetitive extragenic palindromic-PCR (Rep-PCR), rely on indirect measures of bacterial genetic composition (22–26). Of these, PFGE arguably remains the gold standard for A. baumannii outbreak investigations (19). More recently, partial direct genetic sequencing techniques have been developed, such as MLST and bla_oxa-51-like sequencing (20, 27, 28). Although band-based techniques are often highly discriminatory, direct sequence-based methods can provide greater genetic resolution and are more reproducible and portable (29–31).

For years, high-throughput typing of multiple A. baumannii isolates for short-term outbreak investigations or long-term regional or global surveillance was limited to these band-based or partial sequence-based techniques (32). The decreasing cost, complexity, and turnaround time of whole-genome sequencing (WGS) may soon allow the application of this technology to routine bacterial strain typing in clinical microbiology laboratories (33, 34). A number of recent studies have used WGS to characterize the genetic relatedness of clinical bacterial isolates for various purposes. Some studies have used WGS for bacterial species identification and investigations of the population structure and global spread of bacteria (31, 35, 36). Other studies have analyzed hospital outbreaks in greater detail and have used WGS to distinguish between possible transmission scenarios, suggest alternative transmission links, and identify previously unrecognized colonized patients (37–39). Although WGS may be the most powerful and adaptable tool for bacterial species identification, epidemiologic surveillance, and outbreak analysis, the application of WGS to these endeavors has yet to be fully explored. Careful characterization of the advantages of WGS is necessary to determine whether they outweigh the costs of commercially available next-generation sequencing instruments or services. Also, the criteria for discrimination between outbreak and nonoutbreak isolates or between clonal lineages have yet to be clearly defined for WGS. In contrast, the criteria for these purposes have been well described for interpreting PFGE, Rep-PCR, and MLST results (20, 40–42).

In this study, we compared the performance of WGS to that of two band-based typing techniques (PFGE and Rep-PCR) and to a sequence-based typing technique (MLST) to characterize the genetic relatedness and epidemiology of a large collection of ACB complex clinical bloodstream isolates collected over 7 years at our institution. We compared WGS to conventional typing at the level of Acinetobacter species identification, assignment to clonal lineages, inclusion within an intrahospital outbreak, and outbreak transmission mapping. Our results indicate that WGS provides useful information for each of these purposes and may eventually obviate multiple different techniques within clinical microbiology laboratories.

MATERIALS AND METHODS

Ethics, consent, and permissions.

This study was approved by the institutional review board of Northwestern University, Chicago, IL, with a waiver of informed consent. Informed consent was not obtained because this was a retrospective study, and data collection occurred after patients left the hospital and/or died from their illness. No additional specimens were collected beyond those in routine clinical care, and no diagnostic or treatment decisions were affected by our study.

Bacterial isolates and identification.

One hundred fifty-four bloodstream isolates (designated ABBL isolates) collected between January 2005 and November 2012 from hospitalized patients at Northwestern Memorial Hospital (NMH), a 900-bed tertiary-care academic medical center in Chicago, IL, were analyzed in this study. These isolates were previously reported (13). In addition, 13 blood and respiratory isolates collected during the investigation of a carbapenem-resistant A. baumannii (CRAB) intensive care unit (ICU) outbreak from June 2013 to December 2013 (designated ABOB isolates) were included. Phenotypic identification of isolates to the ACB complex level was performed using the Vitek 2 system (bioMérieux, Marcy l'Etoile, France). Antimicrobial susceptibility testing was performed using a combination of the Vitek 2, disk diffusion test, and Etest (bioMérieux) and was interpreted in accordance with the Clinical and Laboratory Standards Institute guidelines (43). For the purposes of this study, all isolates were identified to the species level using the partial rpoB gene sequence. Zone 1 rpoB sequences were identified from the assembled whole-genome sequences using an in-house in silico PCR script (44). Briefly, assembly contigs were screened to identify sequences flanked by the virtual primers Ac696F (TAYCGYAAAGAYTTGAAAGAAG) and Ac1093R (CMACACCYTTGTTMCCRTGA), as described by La Scola et al. (45). In silico amplified sequences were identified by BLAST alignment against the NCBI nucleotide database (46).

Pulsed-field gel electrophoresis.

During the period of collection of these strains, the NMH clinical microbiology laboratory performed PFGE surveillance on all CRAB isolates in our hospital, as previously described (25). Banding patterns were interpreted and strain types assigned according to the criteria of Tenover et al. (40). Isolates with indistinguishable, closely related, and possibly related PFGE patterns were considered the same type, as they are often analyzed together during outbreak investigations.

DNA extraction.

Bacterial isolates were streaked from −80°C frozen stocks, inoculated in Luria-Bertani (LB) broth, and grown with shaking overnight at 37°C. Genomic DNA was extracted from the cultures using either the Qiagen EZ1 Advanced robotic workstation (Valencia, CA) or the Promega Maxwell 16 instrument (Madison, WI), according to each manufacturer's instructions. Genomic DNA extracts were either kept on ice for immediate use or were stored at −20°C for future use.

Repetitive extragenic palindromic-PCR.

For Rep-PCR analyses, approximately 300 ng of genomic DNA was used in the PCRs, along with the REP1+REP2 oligonucleotide primer set (47). The reaction mixtures contained 100 pmol each primer and 25 μl of AccuStart II PCR Supermix (Quanta Biosciences, Gaithersburg, MD), which includes 2× reaction buffer, 3 mM MgCl₂, 0.3 mM each deoxynucleoside triphosphate (dNTP), and AccuStart II Taq DNA polymerase. Sterile distilled water was added to a final volume of 50 μl. The PCR conditions were as follows: initial denaturation of 94° for 3 min, 30 cycles of 94°C for 1 min, 45°C for 1 min, and 65°C for 8 min, and a final elongation of 65°C for 16 min. Samples (20 μl) of each PCR end product were analyzed on a 1% agarose gel with ethidium bromide added. Rep-PCR patterns were visualized with UV transillumination using the AlphaImager system (ProteinSimple Biosciences, Santa Clara, CA). Gel images were uploaded into the BioNumerics version 7.0 program (Applied Maths, Austin, TX) to perform clustering analysis. Briefly, after normalization of inter- and intragel variation using molecular weight standards, individual bands were manually chosen, and a clustering dendrogram was created using the unweighted pair group method using average linkages (UPGMA). The Dice statistic was used, and band tolerance was set to 1%. Isolates clustering together at >90% similarity based on the UPGMA dendrogram were considered to be the same type (23, 48, 49).

Whole-genome sequencing and assembly.

Genomic DNA libraries were prepared and indexed using the Nextera XT kit (Illumina, San Diego, CA). DNA library concentrations were quantified using the Qubit 2.0 fluorometer (Life Technologies, Grand Island, NY). Equal amounts (200 ng) of each library were pooled and run on either the Illumina HiSeq 2000 system with 100-bp paired-end reads or on the Illumina MiSeq system with 250-bp paired-end reads. Sequencing was performed by staff at the University of Maryland Institute for Genome Sciences, Baltimore, MD, who were blinded to all clinical data. Raw sequence reads were then assembled de novo using Ray version 1.7.0 (50). Six bloodstream isolates with low-quality assemblies, defined as a sequence size of assembled contigs of ≥500 bp totaling >4.5 Mb, were excluded from further analysis. The assembly statistics are displayed in Table 1.

TABLE 1.

Date of isolation, sequencing, and assembly statistics for new Acinetobacter bloodstream and outbreak isolates

Isolate	Culture date (mo/day/yr)	Avg read length (bp)	No. of raw reads	Predicted coverage (fold)	No. of raw reads after downsampling	No. of contigs >500 bp	Total size of contigs >500 bp (bp)	Contig N50 (bp)	SRA accession no.	GenBank WGS accession no.a
ABBL001	1/30/2005	101	5,751,032	145	3,168,318	79	3,871,836	122,252	SRR2558732	LLCI00000000
ABBL003	2/25/2005	101	5,878,804	148	3,168,318	72	3,959,187	114,915	SRR2558733	LLCJ00000000
ABBL004	5/17/2005	101	4,935,656	125	3,168,318	76	3,857,255	129,801	SRR2558836	LLCK00000000
ABBL005	6/21/2005	101	5,456,552	138	3,168,318	66	4,105,919	146,012	SRR2558848	LLCL00000000
ABBL006	7/6/2005	101	5,276,928	133	3,168,318	79	4,070,748	92,563	SRR2558860	LLCM00000000
ABBL007	8/1/2005	234	4,672,638	273	1,367,522	770	3,762,155	8,012	SRR2558871	LLCN00000000
ABBL008	8/11/2005	101	6,100,156	154	3,168,318	89	4,066,927	139,396	SRR2558883	LLCO00000000
ABBL010	8/15/2005	101	5,917,234	149	3,168,318	62	4,050,651	126,884	SRR2558907	LLCQ00000000
ABBL011	9/4/2005	101	5,431,532	137	3,168,318	71	4,006,992	118,605	SRR2558919	LLCR00000000
ABBL012	9/23/2005	101	6,780,124	171	3,168,318	87	3,927,097	110,206	SRR2558734	LLCS00000000
ABBL013	9/19/2005	101	4,344,706	110	3,168,318	138	4,010,638	85,500	SRR2558776	LLCT00000000
ABBL014	9/25/2005	101	6,420,868	162	3,168,318	193	3,857,237	52,131	SRR2558787	LLCU00000000
ABBL015	10/23/2005	101	6,262,008	158	3,168,318	68	4,191,244	186,354	SRR2558808	LLCV00000000
ABBL016	11/7/2005	101	6,536,664	165	3,168,318	35	3,880,175	212,743	SRR2558820	LLCW00000000
ABBL017	11/16/2005	101	5,637,164	142	3,168,318	92	4,028,200	106,423	SRR2558831	LLCX00000000
ABBL018	12/9/2005	101	4,978,776	126	3,168,318	125	4,318,316	105,657	SRR2558832	LLCY00000000
ABBL019	12/13/2005	101	5,935,086	150	3,168,318	24	3,888,930	343,426	SRR2558833	LLCZ00000000
ABBL020	1/17/2006	101	7,809,576	197	3,168,318	83	3,806,309	108,333	SRR2558834	LLDA00000000
ABBL021	1/31/2006	101	7,992,930	202	3,168,318	64	3,952,727	132,629	SRR2558835	LLDB00000000
ABBL022	2/22/2006	101	6,912,504	175	3,168,318	83	3,994,985	105,023	SRR2558837	LLDC00000000
ABBL023	3/21/2006	232	7,041,594	408	1,382.290	579	3,982,278	12,335	SRR2558838	LLDD00000000
ABBL024	4/4/2006	101	4,864,248	123	3,168,318	91	3,892,382	107,246	SRR2558839	LLDE00000000
ABBL025	4/24/2006	101	5,678,228	143	3,168,318	71	4,070,327	171,086	SRR2558840	LLDF00000000
ABBL026	5/1/2006	217	10,727,026	581	1,478,060	602	3,906,420	10,017	SRR2558841	LLDG00000000
ABBL027	5/13/2006	251	2,226,438	140	1,274,900	84	4,051,543	115,039	SRR2558842	LLDH00000000
ABBL028	5/25/2006	239	585,626	35	NAb	909	3,667,117	6,740	SRR2558843	LLDI00000000
ABBL029	6/5/2006	101	5,887,830	149	3,168,318	90	3,922,574	87,612	SRR2558844	LLDJ00000000
ABBL030	7/4/2006	251	2,237,610	140	1,274,900	93	4,022,189	84,636	SRR2558846	LLDK00000000
ABBL031	8/9/2006	151	2,697,432	102	2,119,206	82	4,143,212	122,907	SRR2558847	LLDL00000000
ABBL032	8/31/2006	101	7,469,508	189	3,168,318	111	3,917,819	66,618	SRR2558849	LLDM00000000
ABBL033	11/6/2006	101	7,242,078	183	3,168,318	119	3,800,726	75,277	SRR2558850	LLDN00000000
ABBL034	1/9/2007	101	6,643,438	168	3,168,318	89	4,041,141	111,495	SRR2558851	LLDO00000000
ABBL035	1/23/2007	232	3,782,874	219	1,379,310	966	3,811,475	6,705	SRR2558852	LLDP00000000
ABBL036	6/8/2007	101	14,194,932	358	3,168,318	500	3,816,808	15,165	SRR2558853	LLDQ00000000
ABBL037	6/20/2007	101	8,609,660	217	3,168,318	75	4,179,466	109,239	SRR2558854	LLDR00000000
ABBL038	7/7/2007	101	10,957,150	277	3,168,318	86	4,045,591	110,028	SRR2558855	LLDS00000000
ABBL039	8/2/2007	101	8,767,486	221	3,168,318	118	4,005,659	59,483	SRR2558856	LLDT00000000
ABBL040	8/12/2007	101	11,439,386	289	3,168,318	439	3,889,680	16,992	SRR2558857	LLDU00000000
ABBL041	8/12/2007	101	5,683,398	144	3,168,318	105	4,014,785	85,413	SRR2558858	LLDV00000000
ABBL042	8/13/2007	251	2,148,934	135	1,274,900	104	4,023,065	95,503	SRR2558861	LLDW00000000
ABBL043	8/28/2007	251	2,304,972	145	1,274,900	113	4,193,893	79,829	SRR2558862	LLDX00000000
ABBL044	9/20/2007	251	2,114,926	133	1,274,900	107	4,108,999	64,973	SRR2558863	LLDY00000000
ABBL045	9/20/2007	101	8,425,048	213	3,168,318	103	4,052,607	101,492	SRR2558864	LLDZ00000000
ABBL046	10/14/2007	101	11,383,666	287	3,168,318	117	3,917,523	70,298	SRR2558865	LLEA00000000
ABBL047	10/28/2007	101	6,941,066	175	3,168,318	57	4,100,149	154,707	SRR2558866	LLEB00000000
ABBL048	11/27/2007	251	2,144,992	135	1,274,900	75	4,161,657	120,446	SRR2558867	LLEC00000000
ABBL049	12/12/2007	251	2,480,548	156	1,274,900	88	4,142,974	116,905	SRR2558868	LLED00000000
ABBL050	1/22/2008	101	13,089,350	331	3,168,318	292	4,173,128	27,050	SRR2558869	LLEE00000000
ABBL051	1/31/2008	101	13,714,298	346	3,168,318	858	3,801,250	7,151	SRR2558870	LLEF00000000
ABBL052	2/24/2008	101	8,137,460	205	3,168,318	106	4,109,803	75,223	SRR2558873	LLEG00000000
ABBL053	3/17/2008	101	7,715,994	195	3,168,318	88	4,085,835	90,003	SRR2558874	LLEH00000000
ABBL054	3/22/2008	101	11,043,836	279	3,168,318	108	3,435,547	70,480	SRR2558875	LLEZ00000000
ABBL055	4/5/2008	101	9,541,470	241	3,168,318	110	4,102,482	60,720	SRR2558876	LLFA00000000
ABBL056	7/11/2008	101	7,449,168	188	3,168,318	84	4,012,566	80,470	SRR2558877	LLFB00000000
ABBL057	7/22/2008	101	7,511,856	190	3,168,318	78	4,016,266	236,259	SRR2558878	LLFC00000000
ABBL058	7/25/2008	101	9,584,202	242	3,168,318	117	4,010,913	80,750	SRR2558879	LLFD00000000
ABBL059	8/7/2008	101	5,781,700	146	3,168,318	226	3,989,506	37,323	SRR2558880	LLFE00000000
ABBL060	9/15/2008	101	20,003,004	505	3,168,318	198	4,033,351	37,283	SRR2558881	LLFF00000000
ABBL061	9/29/2008	101	5,134,402	130	3,168,318	93	3,960,869	101,526	SRR2558882	LLFG00000000
ABBL062	10/1/2008	101	5,970,402	151	3,168,318	72	3,873,658	103,774	SRR2558884	LLFH00000000
ABBL063	10/23/2008	101	10,606,350	268	3,168,318	263	3,680,335	27,301	SRR2558885	LLFI00000000
ABBL064	12/8/2008	101	12,685,308	320	3,168,318	479	3,871,975	15404	SRR2558886	LLFJ00000000
ABBL065	12/9/2008	101	8,501,444	215	3,168,318	75	4,039,674	154,921	SRR2558887	LLFK00000000
ABBL066	12/11/2008	101	8,860,706	224	3,168,318	72	4,319,668	106,135	SRR2558889	LLFL00000000
ABBL067	12/21/2008	101	4,584,514	116	3,168,318	78	4,032,722	114,946	SRR2558890	LLFM00000000
ABBL067a	5/12/2005	101	12,508,224	316	3,168,318	320	3,992,826	22,794	SRR2558891	LLFN00000000
ABBL067b	9/14/2005	101	13,860,118	350	3,168,318	357	4,059,209	22,441	SRR2558892	LLFO00000000
ABBL067c	12/25/2005	101	12,256,156	309	3,168,318	396	4,013,666	19,274	SRR2558893	LLFP00000000
ABBL067e	12/25/2005	101	13,185,332	333	3,168,318	370	3,999,590	21,308	SRR2558894	LLFQ00000000
ABBL067f	1/11/2006	101	12,775,890	323	3,168,318	444	3,969,780	16,568	SRR2558896	LLFR00000000
ABBL067g	1/22/2006	101	12,501,658	316	3,168,318	431	3,971,931	16,858	SRR2558897	LLFS00000000
ABBL067h	4/3/2006	101	12,791,622	323	3,168,318	404	4,002,856	19,517	SRR2558898	LLFT00000000
ABBL067i	4/23/2006	101	6,390,358	161	3,168,318	418	3,982,233	17,399	SRR2558900	LLFU00000000
ABBL067j	5/20/2007	101	10,486,564	265	3,168,318	136	4,088,444	52,671	SRR2558901	LLFV00000000
ABBL067k	12/12/2007	101	6,091,926	154	3,168,318	98	4,002,082	82,155	SRR2558902	LLFW00000000
ABBL067l	1/16/2008	101	8,391,138	212	3,168,318	102	4,001,051	86,209	SRR2558903	LLFX00000000
ABBL068	1/14/2009	101	6,696,118	169	3,168,318	215	4,134,094	35,378	SRR2558904	LLFY00000000
ABBL069	2/2/2009	101	32,813,908	829	3,168,318	319	4,091,029	25,495	SRR2558905	LLFZ00000000
ABBL070	2/5/2009	101	26,881,038	679	3,168,318	265	4,020,734	31,563	SRR2558906	LLGA00000000
ABBL071	4/5/2009	101	6,291,266	159	3,168,318	116	4,093,661	63,583	SRR2558908	LLGB00000000
ABBL072	4/9/2009	101	23,436,746	592	3,168,318	198	4,324,713	45,835	SRR2558909	LLGC00000000
ABBL073	4/22/2009	101	16,950,024	428	3,168,318	131	4,092,205	78,381	SRR2558910	LLGD00000000
ABBL074	5/4/2009	101	7,723,174	195	3,168,318	211	3,975,531	36,237	SRR2558911	LLGE00000000
ABBL075	5/12/2009	101	26,829,806	677	3,168,318	288	4,159,435	30,922	SRR2558912	LLGF00000000
ABBL076	5/25/2009	101	13,587,060	343	3,168,318	386	3,868,391	19,568	SRR2558913	LLGG00000000
ABBL077	5/26/2009	101	35,514,828	897	3,168,318	305	3,957,932	27,593	SRR2558914	LLGH00000000
ABBL078	6/15/2009	101	16,773,812	424	3,168,318	297	3,973,781	24,307	SRR2558916	LLGI00000000
ABBL079	7/3/2009	101	28,887,508	729	3,168,318	342	4,265,921	22,807	SRR2558917	LLGJ00000000
ABBL080	8/7/2009	101	26,434,436	667	3,168,318	428	3,976,763	19,694	SRR2558918	LLGK00000000
ABBL082	9/1/2009	101	25,406,536	642	3,168,318	292	3,852,334	25,470	SRR2558920	LLGL00000000
ABBL083	8/30/2009	101	30,989,798	782	3,168,318	301	3,940,840	25,555	SRR2558921	LLGM00000000
ABBL085	9/11/2009	101	27,206,694	687	3,168,318	241	3,813,467	29,772	SRR2558922	LLGN00000000
ABBL086	9/10/2009	101	16,590,444	419	3,168,318	348	4,058,485	22,997	SRR2558923	LLGO00000000
ABBL088	9/13/2009	101	17,105,376	432	3,168,318	330	4,021,290	25,565	SRR2558924	LLGP00000000
ABBL089	10/6/2009	101	34,503,130	871	3,168,318	327	3,997,927	26,072	SRR2558925	LLGQ00000000
ABBL090	10/9/2009	101	38,910,236	982	3,168,318	310	3,817,936	24,138	SRR2558926	LLGR00000000
ABBL091	11/7/2009	101	28,234,296	713	3,168,318	436	4,115,100	19,684	SRR2558927	LLGS00000000
ABBL092	11/13/2009	101	30,855,824	779	3,168,318	399	3,883,553	19,303	SRR2558928	LLGT00000000
ABBL093	11/12/2009	101	19,490,714	492	3,168,318	309	3,874,412	25,329	SRR2558929	LLGU00000000
ABBL094	12/14/2009	101	18,463,954	466	3,168,318	369	3,903,234	19,927	SRR2558737	LLGV00000000
ABBL095	12/19/2009	101	12,628,086	319	3,168,318	218	3,885,321	30,203	SRR2558739	LLGW00000000
ABBL096	1/20/2010	101	37,281,074	941	3,168,318	306	3,898,693	23,502	SRR2558766	LLGX00000000
ABBL097	2/19/2010	101	29,085,722	734	3,168,318	261	3,905,810	28,228	SRR2558767	LLGY00000000
ABBL098	3/6/2010	101	25,439,548	642	3,168,318	269	3,853,967	27,320	SRR2558768	LLGZ00000000
ABBL099	3/6/2010	101	17,538,744	443	3,168,318	231	3,840,161	27,302	SRR2558769	JPDG00000000
ABBL100	4/9/2010	101	27,451,932	693	3,168,318	241	3,939,568	30,882	SRR2558772	LLHA00000000
ABBL101	5/3/2010	101	19,403,912	490	3,168,318	382	3,931,279	20,041	SRR2558773	LLHB00000000
ABBL102	6/4/2010	101	14,401,494	364	3,168,318	231	3,776,367	31,608	SRR2558774	LLHC00000000
ABBL103	7/29/2010	101	12,252,832	309	3,168,318	207	3,987,056	37,009	SRR2558775	LLHD00000000
ABBL105	9/6/2010	101	37,753,256	953	3,168,318	376	4,118,335	21,631	SRR2558777	LLHE00000000
ABBL106	9/6/2010	101	40,287,706	1,017	3,168,318	370	4,124,964	21,780	SRR2558778	LLHF00000000
ABBL107	9/6/2010	101	22,537,832	569	3,168,318	292	3,979,777	27,143	SRR2558779	LLHG00000000
ABBL109	9/6/2010	101	22,392,882	565	3,168,318	279	3,954,675	29,282	SRR2558780	LLHH00000000
ABBL110	9/27/2010	101	15,009,192	379	3,168,318	250	3,694,263	30,739	SRR2558781	LLHI00000000
ABBL111	11/16/2010	101	28,853,656	729	3,168,318	419	4,108,846	24,331	SRR2558782	LLHJ00000000
ABBL113	12/15/2010	101	14,977,074	378	3,168,318	308	4,019,479	26,483	SRR2558783	LLHK00000000
ABBL114	1/20/2011	101	39,589,430	1,000	3,168,318	337	4,137,977	25,732	SRR2558784	LLHL00000000
ABBL115	2/21/2011	101	32,786,960	828	3,168,318	487	3,942,056	15,490	SRR2558785	LLHM00000000
ABBL116	2/26/2011	101	12,053,200	304	3,168,318	377	4,044,273	23,431	SRR2558786	LLHN00000000
ABBL117	3/24/2011	101	16,212,654	409	3,168,318	550	4,115,897	14,029	SRR2558788	LLHO00000000
ABBL118	4/5/2011	101	28,120,660	710	3,168,318	419	3,938,213	18,512	SRR2558790	LLHP00000000
ABBL120	6/16/2011	101	13,437,746	339	3,168,318	536	3,979,007	14,187	SRR2558791	LLHQ00000000
ABBL121	7/25/2011	101	6,002,822	152	3,168,318	466	4,039,842	17,305	SRR2558792	LLHR00000000
ABBL122	7/30/2011	101	40,444,704	1,021	3,168,318	306	3,922,223	23,597	SRR2558797	LLHS00000000
ABBL123	8/23/2011	101	14,220,382	359	3,168,318	596	3,795,535	12,218	SRR2558798	LLHT00000000
ABBL124	9/3/2011	101	4,054,356	102	3,168,318	197	4,013,619	34,239	SRR2558799	LLHU00000000
ABBL125	11/8/2011	101	20,151,532	509	3,168,318	456	3,780,263	15,263	SRR2558800	LLHV00000000
ABBL126	11/8/2011	101	20,027,030	506	3,168,318	349	3,821,336	20,964	SRR2558805	LLHW00000000
ABBL127	11/13/2011	101	9,442,450	238	3,168,318	356	4,152,329	22,917	SRR2558807	LLHX00000000
ABBL128	11/23/2011	101	20,775,318	525	3,168,318	458	3,950,643	19,092	SRR2558809	LLHY00000000
ABBL129	1/5/2012	101	5,353,678	135	3,168,318	186	4,027,647	38,732	SRR2558810	LLHZ00000000
ABBL130	2/12/2012	101	22,383,756	565	3,168,318	343	3,932,290	21,064	SRR2558811	LLIA00000000
ABBL131	2/14/2012	101	16,386,078	414	3,168,318	394	3,989,817	19,923	SRR2558812	LLIB00000000
ABBL132	3/2/2012	101	23,009,154	581	3,168,318	383	3,983,304	19,969	SRR2558813	LLIC00000000
ABBL133	3/19/2012	101	14,862,736	375	3,168,318	361	3,930,248	21,247	SRR2558814	LLID00000000
ABBL134	3/22/2012	101	19,728,578	498	3,168,318	369	4,054,261	23,792	SRR2558815	LLIE00000000
ABBL135	5/1/2012	101	24,722,876	624	3,168,318	495	3,915,368	17,220	SRR2558816	LLIF00000000
ABBL137	5/14/2012	101	17,330,758	438	3,168,318	428	3,919,186	18,907	SRR2558818	LLIG00000000
ABBL138	5/18/2012	101	16,430,912	415	3,168,318	435	3,882,593	17,739	SRR2558819	LLIH00000000
ABBL140	6/9/2012	101	49,913,702	1,260	3,168,318	543	3,869,396	13,065	SRR2558822	LLII00000000
ABBL141	6/12/2012	101	13,875,892	350	3,168,318	367	4,236,436	23,187	SRR2558824	LLIJ00000000
ABBL142	7/31/2012	101	11,970,574	302	3,168,318	259	3,978,019	29,463	SRR2558825	LLIK00000000
ABBL143	8/31/2012	101	79,935,976	2,018	3,168,318	478	3,908,504	16,117	SRR2558826	LLIL00000000
ABBL144	9/16/2012	101	14,559,986	368	3,168,318	381	3,943,688	20,834	SRR2558827	LLIM00000000
ABBL147	10/10/2012	101	28,332,702	715	3,168,318	395	3,716,396	19,347	SRR2558828	LLIN00000000
ABBL148	10/22/2012	101	20,550,420	519	3,168,318	423	3,549,950	15,415	SRR2558829	LLIO00000000
ABBL149	11/10/2012	101	30,202,168	763	3,168,318	373	3,794,913	19,926	SRR2558830	LLIP00000000
ABOB01	6/20/2013	101	7,652,372	193	3,168,318	139	4,062,889	58,695	SRR2559322	LLIQ00000000
ABOB02	6/29/2013	101	14,825,292	374	3,168,318	389	3,963,019	18,994	SRR2559323	LLIR00000000
ABOB03	7/2/2013	101	16,796,756	424	3,168,318	370	3,977,012	21,240	SRR2559352	LLIS00000000
ABOB04	8/31/2013	101	19,297,440	487	3,168,318	304	3,971,774	23,905	SRR2559353	LLIT00000000
ABOB04_a	9/9/2013	101	16,698,092	422	3,168,318	364	3,978,511	21,753	SRR2559354	LLIU00000000
ABOB04_b	11/16/2013	101	15,920,118	402	3,168,318	432	4,049,068	19,216	SRR2559355	LLIV00000000
ABOB06	10/25/2013	101	13,923,144	352	3,168,318	278	3,993,533	27,005	SRR2559356	LLIW00000000
ABOB06_a	11/21/2013	101	20,292,790	512	3,168,318	602	3,891,945	10,555	SRR2559357	LLIX00000000
ABOB07	10/30/2013	101	25,414,296	642	3,168,318	1,081	3,836,895	5,674	SRR2559358	LLIY00000000
ABOB08	10/31/2013	101	15,633,066	395	3,168,318	418	3,993,223	19,275	SRR2559359	LLIZ00000000
ABOB09	11/8/2013	101	19,274,076	487	3,168,318	375	3,945,737	20,698	SRR2559324	LLJA00000000
ABOB10	11/15/2013	101	17,320,904	437	3,168,318	460	3,948,321	17,481	SRR2559325	LLJB00000000
ABOB11	11/16/2013	101	20,312,352	513	3,168,318	401	3,953,603	20,420	SRR2559326	LLJC00000000
ABOB12	11/18/2013	101	17,876,054	451	3,168,318	284	4,086,064	27,116	SRR2559327	LLJD00000000
ABOB15	Unknown	101	15,080,752	381	3,168,318	89	4,021,456	84,258	SRR2559328	LLJE00000000
ABOB16	Unknown	101	16,396,238	414	3,168,318	324	3,852,656	22,273	SRR2559329	LLJF00000000
ABOBEN	Unknown	101	18,111,886	457	3,168,318	381	3,940,575	20,023	SRR2559351	LLJG00000000

^aWGS, whole-genome sequencing.

^bNA, not available.

Multilocus sequence typing.

The sequences for the Institut Pasteur MLST genes (20) were extracted from the assembled contigs for all isolates, concatenated, and aligned with MUSCLE (51). A phylogeny was inferred with MEGA version 5.2.2 (52), using the maximum-likelihood method, and exported to FigTree version 1.4.2 for visualization (53). In addition, sequence types were determined for all A. baumannii isolates using the database available on the Institut Pasteur MLST website (http://www.pasteur.fr/mlst).

Whole-genome phylogeny and SNP detection.

Unless otherwise stated, single nucleotide polymorphism (SNP) analyses were based on the core genome, which was defined as sequences found in at least 95% of the ABBL isolates (54, 55). The value of >95% was used to avoid substantial changes in the core-genome definition due to one or two isolates that might have undergone gene deletion or for which assembly errors may have resulted in the omission of genes. The kSNP version 2.1.2 program, which uses k-mers (all possible stretches of k-consecutive nucleotides) from input genomes to identify SNPs, was used for this purpose (56). This program has the advantage of not requiring multiple sequence alignments or comparisons with a reference genome. Thirty-one-base-pair k-mers were used, as suggested by the Kchooser script included with kSNP. SNPs were identified by comparing orthologous k-mers from distinct isolates that were identical except for the central nucleotide (nucleotide 16). The kSNP software has the option of searching for k-mers in either assembled genomic sequences or raw sequencing reads. We chose to use assembled sequences, as the de novo assembly process filters out most nonspecific and low-quality reads. However, de novo assembly of short Illumina reads can also produce errors in some genomes, such as the omission of regions from the final assembled contigs or the collapse of repeat genomic regions with one or more nucleotide differences into a single contig. To correct for these errors, we applied a supplemental bioinformatics approach to the kSNP output. Briefly, the Jellyfish k-mer counting software version 1.1.5 (57) was used to directly extract all possible k-mers from each set of unassembled sequencing reads. A Perl script was then used to query these k-mers against the list of all k-mer outputs by kSNP analysis of assembled contigs. For each isolate, if a particular k-mer was not found in the assembled contigs but was found in five or more sequencing reads, the base at the SNP position of that k-mer was added to the kSNP output matrix for that genome. To avoid miscalling sequencing errors as SNPs, a base was removed from the kSNP output matrix for a genome and replaced with a gap (“-”) if it met the following criteria: the k-mer was found to be mixed in the sequencing reads (i.e., two or more k-mer sequences were identified that differed at the central SNP position only), and the occurrence of the most abundant minority k-mer was at least 10% of the occurrence of the majority k-mer. Likewise, if a core-genome k-mer was missing from a strain, the corresponding gap (“-”) was treated as missing data and did not contribute to the placement of that strain in the phylogenetic tree to avoid potentially false inferences resulting from sequencing errors. When indicated, kSNP results were confirmed by the alignment of raw sequencing reads against a reference sequence using the bwa alignment program version 0.7.6a-r433 (58). SNPs were sometimes examined using a combination of automated variant calling with the programs SAMtools and bcftools (both version 0.1.19-44428cd) (59) and manual examination of regions of interest using the Tablet alignment visualization program version 1.13.05.17 (60). To minimize false-positive SNP calls from raw read alignments to sequence contigs due to ambiguous mapping of reads to repeat regions, SNP calls produced by SAMtools and bcftools from the alignment of reads from the query genome to reference contigs were filtered using SNP calls produced from the alignment of reads from the reference genome back to the reference contigs. Any SNP calls found in both alignments were removed. To visualize the relative SNP density, any remaining high-quality SNPs (quality score, 222) between the query genome and reference genome were plotted using CGView (61). The scripts developed for these and other analyses used in this study can be found at the GitHub website (https://github.com/egonozer/snp_tools).

To determine whether the acquisition of DNA sequences through recombination influenced the phylogenetic relationships we obtained, a subset of 23 arbitrarily selected isolates was chosen for further analysis. Whole-genome alignments were generated by aligning sequencing reads of these 23 isolates to the reference genome of A. baumannii ATCC 17978 (GenBank accession no. CP000521) using the bwa program (58). SNPs were called using SAMtools and bcftools. Indels, positions with variant quality <30, positions covered by <8 reads, or positions with <0.7 of the reads representing either the reference or an alternate base were filtered. A tree was then produced from the alignments using ClonalFrameML (62) to remove potential recombinant regions. A comparison of this tree to one using the kSNP approach described above showed no significant differences in topology (data not shown), demonstrating that SNPs present in recombination regions had a minimal impact on the tree structures we obtained.

To determine pairwise core-genome SNP counts, we developed software based on kSNP to (i) identify a core genome of k-mers from genomic assemblies and sequencing reads, and (ii) count SNPs between pairs of these core-genome k-mer sets. First, the set of core-genome k-mers was identified using the kmer_core.pl software. This software compares k-mers among a set of input genomic DNA sequences and outputs all k-mers with identical 15-nucleotide flanking arms (e.g., identical except perhaps at the central nucleotide) that were present in a specified subset of the genomes. For the purposes of this study, a subset cutoff of k-mers present in ≥95% of the input genomes was chosen (54, 55). The assembled genomes of the A. baumannii ABBL strains were used as input for kmer_core.pl to identify a core k-mer set. A second software package (kmer_compare.pl) was then used to search for these core k-mers in query sequences and perform pairwise comparisons to determine the SNP totals between the genomes. To maximize the sensitivity of identification of core k-mers in the strains sequenced for this study, raw sequencing reads (rather than assembled genomes) were used as input to kmer_compare.pl, and a k-mer was considered present if at least 5 sequencing reads were found to contain the k-mer. For reference strains, the genomic sequence was used to search for core k-mers, and k-mers found at least once were considered present. Conflicting k-mers, defined as two or more k-mers present in a single genome that were identical except at the central base, were filtered out of the k-mer set of each genome. To avoid erroneous filtering of k-mers appearing to be in conflict due to sequencing errors in a few reads, conflicting core k-mers were only removed if the number of reads containing k-mers with the most-abundant minority central base was at least 10% of the number of reads containing k-mers with the majority central base. To generate a jitter plot showing the number of core SNPs between pairs of isolates, Microsoft Excel was used to randomly assign an x axis position to each data point within the boundaries of an appropriate column.

For the ABBL isolates, A. baumannii isolates, and ABOB isolates, FastTree2 (63) was used to estimate maximum-likelihood phylogenetic trees (64) from core SNPs, and trees were then visualized in FigTree version 1.4.2 (53). Whole-genome sequences for representative A. baumannii strains from each of the three major IC lineages (65) were downloaded from the NCBI database and also included in some phylogenetic analyses.

To inform the outbreak transmission map, BEAST 2.3.1 (66) was applied to whole-genome SNP, patient of origin, and date of isolation information of the ABOB isolates to generate a Bayesian phylogenetic reconstruction of the outbreak. A Hasegawa, Kishino, and Yano (HKY) substitution model assuming a constant population size was used for the analysis. This information was combined with epidemiological linkages to estimate transmission patterns.

Outbreak analysis.

From June 2013 through December 2013, a CRAB outbreak occurred at NMH and was centered in two ICUs: ICU A, a cardiothoracic ICU that also houses solid-organ transplant recipients, and ICU B, a medical ICU. As part of its investigation, the NMH Department of Infection Prevention and Control performed PFGE typing and assessed epidemiological links. All CRAB isolates deemed to be part of the outbreak had identical or closely related PFGE types. For the purpose of constructing the transmission map, patient-to-patient spread was considered the most likely mode of transmission, and patients were considered to have a direct epidemiologic link if they overlapped in the same unit for >24 h. Indirect links included overlap in the same unit for <24 h, nonoverlapping stays in the same unit in close proximity, and environmental links, such as exposure to contaminated rooms or equipment. As part of the current study, the first CRAB isolates from 10 patients involved in the outbreak were sequenced using the methods described above. Two additional isolates from patient 4 (ABOB04_a and ABOB04_b) and one additional isolate from patient 6 (ABOB06_a) collected during the outbreak were also sequenced. As controls, an environmental isolate taken from the room of patient 4 (ABOBEN) and an isolate with a nonoutbreak PFGE type obtained from a patient in ICU A during the time of the outbreak (ABOB11) were also sequenced. Finally, two carbapenem-susceptible A. baumannii isolates from patients admitted to different locations in the hospital during the outbreak (ABOB15 and ABOB16) were sequenced and also used as controls.

Core- and accessory-genome analyses of outbreak isolates.

As a first step in defining the accessory genome of the outbreak isolates, a core genome was defined by applying the software program Spine (55) to the assembled sequences of the 116 A. baumannii ABBL bloodstream isolates. Briefly, the set of all pairwise whole-genome alignments was used to identify sequences present in 95% of the 116 ABBL isolates (54, 55), and these sequences were defined as the core genome. The software program AGEnt (55) was then used to perform in silico subtractive hybridization of the core genome from the whole-genome sequences of the CRAB ICU outbreak (ABOB) isolates. In this way, the accessory genome of each ABOB isolate was determined. Using BLAST+ version 2.2.24 (44, 46) and in-house Perl scripts, the accessory-genomic sequences of the ABOB strains were aligned and clustered to identify the set of accessory-genomic elements among these strains. To support the accessory element carriage patterns in each strain and correct for false negatives due to misassembly, sequencing reads from each strain were aligned to the set of accessory element sequences using the bwa alignment program (58). The read coverage at each position in the accessory elements was determined from the alignments, and positions with ≥5 aligning reads were considered to be present in the respective genome. A graphical representation of the accessory element composition of each strain was produced using CGView (61). ABOB accessory genetic elements were annotated using Prokka version 1.9 (67). ABOB accessory genomic elements were compared to the nucleotide sequence of A. baumannii plasmid ABKp1 (NCBI accession no. CP001922.1) using BLAST, and gene similarities between these elements and the plasmid were visualized using EasyFig version 2.1 (68).

Comparison of typing techniques.

The adjusted Wallace coefficient, which compares two sets of partitions from different microbial typing methods, was used to determine the congruence between typing techniques (69). Simpson's index of diversity was used to determine the discriminatory power of each individual typing technique (69, 70). For any given typing system, this index measures the probability that two randomly sampled strains will belong to different types. Calculations were performed with the online tool on the Comparing Partitions website (http://darwin.phyloviz.net/ComparingPartitions/). In addition, a sensitivity analysis was performed to evaluate the performance of conventional typing techniques in comparison to WGS in the context of generating a transmission map for a hospital outbreak. For this analysis, WGS was considered to be the gold standard, and a positive result was defined as one that accurately distinguished two distinct isolates as unique.

Nucleotide sequence accession numbers.

All sequence read sets and assembled sequences are deposited in the National Center for Biotechnology Information (NCBI) database (Table 1).

RESULTS

Determining the bacterial species of isolates within the ACB complex.

Species within the ACB complex differ markedly in the severity of illness they cause and in their resistance to antibiotics (7, 13). Accurate identification of Acinetobacter to the species level is becoming increasingly important in the clinical setting. We therefore attempted to determine the species of 154 clinical ACB complex bloodstream isolates cultured from patients at our institution during the study period. One isolate actually contained a mixed culture and was excluded from further analysis. Five isolates gave poor-quality sequencing assemblies, despite repeated attempts, and were also excluded, leaving 148 isolates for analysis. We identified the species of these 148 isolates by rpoB gene sequence analysis, an accepted method of parsing ACB complex bacteria into species (45, 71). This analysis indicated the following Acinetobacter species distribution: A. baumannii (n = 116), A. pittii (n = 28), A. nosocomialis (n = 3), and A. soli (n = 1). Because A. soli is not a member of the ACB complex, the clinical microbiology laboratory presumably misidentified this isolate. These results confirm previous findings that a substantial number of ACB complex isolates acquired from clinical settings are non-baumannii Acinetobacter species (10).

We next examined the utility of using WGS to identify isolates within the ACB complex to the species level. In addition to supplying the sequence of the rpoB gene for each isolate, WGS also allowed the determination of phylogeny via SNP analysis. A phylogenetic tree inferred from core-genome SNPs revealed well-delineated monophyletic lineages corresponding to each species within the ACB complex (Fig. 1). (An alternate version of the tree more clearly demonstrating the species clustering is shown in Fig. S1 in the supplemental material.) Because A. soli is genetically distant from the ACB complex (72, 73), it was used to root the tree. Similar to prior studies, our results showed that A. nosocomialis diverges first from the A. baumannii clade, followed by A. pittii (20). These results confirm that phylogenic analysis based upon SNPs in conserved sequences is capable of distinguishing species within the ACB complex.

FIG 1.

Phylogenetic analysis based on core-genome sequences of all Acinetobacter isolates. A phylogenetic tree for all Acinetobacter isolates was inferred from core SNPs using kSNP version 2. The tree was rooted on isolate ABBL054 (A. soli). The colors correspond to individual ACB complex species, as determined by rpoB gene sequence analysis, and these are indicated in the key. The scale bar indicates branch length, expressed as the number of changes per total number of SNPs.

Next, a band-based typing method was used to examine the collection of ACB complex isolates. Because the NMH clinical microbiology laboratory routinely performed PFGE only on carbapenem-resistant ACB complex isolates (nearly all of which were A. baumannii), we focused on Rep-PCR. This methodology identified 50 unique types among the clinical isolates (Table 2). A dendrogram created from the Rep-PCR fingerprints for all ACB complex isolates demonstrated poor delineation of the isolates into distinct species (see Fig. S2 in the supplemental material). These results demonstrate that Rep-PCR is not a useful genotyping method for determining distant phylogenetic relationships among isolates within the ACB complex.

TABLE 2.

Comparison of typing methods used to characterize CRAB isolates^a

Typing methodb	No. of types	Simpson's index (95% CI)c	Sensitivity (%)	Specificity (%)
PFGE	40	0.892 (0.834–0.950)	90	53
Rep-PCR	50	0.970 (0.950–0.990)	97	57
MLST	10	0.758 (0.706–0.810)	81	100
WGSd	75	0.997 (0.995–1.000)

^aIn this analysis, sensitivity refers to the ability of the technique to correctly discriminate two unique isolates, and specificity refers to the ability of the technique to correctly group two identical isolates. Sensitivity and specificity were calculated in comparison to WGS as the gold standard.

^bPFGE, pulsed-field gel electrophoresis; Rep-PCR, repetitive extragenic palindromic-PCR; MLST, multilocus sequence typing; WGS, whole-genome sequencing.

^cCI, confidence interval.

^dIsolates were considered the same type by WGS if there were 0 SNPs between them.

Assignment of A. baumannii clinical isolates to ST lineages.

Previous investigations established that certain clonal lineages of A. baumannii have spread widely across and between continents. Some of these lineages have been referred to as international clones (ICs) or European clones (ECs) (17–19, 21). The widespread distribution of these clonal lineages implies that they are highly adapted for persistence and transmission in health care environments. MLST has become the gold standard for assigning A. baumannii isolates to STs that correspond to these lineages. Multilocus sequence types were therefore determined for all A. baumannii strains using the Institut Pasteur database. Twenty-four unique STs were identified among the 116 A. baumannii isolates, including seven new STs, four of which have been assigned: ST496, ST497, ST498, and ST499. However, 90 of 116 (78%) isolates were assigned to one of four STs: ST2, ST79, ST406, and ST499. A phylogeny inferred from the concatenated sequences of the MLST genes for A. baumannii isolates confirmed this clustering (see Fig. S3 in the supplemental material). These results indicated that both previously reported and novel A. baumannii clonal lineages circulated throughout our hospital.

In examining the utility of PFGE and Rep-PCR for identifying clonal lineages of A. baumannii, we found that neither band-based technique accurately grouped strains together in agreement with the MLST data. A dendrogram created using the Rep-PCR results demonstrated only one distinct cluster with >90% similarity (see Fig. S4 in the supplemental material), a cutoff level above which isolates can be considered genetically related (21, 46, 47). This cluster included eight A. baumannii isolates involved in a prolonged multistate outbreak from 2005 to 2006. All of these isolates were ST79, suggesting some concordance between Rep-PCR and MLST in a regional outbreak setting. However, this Rep-PCR cluster also included 11 other isolates that were collected at various times throughout the study and were represented by eight different STs, confirming a lack of agreement with MLST. PFGE likewise showed poor agreement with MLST results (see below).

We next examined the ability of WGS to partition A. baumannii isolates into clonal lineages. The core genomes of these isolates were analyzed for the presence of discriminating SNPs, and a phylogenetic tree was created. Four distinct primary clades (labeled A to D) were identified, and also a number of phylogenetically distinct isolates (Fig. 2). Each of the four major clades corresponded to one of the four predominant STs (ST79, ST2, ST406, and ST499 for clades A to D, respectively). Each clade contained isolates collected throughout the 8-year study from patients at differing locations in the hospital, suggesting that they did not represent confined outbreaks within our hospital. To better examine the relationships between the isolates in our hospital and globally disseminated lineages, we added several published strains to our phylogenetic analysis (see Fig. S5 in the supplemental material). Clade B (ST2) strains were closely related to published IC-II strains. Clade C (ST406) is most closely related to IC-I strains, although the tight clustering of the clade C isolates and the phylogenetic distance separating them from the IC-I reference strains suggest that clade C may be a distinct clonal group. Clade A (ST79) isolates do not clearly cluster with any reported IC lineage but are closely related to published strains from the United States and other parts of the world (74–77). Group D (ST499) represents a collection of isolates distinct from known clonal lineages or reported strains. Interestingly, our WGS analysis indicated that IC-III strains did not represent a distinct lineage of strains but rather were phylogenetically diverse. Many of the remaining isolates from our study were interspersed with these published IC-III strains. These findings indicate that WGS accurately assigns A. baumannii isolates to ST lineages and provides additional information regarding the genetic relationships between different lineages.

FIG 2.

Phylogenetic analysis based on core-genome sequences of A. baumannii isolates. A phylogenetic tree for all A. baumannii isolates was inferred from core SNPs using kSNP version 2. Individual taxa and branches of the four major clades are identified by shades of gray and letters (A to D). Isolates collected during the ICU outbreak (those with ABOB) are also shown and colored red. The scale bar indicates branch length, expressed as the number of changes per total number of SNPs.

Use of WGS to define a hospital outbreak of A. baumannii.

A. baumannii frequently causes outbreaks within health care settings, in which a single strain is transferred from one patient to another. Because of the high level of antibiotic resistance associated with some A. baumannii strains, these outbreaks have important consequences for patient care and require the use of considerable resources by infection control personnel. PFGE is routinely used by many hospitals to track nosocomial outbreaks, but several studies indicate that WGS is better suited for this purpose (38, 78, 79). In 2013, a CRAB outbreak occurred in our hospital and was investigated using PFGE and contact tracing. We reexamined this outbreak using WGS to determine whether outbreak isolates and transmissions could be more accurately identified with this technique.

In June 2013, two patients were admitted to ICU B from the same skilled nursing facility within 8 days of each other. Patient 1 grew CRAB (isolate ABOB01) from a respiratory sample 9 days after admission, and patient 2 grew CRAB (isolate ABOB02) with a closely related PFGE type from the blood 10 days after admission. Neither patient was in contact isolation prior to CRAB growth, nor had they overlapped previously with a CRAB patient during their hospital stay. Over the next 5 months, an additional eight patients were identified in ICU A and B who grew CRAB (isolates ABOB03, ABOB04, ABOB06, ABOB07, ABOB08, ABOB09, ABOB10, and ABOB12) with identical or closely related PFGE types to the index cases and were classified as part of the outbreak (see Fig. S6 in the supplemental material). This represented a substantial increase in the monthly number of CRAB cases at the NMH. Three isolates were available from patient ABOB04 (isolates ABOB04, ABOB04_a, and ABOB04_b) and two isolates were from patient ABOB06 (isolates ABOB06 and ABOB06_a). All isolates were susceptible to colistin, and most were susceptible to either doxycycline or minocycline but were otherwise resistant to all antibiotics tested. At the time, epidemiologic contact tracing was performed to estimate a transmission map, which revealed multiple potential infection routes among outbreak patients (Fig. 3A). Interestingly, the PFGE patterns of the two initial isolates ABOB01 and ABOB02 (from patients 1 and 2, respectively) were found to be closely related but not indistinguishable, suggesting that the outbreak may have begun with the introduction of two unique but closely related CRAB isolates into the hospital.

FIG 3.

Possible transmission maps of an ICU CRAB outbreak. Transmission maps were created using patient epidemiologic trace data and PFGE results by the infection control team investigating the outbreak at the time it occurred (A), and additional phylogenetic information inferred from whole-genome sequences (B). (B) Data from whole-genome SNPs were taken into account, as shown in Fig. 5 and Fig. S7 in the supplemental material. Specifically, the most likely patient-to-patient transmissions suggested by BEAST software analysis of WGS data (as shown in Fig. S7 in the supplemental material) were used to resolve ambiguous transmission events in panel A, which in turn yielded the transmission map in panel B. In all maps, nodes represent individual patients, with shaded nodes indicating patients in ICU A and white nodes indicating patients in ICU B. (A) Solid arrows represent direct epidemiological links, while dashed arrows represent indirect or environmental links. (B) Black arrows indicate transmission links supported by WGS data, while red arrows indicate that the transmission is also supported by epidemiologic data. SNF, skilled nursing facility.

We next applied WGS to investigate this same hospital outbreak. Each outbreak isolate, the control isolates (ABOB11, ABOB15, and ABOB16), and an environmental isolate (ABOBEN) were sequenced, and phylogeny was inferred based on SNPs in the core genome (defined as sequences found in at least 95% of all ABBL isolates) (Fig. 4). As expected, the control isolates were distantly related to the outbreak isolates, which formed a tight cluster. A similar tree including only the clinical outbreak isolates (Fig. 4, first inset) revealed that isolate from patient 9 was genetically distinct, differing by 155 to 157 core SNPs from the other outbreak strains. A closer inspection indicated that these SNPs were not localized to a single small portion of the genome, as would have occurred with a recombination event, but rather were dispersed throughout the chromosome (data not shown). The remaining outbreak isolates differed from each other by 0 to 2 core SNPs. Thus, although patient 9 was infected with a CRAB strain in ICU B during the outbreak, and this CRAB isolate had a PFGE pattern similar to that of the other outbreak isolates, WGS indicated that the patient was not infected with an outbreak strain. ABOB09 was therefore removed from subsequent phylogenetic analyses (Fig. 4, smaller inset).

FIG 4.

Phylogenetic analysis based on core-genome sequences from the ICU CRAB outbreak isolates. A phylogenetic tree for the outbreak and control isolates was inferred from core-genome SNPs using kSNP version 2 and the maximum-likelihood method. Thirteen outbreak isolates (red), three control nonoutbreak isolates (blue), and one environmental outbreak isolate (green) were included. The large inset is the same tree showing only the clinical outbreak (red) isolates and excluding the control and environmental isolates. The small inset excluded isolate ABOB09 to better show the relationships between the remaining isolates. In this analysis, the core genome was defined as sequences present in 95% of the ABBL isolates.

Generation of an intrahospital outbreak transmission map.

The identification of factors allowing an A. baumannii strain to move from patient to patient during a hospital outbreak is necessary to determine the source and mechanisms by which transmission is facilitated in an outbreak. High-resolution genetic information can help distinguish isolates even within the context of a nearly clonal outbreak. We therefore examined whether WGS could discriminate isolates within our ICU outbreak and help generate a more-refined transmission map. To accomplish this, we performed a phylogenetic analysis of whole-genome SNPs in the outbreak isolates (Fig. 5). The core-genome analysis used only SNPs located in the core genome, whereas the whole-genome analysis also included SNPs located in accessory-genome sequences and therefore used more of the information present in the bacterial genomes. In contrast to the core-genome tree, the whole-genome tree placed ABOB02 at an increased distance from the remaining outbreak isolates (Fig. 5). Comparing whole-genome sequences, ABOB02 differed by 98 to 103 SNPs from the remaining outbreak isolates, in contrast to 1 to 2 SNP differences found in the core-genome sequences. At first glance, this difference in the number of whole-genome SNPs suggested that ABOB02 was also not part of the outbreak. However, further examination revealed that most of these SNPs were localized to an ∼14-kb region that is accessory in the ABBL bloodstream isolates but found in each of the outbreak strains (data not shown). Alignment of raw reads from ABOB02 against the ABOB01 sequence and manual examination confirmed these were likely true SNPs and not sequencing errors, with an average of 311 (range, 51 to 696) sequencing reads at each SNP position containing the variant base, and an average of less than one read (range, 0 to 7 reads) containing the reference base (data not shown). The concentration of SNPs in a small region of the genome suggested that this 14-kb accessory region had been acquired by a recent recombination event and therefore did not represent the true genetic history of ABOB02. For this reason, ABOB02 was considered likely to be related to the outbreak. A Bayesian phylogenetic reconstruction of the outbreak (see Fig. S7 in the supplemental material) was generated using the whole-genome sequence SNP information along with isolation dates (see Materials and Methods). The incorporation of these results into the transmission map defined many of the transmission links that were previously ambiguous (Fig. 3B).

FIG 5.

Phylogenetic analysis based on whole-genome sequences from the ICU CRAB outbreak isolates. A phylogenetic tree for 12 outbreak isolates was inferred from SNPs present in the whole genome using kSNP version 2 and the maximum-likelihood method. In the inset, ABOB02 is excluded to better show the genetic distances between the remaining outbreak isolates.

To determine whether there were differences in the accessory genomes of the outbreak isolates beyond SNPs, we looked for accessory sequences that were present in some ABOB isolates but absent from others. We identified four accessory-genome elements (AGEs) present only in isolates ABOB04_b, ABOB07, ABOB08, and ABOB12 and one small AGE present only in isolates ABOB07 and ABOB08 (Fig. 6). These AGEs ranged in size from 1,337 bp to 56,215 bp, totaling 71.8 kb of sequence. These five AGEs are difficult to assemble de novo using short reads, likely due to repeat sequences or ambiguity; however, their presence or absence as a defined set in the outbreak isolates suggested they were in close proximity to or contiguous with each other. This was confirmed by BLAST search, which indicated that together, they were closely homologous with ABKp1, a plasmid found in A. baumannii strain 1656-2 (see Fig. S8 in the supplemental material). Patient 4, who had a prolonged stay in ICU A during the outbreak, harbored isolates both with and without this plasmid. Although plasmids have been shown to be important in the dissemination of antibiotic resistance during outbreaks (80, 81), they are of limited value for defining transmission pathways, because they are frequently gained and lost by closely related circulating strains (74, 82, 83). These results indicated that a plasmid circulated among the outbreak isolates and was the only gross difference in their accessory genomes.

FIG 6.

Accessory-genome differences between ICU CRAB outbreak isolates. All AGEs of >1,000 bp are represented in the circular map, which is organized by size from largest to smallest AGE in a clockwise direction. The outer green and orange ring shows the cumulative collection of AGEs found in at least one of the outbreak isolates. The inner rings indicate the presence or absence of AGEs in each outbreak isolate, colored according to the key. Gaps in color represent AGEs or parts of AGEs that are missing in those isolates. ABOB04, ABOB04_a, and ABOB04_b represent three different isolates from patient 4; ABOB06 and ABOB06_a represent two different isolates from patient 6.

The incorporation of the total information generated by WGS allowed us to refine and remove ambiguities from the previous transmission map based solely on PFGE patterns and epidemiologic links (compare Fig. 3A and B). It also suggested possible transmissions (e.g., from patient 3 to patient 4) that had not been identified by the epidemiological investigation. Such transmission may have occurred indirectly through colonized patients that had not been identified as part of the outbreak. These results suggest that WGS is superior to PFGE for distinguishing A. baumannii outbreak transmission routes, but the quantification of SNPs alone may be misleading, and an examination of both core- and accessory-genome content is beneficial.

SNP thresholds that define isolates as part of a clonal lineage or intrahospital outbreak.

Well-defined criteria are available for interpreting conventional typing techniques for the purpose of parsing clinical isolates into the same clonal lineage or the same hospital outbreak (20, 40–42). Corresponding thresholds of relatedness would be helpful for interpreting WGS results. We hypothesized that pairwise core SNP counts might serve as such thresholds and provide an easy and straightforward method by which infection control practitioners could use WGS data to quickly determine the likelihood of whether any two isolates were part of the same clonal lineage or outbreak. To this end, the total number of core-genome SNPs between any two of the 116 ABBL and 15 ABOB isolates was quantified. These comparisons yielded a minimum of 0 and a maximum of 10,360 core-genome SNPs (median, 9,523 SNPs). We next compared the number of core-genome SNPs between ABBL isolates from different clonal lineages to those from the same clonal lineages (Fig. 7). For the purposes of this analysis, we defined two strains as belonging to the same clonal lineage if they had the same ST. A discontinuity in the distribution of core-genome SNPs was observed between these two groups of isolates. The median number of core-genome SNPs between isolates within the same ST lineage was 261 (interquartile range, 67.5 to 393.5 SNPs) versus 9,613 SNPs (interquartile range, 9,443 to 9,793.75 SNPs) between isolates not within the same ST lineage (Fig. 7). With only two exceptions, a threshold of 2,500 core SNPs distinguished isolates within the same ST lineage from those in different ST lineages (Fig. 7). We next performed a similar analysis on the ABOB outbreak isolates. A maximum of 2 SNPs were found among all pairwise comparisons of the ABOB outbreak isolates. Thus, a threshold of 2.5 SNPs distinguished the vast majority of outbreak isolates from isolates of the same ST lineage that were not part of the outbreak (Fig. 7). Although a number of non-ABOB isolates also differed by <2.5 SNPs, it is possible that some of these represent episodes of unrecognized patient-to-patient transmissions that occurred within our hospital. This analysis needs to be expanded to include other outbreaks at different institutions and of different durations, but these results suggest that core-genome SNP thresholds are useful in preliminarily determining whether an isolate is within or distinct from an ST lineage or intrahospital outbreak.

FIG 7.

Core-genome SNP counts among isolates of the same or different clonal lineage or the same intrahospital outbreak. To generate a jitter plot of core SNP differences, all pairwise core-genome SNP counts were compared among isolates belonging to different clonal lineages (as determined by different STs; column 1, n = 5,026), to the same clonal lineage (the same ST) but not part of a recognized outbreak (column 2, n = 1,079), or to a recognized outbreak (column 3, n = 66). Each dot represents the number of core SNPs between one pair of isolates. The red dashed line indicates a threshold of 2,500 SNPs, and the green dashed line indicates a threshold of 2.5 SNPs.

Comparison of typing methods.

Using the WGS threshold of <2,500 core SNPs, we next compared the four typing techniques (MLST, Rep-PCR, PFGE, and WGS) for their ability to distinguish A. baumannii clonal lineages. Comparisons were restricted to CRAB isolates, because typing results from all methods were available for these isolates. Wallace coefficients, which quantify the congruence between two methods, were calculated for a pairwise comparison of each typing technique (Table 3). Since MLST is currently accepted as the gold standard for the identification of A. baumannii clonal lineages (19), we paid particular attention to how the results of the other techniques compared to MLST. PFGE performed poorly in this regard; two isolates grouped together by PFGE had only a 39% probability of having the same MLST. Rep-PCR performed somewhat better; two isolates grouped together by Rep-PCR had a 65% probability of having the same MLST. As expected based on how the 2,500-core-SNP threshold was derived, two isolates placed in the same clonal lineage by WGS were grouped together by MLST 100% of the time. These results indicate that the band-based techniques PFGE and Rep-PCR differ significantly from MLST in assigning isolates to clonal lineages, but that WGS performed well when a threshold of 2,500 core SNPs was used.

TABLE 3.

Concordance of typing techniques using the adjusted Wallace coefficient for nonoutbreak (ABBL) CRAB isolates

Typing methoda	Adjusted Wallace coefficient (95% CI)b
	PFGE	Rep-PCR	MLST	WGSc
PFGE		0.168 (0.013–0.323)	0.387 (0.120–0.654)	0.387 (0.120–0.654)
Rep-PCR	0.659 (0.603–0.715)		0.651 (0.415–0.887)	0.651 (0.415–0.887)
MLST	0.147 (0.014–0.279)	0.063 (0–0.131)		1.0 (1.0–1.0)
WGSc	0.147 (0.014–0.279)	0.063 (0–0.131)	1.0 (1.0–1.0)

^aPFGE, pulsed-field gel electrophoresis; Rep-PCR, repetitive extragenic palindromic-PCR; MLST, multilocus sequence typing; WGS, whole-genome sequencing.

^bCI, confidence interval.

^cIsolates were considered the same type by WGS if there were ≤2,500 single nucleotide polymorphisms between them.

We next determined how WGS with the threshold of 2.5 core SNPs compared to PFGE and MLST in correctly identifying A. baumannii isolates as part of an intrahospital outbreak. We compared these techniques using our collection of CRAB isolates, some of which comprised a hospital outbreak (ABOB isolates), but most of which did not (ABBL isolates). Of isolates identified by WGS as having genetic similarity consistent with patient-to-patient transmission (defined as <2.5 core SNPs), 88% had the same PFGE pattern (Table 4). In contrast, of isolates with the same or similar PFGE patterns, only 14% were found to be genetically similar by WGS (<2.5 core SNPs). In other words, PFGE grouped many more isolates together as similar than did WGS, suggesting that WGS has higher discriminatory power. Due to its limited ability to discriminate related strains, MLST showed poor congruence with WGS and PFGE (Table 4).

TABLE 4.

Concordance of typing techniques using the adjusted Wallace coefficient for nonoutbreak (ABBL) CRAB and outbreak (ABOB) CRAB isolates

Typing methoda	Adjusted Wallace coefficient (95% CI)b
	PFGE	MLST	WGSc
PFGE		0.462 (0.228–0.696)	0.136 (0.087–0.186)
MLST	0.122 (0.038–0.205)		0.041 (0.000–0.091)
WGSc	0.882 (0.841–0.923)	1.000 (1.000–1.000)

^aPFGE, pulsed-field gel electrophoresis; MLST, multilocus sequence typing; WGS, whole-genome sequencing.

^bCI, confidence interval.

^cIsolates were considered the same type by WGS if there were ≤2 single nucleotide polymorphisms between them.

We next directly examined the discriminatory power (i.e., the ability to distinguish any two isolates as distinct) of PFGE, Rep-PCR, MLST, and WGS. The ability to discriminate between two nearly clonal strains is highly important for developing transmission maps during an outbreak. We calculated the Simpson's index, a measure of discriminatory power, of the four typing techniques (Table 2). Overall, both WGS and Rep-PCR were highly discriminatory, with Simpson index values of 0.997 and 0.970, respectively (Table 2). PFGE and MLST were somewhat less discriminatory, with Simpson index values of 0.892 and 0.758, respectively. We also calculated the sensitivity and specificity of the techniques for distinguishing isolates, using WGS as the gold standard. The band-based techniques were highly sensitive for discriminating distinct isolates but suffered from low specificity. In other words, they correctly identified distinct isolates as different but often erroneously distinguished strains that WGS categorized as being the same. MLST was less sensitive in its ability to distinguish isolates but had a specificity of 100%.

DISCUSSION

Hospital-based clinical microbiology laboratories provide a number of services, including the identification of pathogens in clinical specimens and the typing of strains to facilitate regional epidemiological surveillance and hospital outbreak investigation. These tasks require typing techniques with a broad spectrum of discriminatory powers, from distinguishing species of related bacteria to distinguishing individual clones. As a result, different typing techniques have been found to be optimal for these different tasks. In this study, we used conventional band- and sequence-based typing techniques and WGS to characterize the genetic diversity and epidemiology of a large collection of clinical ACB complex isolates and to investigate an ICU CRAB outbreak. Our study adds to the growing body of literature supporting the use of WGS to define Acinetobacter species, characterize the population structure of A. baumannii, clarify the extent of a hospital outbreak, and delineate an outbreak transmission map (84).

Commercially available platforms for the identification of bacteria perform poorly in identifying ACB complex bacteria to the species level, yet distinguishing these species has clinical importance. Compared with non-baumannii ACB complex species, A. baumannii is more frequently multidrug resistant, more often cultured from critically ill patients, and associated with poorer clinical outcomes, likely due to delays in the receipt of appropriate antibiotic therapy (7, 13, 85). A strain typing method that rapidly identifies bacteria within the ACB complex to the species level and identifies resistance genes would lead to a timelier administration of appropriate antibiotic therapy and, presumably, improved patient outcomes. WGS allowed us to assign ACB complex bacterial isolates to an individual species in two ways. First, WGS provided the sequence of the rpoB gene, which by itself can accurately categorize Acinetobacter species (45, 71). Second, phylogenies based on WGS demonstrated distinct clades representing individual Acinetobacter species. The inclusion of the strain types for the relevant Acinetobacter species within these phylogenies would allow for easy assignment of an individual isolate to a species. Although not examined in this study, WGS can also provide ancillary information, such as the presence or absence of virulence factors and antibiotic resistance islands, which are important for developing effective and timely treatment strategies (74, 86). In contrast, the band-based technique Rep-PCR was not able to separate ACB complex isolates into distinct species.

Certain A. baumannii clonal lineages have disseminated widely and are of particular clinical importance. In recent years, MLST has emerged as the gold standard for categorizing A. baumannii isolates into clonal lineages and characterizing their continental or global spread. Indeed, MLST indicated that four major STs (ST2, ST79, ST406, and ST499) were endemic at NMH over the course of our study. A phylogenetic analysis based on WGS also clustered the majority of A. baumannii strains into four groups corresponding to these STs. ST2 (clade B) is part of the IC-II lineage, which is widely prevalent throughout Europe and the United States and is frequently implicated in nosocomial outbreaks (20, 74, 75). ST406 (clade C) was first reported in an epidemiological study of CRAB clinical isolates from Japan (87) and was subsequently identified in Brazil (88). It is closely related to IC-I strains, which are globally disseminated. ST79 (clade A) was first described by Villalón and colleagues (21) during a study of epidemic nosocomial A. baumannii isolates collected over an 11-year period in Spain and has subsequently been described in the United States, Canada, and Argentina (74–77). Although not assigned an IC number, it appears to represent an A. baumannii lineage that has spread globally. We also identified a novel ST, ST499 (clade D), which is closely related to ST123, which was first identified in Las Vegas, NV (75). Many of the remaining isolates from our study were phylogenetically interspersed with strains reported in the literature as belonging to IC-III (see Fig. S5 in the supplemental material). Interestingly, our analysis indicates that many strains previously published as IC-III are in fact quite heterogeneous, with deep branches in the phylogenetic tree. These IC-III strains are more distantly related to each other than IC-I or IC-II strains are to each other and may not represent closely related clonal lineages in the same manner as IC-I and IC-II strains. Consistent with this conclusion, these IC-III strains represented a variety of different STs. Our findings, which are in agreement with the population structure reported in other studies (20, 74, 89), demonstrate that both well-characterized IC lineages and novel clonal lineages of A. baumannii are endemic in NMH, and that WGS accurately identifies these lineages.

In contrast to WGS, the band-based techniques PFGE and Rep-PCR performed poorly in delineating isolates into distinct clonal lineages. In particular, isolates with similar PFGE patterns were of the same ST only 39% of the time. Other studies have also reported poor agreement between PFGE and MLST for A. baumannii (90, 91) and have demonstrated that isolates from a single ST yield markedly different PFGE patterns (21, 89). These disparate results likely reflect the sensitivity of PFGE to accessory-genome changes (such as genomic island acquisition and deletion), which appear to occur relatively frequently and are not limited to isolates with similar evolutionary histories (80). In contrast, MLST reflects SNPs in conserved housekeeping genes and therefore more accurately represents population phylogeny. Although PFGE has improved with standardization, it remains subject to a number of factors that decrease reliability and reproducibility, such as DNA yield and purity, the type of enzyme used, and errors in gel resolution, intergel standardization, and band visualization and identification (25, 26, 49, 92). These findings suggest that PFGE should not be used to determine relationships between distantly related bacterial isolates (93). In our study, Rep-PCR performed better than PFGE but was still poorly congruent with MLST (Table 3). This was somewhat surprising, since Rep-PCR has been used to define globally disseminated A. baumannii clonal lineages, and others have found good agreement between Rep-PCR and MLST when used in this way (22). A possible explanation is that many of the published studies utilized the DiversiLab system, an automated Rep-PCR platform to which we did not have access. DiversiLab uses microfluidic capillary electrophoresis, which allows for high degrees of resolution and reproducibility (94). Our Rep-PCR assays involved individually performed PCRs, traditional gel electrophoresis, and subjectively interpreted banding patterns, which may have resulted in decreased reproducibility, a recognized disadvantage intrinsic to nonautomated band-based techniques (95). Furthermore, it has been shown that Rep-PCR results are affected by the DNA extraction method used (96). We note that these issues are not specific to our laboratory and may make our results more applicable to other clinical microbiology laboratories that do not use the DiversiLab system.

We also investigated the performance of WGS in the context of an intrahospital A. baumannii outbreak. In this setting, analysis of core-genome SNPs via WGS provided greater resolution than conventional PFGE typing and clarified which isolates were part of the outbreak. One isolate (ABOB09) deemed part of the outbreak by epidemiologic tracing and PFGE typing was excluded by WGS. The incorporation of whole-genome SNP information confirmed that the outbreak began with the near-simultaneous introduction into our hospital of two discrete strains from the same skilled nursing facility. Such facilities are known reservoirs for health care-associated bacteria, such as A. baumannii (97, 98). WGS also clarified some of the possible transmission routes suggested by epidemiologic data. In this regard, WGS was superior to PFGE due to its enhanced discriminatory power (Table 2) (0.997 versus 0.892), which has been noted in other studies of bacterial outbreaks (31, 79, 99, 100). In particular, our results are similar to those of Salipante and colleagues (79), who used a different approach to identify SNPs and define outbreak strains but arrived at the same conclusions regarding the deficiencies of PFGE in the context of an A. baumannii outbreak. At the molecular level, this reflects the relatively large number of SNPs that can accumulate in genomes before they are detectable by PFGE (79). In addition, our study found three variants of the same strain within a single patient (ABOB04, ABOB04_a, and ABOB04_b). The existence of multiple strain variants within a single patient, a “cloud of diversity” (101), has been noted by others (82, 102) and complicates the construction of transmission maps. A single patient is capable of spreading different strain variants to multiple other patients, and several strain variants can evolve over time while colonizing or infecting a single patient. These occurrences may confuse transmission events and require that many bacterial isolates from each patient be sequenced if the full complexity of an outbreak is to be captured (101). However, even sequencing a single isolate from most patients allowed us to improve the resolution of our transmission map. Overall, our results agree with prior studies, which have also demonstrated enhanced outbreak analysis with WGS (38, 83, 103). Snitkin and colleagues (38) integrated data from WGS and epidemiologic tracings to more accurately reconstruct transmissions and dissect the progression of a carbapenem-resistant Klebsiella pneumoniae outbreak. Other studies have revealed that WGS is more accurate than conventional techniques in discriminating among alternate transmission scenarios during outbreaks of MDR A. baumannii (78) and carbapenem-resistant Enterobacter cloacae (37). The results of these studies in conjunction with our own reveal how WGS can be a powerful tool in outbreak analysis and can better direct infection prevention efforts, leading to earlier and more successful control of hospital outbreaks.

Before WGS can widely be used by infection control practitioners, user-friendly approaches must be developed that rapidly analyze sequence data to segregate isolates. As stated by Sabat and colleagues (94) in their discussion of the use of WGS for epidemiological surveillance, “…the key challenge will not be to produce the sequence data, but to rapidly compute and interpret the relevant information from large data sets” (94). Easy-to-use criteria to define which isolates are within the same clonal lineage or within the same hospital outbreak are currently available for band-based techniques (23, 40, 48) but not for WGS (104). For this reason, we identified a threshold of 2,500 core SNPs that distinguished isolates of the same clonal lineage from those of different clonal lineages. Using this threshold, WGS was highly congruent with MLST. Similarly, we used a threshold of 2.5 core SNPs to segregate outbreak from nonoutbreak isolates. We acknowledge that several apparently nonoutbreak isolate pairs in our study differed by <2.5 core SNPs (Fig. 7). However, since many of these isolate pairs were collected within 2 weeks of each other, it is possible they were not false positives but rather represent unrecognized patient-to-patient transmissions (data not shown). A threshold of 2.5 core SNPs agrees with other studies of A. baumannii and other bacterial species (39, 79, 105). For example, Salipante and colleagues (79) defined ≤3 SNPs as a threshold for interpreting isolates as clonal, although their value was based on the reproducibility of WGS using technical replicates. To similarly avoid calling sequencing errors as SNPs, we sequenced to a much-higher read number (139 versus ∼50 reads) and filtered low-frequency reads. Our threshold value was also in agreement with sequencing of distinct isolates from the same patient in our ICU outbreak; such isolates had a maximum difference of 1 SNP, demonstrating that A. baumannii accumulated SNPs slowly over the time frame of an intrahospital outbreak. In contrast to our results, some studies have identified greater SNP differences between A. baumannii isolates recovered at different times from the same patient (106, 107). In one study (107), the authors detected SNPs via alignment to a reference genome, followed by SNP filtering based on a “probability score.” Alignment-based methods can result in false-positive SNP calls. Filtering based on multiple parameters with manual verification of SNPs may be more accurate in identifying SNPs (30, 38, 106, 108, 109). Other studies using alignment-based methods with more-stringent filtering have demonstrated far fewer SNPs among similar isolates identified in an outbreak (82). In contrast to reference genome alignment, the kSNP program identifies SNPs by a direct comparison of short oligomers, minimizing false-positive SNP identification resulting from poor alignment. It also filters out high-density SNPs, which are more likely to result from misalignment or misassembly. Whatever method is used, it is important to exclude SNPs acquired through recombination events, as such events can dramatically increase SNP counts and erroneously classify two closely related strains as being genetically divergent (109).

The approach of counting SNPs to categorize A. baumannii isolates does have several limitations. First, it does not take into account the full extent of evolutionary information intrinsic to the locations and types of SNPs present in a collection of isolates. Powerful and sophisticated methodologies are being developed and applied to WGS data to capture this information and more accurately determine transmission maps (110–112). The core-genome SNP thresholds determined in our study should be viewed as rapid and easily obtained first approximations that may require subsequent verification with more sophisticated methodologies. Second, it is likely that core SNP thresholds will vary with the duration of an outbreak. Regional and multistate outbreaks and outbreaks of longer durations will likely require different thresholds. For example, it is anticipated that more SNPs will accumulate in outbreaks lasting a year than in those lasting only a month. Third, our thresholds were based on a single population of isolates collected at one hospital. Studies of additional hospitals and outbreaks are necessary to iteratively refine these values and to enhance their robustness. For this reason, we view the thresholds presented here as starting points for further refinements.

As clinical microbiologists and infection control practitioners consider changing from conventional genotyping techniques to WGS, they must carefully weigh the advantages and disadvantages of both approaches. This study and other recently published studies provide a framework for doing so. We found that WGS was superior to conventional techniques for typing Acinetobacter at the levels of species identification, population structure, and intrahospital outbreak analysis. WGS is also capable of generating reproducible results regardless of the laboratory location or reagents used, and sequencing data can be easily distributed to other epidemiologists, laboratories, and researchers. These benefits of WGS must be balanced against the cost of commercially available next-generation sequencing instruments or services and the current complexity of sequence analysis. Published estimates of per-strain WGS costs approach those for conventional typing techniques (WGS, $35 to 300; PFGE, $150; MLST, $65 to 120; and Rep-PCR, $26 [79, 94, 113]), although these estimates are highly dependent upon how instrument costs are amortized. WGS may allow additional cost savings resulting from earlier control of hospital outbreaks and infection prevention. As sequencing costs continue to fall and resources for fast and accurate sequence interpretation improve, we anticipate that WGS will be regularly incorporated into routine microbiological surveillance and infection control programs.

Funding Statement

Funding was provided to Margaret A. Fitzpatrick through the Eleanor Wood Prince Grants Initiative, a project of the Woman's Board of Northwestern Memorial Hospital.