Genomic Analysis of Factors Associated with Low Prevalence of Antibiotic Resistance in Extraintestinal Pathogenic Escherichia coli Sequence Type 95 Strains

Antibiotic resistance in bacterial pathogens is a major public health concern. This work was motivated by the observation that only a small proportion of ST95 isolates, a major pandemic lineage of extraintestinal pathogenic E. coli, have acquired antibiotic resistance, in contrast to many other pandemic lineages. Understanding bacterial genetic factors that may prevent acquisition of resistance could contribute to the development of new biological, medical, or public health strategies to reduce antibiotic-resistant infections.

types 131 (ST131), ST95, ST69, ST73, and ST393. These STs are responsible for nearly half of all E. coli UTIs or BSIs in many regions of the world (4). Comparative genomic analysis of ExPEC strains to date has not conclusively identified essential "pandemicity" genes in these lineages (5), but there is considerable interest in understanding virulenceassociated genetic features of pandemic ExPEC lineages.
In several recent surveys of ExPEC isolates from the United States, ST95 was the second most common ST, after ST131. These include reports by Adams-Sapper et al. (6), who examined bloodstream isolates from a San Francisco, CA, hospital; Bannerjee et al. (7), who examined isolates from blood, urine, and other extraintestinal sites from Olmsted County, Minnesota; and Salipante et al. (5), who examined urinary and bloodstream isolates from the University of Washington Hospital system (Seattle, WA). ST131 clinical isolates typically are multidrug resistant (MDR), expressing fluoroquinolone resistance due to chromosomal mutations and, in many instances, extendedspectrum ␤-lactam resistance due to acquired plasmid-borne or chromosomal genes (8,9). In contrast, ST95 isolates are much less frequently antibiotic resistant than ST131 strains and many other clonal groups (4). Furthermore, Adams-Sapper et al. (6) noted that, at least in San Francisco, the fimH6 sublineage of the ST95 group was nearly always pansusceptible and yet very common. This implies, first, that antimicrobial resistance may not be a major factor contributing to the pandemicity of ST95 strains, and second, that there may be something distinctive about fimH6 strains with respect to acquisition or maintenance of antibiotic resistance.
We describe here the use of whole-genome sequence analysis to explore these implications. We sequenced 44 ST95 bloodstream isolates from San Francisco General Hospital (SFGH) that had been recovered between 2007 and 2011 (6,10). These initial strains encompassed the four major sublineages of ST95, as defined by alleles of the fimH marker. We added more geographic diversity with four ST95 isolates from Minnesota (11) and five from other locations in the United States (12). Finally, we included archived sequences of 33 ST95 isolates from the Seattle area (5) in our analysis. Through this approach, we were able to identify the genetic basis for acquired antibiotic resistance genes in a large United States-based sample of ST95 isolates and to identify genomic differences between pansusceptible and resistant isolates that help explain disparities in resistance within the ST95 group.

RESULTS
Genome sequencing and isolate characterization. A total of 53 E. coli ST95 isolates were subjected to whole-genome sequencing on the Illumina MiSeq platform, followed by de novo assembly and analysis (Table 1; see Table S1 in the supplemental material). Most (83%) of the isolates were from San Francisco, CA. We also examined archived whole-genome sequences of 33 ST95 isolates from the Seattle, WA, region (5). The fimH gene, which encodes an adhesin critical for urinary tract pathogenesis, has proven to be useful as a genetic marker to increase the resolution of MLST with ExPEC strains (13,14). Examination of fimH genotypes from genome sequences allowed assignment of the ST95 isolates to sublineages ( Table 1). Alignment of scaffolded assemblies from the draft genome sequences and subsequent tree building ( Fig. 1) showed the fimH sublineages to be phylogenetically coherent, with the major division being between fimH1 and fimH6 isolates, and with the fimH9 and fimH47 clusters emerging from separate branches of the fimH1 group. Among the isolates we sequenced, the fimH6 genotype was most abundant (48%), followed by the fimH1 (31%), fimH47 (17%), and fimH9 (3%) genotypes.
Based on their fimH genotype and antibiotic resistance phenotypes, four ST95 isolates from San Francisco were selected for long-read, single-molecule real-time (SMRT) sequencing (Pacific Biosciences) to develop high-quality, fully assembled reference genomes (15 (11), USVAST isolates were from other states in the United States (10), and "Blood" and "UPEC" isolates were from Seattle (5). b Antibiotic resistance phenotypes are shown for all strains other than those from Seattle, with the ResFinder-identified gene(s) (20) presumed responsible shown in parentheses. For Seattle isolates, only the genes identified are shown, as phenotypes were not independently confirmed in this study. Antibiotic abbreviations: Amp, ampicillin; Azm, azithromycin; Cef, cephalothin; Chl, chloramphenicol; Gen, gentamicin; Nor, norfloxacin; Str, streptomycin; Tet, tetracycline; Sul, sulfamethoxazole; Tmp, trimethoprim. c Plasmid replicons were predicted with PlasmidFinder (25). d The number of small plasmids was determined from sequence assemblies, as described in Materials and Methods. Seattle isolates are listed as "ND" (no data), as these genomes were downloaded as assembled contigs without topology or coverage data, which precluded identification of small plasmids. e Prediction of the presence of pUTI89 is described in Materials and Methods. f The presence of resistance genes is discordant with phenotypes reported in reference 5: UPEC-007 (reported as Tet r ), UPEC-098 (reported as Amp r Sul r Tmp r ), UPEC-124 (reported as Amp r Sul r Tmp r ), UPEC-131 (reported as Amp r Cef r Tet r ), UPEC-144 (not reported as Sul r Tmp r ), UPEC-157 (reported as Amp r ), UPEC-185 (not reported as Amp r Sul r Tmp r ).
SMRT sequencing allowed identification of DNA methylation patterns (19). All four genomes displayed adenine methylation (m6A) at GATC sites, attributable to Dam DNA methyltransferase (see Table S2 in the supplemental material). The SF-468 and SF-173 isolates shared an additional m6A motif, as did the SF-166 and SF-088 isolates. SF-088 showed m6A methylation at two additional sites. All motifs showing m6A methylation were nearly 100% methylated. The SF-166 genome was the only one to show cytosine (m4C) methylation, at roughly one out of eight RCCGGY sequences. The SF-166 genome contains three annotated DNA-cytosine DNA-methyltransferase genes: a homolog of dcm (located at~2.1 Mb) that was present in all of the other ST95 genomes sequenced, a component of a restriction-modification system (located at 4.12 Mb) found in all fimH6 and fimH9 isolates, and a 1.2-kb gene located at 4.16 Mb in SF-166 that was universal in fimH6 isolates but absent in the non-fimH6 ST95 genomes examined. The latter gene is hypothesized to be responsible for the unique m4C methylation observed in SF-166, but this was not confirmed experimentally.
Antibiotic resistance. Thirty-four (40%) of the 86 genomes we examined contained at least one resistance gene, as identified by ResFinder (20). Resistance genes were detected for seven classes of antimicrobial agents ( Table 2). For the 53 isolates we The resulting genome sequences were aligned using progressiveMauve 2.4.0 (38), and an unrooted tree based on the alignments was generated using Archaeopteryx 0.9920. Isolate names are shown to the right, along with fimH type and whether the isolate also contains a pUTI89* plasmid. sequenced and assessed for resistance, predictions by ResFinder correlated well with observed phenotypes. For the Seattle isolates, intact resistance genes were assumed to confer the expected phenotype. Resistance genes were not evenly distributed among fimH sublineages: only 8 (20%) of 41 fimH6 isolates contained one or more acquired resistance genes, versus 1 (33%) of 3 fimH9 isolates, 7 (47%) of 15 fimH47 isolates, and 18 (67%) of 27 fimH1 isolates (for fimH6 versus non-fimH6 isolates, P Ͻ 0.001). With respect to geographic origin, the Seattle ST95 isolates were significantly less likely to contain genes for ␤-lactam and tetracycline resistance (P ϭ 0.01 and 0.02, respectively) than the San Francisco isolates, a result attributable to the non-fimH6 component of the population.
The aminopenicillin resistance gene bla TEM-1 , found in 38% of isolates, was the most common acquired resistance gene. Most bla TEM-1 genes were located in the context of Tn3 mobile elements. The only extended-spectrum ␤-lactamase (ESBL) gene identified, bla CTX-M-14 , was on plasmid pSF-468-2 in strain SF-468, which also contained bla TEM-1 on a separate plasmid (pSF-468-1). Among the 33 ampicillin-resistant isolates, 15 were resistant to at least three classes of antibiotics. The MDR isolates accounted for nearly all detected genes conferring resistance to aminoglycosides, tetracyclines, and sulfamethoxazole/trimethoprim (Table 1). Sulfonamide and trimethoprim resistance genes were typically coresident (occurring together in 12 of 15 isolates with either type of gene) and in close proximity on the same contig. One strain (SF-264) was also resistant to fluoroquinolones due to the presence of typical chromosomal mutations (in gyrA, Ser83-to-Leu and Asp87-to-Asn; in parC, Ser80-to-Ile) encountered in fluoroquinoloneresistant clinical E. coli isolates (21).
Plasmid content of ST95 genomes. Each of the SMRT-sequenced and fully assembled ST95 genomes contained at least one circular plasmid larger than 90 kb. These large plasmids all contained one or more known Inc replicons (Table S1), and with one exception (pSF-166-1), all of the large plasmids contained antibiotic resistance genes. Furthermore, all of the resistance genes in the three sequenced MDR isolates were contained on large plasmids. pSF-468-1 carried bla TEM-1 in a Tn3-type transposon adjacent to a 17-kb region containing genes for resistance to streptomycin (strAB), sulfonamides (sul2), trimethoprim (dfrA17), and tetracycline (tetA). pSF-088-1 carried bla TEM-1 , sul2, dfrA5, strA, and strB in an 11-kb region, while pSF-173-1 included dfrA12, aadA2, sul1, and mphA in a 10-kb region. The large IncF plasmids (pSF-468-1, pSF-088-1, pSF-173-1, and pSF-166-1) shared roughly 40 kb of sequence encoding conserved replication and conjugation functions. pSF-468-1 also shared with pSF-088-1 another 45 kb of sequence that included genes encoding potential virulence-associated factors, such the Sit manganese-iron transport system and genes for aerobactin siderophore production and uptake. Virulence factors in the ST95 genomes will be reported in greater detail elsewhere.
pSF-166-1, the 114-kb plasmid from pansusceptible strain SF-166, was the only large plasmid not encoding antibiotic resistance and the only one with a full-length match (Ͼ98% identity, 100% coverage) to previously described plasmids: pUTI89 from uro-  (26) and a rat model for neonatal meningitis (24). For practical purposes, we will refer to this group of plasmids and the close relatives identified in this work as pUTI89* unless greater specificity is needed. Examination of our draft genomes showed that pUTI89* was common. When pUTI89 was used as the scaffold for reference-guided assemblies of reads from all 53 isolates we sequenced, a complete version of pUTI89* (defined as Ͼ98% of pUTI89 assembled) was likely present in 16 isolates-14 from San Francisco, one from Sacramento, and one from Minnesota ( Because most pUTI89*-containing isolates were also in the fimH6 sublineage, and fimH6 isolates were also less likely to be resistant than non-fimH6 isolates, we examined whether these traits were independently associated with resistance frequency in this population of isolates (Table 3). Among fimH6 isolates, resistance genes were less common among isolates with pUTI89* than among those without pUTI89* (2/21 [10%] versus 6/20 [30%]), but the difference was not significant (P ϭ 0.10). Among non-fimH6 isolates, pUTI89* was also associated with reduced resistance frequency (0/6 resistant isolates containing pUTI89 versus 26/39 resistant isolates lacking pUTI89; P ϭ 0.002). Comparing all isolates lacking pUTI89*, fimH6 isolates were still less likely to be resistant than non-fimH6 isolates (6/20 [70%] versus 26/39 [33%]; P ϭ 0.007). These data suggest that both the fimH genotype (specifically fimH6) and the presence of pUTI89* may be associated with reduced carriage of antibiotic resistance genes.
The published sequences of pUTI89 and pRS218 differ by only 16 single nucleotide polymorphisms (SNPs). Alignment of the assembled sequences from this study showed that pSF-166-1 differs from pUTI89 by 32 SNPs and from pRS218 by 40 SNPs. Many of the plasmids from San Francisco isolates were more closely related, including pSF-403-1, pSF-420-1, and pSF-440-1 (7 SNPs), which were obtained from separate patients in late 2009 and early 2010, and pSF-560-1 and pSF-567-1 (6 SNPs), obtained from separate patients in October 2010. Given the expected rate of false-positive SNPs in assemblies from MiSeq data (1 ϫ 10 Ϫ4 ) (27) at comparable coverage, the plasmids in these isolates may in fact be identical.
Although large plasmids were assembled readily from SMRT sequencing data, they rarely assembled as circular contigs from shorter MiSeq reads, so it was not possible to reliably assess the presence of such plasmids in draft genomes. As an alternative, we used PlasmidFinder (28) to identify replicons from various incompatibility (Inc) groups in draft genome sequences. A total of 190 IncF (IA, IB, IC, or II), IncB, IncI, IncP, or IncQ plasmid replicons were detected in the 86 ST95 genomes examined (mean, 2.2 per Because in the draft genomes most large plasmids did not assemble as single contigs, whether antibiotic resistance genes were plasmid associated could not always be determined definitively. However, 35 (64%) of 55 identified antibiotic resistance genes were located on the same contigs as Inc replicons or on pAnkS (see below), suggesting that most resistance genes were plasmid-borne, similar to the pattern observed in the fully assembled genomes.
Genomes of pansusceptible ST95 isolates were less likely to contain plasmids than those of resistant strains: the mean number of Inc replicons was lower for isolates lacking resistance genes (mean, 1.9) than for those containing one or more resistance genes (mean, 2.7) (P Ͻ 0.001 by Student's t test). In the genomes we sequenced, the number of small plasmids was also lower in isolates lacking antibiotic resistance (mean of 0.9 small plasmids/isolate) (P ϭ 0.002). Included among the 52 pansusceptible strains were five that lacked Inc replicons and small plasmids altogether.
Small plasmids in ST95 genomes. Of the 53 ST95 genomes we sequenced, 26 (49%) contained small plasmids (Table 1), appearing in the draft genome assemblies as circular contigs of less than 10 kb (see Materials and Methods). PlasmidFinder identified some of these as containing replicons of colicin-producing plasmids, but most were not flagged by PlasmidFinder and lacked the Inc replicons enumerated above. Verification that these contigs represented small plasmids was demonstrated by isolation of plasmid DNA in the laboratory (see Fig. S2 in the supplemental material). fimH1 isolates were much more likely to contain at least one small plasmid (17/18 [94%]) than fimH6 isolates (6/26 [23%]) (chi-square test, P Ͻ 0.001). fimH1 isolates also contained more small plasmids per isolate (3.8/isolate; SEM, 0.4; range, 2 to 7) than fimH6 isolates (mean, 0.5; SEM, 0.2; range, 0 to 3). Isolates containing pUTI89* (many of which are fimH6 strains) also contained significantly fewer small plasmids than isolates lacking pUTI89* (Table 4). The vast majority of small plasmids identified had no antibiotic resistance genes or known virulence factor genes. An exception was an 8.3-kb plasmid encoding ampicillin resistance that was present in six fimH1 isolates (SF-94, -149, -305, -457, -522, and -626), isolated over a 4-year period (2007 to 2011). This plasmid and the accompanying ␤-lactam resistance could be moved into laboratory E. coli strains by transformation, allowing it to be separated from other small plasmids present in the clinical isolates (Fig. S2). All ST95 isolates containing the 8.3-kb plasmid showed an elevated cephalothin MIC (32 g/ml versus 12 g/ml for other strains containing bla TEM-1 ), perhaps due to an increased bla TEM-1 gene dosage relative to larger, lower-copy-number plasmids in other ST95 strains.

DISCUSSION
The work presented here explores genomic diversity and antibiotic resistance in the E. coli ST95 lineage. Our results suggest that most antibiotic resistance in ST95 strains is associated with large conjugal plasmids. In the SMRT-sequenced SF-088, SF-173, and SF-468 strains, all resistance genes were borne on large multireplicon plasmids. This likely was true also for isolates sequenced to the draft level, since in them most resistance genes were found on contigs containing IncF replicons, as was the case for resistance plasmids in the SMRT-sequenced strains. Intriguing exceptions included six San Francisco isolates containing an 8.3-kb plasmid nearly identical to pAnkS, previously reported in Salmonella enterica isolates in Turkey (29) and Uruguay (30). None of the non-San Francisco isolates examined contained a pAnkS-like plasmid, but the temporal and geographic distribution of isolates was far from comprehensive. The pAnkS-containing strains were all isolated between 2007 and 2011; determination of whether this plasmid is still contributing to aminopenicillin resistance in San Francisco or elsewhere warrants further effort.
We also sought to shed light on why ST95 strains are less frequently antibiotic resistant than many other major ExPEC lineages. In previous work, among ST95 clinical isolates from San Francisco the fimH6 sublineage was most closely associated with pansusceptibility (6). The present analysis of an expanded set of isolates confirmed that fimH6 isolates were significantly less likely than other ST95 isolates to contain antibiotic resistance genes. Although antibiotic resistance in ST95 isolates derived largely from acquired, plasmid-borne genes, it is conceivable that both chromosomal and plasmid composition could impact the likelihood of acquiring resistance. Indeed, analysis of data in Table 3 suggests that both the fimH6 genotype and pUTI89* are associated with a reduced likelihood of resistance.
Chromosomal elements in fimH6 strains such as CRISPR or restriction-modification systems might restrict entry of new plasmids. The ST95 genomes examined here lacked CRISPR1 and contained only a rudimentary CRISPR2, as is common in the E. coli B2 group that includes ST95 (18). The CRISPR3 and CRISPR4 loci varied somewhat within and between the fimH lineages within ST95. No correlation was observed between the number of spacers or the sequences of spacers in ST95 isolates and their resistance to antibiotics or the presence of particular plasmids. Thus, there is currently no evidence to suggest that CRISPR-mediated effects on DNA acquisition could account for the patterns of antibiotic resistance observed in this work.
Multiple restriction-modification systems are found in each of the completely sequenced ST95 isolates according to REBASE (31). The genome of pansusceptible isolate SF-166 contained two fimH6-specific chromosomal regions: an~12-kb segment at 4.11 Mb that includes a homolog of the Eco31 restriction-modification system (including separate adenine-and cytosine-specific DNA methyltransferases) and a 31-kb segment at 4.15 Mb that includes a DNA-cytosine methyltransferase. The pUTI89* plasmid in SF-166 is also annotated to encode a putative DNA methyltransferase of unknown specificity. Analysis of genome methylation patterns showed no adenine methylation unique to SF-166, but this was the only isolate exhibiting cytosine methylation, and a chromosomally encoded putative cytosine DNA-methyltransferase unique to the fimH6 lineage was identified. Whether cytosine methylation in fimH6 strains has any role in limiting plasmid entry, and thereby affecting acquisition of antibiotic resistance, is an open question warranting further study.
Many ST95 isolates, particularly within the fimH6 sublineage, contained a large multireplicon plasmid nearly identical to pUTI89, designated here generically as pUTI89*. Our data showed that the presence of pUTI89* was correlated with a lower likelihood of antibiotic resistance (Table 3) and a reduced content of other plasmids ( Table 4). The preferential association of pUTI89* with the fimH6 sublineage seems likely to at least partially explain why pansusceptibility is so common in this group (6).
We hypothesize that the pUTI89* plasmid (IncF) inhibits the acquisition or maintenance of other plasmids in the same cell. Incompatibility due to replicon competition could reduce the pool of resistance-bearing plasmids able to establish themselves in cells already containing the multireplicon pUTI89*. For example, in the ST95 strains we examined, bla TEM-1 was usually located on contigs containing IncF replicons, and such plasmids would be incompatible with a resident pUTI89*. Resistance genes also could be introduced on plasmids from incompatibility groups other than IncF: e.g., the bla TEM-1 -carrying IncB plasmid in pUTI89*-containing isolate SF-335. However, strains containing pUTI89* also were also less likely to harbor plasmids from non-IncF incompatibility groups (Table 4). Thus, factors other than replicon competition may be operative and should be tested experimentally.
Several pUTI89-like plasmids were known previously from the literature. Plasmid pRS218 is derived from E. coli strain RS218, which was originally isolated from a neonatal meningitis case in San Francisco in 1974 (32), roughly three decades earlier and thousands of miles separated from the UPEC strains bearing the closely related pUTI89 and pEC14_114 plasmids, which were isolated in St. Louis, MO, and St. Paul, MN, respectively. Both DebRoy et al. (25) and Cusumano et al. (26) have presented PCRbased evidence that pUTI89-like plasmids are common in UTI isolates. pUTI89 and the closely related pRS218 have also been shown to have roles in pathogenesis. A UTI89 derivative cured of pUTI89 was impaired in a mouse UTI virulence model (26), and an RS218 derivative cured of pRS218 was impaired in a rat pup model of neonatal meningitis (23). Genes on pUTI89* that could be involved in pathogenesis include those encoding a potential cytotoxin (senB) and systems possibly involved in iron uptake (cjr) (26). More detailed dissection of this plasmid has not yet been done due to difficulties working with it. pUTI89 was found to be highly stable during laboratory culture, and presumably is "in the wild" as well, due at least in part to a stabilization system related to the ParM ATPase (33) that prevents plasmid loss (26).
In summary, genomic analysis of ST95 ExPEC strains revealed an extensive plasmidome accounting for acquired antibiotic resistance and an intriguing phenomenon in which a large plasmid (pUTI89), perhaps in collaboration with chromosomal genes, may inhibit a subset of ST95 strains from acquiring other plasmids and associated antibiotic resistance. A deeper understanding of these pansusceptible strains could potentially be exploited to devise biological strategies to combat drug-resistant Gram-negative bacterial infections.
Genomic DNA sequencing. Genomic DNA was prepared from all strains using the Qiagen blood and tissue DNeasy kit. Library preparation and sequencing using single-molecule real-time (SMRT) sequencing technology (Pacific Biosciences) have been described elsewhere (15). Analysis of DNA methylation patterns in the SMRT data (19) was done using analysis tools provided by the manufacturer. Library preparation for the MiSeq platform followed a standard protocol for Illumina-compatible libraries (Wafergen Biosystems). Samples were fragmented by a Covaris S220 ultrasonicator to generate an average insert size of 800 bp. After appropriate fragmentation was verified by an Agilent Bioanalyzer, samples were loaded on the Wafergen Apollo 324 NGS Library Prep system. Wafergen PrepX library kits were used for end repair, A-tail addition, adapter ligation, and size selection using AMPure XP beads. Sample concentration was quantified with a Qubit fluorometer. Libraries were PCR amplified to incorporate index tags and flow cell-binding regions. Final libraries were quantified by Qubit, Bioanalzyer, and quantitative PCR (qPCR) and then sequenced via a 300-bp paired-end run on a MiSeq instrument using V3 chemistry and standard Illumina analysis software.
Bioinformatic analysis. For draft genomes, MiSeq reads were screened and trimmed based on length and quality using BBDUK within the Geneious software package, version R9 (Biomatters, Ltd.). The trimming process also removed any residual adapter sequences. Trimmed paired reads (4 ϫ 10 5 to 1.2 ϫ 10 6 ) were assembled de novo into contigs with the Geneious assembler. The number of reads used in each assembly was sufficient to give a minimum of 25-fold coverage, averaged across all contigs. (A maximum of 45-fold coverage was used.) Contigs that were Ͻ1 kb in length, with Ͻ2-fold read coverage, or with Ͻ80% high-quality base calls were eliminated from subsequent analysis. Annotation of contigs for the analysis presented here was done by the RAST server (34), but sequences submitted to GenBank were annotated by the NCBI prokaryotic genome annotation pipeline. Previously assembled genomes of ST95 ExPEC isolates from the Seattle area (5) were downloaded from NCBI for analysis. Prediction of acquired antibiotic resistance genes employed ResFinder v.2.1 (20). Identification of potential plasmid replicons was done with PlasmidFinder v.1.3 (28). Identification of insertion elements was done with ISFinder (35), and identification of lysogenic prophage used PHAST (36).
To determine whether a pUTI89-like plasmid was present in ST95 isolates from Seattle (5), pUTI89 was used to query the archived draft genomes using BLAST. Contigs or fragments greater than 500 bp (allowing for the possibility of misassembly) that were Ͼ98% identical to the pUTI89 sequence were aggregated. If Ͼ98% of the total 114-kb pUTI89 sequence could be generated by combining nonoverlapping contigs or fragments from an isolate, we considered that isolate to be positive for the pUTI89 plasmid.
Plasmids less than 10 kb were not well represented in the long-read data, as the template DNA had been size selected to be greater than 5 kb. These plasmids emerged from de novo assembly of 300-base MiSeq reads. Contigs were designated "small plasmids," analogous to the "small cryptic plasmids" identified in ESBL-producing E. coli strains by Brolund et al. (37), if (i) they assembled de novo as circular sequences, (ii) they showed read coverage at least 50% higher than the mean coverage of the top five chromosomal contigs in the same assembly, (iii) they were Ͻ10 kb in size, and (iv) BLAST searches of GenBank identified primarily known plasmids, and the contig used as the query covered 80% or more of the plasmid sequence in GenBank. (The Geneious de novo assembler can produce a circular contig if the reads on each end of a linearly assembled contig match and those reads do not intersect with each other anywhere else in the contig.) The archived genomes of the Seattle isolates could not be used for identification of small plasmids through this approach, as topology and read coverage data were not available for these assemblies.
Statistical analysis. Comparison of population frequencies (e.g., frequency of pansusceptibility) and calculated P values used the chi-square test, with a significance cutoff of P ϭ 0.05. Comparison of population means (e.g., mean number of small plasmids per isolate) used Students t test, with a significance cutoff of P ϭ 0.05.
Plasmid DNA isolation and transformation. Plasmid DNA was isolated from E. coli cultures using the ZR Plasmid Miniprep Classic kit (Zymo Research). Plasmid DNA was analyzed on 1% agarose gels. DNA was transformed in chemically competent OneShot E. coli TOP10 cells (Invitrogen) by the manufacturer's protocol, and colonies were selected on LB agar plus ampicillin (50 g/ml).
Accession number(s). Complete genome sequences have been deposited in DDBJ/EMBL/GenBank under the following accession numbers: SF-468 chromosome and plasmids, CP012625 to CP012630; SF-173 chromosome and plasmid, CP012631 and CP012632; SF-166 chromosome and plasmid, CP012633 and CP012634; and SF-088 chromosome and plasmids, CP012635 to CP012638. The whole-genome shotgun sequencing projects described here have been deposited in DDBJ/ENA/GenBank under the accession numbers shown in Table S3.