We collected nonduplicate Enterobacter strains recovered from blood cultures during January 2016–June 2018 at West China Hospital of Sichuan University (Appendix 2 Table 1). West China Hospital is a 5,000-bed major referral hospital in western China. Initial species identification and in vitro susceptibility testing were performed by using Vitek II. We determined MICs of colistin by using the broth microdilution method of the Clinical and Laboratory Standards Institute (CLSI) and interpreted susceptibility following CLSI guidelines (11). For colistin and tigecycline, no CLSI breakpoints are available, so we used breakpoint standards defined by the European Committee on Antimicrobial Susceptibility Testing (https://www.eucast.org). Multidrug resistance was defined based on the criteria for Enterobacteriaceae (12).

West China Hospital has a comprehensive hospital information system, which allowed us to retrieve patient data including age, sex, length of hospitalization, and clinical outcomes (death or discharge) from electronic medical records. One patient (with strain 090040) had an unusually long hospital stay (578 d) because of a medical dispute and was removed from our analysis of length of stay. According to social customs in China, dying at home is preferred over the hospital; it is likely many patients chose to stop treatment and return home if treatment was not working and patients felt death was imminent. We categorized patients who chose to be discharged but were likely to die in the hospital (judged by the consensus of 2 physicians reviewing blind data) as patients with predicted death. BSI, the type of BSI (primary or secondary to infection of other sites), central line–associated BSI (CLABSI), and healthcare-associated infection were determined by using criteria established by the Centers for Disease Control and Prevention’s National Healthcare Safety Network (13,14). We conducted the study in accordance with the amended Declaration of Helsinki. The Ethics Committee of West China Hospital approved the study and waived informed consent.

All strains underwent whole-genome sequencing by using the HiSeq X10 platform (Illumina, https://www.illumina.com). Genomic DNA was prepared by using the QIAamp DNA mini kit (QIAGEN, https://www.qiagen.com). We used Unicycler version 0.4.3 (15), in the conservative mode for increased accuracy, to perform a de novo hybrid assembly. Precise species identification was established by determining the pairwise isDDH between the genome sequence of the query strain and those of type strains of Enterobacter spp., including the validly published species and the species awaiting validation (Appendix 1 Table 1). This process was performed by using the Genome-to-Genome Distance Calculator, formula 2 (16). A >70% cutoff was applied to define a species. In addition, we determined the pairwise ANI of the genome sequence of the query strain and those of type strains of Enterobacter spp. (Appendix 1 Table 1) by using JSpecies software (https://imedea.uib-csic.es/jspecies) with a >96% ANI cutoff to define a species (10). Sequence types (STs) were determined by using the genomic sequences to query the multilocus sequence typing database of E. cloacae (https://pubmlst.org/ecloacae). Antimicrobial resistance genes were identified from genome sequences by using the ABRicate program (https://github.com/tseemann/abricate) to query the ResFinder database (https://genomicepidemiology.org).

The genome sequences of all Enterobacter strains recovered from human blood (n = 349, Appendix 2 Table 2) were retrieved from GenBank (accessed 2018 Nov 1). These Enterobacter genomes were subjected to precise species identification by using the Genome-to-Genome Distance Calculator as described.

We performed single-nucleotide polymorphism (SNP) calling for genome sequences to untangle the clonal relatedness of ST171 strains and to investigate whether the ST171 strains in this study are clonally related to strains recovered elsewhere. All genome sequences of ST171 Enterobacter strains (n = 102), regardless of types of the host source (human, nonhuman, or unknown) and source (blood, nonblood, or unknown), were retrieved from GenBank. We used the complete chromosome sequence of ST171 strain 34798 (GenBank accession no. CP012165), which was recovered from a bile sample in the United States in 2011, as our reference for mapping. Genome sequences of the strains were mapped against the reference genome by using Snippy version 4.3.6 (https://github.com/tseemann/snippy) at default settings. The resulting core SNPs (n = 1,918) were concatenated and used to infer a phylogenomic tree by using RAxML version 8.2.12 (17) under the general time-reversible plus gamma model with a 1,000-bootstrap test.

We determined plasmid replicons of all ST171 strains in this study by using PlasmidFinder version 2.0 (https://cge.cbs.dtu.dk/services/PlasmidFinder). The first bla_NDM-5–harboring ST171 strain (090011) in this study and the only bla_NDM-1–harboring ST171 strain (045001) were selected for whole-genome sequencing by using the long-read MinION sequencer (Nanopore, https://nanoporetech.com) to obtain complete chromosomal and plasmid sequences. De novo hybrid assembly of short Illumina reads and long MinION reads was performed by using Unicycler version 0.4.3 (15) in conservative mode for increased accuracy. Complete circular contigs were then corrected by using Pilon version 1.22 (18) with Illumina reads for several rounds until no further improvements were reported. Short reads of the remaining four bla_NDM-5–harboring ST171 strains (090022, 090023, 090055, and 090059) were mapped against the bla_NDM-5-carrying plasmid (designated pNDM5_090011) of strain 090011 by using BWA version 0.7.17 (H. Ling, unpub. data, https://arxiv.org/abs/1303.3997) at default settings.

Continuous variables were presented as median and interquartile range and were compared by using rank-sum test. We used the Pearson χ² test, Yates correction for continuity, or Fisher exact test to compare disparities between different groups for categorical variables. Pearson χ² test was used when sample size (n) was >40 and theoretical frequency (T) >5, Yates correction for continuity when n>40 and 1<T<5, and Fisher exact test when n<40 or T<1. We used SPSS Statistics 21.0 (IBM Inc., https://www.ibm.com) to perform statistical analyses. All p values were 2-tailed, and p<0.05 was considered statistically significant. We used PASS version 11.0 (NCSS, https://www.ncss.com) to calculate statistical power after using the Wilcoxon test to conduct nonparametric adjustment.

Draft genome sequences of the strains in this study were deposited into GenBank (accession numbers in Appendix 2 Table 1). The complete genome sequence of strain 090011 was deposited into GenBank under accession nos. CP036310–2. and the sequence for strain 045001 was deposited under accession nos. CP043382–5.

A total of 48 nonduplicated Enterobacter strains were recovered from blood during our 2.5-year study period and were collected for study (Table 1; Appendix 2 Table 1). Whole-genome sequencing results, isDDH, and ANI values of these strains are summarized in Appendix 2 Table 1. The 48 strains were identified as E. cloacae (n = 42), E. asburiae (n = 3), and E. kobei (n = 3) by Vitek II. However, precise species identification on the basis of isDDH revealed that the most common species was actually E. xiangfangensis (n = 21) (Table 2); the next most common was an Enterobacter sp. (n = 8) that has no assigned species name but was previously known as Enterobacter cluster III, as defined by Hoffman et al. (6). This species is most closely related to E. xiangfangensis with a 66.6% isDDH value and is temporarily assigned taxon 1 here (Appendix 2 Table 2). The remaining strains were assigned to 10 Enterobacter species: E. bugandensis (n = 4), E. cloacae (n = 3), E. asburiae (n = 2), E. hormaechei (n = 2), E. huaxiensis (n = 2), E. roggenkampii (n = 2), E. chuandaensis (n = 1), E. ludwigii (n = 1), E. sichuanensis (n = 1), and an unnamed Enterobacter sp. (n = 1) (Table 2). The unnamed species is most closely related to E. roggenkampii with a 65.4% isDDH value (Appendix 1 Table 2) and is temporarily assigned taxon 2 here.

Of 349 Enterobacter strains in GenBank that were recovered from human blood, most (221, 63.3%) are also E. xiangfangensis; taxon 1 is the second most common species with 69 strains (19.8%) (Table 2). The remaining 59 strains (16.9%) belong to 14 species: E. asburiae (n = 14 [4.0%]), E. kobei (n = 10 [2.9%]), E. bugandensis (n = 7 [2.0%]), E. roggenkampii (n = 7 [2.0%]), E. ludwigii (n = 6 [1.7%]), E. cloacae (n = 4 [1.1%]), taxon 1 (n = 2), E. chengduensis (n = 1), E. mori (n = 1), E. sichuanensis (n = 1), and 4 species without assigned species names (n = 1 or 2 for each species, 6 in total; Table 2). The 4 unnamed species were assigned taxon 3–6 (Table 2); the closest species of taxon 3–6 are listed in Appendix 1 Table 2 and are also shown in a phylogenomic tree in Appendix 1 Figure 1. E. xiangfangensis and taxon 1 are closely related as shown by their phylogenetic position in the phylogenomic tree of Enterobacter spp. and by their common 66.6% isDDH value (close to the 70% cutoff to define a species). We therefore combined the 2 species in the following analysis.

Most of the 48 BSI cases (n = 36, 75%) were primary BSIs, including 16 cases of CLABSI. BSIs in the remaining 12 cases were secondary; original sources were intraabdominal infection (n = 5), cholangitis (n = 3), urinary tract infection (n = 2), wound infection (n = 1), and gastrointestinal tract infection (n = 1). Most (n = 41, 85.4%) BSIs caused by Enterobacter spp. were healthcare-associated infections. E. xiangfangensis and taxon 1 were more likely to cause primary BSI (82.8% vs. 63.2%), CLABSI (27.9% vs. 17.2%), and healthcare-associated BSI (93.1% vs. 73.7%) than were other Enterobacter spp. However, the differences were not statistically significant (Table 3).

Two patients who had Enterobacter BSIs (1 E. xiangfangensis and 1 taxon 1) died in the hospital. In addition, 7 patients with Enterobacter BSIs (4 E. xiangfangensis, 2 E. bugandensis, and 1 E. sichuanensis) did not respond to treatment and were discharged in critical condition. These 7 case-patients were categorized as patients in whom death was predicted. The death rate for Enterobacter BSI was 18.8% (9/48); the death rate (20.7% [6/29]) of BSI caused by E. xiangfangensis or taxon 1 was not statistically different (15.8% [3/19] p>0.05) (Table 3) from that of BSI caused by other Enterobacter spp. Of note, 2 of the 4 patients with E. bugandensis–caused BSI had poor outcomes (predicted death) (Table 1). BSIs caused by E. xiangfangensis or taxon 1 were more common in younger patients and resulted in lengthier overall hospitalizations (median 33 vs. 19.5 d; p>0.05) (Table 3) than BSIs caused by other Enterobacter spp. This difference was largely because of the prolonged length of stay (median 11 d vs. 4.5 d; p>0.05) before the episode of BSI.

The antimicrobial susceptibility and antimicrobial resistance gene repertoire of the 48 Enterobacter strains are shown in Appendix 2 Table 1. E. xiangfangensis and taxon 1 had substantially higher rates of resistance to aztreonam (48.3% vs. 10.5%), cefepime (41.4% vs. 10.5%), ceftriaxone (58.6% vs. 15.8%), piperacillin/tazobactam (41.4% vs. 10.5%), and tobramycin (44.8% vs. 5.3%) (Table 4) and were substantially more likely to be multidrug resistant (55.2% vs. 10.6%) (Table 3). There were 10 carbapenem-resistant strains (8 E. xiangfangensis, 1 E. bugandensis, and 1 E. cloacae), all of which carried a bla_NDM gene (bla_NDM-5, n = 7; bla_NDM-1, n = 3) (Table 1; Appendix 2 Table 1), and they belonged to 5 STs (ST171 [n = 6], ST718 [n = 1], ST1 [n = 1], ST418 [n = 1], and a new ST [n = 1]). No carbapenemase genes were identified in carbapenem-susceptible strains.

The 48 Enterobacter strains belonged to 37 STs (Table 1; Appendix 1 Table 3) but only ST171 (E. xiangfangensis) and ST78 (taxon 1) contained >3 strains (6 for ST171 and 3 for ST78). We performed analysis of clonal relatedness based on SNPs for ST78 and ST171 strains. In the 3 ST78 strains, there were 306 to 1,052 SNPs difference, suggesting no recent shared origins (Appendix 1 Table 4). The 6 ST171 strains were all resistant to carbapenems and carried bla_NDM-5 (n = 5) or bla_NDM-1 (n = 1). Among the 6 ST171 strains, 4 (all carrying bla_NDM-5) had 0–2 SNPs difference (Table 5) and were recovered from patients in the same ward (cardiac surgery). One patient infected with strain 090011 died but the remaining 3 patients recovered.

The remaining 2 ST171 strains, 045001 (carrying bla_NDM-1) and 090055 (carrying bla_NDM-5), were 81–82 SNPs different from the 4 previously mentioned strains and were 38 SNPs different from each other. These 2 strains were recovered from patients in 2 different wards (medical and respiratory intensive care units). The 6 ST171 strains isolated in our study formed a phylogenetic cluster with strain CCBH10892, which was isolated from a rectal swab sample in Brazil in 2012 (GenBank accession no. JSBO00000000), and strain EC_849, which was isolated from a sputum sample in South Africa in 2012 (GenBank accession no. LRIZ00000000) (Appendix 1 Figure 2). The cluster contained 98 to 107 SNPs difference (Table 5; Appendix 1 Figure 2). Of note, both CCBH10892 and EC_849 carried bla_NDM-1. By contrast, the 6 strains were >300 SNPs different from other ST171 strains with genome sequences deposited in GenBank (Appendix 2 Table 3).

The complete genome sequences of bla_NDM-5–harboring strain 090011 and the bla_NDM-1–harboring strain 045001 were obtained. Strain 090011 had a 4.64-Mb circular chromosome and 2 plasmids (a 102.5-kb plasmid containing IncFIA, IncFIB, and IncR replicons and a 46.1-kb plasmid containing an IncX3 replicon) (Appendix 1 Table 5). The bla_NDM-5 gene in strain 090011 was carried on the 46.1-kb IncX3 plasmid, designated pNDM5_090011. The short reads of the remaining 4 bla_NDM-5–harboring strains were then mapped against pNDM5_0900117. The 4 strains had contigs showing 100% coverage and 100% identity with pNDM5_090011, suggesting a common plasmid in all 5 bla_NDM-5–harboring ST171 strains in our study. Strain 045001 had a 4.70-Mb circular chromosome and 3 plasmids (an 85.7-kb IncFII plasmid, a 78.2-kb plasmid, and a 2.5-kb plasmid) (Appendix 1 Table 5). The replicon type of the latter 2 plasmids could not be determined by the current replicon-typing scheme. The bla_NDM-1 gene in strain 045001was carried on the 85.7-kb IncFII plasmid.