Materials and Methods

Bacterial Isolates

We obtained ethics approval for this study through the University of Calgary Conjoint Health Research Ethics Board (approval no. REB17-1010). We included 229 clinical, nonrepeat E. coli isolates collected from 2 global surveillance programs (SMART and INFORM) during 2015–2017 (Appendix). Isolates had undergone identification and susceptibility testing using Clinical Laboratory and Standards Institute guidelines^[4,5,11]. Carbapenem nonsusceptible isolates underwent molecular screening for bla_KPC, bla_VIM, bla_NDM, bla_OXA-48-like, bla_IMP, and bla_GES, as described previously^[4,5]. Overall, we collected 87,182 Enterobacterales for the period 2015–2017 from 62 countries: 27,444 were identified as E. coli and 275 (1%) tested nonsusceptible to ≥1 of the carbapenems. Most (229 [83%]) were positive for either bla _KPC, bla _OXA-48-like, bla _NDM, bla _VIM, or bla _IMP and were included in this study. The remaining 46 were negative for bla _KPC, bla _VIM, bla _NDM, bla _OXA-48-like, bla _IMP, and bla _GES.

We defined major STs as representing >10% and minor STs as representing 5%–10% of the total E. coli carbapenemase population^[12]. Dominant STs were both major and minor STs.

Genomic Analysis

We subjected the carbapenemase-producing E. coli (n = 229) to short-read WGS by using NovoSeq (Illumina, https://www.illumina.com) with 151 × 2 paired-end reads^[13,14]. We obtained draft genomes by using SPAdes 3.15^[15]. We used BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi) to determine AMR genes, plasmid replicons, and virulence genes against the following databases or typing schemes: National Center for Biotechnology Information Bacterial Antimicrobial Resistance Reference Gene Database (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA313047), ResFinder^[16], PlasmidFinder^[17], multilocus sequence typing^[18], and virulence finder^[19]. We conducted multilocus sequence typing by using mlst 2.19 (https://github.com/tseemann/mlst). We identified ST410 and ST131 clades and subclades as described previously^[20,21].

For phylogenetic analyses, we mapped trimmed raw reads from each genome to a reference genome sequence (EC958 [GenBank accession no. HG941718] for ST131, JS316 [GenBank accession no. CP058618] for ST410, WCHEC005237 [GenBank accession no. CP026580] for ST167, and AR_0015 [GenBank accession no. CP024862] for ST405) by using snippy (https://github.com/tseemann/snippy). We filtered single-nucleotide polymorphisms (SNPs) among prophages, repeated sequences, or insertion sequences as previously described^[22], and we generated a maximum-likelihood phylogenetic tree inferred from the resulting SNP alignment by using RAxML 8.2.12 by using a general time-reversible model of nucleotide substitution and 4 discrete γ categories of rate heterogeneity^[23]. We identified phylogenetic clades by using hierarchical Bayesian analysis of the population structure in R by using RhierBAPS with 10 initial clusters at 2 clustering levels^[24]. We defined clades by using the first level of clustering and subclades at the second level of clustering^[25]. We annotated the phylogenetic trees in iTOL^[26]. We deposited all sequencing data in the National Center for Biotechnology Information database (BioProject PRJNA780590).

Statistical Analyses

We conducted all analyses in R 3.6.1^[27]. Initially, we attempted to fit generalized linear mixed models with country-level random effects to summarize comparisons between dominant STs with respect to antimicrobial and virulence genes. Most models failed to converge, possibly because of the low number of isolates for some STs and the large number of countries involved. Thereafter, we attempted to use exact logistic regression models for clustered data, as previously described^[28]. Similarly, most models failed to converge, particularly for comparisons involving ST1284 where all isolates were obtained from a single country. We then used Fisher exact tests to perform pairwise comparisons of antimicrobial and virulence genes among dominant STs. We used Mann-Whitney tests for comparison of virulence scores between dominant STs. We adjusted p values for multiple comparisons within each outcome by using the false discovery rate^[29]. We defined statistical significance as p≥0.05.