We collected 450 bacterial isolates (including 195 Enterobacteriaceae) in Pakistan during February 2012–March 2013 from Pakistan Railway General Hospital in Rawalpindi and the Pakistan Institute of Medical Sciences in Islamabad. From this collection, we randomly selected 55 Enterobacteriaceae isolates for whole-genome sequencing. We then selected 23 isolates from samples collected in the United States during January 2010–June 2013 from patients in Barnes Jewish Hospital in St. Louis, Missouri, that had similar proportions of β-lactam susceptibility and resistance to the isolates collected in Pakistan for sequencing. All isolates were de-identified and retrieved from existing strain banks. The combined set included 33 E. coli, 30 K. pneumoniae, 9 Enterobacter cloacae, and 6 Enterobacter aerogenes (Technical Appendix Table 1). We extracted plasmid DNA from 9 isolates encoding NDM-1, 11 isolates encoding KPC, and 3 isolates encoding CTX-M-15 and performed shotgun sequencing on those plasmid preparations. Detailed methods are described in the Technical Appendix.

Using antibiotic resistance gene predictions from the Resfams database (13) and core genome alignment, we constructed a phylogenetic tree for each species in our set, overlaid by the β-lactamases encoded by each isolate (Figure 1). Isolates from both locations were found to be members of the same subspecies clades (Technical Appendix Figure 1) and to contain similar repertoires of β-lactamases (Figure 1), indicating that geography is not a discriminating variable for these isolates. Many of these isolates were also MDR: resistance to ciprofloxacin, trimethoprim/sulfamethoxazole, gentamicin, doxycycline, and chloramphenicol occurred in 63%, 65%, 45%, 54%, and 56% of isolates, respectively. As expected from results of previous work (8), E. coli ST131 isolates had high rates of CTX-M carriage (82%; Figure 1, panel A) and ciprofloxacin resistance (100%).

The variety of strains that we discovered encoding KPC and NDM-1 is consistent with existing evidence that HGT is a major factor in their spread. All KPC genes were proximal to Tn4401 and all NDM-1 genes were carried on ISAba125, mobile elements with which each gene has respectively been previously associated (14). We observed multiple examples of NDM-1 within the K. pneumoniae ST11 clade (15) (Figure 1, panel B; Technical Appendix Figure 1, panel B), a close relative of ST258. This association could be caused by clonal expansion or multiple HGT events and emphasizes that lineages known to encode KPC are now also acquiring NDM-1. We also observed high rates of NDM-1 carriage in Enterobacter isolates (Figure 1, panels C and D), which in general showed a high number (maximum 8) and wide variety of β-lactamases. These isolates were also MDR: 57% of the Enterobacter isolates were resistant to all or all but 1 of the antimicrobial drugs tested. At best, these Enterobacter strains are a reservoir for resistance in Pakistan; at worst, they are the vanguard of an expansion of carbapenem-resistant Enterobacter infections.

Previous observations have predominantly found KPC and NDM-1 to be expressed from plasmids (6,11). To characterize the sequence similarity of plasmids within the NDM- and KPC-carrying plasmid populations, we purified and sequenced plasmid DNA from 9 isolates encoding NDM-1, 11 encoding KPC, and 3 encoding CTX-M-15. Sequencing showed that these plasmids include representatives from IncHI2, IncY, IncN, IncFIA, IncFIB, IncFIC, and IncI1 incompatibility groups. Using reciprocal BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi) alignment between each pair of plasmid preparations, we calculated the percentage of each plasmid shared using a 99% identity threshold. We performed this same analysis for all sequenced plasmids containing NDM-1, KPC, or CTX-M available in the National Center for Biotechnology Information database (http://www.ncbi.nlm.nih.gov) together with our set (Figure 2) and separately (Technical Appendix Figure 2). Certain components, primarily mobile elements, were abundant within these plasmids: the average plasmid shared 500 contiguous bases with 58 of the other plasmids; however, median BLAST identity for this pairwise comparison was <12%, even when considering plasmids with the same β-lactamase, suggesting that both carbapenemases exist within a variety of plasmid configurations.

To visualize this comparison of carbapenemase plasmids, we generated a network diagram in which each node represented a plasmid and each line represented shared sequence between 2 plasmids (Figure 2, panel B). Node size and line width correlate to the number of nucleotides contained in the plasmid or sharing interaction. This visualization shows the abundant small, shared regions that exist between most plasmid pairs, represented as thin background lines. This visualization also highlights the larger shared regions that indicate highly similar plasmids, represented by the few wide lines. These outliers were often between pairs of plasmids encoding the same β-lactamase but were also observed between NDM-1 and KPC containing plasmids (maximum 79% of smaller plasmid length).

Together, this evidence supports our hypothesis that strains and plasmids known to carry either carbapenemase also have access to the other. Given the similarity of carbapenemase-negative strains to those carrying KPC or NDM-1 and the high diversity of plasmids in which they can be found, we anticipate that global carbapenem usage will encourage HGT of both of these carbapenemases into additional strain and plasmid backgrounds. Because KPC and NDM-1 are poised to cross genetic and geographic boundaries, we recommend that hospitals routinely screen Enterobacteriaceae strains for both genes, even in regions where they are not yet endemic. We further advocate reduced carbapenem use to limit the selection for resistance against this vital antibiotic class.

Phylogenetic analysis, sequence conservation between plasmids, and antibiotic drug susceptibility profiles of clinical isolates from Pakistan and the United States.