We described South Africa’s health system and the role of Ampath-MDRC (Appendix). We also summarized the study workflow (Appendix Figure 1).

We screened all carbapenem-nonsusceptible K. pneumoniae at Ampath-MDRC using a PCR for bla_NDM, bla_KPC, bla_OXA-48–like, bla_IMP, bla_VIM, and bla_GES. We performed pulsed-field gel electrophoresis (PFGE) on 471 K. pneumoniae isolates positive for OXA-48–like enzyme to determine if some of the isolates were genetically related. We identified 1 dominant pulsotype (pulsotype A) and a related pulsotype (AR) and selected isolates representing them from different geographic sites over various time periods. We used these isolates for initial Illumina short-read sequencing (n = 28) (Illumina, http://www.illumina.com) and PacBio long-read sequencing (n = 1) (Pacific Biosciences, http://www.pacb.com) to determine if they belong to the same sequence type and to identify the OXA-48–like enzyme.

We identified the pulsotypes as sequence type (ST) 307 containing OXA-181 on IncX3 plasmids. We then compared this whole-genome sequencing (WGS) data with the sequences from the US National Center for Biotechnology Information (NCBI) genome database (ftp://ftp.ncbi.nih.gov/genomes) to design PCR primers for the detection of ST307, IncX3, and OXA-181-IS3000 mobile genetic element (MGE). We used previously characterized K. pneumoniae STs with different plasmid replicons to verify these primers (Appendix Table 1). We screened K. pneumoniae producing OXA-48–like enzyme (n = 471) with PCR primers to identify ST307, IncX3 plasmids, and OXA-181-IS3000 MGE. We selected additional K. pneumoniae ST307 isolates (n = 60) from different geographic locations, time points, and specimens to undergo Illumina short-read WGS to elucidate the evolution of ST307 in South Africa.

During January 2014–December 2016, various Ampath regional clinical laboratories referred 1,247 unique clinical, carbapenem (i.e., ertapenem, meropenem, or imipenem) nonsusceptible K. pneumoniae isolates to Ampath-MDRC for PCR confirmation of carbapenemases (Appendix Figure 1). We performed PCR screening for bla_NDM, bla_KPC, bla_OXA-48–like, bla_IMP, bla_VIM, and bla_GES using LightMix modular carbapenemase kits (TIB Molbiol, https://www.tib-molbiol.com) on a LightCycler 480 II instrument (Roche Diagnostics, https://www.roche.com). Details on the identification and susceptibility testing of the bacterial isolates are provided in the Appendix.

We performed PFGE on the K. pneumoniae isolates with OXA-48–like enzymes (n = 471) to determine if there was a dominant pulsotype among them. The Appendix describes the methodology used and results obtained with PFGE.

We designed 3 sets of PCR primers for the detection of ST307, IncX3 plasmid, and the IS3000-OXA MGE. We screened K. pneumoniae isolates with OXA-48–like enzymes (n = 471) using different primer sequences (Appendix Table 1).

We sequenced the K. pneumoniae isolates that tested positive by PCR for ST307 (n = 88) using the Illumina NextSeq platform (Appendix). We prepared libraries with the Illumina Nextera XT kit to produce paired end reads of 150 bp for a predicted coverage of >75×. We chose K. pneumoniae I72 (isolated from urine obtained in Benoni, Gauteng Province, during December 2014) that was positive for bla_OXA-181 and bla_OXA-48 for long-read WGS using the RSII platform (Pacific Biosciences) to characterize plasmids.

We compared the South Africa genomes sequenced in this study (deposited in the NCBI Bioproject database [https://www.ncbi.nlm.nih.gov/bioproject] under accession no. PRJNA488070) with 620 ST307 genomes previously deposited in the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra) and the genome database. The global ST307 genomes were from the United States (n = 488), the United Kingdom (n = 45), Norway (n = 30), Italy (n = 10), Thailand (n = 9), Australia (n = 6), Brazil (n = 3), Colombia (n = 3), China (n = 3), Nepal (n = 3), Cambodia (n = 2), France (n = 2), Nigeria (n = 2), Cameroon (n = 1), Guinea (n = 1), Iran (n = 1), Netherlands (n = 1), Pakistan (n = 1), and other unspecified regions (n = 9).

We mapped all genomes to the ST307 reference genome NR5632 (GenBank accession no. CP025143) using Snippy (https://github.com/tseemann/snippy). We predicted prophages using PHASTER (6), examined repeated regions using MUMmer (7), and predicted putative regions of recombination with Gubbins (8), followed by filtering using vcftools. We generated a recombination-free single-nucleotide polymorphism phylogenetic tree, using a general time-reversible model of nucleotide substitution with a gamma model of rate heterogeneity and 4 rate categories, by using RAxML version 8.2.4 (9). We conducted hierarchical Bayesian analysis of population structure with 3 nested levels and 10 independent runs of the stochastic optimization algorithm with the a priori upper bound of 10–30 clusters varying across the runs to identify phylogenetic clades. We defined clades using the first level of clustering (10) and annotated the phylogenetic tree in iTOL (11).

We used BEAST version 2.4.7 (12) to estimate a timed phylogeny with concatenated recombination-free core single-nucleotide polymorphism alignment. To increase the accuracy of the time to most recent common ancestor, we included 17 additional ST307 isolates from a Pretoria academic hospital in this analysis. We identified the bla_OXA181 harboring IncX3 plasmids by de novo assembly using plasmidSPAdes (13) and manually investigated the findings using Bandage assembly graph viewer (14) and blastn (https://blast.ncbi.nlm.nih.gov/Blast.cgi), as well as S1-PFGE in combination with Southern blotting (Appendix).

Ampath-MDRC experienced an increase in OXA-48–like K. pneumoniae isolates during July 2015–June 2016, especially from private hospitals in Gauteng Province. We identified by PFGE a dominant pulsotype named A (and a related pulsotype AR) among the OXA-48–like collection. Initial WGS on 28 isolates that belong to pulsotypes A and AR identified them as ST307 with OXA-181 on IncX3 plasmid that was shown using PacBio long-read sequencing. Using WGS, we used PCR primers specific to ST307, IncX3, and OXA-181-IS3000 MGE to screen the OXA-48–like K. pneumoniae and showed that 74% belonged to ST307 containing IncX3 plasmids.

WGS of an additional 60 PCR-positive ST307 illustrated, when compared with ST307 from other countries, that the South Africa ST307 belonged to a different clade. The South Africa clade emerged around March 2013 in Gauteng Province, then evolved during 2014 into 2 distinct lineages that spread across northeastern South Africa, affecting 350 patients.

Overall, 574 carbapenemase gene PCR-positive K. pneumoniae were detected at Ampath-MDRC during January 2014–December 2016; the total included NDM (n = 58), KPC (n = 10), VIM (n = 35), and OXA-48–like (n = 471) carbapenemases (Appendix Table 2). The OXA-48–like isolates from 2014 came mainly from different private hospitals in Johannesburg and Alberton, Gauteng Province. The numbers of OXA-48–like K. pneumoniae increased exponentially toward the end of 2015 and peaked during the first 6 months of 2016 (e.g., we detected 349/471 [74%] of OXA-48–like K. pneumoniae during July 2015–June 2016). This increase was especially evident in various hospitals across the province. Private hospitals from other areas in northern and eastern South Africa also experienced increases during 2015–2016 (Appendix Table 2). Throughout the study period, the numbers of K. pneumoniae with KPC, NDM, and VIM remained relatively low and stable compared with OXA-48–like K. pneumoniae. PCR testing did not detect IMP or GES during this period.

We obtained OXA-48–like K. pneumoniae from intraabdominal specimens (n = 78), urine (n = 196), skin and soft tissues (n = 11), blood (n = 76), central venous catheter tips (n = 8), respiratory specimens (n = 99), and rectal specimens (n = 3). The isolates tested nonsusceptible (intermediate or full resistance) to ampicillin, amoxicillin/clavulanic acid, piperacillin/tazobactam, cefotaxime, ceftazidime, cefepime, and ertapenem. Most isolates were also nonsusceptible to trimethoprim/sulfamethoxazole (98%), gentamicin (98%), ciprofloxacin (92%), and meropenem (52%), whereas 44% of isolates were nonsusceptible to imipenem and 11% to amikacin. The tigecycline (TGC) and colistin MICs were each <1 µg/mL, except for 4 isolates that had TGC MICs of 2, 4, 4, and 8 µg/mL.

PCR screening of OXA-48–like K. pneumoniae showed that 350/471 (74%) isolates were positive for ST307, IncX3, and IS3000-OXA-181 (cited as ST307_X3-OXA-181). ST307_X3-OXA-181 appeared in different private hospitals in Johannesburg and Alberton during January–June 2014 (Figures 1, 2). During July 2014–June 2016, ST307_X3-OXA-181 subsequently spread to other private hospitals in Gauteng Province, especially in Pretoria, and to Mpumalanga, North West, Limpopo, Free State, and Eastern Cape Provinces (Figures 1, 2). Gauteng Province was the epicenter of the dissemination of ST307_X3-OXA-181, showing a substantial increase in numbers from 2014 (n = 18) to 2015 (n = 138) (Figure 1).

We isolated ST307_X3-OXA-181 mostly from urine (41%), respiratory (20%), intraabdominal (18%), and blood (16%) specimens; 96% of specimens were submitted from hospital settings. Ampath-MDRC receives ≈80% of community, nursing home, and hospital specimens from the private sector in Gauteng, Mpumalanga, North West, Limpopo, Free State, and Eastern Cape Provinces. We isolated ST307_X3-OXA-181 exclusively from hospital specimens; therefore, it is unlikely that the community or nursing home sectors were important reservoirs of ST307.

The overall characteristics of the ST307 sequences from this study (n = 88) were similar to ST307 with bla_KPC and bla_CTX-M-15 sequences obtained previously in Colombia, Italy, and the United Kingdom, and those in this study contained the capsular loci wzi-173, capsule 2, π-fimbrial cluster, Type IV secretion system (15). The hierarchical Bayesian clustering analysis of 88 South Africa ST307 and additional 620 genomes from publicly available databases divided ST307 into 6 distinct clades: clades I–IV, from the United States (mainly Texas) (16); clade V, from various countries (Australia, Brazil, Cambodia, Cameroon, China, Colombia, France, Guinea, Iran, Italy, Nepal, Netherlands, Nigeria, Norway, Pakistan, Thailand, the United Kingdom, and the United States); and clade VI, which consisted of the isolates from South Africa (Figure 3). Clade V also includes a subset of isolates from Texas. Sequence analysis of the mutations in the quinolone resistance–determining regions in chromosomal gyrA and parC indicated that most of the isolates from clades V and VI contained the ParC-80I and GyrA-83I mutations, whereas isolates from clades I–IV harbored an additional GyrA-87N mutation (Figure 3).

Genomic analysis revealed that clade VI isolates contained some unique characteristics when compared with clades I–V. All of clade VI harbored the plasmid p72_X3_OXA181 with bla_OXA-181 that contained the IncX3 and truncated ColKp3 replicons (Appendix Figure 2). Approximately 30% of isolates in clades I–V contained various carbapenemase genes, namely bla_KPC-2 (n = 184), bla_KPC-3 (n = 4), bla_NDM-1 (n = 3), bla_NDM-5 (n = 2), and bla_OXA-48 (n = 2). KPC-2 was distributed within clades I–V, whereas KPC-3, NDM-1, ΝDΜ-5, and OXA-48 were restricted to clade V (Figure 3). Most ST307 isolates in clades I–VI contained the extended-spectrum β-lactamase gene bla_CTX-M-15 (684/708, 96.6%) that was adjacent to ISEcp1 (682/684, 99.7%) situated on a FIB-like plasmid and showed high similarities to the previously described pKPN3–307_type A plasmid (15).

We identified 5 prophages (72_phage 1–5) and 1 novel integrative conjugative element (72_ICE) within the 88 sequenced ST307 that belonged to clade VI (Figure 3); the 72_phages 2, 3, and 4 correlated with the previously described ST307 phages 1, 3, and 4 (15). The 72_phage1, 72_phage5, and 72_ICE were mainly restricted in clade VI; 72_ICE was unique to clade VI, whereas 72_phage1 was also present in 5 isolates and 72_phage5 in 23 isolates from clade V. The 72_phage3 and 72_phage4 were found in clade VI and a subset of clade V, and the 72_phage 2 was part of ST307 isolates from clusters I–VI (Figure 3).

Bayesian evolution analysis conducted with BEAST estimated the time to most recent common ancestor of clade VI around March 2013, ≈2 months before the first isolate (C6) that was collected in May 2013 from a Johannesburg hospital, J1 (Figure 4). We estimated the mean evolutionary rate as 1.16 × 10⁻⁶ substitutions/site/year (95% highest posterior density 7.7 × 10⁻⁷ to 1.6 × 10⁻⁶ substitutions/site/year), corresponding to 5.7 substitutions/genome/year among clade VI. This analysis placed the outbreak node, from which all outbreak isolates were derived, at April 2014 (95% highest posterior density September 2013–September 2014).

Two distinct lineages evolved within Gauteng Province from the outbreak node. Lineage A originated in Johannesburg and lineage B in Alberton, and they spread to different hospitals across Gauteng (both lineages), Mpumalanga (both lineages), Free State (lineage A), Eastern Cape (lineage A), Limpopo (lineage A), and North West Provinces (lineage B) (Figure 4).

Genomics, combined with admission, discharge, and transfer data, showed evidence of intrahospital spread. For example, patient I51 in hospital P5 was transferred during December 2015 from the medical intensive care unit to the high-risk unit. Subsequently a highly related ST307 was isolated during January 2016 from patient I68 in the high-risk unit (Figure 4). We also observed interhospital spread when patient I14 was transferred from the surgical intensive care unit at hospital P1 to hospital P4 during September 2016. In December 2016, a highly related ST307 was isolated at hospital P4 from patient K43 (Figure 4). Intercity and interprovince spread were illustrated by patient I2 from the city of Springs, patient K61 from the town of Heidelberg, and patient K22 from the North West Province, all with highly related ST307 isolates, having previously received medical care at hospital J2 in Johannesburg (Figure 4).