Plasmid Dissemination and Selection of a Multidrug-Resistant Klebsiella pneumoniae Strain during Transplant-Associated Antibiotic Therapy

Antibiotic-resistant bacteria are a serious threat to medically fragile patient populations. The spread of antibiotic resistance through plasmid-mediated mechanisms is of grave concern as it can lead to the conversion of endogenous patient-associated strains to difficult-to-treat pathogens.

T he increase in carbapenem resistance among Enterobacteriaceae species is a worldwide threat (1,2). Horizontal gene transfer (HGT) plays an important role in the spread of antibiotic resistance genes in clinical settings. Plasmid-borne resistance genes can turn infections with common organisms, like Escherichia coli into serious complications. Although guidelines to prevent transmission of carbapenem-resistant Enterobacteriaceae (CRE) exist (https://www.cdc.gov/hai/organisms/cre/cre-toolkit/index.html), they do not explicitly take into account HGT, which may complicate health care facility surveillance. Genomic studies like this one, and those conducted at other health care facilities (3)(4)(5), are critical to our understanding of plasmid-mediated antibiotic resistance and to improving local and global sequence databases. Here we investigate an exceptional case of a pair of plasmids that are widely distributed across isolates from a single patient, using both whole-genome sequencing (WGS) and metagenomic sequencing. We have combined treatment data, WGS, and metagenomics to explore the distribution and persistence of carbapenem resistance plasmids and changes to the microbiome associated with this patient's antibiotic chemotherapy.

RESULTS
Case history. In 2014, a patient entered care at the National Institutes of Health (NIH) Clinical Center for management of a hematologic malignancy. From December 2014 through April 2016, this patient was admitted on eight separate occasions, ranging in duration from 6 to 119 days. During the first admission (H1), the patient underwent a hematopoietic stem cell transplantation (HSCT) and started immunosuppressive therapy and supportive antibiotic therapy (Fig. 1). During that time, a carbapenem-susceptible Klebsiella pneumoniae strain of sequence type 45 (ST45) was cultured from blood and treated with piperacillin-tazobactam followed by ceftazidime. Ten days later, a carbapenemase-producing, bla KPC -positive K. pneumoniae strain (ST340) was cultured from an abscess. The patient continued to test positive for carbapenemase-producing K. pneumoniae across the patient's remaining hospitalizations. In addition, surveillance and clinical cultures positive for Enterobacteriaceae, Pseudomonas, Staphylococcus, and Streptococcus spp. were detected (see Table S1 in the supplemental material). In March, during the patient's third hospitalization, 68 days after the first bla KPC -positive K. pneumoniae strain was cultured, three bla KPC -positive  Fig. 2). The four K. pneumoniae genomes belong to three different sequence types (susceptible, ST45; resistant, ST340 and ST35). Carbapenemaseproducing Citrobacter freundii (ST22), K. aerogenes (ST197), and Enterobacter cloacae (ST1260) strains were sequenced from the third and fourth hospitalizations. Finally, a carbapenem-susceptible strain of C. freundii was also sequenced.

FIG 2
Plasmid composition of sequenced Enterobacteriaceae. WGS assemblies were scaffolded onto reference sequences. Plasmids shared across genomes are shown in the same color. The Tn4401b-KPC3 cassette is shown as a magenta circle. Plasmids that are unique to a strain are in black. Plasmids that were detected but could not be scaffolded on an existing reference sequence are shown in gray. The two K. pneunomiae ST340 isolates with chromosomal bla KPC genes are marked with "^". Hospitalization number is indicated below each isolate (H1 to H7). Plasmid incompatibility groups were determined using PlasmidFinder v2.0. Plasmid mobility was determined using MOB-suite. Conjugative plasmids are denoted with "(c)," and mobilizable plasmids are marked with "(m)." Nonmobilizable plasmids are unmarked.
Two plasmids are found across CPOs. Analysis of the whole-genome sequence data from the carbapenemase-producing organisms (CPOs) in Fig. 1 determined that they all carry the KPC-3 allele in the context of a Tn4401b cassette on one or two plasmids. The first plasmid (pKPC-e4b7) is nearly identical to pMNCRE78_3 from a wound-associated Klebsiella pneumoniae strain (2013, Minnesota Department of Health, unpublished data [GenBank accession no. CP018432.1]) aligning over 99% of its length with only 3 single nucleotide polymorphisms (SNPs) and 2 indels (see Fig. S1 in the supplemental material). The SNPs are in a transglycosylase (g.10957CϾT, p.Asnϭ), Tn3-family transposase (g.27840TϾC, p.Phe984Leu), and in the bla KPC gene (g.31028CϾT, p.His272Tyr). The bla KPC mutation reflects the KPC-3 allele in our isolates compared to a KPC-2 allele in the reference plasmid. Our lab and others have observed the phenomenon of essentially identical plasmids with different KPC alleles previously with pKpQIL (6, 7). An insertion (g.31_32insA) corrects the reading frame of an annotated pseudogene for a type II/IV secretion system family protein. Finally, there is a 490-nucleotide (nt) deletion in a pseudogene (g.13534_14023del). There were no additional shared SNPs that could be used to identify transmission links between bacteria.
The second KPC-positive plasmid, carried by a subset of sequenced isolates, is related to pNJST258C2 (8) and occurs in two variants (pKPC-610e and pKPC-5fbf) differentiated by sequence inversions (Fig. S1). Long-read sequencing of the ST340 K. pneumoniae isolate also revealed a chromosomal copy of the bla KPC-3 gene embedded in a Tn4401b cassette. The ST340 K. pneumoniae and K. aerogenes isolates share an additional plasmid from the IncR incompatibility group ( Fig. 2; blue plasmid). It carried genes for resistance to a variety of antibiotics (e.g., aminoglycosides, chloramphenicol, and macrolides), two class 1 integrons (In27 and In641) (9), and an iron transport locus.
Detection of putative C. freundii recipient strain. A carbapenem-susceptible C. freundii strain was identified on day 39, prior to detection of the carbapenemaseproducing C. freundii strain on day 68 (Fig. 1). We sequenced the susceptible strain and determined that it was the same sequence type as the resistant isolate and that the two isolates differed by only 15 SNPs across 99.2% of the chromosome. As expected, the susceptible strain did not carry either bla KPC -containing plasmid but it did carry a plasmid with similarity to PCN061p4 (CP006640.1). The carbapenemase-producing C. freundii strain lacked the PCN061p4 plasmid but carried both pKPC-e4b7 and pKPC-5fbf, which were previously detected in other Enterobacteriaceae.
bla KPC plasmid-containing isolates persist for over a year. Of the five carbapenemase-producing isolates sequenced from the third and fourth hospitalizations, all of them carried the bla KPC -positive pMNCRE78_3 plasmid, and two carried the bla KPC -positive pNJST258C2-like plasmid. Using PCR primers specific to each plasmid (see Materials and Methods), we found that both bla KPC -positive plasmids persisted through the seventh hospitalization, 435 days after the first bla KPC -positive culture (Fig. 1). Of the 10 isolates tested from the sixth and seventh hospitalizations, seven carried the pMNCRE78_3 plasmid and nine carried one of the pNJST258C2 plasmid variants. An ST340 K. pneumoniae isolate from day 435 was sequenced, found to be nearly identical (Ͻ10 chromosomal SNPs) to the initial ST340 isolate, and carried all four plasmids. However, it now carries the pNJST258C2-like plasmid with sequence inversions and an additional small IncN3 plasmid (Fig. 2).
Plasmid mobility. Plasmids may move between bacterial hosts by a number of mechanisms, but conjugation and mobilization (MOB) are the most relevant in the current context. Conjugation requires genes for mating pair formation (MPF) and mobilization (MOB), while mobilizable plasmids only require MOB genes and are thought to use MPF systems encoded by other plasmids (10). We used MOB-suite (11) and PlasmidFinder (12) to characterize all the replicon type and predicted mobility of all 12 plasmids in this study. pKPC-e4b7 and the reference pMNCRE78_3 plasmid are both predicted to be nonmobilizable, despite being annotated to encode some genes associated with a type IV secretion system. pKPC-610e and pKPC-5fbf are predicted to be mobilizable, since they carry a MOBC type relaxase gene. Overall, of the 12 unique plasmid types in this study, six are predicted to be conjugative, 4 are mobilizable, and 2 are nonmobilizable (Fig. 2).
Antibiotics alter the gut microbiome and select for a drug-resistant strain. In previous studies of plasmid carriage in patients (13), we have noted that culture-based studies are limited in that single isolates are a nonrepresentative subsample of the patient microbiome. While we have been successful in using culture-enriched metagenomes (i.e., "plate swipes") to characterize microbial communities (14), here we sought to explore the utility of unbiased fully metagenomic sequencing to characterize the changes in the microbiome associated with a small window of time immediately following administration of a course of meropenem.
Stool and perirectal swabs were collected at three time points during March through April 2015 (T1, T2, and T3 [ Fig. 1]) during the patient's third hospitalization and were subjected to shotgun sequencing to capture the metagenome. Immediately following enrollment in the study, baseline samples were taken from stool and perirectal swabs on the same day as a 4-day course of meropenem and a single-day course of gentamicin were initiated (T1). One day later, the patient began a 13-day course of levofloxacin. The patient also received an extended 42-day course of daptomycin, initiated the previous month. Additional perirectal and stool samples were acquired approximately 1 week (T2) and 2 weeks (T3) after T1. While somewhat complicated by previous antibiotic exposures and the patient's multiple overwhelming infections, these samples provide an opportunity to examine the response of the microbiome and target organisms to the stress of combination antibiotic therapy.
We calculated the relative abundances of bacterial taxa, with special emphasis on taxa matching cultured isolates (Fig. 3). As expected for gut-associated communities, all samples were dominated by Bacteroidetes, Firmicutes, and Proteobacteria. The Proteobacteria present were predominantly (Ͼ98%) Gammaproteobacteria, and specifically Enterobacteriaceae. At T1, the Enterobacteriaceae were a mixture of Klebsiella, Citrobacter, Enterobacter, and Escherichia species. After the administration of meropenem, gentamicin, and levofloxacin, K. pneumoniae became the dominant Enterobacteriaceae species (T2 and T3).
The loss of Enterobacteriaceae diversity may be explained by the antibiotic resistance profiles of each strain. Specifically, the ST340 K. pneumoniae strain is resistant to levofloxacin, which was given immediately after T1. The resistance of this strain to quinolones, like levofloxacin, is likely due to mutations in gyrA (D87Y) and parC (S80I) (15 (10 and 6 reads, respectively). The ST340 strain carried additional resistance genes for aminoglycosides: ANT(3Љ)-Ia family aminoglycoside nucleotidyltransferase genes aadA1 and aadA2, as well as aminoglycoside O-phosphotransferase APH(3=)-Ia, macrolide resistance, and chloramphenicol resistance genes.

DISCUSSION
HSCT offers patients with hematologic malignancies the chance for a cure or remission. However, conditioning regimens, long-term immunosuppressive therapy, and frequent hospitalization are all risk factors that complicate treatment. We investigated the impact of HSCT and the follow-up treatments and therapies on the patient's microbiome and several target organisms with pathogenic potential. Because of the patient's fragile medical condition and colonization with organisms of infection control concern, we conducted close clinical and microbiological monitoring that made it possible to detect the HGT. Despite infections with multidrug-resistant organisms and severe immunosuppression, the patient ultimately recovered, was discharged home, and continues to return for clinic visits.
Multidrug-resistant bacteria are an unavoidable consequence of decades of antibiotic use. As such, combination therapies are commonly used to both prevent and treat bacterial infections in vulnerable patients. In the current case, treatment with multiple classes of antibiotics following the detection of a single carbapenemase-producing K. pneumoniae strain resulted in the detection of multiple carbapenemase-producing organisms, including Citrobacter and Enterobacter spp. While we cannot rule out that additional resistance mechanisms may be at play (e.g., efflux pumps or outer membrane protein mutations), we did not detect disruptive mutations in the K. pneumoniae OmpK35/36 porins, which are known to be associated with carbapenem resistance (16). Furthermore, the presence of two plasmids carrying the bla KPC gene makes it likely that HGT underlies the rapid spread of carbapenem resistance. While HGT within the gut of the patient is one possible explanation of these findings, without additional studies we cannot rule out other hypotheses. Limited dissemination of plasmids in vivo has been described previously in patients (13,(17)(18)(19) and in a mouse model (20)-however, not to the extent seen in the present study with two plasmids distributed across five strains. Stecher and colleagues describe high rates of HGT during blooms of Enterobacteriaceae in a mouse model of Salmonella infection (21).
Plasmids can be assigned to three mobility groups: conjugative, mobilizable, and nonmobilizable. Surprisingly, a prevalent bla KPC plasmid (pKPC-e4b7) associated with multiple species isolated from this patient is predicted to be nonmobilizable. While it is possible that pKPC-e4b7 carries a relaxase (mob) gene that is not detected by the MOB-suite analysis, we did not detect one using additional blastp/blastx searches against a database of Mob proteins. Other mechanisms of HGT like transformation and transduction are predicted to be low-frequency events that would not explain this plasmid's broad distribution. Despite this, it is clear that pKPC-e4b7 is present in five different hosts isolated over a relatively short period of time.
One limitation of this work is that we did not demonstrate in vitro plasmid transfer. Previously, we have shown that in vitro plasmid conjugation experiments often do not recapitulate what is seen in vivo (22), and we expect that will be true in this case. That said, several lines of evidence support recent dissemination of these plasmids, possibly in the patient. The pKPC-e4b7 plasmid is essentially identical across strains, supporting recent transmission events. We routinely sequence all bla KPC -positive isolates from the NIH Clinical Center, and we have not observed this plasmid in patients or the environment (7, 23), suggesting it is not an endemic plasmid. Finally, we identified the probable recipient Citrobacter strain in the patient prior to the detection of the plasmid-containing carbapenemase-producing strain.
We investigated the changes in the patient's microbiome using shotgun metagenomics. Stool and perirectal surveillance swabs provided complementary and mostly correlative measures of microbial relative abundance. That said, metagenomes prepared from perirectal swabs were dominated by host-derived reads, making them a suboptimal choice for characterizing the bacterial component of the microbiome. Furthermore, in cases such as this one where target bacteria are easily cultivated, we would recommend the plate swipe method (14) for assessing bacterial diversity. Unfortunately, that was not possible in the current study because the primary samples were frozen immediately upon collection.
The bacterial component of the microbiome showed a number of changes as a result of antibiotic therapy. Specifically, there was a reduction in the diversity of Enterobacteriaceae, with Klebsiella pneumoniae becoming the dominant species. The dominance of K. pneumoniae was driven by the clonal expansion of a single multidrugresistant ST340 strain, part of clonal complex 258. In particular, the apparent fixation of fluoroquinolone-resistant alleles highlights the potential consequences of multidrug antibiotic therapies. All inpatients in our hospital undergo screening for carbapenemase-producing organisms. However, this patient's higher clinical acuity exposed the patient to a greater intensity of microbial scrutiny, which may have made the HGT more apparent by greater sampling. In addition, the patient's immunosuppressive therapy as well as effects of antimicrobial pressure from treatment of multiple concurrent multidrug-resistant infections may have played roles in these findings.

MATERIALS AND METHODS
Sample collection and bacterial culturing. The patient provided written consent to participate in a study approved by the institutional review board of the National Institutes of Health Clinical Center (https://clinicaltrials.gov/ct2/show/NCT01933620). Perirectal surveillance cultures were plated directly onto CRE Chromagar (Hardy Diagnostics) and incubated overnight at 35°C in ambient air. Blood cultures were collected into Bactec Plus bottles (Becton, Dickinson). Other clinical cultures were plated onto routine clinical microbiology media, which included a combination of sheep blood agar (Remel, Inc.), chocolate agar (Remel, Inc.), MacConkey agar (Remel, Inc.), and/or fastidious broth (Hardy Diagnostics) and incubated at 35°C in 5% CO 2 . Organisms were identified by matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF MS) (BioTyper; Bruker, Billerica, MA). Susceptibility testing was performed via broth microdilution (Trek Diagnostics) or disk diffusion (Thermo Fisher Scientific, Lenexa, KS) following Clinical and Laboratory Standards Institute M100 interpretive breakpoints that were current for the time (24)(25)(26). Isolates meeting internal laboratory testing criteria were assessed by real-time PCR for the presence of bla KPC or bla NDM (http://www.cdc.gov/HAI/pdfs/labSettings/KPC -NDM-protocol-2011.pdf). Primary stool and perirectal samples were frozen at -80°C for subsequent DNA extraction and metagenomic analyses.
Genome sequencing and assembly. Genomic DNA was purified for each isolate, from which Nextera XT (Illumina) libraries were generated. Each isolate was sequenced using a paired-end 300-base dual index run on an Illumina MiSeq. The reads were subsampled to achieve 80ϫ coverage and then assembled with SPAdes (version 3.6.0) (27) and polished using Pilon (28). To achieve full reference genomes for select isolates, genomic DNA was sequenced on PacBio RSII SMRT platform (P6 polymerase binding and C4 sequencing kit; 240-min collection), and assembled and polished using Canu v1.3 (29). Genome annotation was performed using National Center for Biotechnology Information (NCBI) Prokaryotic Genome Annotation Pipeline (PGAP [https://www.ncbi.nlm.nih.gov/genome/annotation_prok/]). The multilocus sequence type (MLST) databases at https://pubmlst.org (30) were used to sequence type isolates; K. aerogenes ST197 and E. cloacae ST1260 were identified in this study. Whole-genome shotgun assemblies were scaffolded on existing references using ABACAS (31) and the mummer package (32). Plasmid incompatibility groups were predicted using PlasmidFinder 2.0 (12), and plasmid mobility was predicted using MOB-suite (11).
Metagenome analysis. Metagenomic samples were processed and analyzed as described previously (33,34). Briefly, Nextera XT library kits were used to generate Illumina libraries per the manufacturer's instructions. Libraries were sequenced on an Illumina HiSeq at the NIH Intramural Sequencing Center to a target depth of 15 to 50 million clusters of 2 ϫ 125-bp reads. Human reads were removed based on alignment to the human reference genome (hg19) using bowtie2, bases with quality scores below 20 were trimmed, and remaining reads less than 50 bp were removed. Reads were assigned to a microbial reference genome database of bacteria, fungi, viruses, and archaea using Clinical Pathoscope as described previously (34,35).
Primer design. Primers were designed using Primer3 (36) for the pMNCRE78_3 plasmid (CP018432.1) in the BTE52_30285 (Tn) and BTE52_30260 (hypothetical) genes. Primers were designed for the Plasmid Spread Associated with HSCT and Antibiotics ® pNJST258C2 plasmid (CP006919) in the KPNJ2_05469 (hypothetical) gene and flanking one of the recombination junctions (see Table S2 and Fig. S1 in the supplemental material).