Molecular Epidemiology of Colonizing and Infecting Isolates of Klebsiella pneumoniae

K. pneumoniae commonly infects hospitalized patients, and these infections are increasingly resistant to carbapenems, the antibiotics of last resort for life-threatening bacterial infections. To prevent and treat these infections, we must better understand how K. pneumoniae causes disease and discover new ways to predict and detect infections. This study demonstrates that colonization with K. pneumoniae in the intestinal tract is strongly linked to subsequent infection. This finding helps to identify a potential time frame and possible approach for intervention: the colonizing strain from a patient could be isolated as part of a risk assessment, and antibiotic susceptibility testing could guide empirical therapy if the patient becomes acutely ill.

. There were no significant differences in age, sex, or self-reported race/Hispanic ethnicity. Antibiotic exposure was numerically higher in the uninfected group (26.2% versus 12.8% of infected patients), but the difference did not reach statistical significance (P ϭ 0.067). Neurologic disorders  and fluid and electrolyte disorders were significantly more frequent in infected patients than in noninfected patients. Baseline albumin levels were significantly lower in the infected group (P ϭ 0.009), and length of stay was significantly longer in the infected group (14.9 versus 11.6 days for uninfected; P ϭ 0.01).  Table 3). For 20 of 21 colonized patients who became infected, colonization was detected on their initial rectal swab; 1 patient became positive on their second rectal swab 6 days later.

Association of colonization with
In the final multivariable model, colonization with K. pneumoniae had the highest association with infection (all sites) after adjustment for other potential confounders (OR, 4.01; 95% CI, 2.08 to 7.73; P Ͻ 0.001) ( Table 4). In addition, fluid and electrolyte disorder, neurologic disorder, and previous hospital admissions within the past 28 days were independently associated with infection. Low baseline platelet levels approached but did not reach significance (P ϭ 0.058); however, this variable was retained, as it significantly improved the performance of the model (P ϭ 0.046 for the likelihood ratio test) without significantly altering the other variables' estimates. The area under the receiver-operator characteristic (AUROC) demonstrated acceptable performance of the model (0.78; 95% CI, 0.718 to 0.842) (Fig. 2), and the Hosmer-Lemeshow test did not indicate poor model fit (P ϭ 0.135).

Concordance of colonizing and infecting isolate pairs based on molecular strain typing.
To determine if patients become infected with strains with which they were previously colonized, we first screened for genetic differences using wzi gene sequencing. Preliminary results from 17 patients indicated that wzi sequencing had a similar discriminatory power to MLST, with both distinguishing 16 sequence types among 20 isolates (see Fig. S1 in the supplemental material). In order to assess the diversity of strains with which patients were colonized, we determined the wzi types of colonizing isolates from 40 patients. Sixteen of these colonized patients had subsequent positive clinical cultures and met case definitions for BSI, pneumonia, or UTI; 24 patients did not. A total of 110 rectal swab isolates were tested; up to three isolates were obtained from each patient. From these 40 patients, only 8 patients had two wzi types detected, and no patients had three types within a sample. Despite the homogeneity within individual patients, 43 different wzi types were identified among these  40 patients, suggesting high genetic diversity of colonizing K. pneumoniae in this patient population (Fig. 3). Of 21 colonized patients who developed infection, 16 sets of colonizing and infecting isolates were available for analysis. Two out of five patients with BSI (40%) had concordant pairs based on wzi sequencing of blood and rectal swab isolates. Respiratory and rectal swab isolates from patients with pneumonia (n ϭ 7) demonstrated perfect concordance (7/7) (see Fig. S2 in the supplemental material). Although two patients with pneumonia were each colonized with 2 different wzi types, one was concordant with each patient's respiratory isolate (stool isolate 1043 matched respiratory isolates 733 and 734, and stool isolate 1967 matched respiratory isolate 2005) (see Fig. S2). Analysis of urine and rectal swab isolates from patients with UTI (n ϭ 4) also demonstrated perfect concordance (4/4).
Despite high concordance of colonizing and infecting isolate pairs by wzi sequencing, using a single gene typing method may not be sufficient to determine true isolate  concordance. To confirm that isolate pairs were the same strain, we performed WGS and determined isolate ST by both 7-gene MLST and 634-gene core genome multilocus sequence typing (cgMLST). We analyzed 13 preliminarily concordant pairs and one unmatched pair as a discordant control (pair 463/1946). MLST analysis showed perfect agreement with wzi sequence typing results in identifying 13 concordant pairs (Fig. 4). Two novel STs were identified, ST2359 and ST2360. For patients in whom one of two colonizing isolates matched the infecting isolate, MLST distinguished between the colonizing isolates and indicated that only one was concordant with the infecting isolate.
The cgMLST method provides a more discriminatory approach to defining concordance, since it is based on allelic similarities of 634 K. pneumoniae genes (24). All concordant pairs, based on wzi and MLST analyses, also clustered together based on cgMLST (Fig. 4). The only pairs that did not group together were our discordant control, and the discordant colonizing isolates (numbers 1045 and 1968) from patients with another colonizing strain that matched the pneumonia isolates. These isolates were also discordant by wzi and MLST analyses. To measure the strength of cgMLST concordance of colonizing-infecting pairs within patients compared to between patients, a minimum spanning tree (see Fig. S3 in the supplemental material) was also generated based on the cgMLST data. For example, there were two allelic differences between stool isolate 1223 and pneumonia isolate 1317 in the same patient, while there were 189 allelic differences between these isolates and their closest neighbors from a different patient (pair 1319/868). Overall, there was an average of 2 allelic mismatches between concordant pairs (range, 0 to 7) and 449 allelic mismatches between patients (range, 189 to 629). Taken together, the wzi, MLST, and cgMLST data indicate that 100% of urinary and pneumonia isolates tested corresponded to the previously colonizing strain of K. pneumoniae.
Categorical agreement of antimicrobial susceptibility of colonizing and infecting isolate pairs. To measure antibiotic susceptibility agreement between concordant and discordant colonizing-infecting isolate pairs, we tested 17 antimicrobials active against Gram-negative bacteria and measured categorical agreement (CA) for susceptible (S), intermediate (I), or resistant (R) isolates based on MIC breakpoints (see Table S1 in the supplemental material) (25). CA was greater than 90% for all antimicrobials tested. However, this isolate collection had a low prevalence of antibiotic resistance. In 12 of 16 patients, both isolates were susceptible to all antimicrobials tested, including the 3 patients with discordant pairs based on sequence type. Two patients with concordant strain types had initially discordant susceptibility results. Only one discrepancy was reproducible by broth microdilution (trimethoprimsulfamethoxazole in pair 1967/2005). The remaining 11 patients with concordant isolate pairs by sequence type had identical susceptibility patterns, including one ESBL K. pneumoniae isolate (pair 767/664).

DISCUSSION
The objective of this study was to examine the association between K. pneumoniae rectal colonization and subsequent extraintestinal K. pneumoniae infection. Based on data from 1,765 intensive care and hematology/oncology patients, we found that approximately 1 in 4 patients were rectal carriers of K. pneumoniae. We also observed a significant association between rectal K. pneumoniae colonization and subsequent infection, even after adjusting for patient variables. Furthermore, there was high concordance among colonizing isolates and subsequent infecting isolates, particularly for pneumonia and UTI, as measured by wzi, MLST, and cgMLST analyses. Taken together, these results implicate colonization as a critical step in the pathogenesis of hospital-acquired infections. These results also identify a possible window for intervention to decolonize patients or characterize their colonizing strain in order to predict risk of disease and inform empirical therapy if infection develops.
Our study has several strengths. First, whereas previous studies focused on drugresistant K. pneumoniae or strains involved in outbreaks (26)(27)(28), we tested all isolates during a 3-month collection period across multiple wards and units in the hospital. This approach provided comprehensive information on K. pneumoniae colonization in the hospital setting and minimized potential selection bias. The large sample size (n ϭ 1,765) provided sufficient power to examine the relationship between colonization and patients that met strict case definitions of infection. Second, we used wzi gene sequencing to rapidly screen for genetic differences between K. pneumoniae isolates (20). We then confirmed concordant pairs by using WGS-based MLST and cgMLST. In a hospital laboratory setting, wzi could be used to screen for a potential outbreak strain as a triage step before more costly WGS (20). Lastly, molecular strain typing indicated high K. pneumoniae strain diversity in our study population. Molecular epidemiology studies show clonal spread of carbapenem-resistant K. pneumoniae in the United States (28,29). If a dominant clone existed in our population, it would obscure the true association between colonization and subsequent infection. In our diverse setting, the high concordance between colonizing and infecting strains suggests a robust pathogenic mechanism in which patients become infected by their colonizing strain.
This study also has some limitations. First, 35 patients with a positive clinical culture had unknown colonization status prior to the culture date and, thus, were excluded. By excluding this subset of data, we potentially lost cases of infection, and we cannot predict in which direction this would bias the results. Although we collected three rectal swab isolates per patient, most extraintestinal K. pneumoniae isolates provided by the clinical lab represented one isolate per site. It is possible that multiple strains may be present at an extraintestinal site but only one isolate was sampled. For wzi sequencing, rectal swab isolates from fewer than 10% of colonized patients were tested. Given 40 unique patients with 43 unique strains, almost every patient was colonized with a different strain. This high level of diversity is unlikely to be maintained in the larger sample set. A limitation of the susceptibility data was that the majority of isolates had no detectable acquired resistance. With a low diversity of resistance phenotypes, we were unable to rigorously test the agreement of susceptibility testing between colonizing isolates and subsequent infecting isolates in the same patient. Future studies should determine if high categorical agreement holds in a larger, more resistant colonized-infected patient population.
We conclude, based on three distinct methods, that there is high concordance between colonizing and infecting isolates, particularly for pneumonia and UTI. The discordance in bloodstream infections could be due to exogenous sources of K. pneumoniae, such as insertion of an intravenous catheter or a healthcare worker's hands. The perfect concordance for UTI is consistent with the paradigm for Escherichia coli UTI, where fecal colonizing strains contaminate the urogenital tract (30). However, the perfect concordance between rectal isolates and pneumonia isolates was striking. This may indicate simultaneous colonization of the respiratory tract at the time of intestinal acquisition of the strain. This strong concordance suggests that infection prevention approaches or guidance of empirical therapies based on detection and characterization of colonizing K. pneumoniae isolates is feasible.
In addition to prior colonization, prior admission, low baseline platelets, and comorbidities of neurologic and fluid and electrolyte disorders were highly predictive of K. pneumoniae infection in a multivariable model. The association between healthcare exposure and subsequent infection is plausible, even after adjustment for colonization, since it likely indicates overall poor health status, itself a risk factor for infection. This is likely also true of the other comorbidities included in the final model. The components of this model include information readily available at admission as part of routine testing and chart review. If validated in an independent cohort, rectal screening paired with these variables could rapidly predict risk of K. pneumoniae infection.
The finding that patients often become infected with their colonizing strain has strong implications for both infection control and patient care interventions. A recent study in a long-term acute care hospital (LTACH) determined that interventions based on screening for KPC decreased both the colonization rate of patients as well as the rate of clinical infections (31). Moreover, characterization of colonizing strains could inform treatment decisions. Understanding the pathogenic mechanisms of progression from K. pneumoniae colonization to disease could enable novel diagnostics and therapeutics to prevent and rapidly treat these common nosocomial infections.

MATERIALS AND METHODS
Patient population and setting. The study was conducted at the University of Michigan Health System (UMHS), a tertiary care hospital with more than 1,000 beds, in Ann Arbor, MI. Approval for this study was granted by the Institutional Review Board of the University of Michigan Medical School. During a 3-month period from 30 July to 31 October 2014, rectal swabs from 1,800 adult (Ն18 years old) patients from the intensive care unit (ICU) or hematology/oncology wards were screened for K. pneumoniae. Concurrently, extraintestinal K. pneumoniae isolates were obtained from the clinical microbiology lab. A total of 1,765 patients had either a rectal swab performed prior to a positive clinical culture or a rectal swab and no positive clinical culture, and these patients were included for analysis of the association between colonization and subsequent infection. Patient demographic characteristics and clinical information was obtained through the electronic medical record (EMR).
Bacterial identification and growth conditions. Rectal swabs were collected during the course of clinical care (upon unit admission, weekly, and at discharge) and were transported and stored in an ESwab transport system (BD, Franklin Lakes, NJ) at room temperature. Within 24 h of receipt, 1 l of inoculated ESwab media was plated onto MacConkey agar (Remel, Lenexa, KS), streaked for quantification, and incubated for 18 to 24 h at 35°C. To ensure collection of the dominant clone in each sample, three mucoid lactose-fermenting (MLF) colonies were isolated as potential K. pneumoniae and subcultured onto blood agar plates (BAP; Remel, Lenexa, KS) (32). If fewer than three MLF colonies were present in a particular sample, all were subcultured. Bacterial identification was performed using matrix-assisted laser desorption ionization-time of flight (MALDI-TOF) analysis. Isolates were stored at Ϫ80°C in Luria-Bertani (LB) broth containing 20% glycerol and were grown on either BAP or LB plates at 30°C overnight unless otherwise indicated.
Definitions. In patients without K. pneumoniae infection, K. pneumoniae colonization was defined as a positive rectal swab culture for K. pneumoniae at any point during the hospital admission. For patients with K. pneumoniae infection, colonization was only considered positive if detected prior to the date of the documented infection. If a patient was identified as colonized on more than one date of collection, we used the first positive sample before infection for wzi sequencing, MLST, cgMLST, and antimicrobial susceptibility testing. Patient EMRs were reviewed for any positive culture for K. pneumoniae within 90 days of rectal swab culture. Bloodstream infection was defined as any positive blood culture for K. pneumoniae. Pneumonia was defined based on a positive K. pneumoniae respiratory culture and other Infectious Diseases Society of America (IDSA) diagnostic criteria (33). Patients with positive urine cultures were identified as cases based on the CDC National Healthcare Safety Network (NHSN) UTI case definitions (34). Patients not meeting a case definition of infection within 90 days of a rectal swab culture were considered uninfected. Comorbid disease definitions were extracted from the EMR based on ICD-9 and DRG codes as specified in the Elixhauser index (35). To ensure that we were capturing antibiotic exposure as a risk factor and not an outcome, this variable was constructed to ensure that exposure was present prior to colonization/infection. Exposure to an antibiotic was defined as true for patients without K. pneumoniae infection or colonization if administered at any point during the admission. For those patients with colonization, the antibiotic exposure variable was true if antibiotics were started at least 48 hours prior to the detected colonization. For those patients with infection but no preceding colonization, antibiotic exposure was positive if started at least 48 hours prior to documented infection.
wzi gene sequencing. DNA preparation and PCR amplification were performed as described by Brisse et al. (20) with the following modifications: PCR products were diluted 1:20 in sterile water, and sequencing primers (wzi_for2 and wzi_rev) were diluted to 1 pM/l prior to submission for Sanger sequencing. Forward and reverse chromatograms were assembled using Lasergene SeqMan (DNASTAR, Madison, WI). Complete alignment was done using ClustalX 2.1 (36). Initial phylogenetic trees were constructed using MEGA 6 (37) based on the neighbor-joining method (500 bootstrap replicates) and Jukes-Cantor distance. The wzi library obtained from Brisse et al. (20) was used as a reference in the current analysis.
Whole-genome sequencing and assembly. Bacterial genomic DNA (gDNA) was purified using the UltraClean microbial DNA isolation kit (MoBio Laboratories, Inc., Carlsbad, CA). Purified gDNA was sent to the University of Michigan DNA Sequencing Core, where it was sheared (320 bp) and prepared as a multiplex library with unique bar codes for each sample. Whole-genome sequencing was performed using the HiSeq 4000 sequencing system (Illumina, San Diego, CA). Reads were preprocessed for each sample by trimming bases at both ends if the quality score was below 10, using Trimmomatic (v0.32) (38), removing read duplicates using FastUniq (39), and performing error correction using SOAPec (v2.01) (40). Preprocessed reads were assembled using VelvetOptimiser (v2.2.5). In this process, the reads were assembled by using Velvet (41) and a stepwise Kmer size at a step of 2, from 51 to 149 (for paired-end samples), or from 25 to 51 (for shared-end samples). The assembly with the largest N 50 value was used for subsequent analysis.
MLST and cgMLST. The gene sequences corresponding to the international MLST scheme of Institut Pasteur (19) were extracted from the genomic assemblies by using the BIGSdb platform (42) and the BLASTN algorithm, and the corresponding allelic number was defined by comparison with the reference nomenclature database (http://bigsdb.pasteur.fr/klebsiella). cgMLST was performed in the same way using the strict core genome MLST scheme defined by Bialek-Davenet et al. (24). Novel alleles and MLST profiles were submitted to the nomenclature database. MLST profiles were compared using the categorical mismatch method within BioNumerics version 6.6 (Applied-Maths, Sint-Martens Latem, Belgium). Uncalled alleles were not considered mismatches in pairwise profile comparisons.
Antimicrobial susceptibility testing. Antimicrobial susceptibility testing was performed on the Vitek 2 automated system (bioMérieux, Marcy-l'Étoile, France), using AST-GN82 cards loaded per the manufacturer's instructions. Isolates were grown on BAP at 37°C overnight. Colonizing and infecting isolates from the same patient were tested in the same batch. Susceptibility testing was performed on one rectal isolate for each patient. ESBL phenotypes were determined with the use of the Vitek 2 Advanced Expert System (AES). Pairs with discrepant results for any antibiotic were tested by Sensititre broth microdilution (Trek Diagnostics Systems, Oakwood Village, OH).
Statistical analyses. Initial tests included examination of variables for out-of-range values, measures of central tendency/spread for continuous variables, and frequencies for categorical variables. These initial analyses assisted in constructing variables, including transformations (for example, length of stay was log transformed prior to analysis, given the nonnormal distribution). Initial bivariable analyses were conducted with Student's t test for continuous variables and the chi-square or Fisher's exact test for categorical variables. Based on these initial analyses, variables with a P value of Ͻ0.2 on bivariable tests were eligible for inclusion in the final multiple logistic regression model. This final model of K. pneumoniae infection was constructed via backwards elimination using a likelihood ratio test for variable retention, with a cutoff ␣ of 0.05. Interactions between variables in the final model were tested and included if significant. Additional model regression assessments included the Hosmer-Lemeshow test for goodness of fit and calculation of the AUROC curve. For interpretation of the results, a P value of 0.05 was considered statistically significant for all analyses. The analyses were performed using SAS 9.3 (SAS Institute, Cary, NC) and R 3.2.2 (R Foundation for Statistical Computing, Vienna, Austria).
Accession number(s). Whole-genome sequencing files have been deposited in the NCBI Sequence Read Archive (PRJNA341404) under accession numbers SAMN05722982, SAMN05722983,

SUPPLEMENTAL MATERIAL
Supplemental material for this article may be found at http://dx.doi.org/10.1128/ mSphere.00261-16. Table S1, DOCX file, 0.02 MB. Figure S1, TIF file, 0.2 MB. Figure S2, PDF file, 1.1 MB. Figure S3, PDF file, 0.6 MB. Text S1, DOC file, 0.1 MB. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funders.

ACKNOWLEDGMENTS
We thank the team of curators of the Institut Pasteur MLST system (Paris, France) for importing novel alleles, profiles, and/or isolates. M.A.B. thanks Michelle Hatto, Vicki Larson, and Pratik Shrestha for assistance with collection and identification of K. pneumoniae isolates.