Emergence of Polymyxin Resistance in Clinical Klebsiella pneumoniae Through Diverse Genetic Adaptations: A Genomic, Retrospective Cohort Study

Abstract

Background

Polymyxins are antimicrobials of last resort for the treatment of carbapenem-resistant Enterobacteriaceae, but resistance in 5% to >40% isolates has been reported. We conducted a genomic survey of clinical polymyxin-resistant (PR) Klebsiella pneumoniae to determine the molecular mechanisms of PR and the role of polymyxin exposure versus transmission in PR emergence.

Methods

We included 88 patients with PR K. pneumoniae from 2011–2018 and collected demographic, antimicrobial exposure, and infection data. Whole-genome sequencing was performed on 388 isolates, including 164 PR isolates. Variant calling and insertion sequence detection were performed, focusing on key genes associated with PR (mgrB, crrAB, phoPQ, and pmrAB). We conducted phylogenetic analyses of key K. pneumoniae multi-locus sequence types (ST258, ST17, ST307, and ST392).

Results

Polymyxin exposure was documented in 53/88 (60%) patients prior to PR detection. Through an analysis of key PR genes, we detected 129 individual variants and 72 unique variant combinations in PR isolates. This included multiple, distinct changes in 36% of patients with serial PR isolates. Insertion sequence disruption was limited to mgrB (P < .001). Polymyxin minimum inhibitory concentrations showed stepwise increases with the number of PR genes affected (P < .001). When clusters containing PR isolates in ≥2 patients were analyzed, 10/14 had multiple genetic events leading to PR.

Conclusions

Molecular mechanisms leading to PR in clinical K. pneumoniae isolates are remarkably heterogenous, even within clusters or individual patients. Polymyxin exposure with de novo PR emergence led to PR in the majority of patients, rather than transmission. Optimizing polymyxin use should be a key strategy in stopping the spread of PR.

Polymyxin resistance (PR) is an emerging threat. We noted a multiplicity of genomic alterations leading to PR in clinical Klebsiella pneumoniae, within both individual patients and isolate clusters. Polymyxin exposure, rather than transmission, was key in the spread of PR.

(See the Editorial Commentary by Doi and Duin on pages 2092–4.)

Polymyxins (including colistin and polymyxin B) are considered antimicrobials of last resort for the treatment of carbapenem-resistant Enterobacteriaceae (CRE) infections. Despite their toxicity, they have been increasingly used in the last decade, leading to concern about the emergence of polymyxin resistance (PR) [1]. Studies focusing on CRE have observed disturbingly high rates, ranging from 5% to >40%, with PR Klebsiella pneumoniae accounting for the majority [1]. Laboratory PR testing is not routinely performed on all gram-negative bacilli, potentially leading to underdiagnoses and the undetected spread of PR [2]. To complicate matters further, plasmid-mediated forms of PR (mcr genes) have been noted since 2015 [3]. While the spread of mcr to CRE has been limited, it underlines the importance of PR as a public health priority.

Despite this emerging threat, how PR develops and spreads in K. pneumoniae remains incompletely understood. Exposure to polymyxins is considered a common risk factor; however, some patients develop PR without exposure [4–6]. Early data on the clinical factors contributing to PR focused on outbreaks [7–9], but it is not known to what extent in-hospital spread contributes to PR outside of outbreak settings. Furthermore, our understanding of the underlying molecular causes of PR has been limited, but has important implications for improving diagnoses and treatments of PR. Whole-genome sequencing (WGS) now provides an unprecedented opportunity to address these questions. To date, previous epidemiologic studies have incorporated limited genomic data [4–6], while genomic studies have included limited numbers of isolates, often with a paucity of clinical metadata [10–15].

In order to address these gaps, we conducted a comprehensive genomic survey of clinical PR K. pneumoniae isolates from 2011–2018. Our central hypothesis was that PR in K. pneumoniae is a complex genetic trait with a previously unrecognized diversity of PR-conferring variants in clinical isolates. Secondly, we hypothesized that exposure to polymyxins, rather than in-hospital transmission, leads to the development of PR in K. pneumoniae.

METHODS

Study Population and Isolate Selection

The study was reviewed and approved by the Columbia University Irving Medical Center Institutional Review Board. Polymyxin B is the polymyxin on formulary at our institution and polymyxin B dosing details are included in the Supplementary Methods. We systematically reviewed all PR K. pneumoniae isolates in an institutional collection spanning 2011–2018 at a health-care system with a tertiary care center and an affiliated community hospital. Patients with PR K. pneumoniae isolates available in this collection were included and all available K. pneumoniae isolates in these patients were analyzed, regardless of polymyxin susceptibility (Supplementary Figure 1).

To provide a genomic context, we included clinical isolates, as well as isolates collected in our study of multidrug-resistant bacteria in liver transplant recipients [16]. These isolates were selected if they belonged to multi-locus sequence types (MLSTs) and their single locus variants were present in PR isolates cultured from ≥2 patients (ST258, ST17, ST307, and ST392).

Clinical data were extracted from the electronic medical record of each patient, including demographics, comorbidities, and antimicrobial exposure. For each patient, the first PR isolate was assessed as being associated with colonization or infection, according to National Healthcare and Safety Network criteria [17]. Crude mortality at 7 and 30 days was recorded and compared to an unmatched cohort of patients from the same time period who had polymyxin-susceptible, carbapenem-resistant, Klebsiella pneumoniae infections, as identified in an institutional database.

Susceptibility Testing

All isolates had antimicrobial susceptibility testing using Vitek2 (bioMerieux; 2011–2017) and Microscan Walkaway (2017–2018). Isolates were initially screened with a polymyxin B Etest performed during routine clinical care. We then performed minimum inhibitory concentration (MIC) determinations with broth microdilution on all isolates and used these as the definitive polymyxin susceptibility data in all analyses. Testing was done according to Clinical and Laboratory Standards Institute guidelines [18]. In vitro PR was defined as an MIC >2 mg/L [18, 19].

Genomic Definitions and Analyses

Firstly, we focused on key genes described in multiple previous studies (crrAB, mgrB, phoPQ, and pmrAB) [10–15, 20]. Henceforth, these genes will be referred to as canonical PR genes. We also conducted a search in other genes implicated in the literature, which can be considered a “secondary resistome” [1] (Supplementary Table 1).

We performed WGS on all included isolates, as described previously [16, 21, 22]. Sequences were deposited under accession number PRJNA557275. SRST2 was used for MLST and resistance determinant detection [23]. There was 1 isolate from each key MLST that had MinION (Oxford Nanopore) long-read sequencing and assembly with Unicycler [24]. For details of variant calling, phylogenetic analyses, and insertion sequence detection, see the Supplementary Methods. The functional impact of SNVs on the protein sequence was predicted using PROVEAN [25].

Statistical Analysis

Categorical variables were compared using χ ² or Fisher’s exact tests and continuous variables were compared using the Student’s t-test or Mann-Whitney-Wilcoxon, as appropriate. An analysis of variance was used for the comparison of means between multiple groups. Statistical analyses were performed in R (v3.4.0).

RESULTS

Study Population and Clinical Characteristics

We identified 665 patients with carbapenem-resistant K. pneumoniae and 106 patients with PR K. pneumoniae over the 7-year study period. Of these, 88 patients had PR isolates available and were included in the study (Supplementary Figure 1). Approximately half of the patients were admitted from home and 22% were solid-organ transplant recipients (Table 1). The median time from admission to detection of the first PR K. pneumoniae isolate was 21 days (interquartile range [IQR] 1–37 days). In 61/88 (69%) patients, the first PR K. pneumoniae isolate was associated with a clinical infection (Table 2), most frequently of the respiratory tract (29/61, 48%). Following the detection of PR K. pneumoniae, 7-day and 30-day all-cause mortality rates were 13% and 32%, respectively, which increased to 15% and 38%, respectively, if the first PR K. pneumoniae isolate was associated with an infection. When compared with an unmatched cohort of 66 patients with polymyxin-susceptible infections, PR infections were associated with all-cause mortality at 7 days (9/61 vs 1/66 patient, respectively; P = .0069), but not at 30 days (23/61 vs 18/66 patients, respectively; P = .29).

Table 1.

Clinical Characteristics of Cohort

Clinical characteristic	(N = 88)
Median age (IQR)	61 years (18–90 years)
Male gender	53 (60%)
Origin
Community	42 (48%)
Hospital	23 (26%)
Nursing facility	23 (26%)
Comorbidities
Diabetes mellitus	25 (28%)
Pulmonary disease	29 (33%)
Liver disease	10 (11%)
Malignancy	17 (19%)
HIV	4 (4.5%)
Solid organ transplant recipient	19 (21%)
Median Charlson Comorbidity Score (IQR)	4 (2–5)
Outcomes
Crude 7-day mortality	11 (13%)
Crude 30-day mortality	28 (32%)

Abbreviations: HIV, human immunodeficiency virus; IQR, interquartile range.

Table 2.

Details of Infection Associated With First Polymyxin-resistant Isolate

Infection Details	Number of Patients
Patients with infection with first polymyxin resistant isolate	61
Site of infection
Urinary tract infection	15 (25%)
Respiratory tract infection	16 (26%)
Ventilator-associated pneumonia	13 (21%)
Skin and soft tissue infection	3 (4.9%)
Intra-abdominal infection	7 (12%)
Multiple sites	5 (8%)
Bloodstream infection associated with primary infection	13 (21%)
Outcome
Death at 7 days	9 (15%)
Death at 30 days	23 (38%)

Patients received combination therapy for most PR infections (45/61, 74%; Supplementary Figure 2), with tigecycline-containing regimens predominating (35/61 patients, 57%). Receipt of a polymyxin-containing regimen was associated with significantly lower clinical response rates (13/34 vs 19/24 patients not on a polymyxin-containting regimen; P = .0048). There were 5 patients (8%) that received a ceftazidime-avibactam–containing regimen initially and 6 others that received ceftazidime-avibactam during follow-up. Ceftazidime-avibactam resistance subsequently developed in 4/11 patients.

Exposure to Polymyxins

Most patients (53/88, 60%) had polymyxin exposure prior to the detection of PR, receiving intravenous (IV) polymyxin for a median of 12 days (range 1–66). While no patients received nebulized colistin alone preceding PR, 11 patients received it with IV polymyxin for a median of 12 days (range 1–49). There were 3 patients that only had exposure to topical formulations of polymyxin prior to the detection of PR. For the 42 patients admitted from home, 27/42 (64%) had been admitted to our institution in the prior year and 23/42 (55%) had received polymyxins. Following the detection of PR, 53/88 patients (60%) had ongoing polymyxin exposure (median 11 days, range 1–76 days).

Isolate Characteristics

For 88 patients with PR K. pneumoniae, we analyzed and sequenced 164 PR K. pneumoniae isolates and 97 polymyxin-susceptible isolates (median 2 per patient, IQR 1–4). An additional 127 polymyxin-susceptible, MLST-matched K. pneumoniae isolates from 89 patients were included for phylogenetic studies. MICs of PR isolates ranged from 4 to >128 mcg/ml (Supplementary Figure 3). In patients with infections caused by the first PR K. pneumoniae isolate, neither 7-day nor 30-day mortality were associated with higher MICs (P = .95 and P = .075, respectively).

All isolates were resistant to multiple classes of antibiotics (Supplementary Table 2), including near-universal carbapenem resistance (377/388, 97%) with bla_KPC-2/3 being the putative cause in 361/377 isolates (97%). We also detected bla_NDM-1 (3/377 isolates) and bla_OXA-48 (2/377 isolates). Tigecycline resistance was noted in 25/148 (17%) PR isolates.

ST258 was the most common isolate (301/388 isolates in 145 patients, 78%), with ST17, ST307, and ST392 being other key sequence types (Supplementary Table 3). Serial isolates were available in 56 patients. Polymyxin-susceptible and PR isolates were often isolated almost concurrently, being detected within 7 days of each other in 18/56 patients (32%). When they were of same sequence type, the median pairwise SNV distance between these isolates was 2.5 SNVs (IQR 0–22.5 SNVs). However, multiple K. pneumoniae sequence types were noted in 8/56 (14%) patients during follow-up, while 6 patients (10.5%) were colonized with different ST258 clades.

Genomic Determinants of PR

Isolates exhibited a remarkable multiplicity of genomic alterations leading to PR. While variants in mgrB were most frequently detected, non–mgrB mediated PR was present in 67/164 (41%) isolates (Figure 1). In total, we detected 129 individual variants in canonical PR genes, of which 83 were noted in PR isolates, with 67/83 (82%) not described previously (Supplementary Table 4). We detected 72 genetic combinations to PR if each combination of variants in canonical PR genes is considered an individual pathway (Figure 2). On a patient level, 20/56 (36%) patients with serial PR K. pneumoniae isolates available were noted to have >1 genetic combination leading to PR. In these patients, ongoing exposure to polymyxin and/or nebulized colistin may have contributed and was noted in 18/20 patients (90%). No isolates harbored mcr genes.

Figure 1.

Summary of variants in canonical polymyxin resistance genes in Klebsiella pneumonia. The variants in polymyxin resistance canonical genes depended upon the gene affected and clonal background of isolates. Insertion sequences often disrupted mgrB, but with the exception of pmrB in 1 isolate, did not affect other canonical genes. Abbreviation: IS, insertion sequence.

Figure 2.

Combinations of genetic alterations leading to polymyxin resistance based on canonical polymyxin resistance gene profile. A heterogeneity of pathways to polymyxin resistance was noted, with 72 unique combinations in 164 polymyxin resistant isolates. To illustrate this, each unique variant in each polymyxin resistance canonical gene was represented by a different color. Specifics of variants encountered can be found in Supplementary Table 4.

Within this diversity, there were important differences between the canonical genes associated with PR. IS were almost exclusively associated with mgrB (40/97 isolates with variants in mgrB vs 1/79 isolates with variants in other genes; P< .001; Figure 1). CrrB was an important contributor (28/164 isolates, 17%), but only 8 isolates had variants in crrB without any concurrent variants in other PR canonical genes. Variants in multiple canonical PR genes were detected in 39/164 (24%) PR isolates and were associated with a significant difference in the mean MIC when isolates were categorized by the number of PR canonical genes affected (Supplementary Figure 4; P< .001). The increasing duration of polymyxin exposure weakly correlated with variants in an increasing number of PR canonical genes (Spearman’s ρ = 0.27; P = .0047), and also with an increasing MIC (Spearman’s ρ = 0.26; P = .0052). Variants in crrA, phoP, and pmrA were infrequently observed (5, 4, and 11 isolates, respectively).

In 15/164 (9%) PR isolates, no variants were detected in canonical PR genes. In our screen of the putative secondary resistome [1] (Supplementary Table 5), we noted 2 variants unique to this group: a V82A variant ramA and IS disruption of ompW, in separate isolates. Despite this screening approach, 13 isolates without an obvious contributor to PR remained.

Variants in PR canonical genes were also noted in 54/218 (25%) polymyxin-susceptible isolates (Supplementary Table 4), highlighting the need for studies to confirm functional impacts. In polymyxin=susceptible isolates, phoQ was the gene most frequently affected (36/218 isolates, 17%; P< .001; Supplementary Figure 5). Importantly, 15/129 (12%) variants in canonical PR genes were noted in both PR and polymyxin-susceptible isolates, suggesting the restoration of polymyxin susceptibility (eg, in isolates with IS in mgrB) or nonfunctional variants with PR resulting from other contributors.

Population Structure of Polymyxin Resistance

We conducted phylogenetic analyses of ST258, ST17, ST307, and ST392 (Figure 3). For ST258, PR isolates were present across the phylogeny, with no predisposition towards a single clade or large clusters suggestive of a generalized PR outbreak. We detected 22 distinct clusters with multiple patients in ST258 and 1 cluster in ST392. There were 17 clusters containing PR isolates, and 14 had ≥2 patients with PR isolates (Supplementary Table 6). In each cluster, we then compared PR genetic combinations between isolates to differentiate the de novo emergence of PR from transmission (Figure 4). In 10/14 clusters, there were PR isolates with different genetic combinations leading to PR, suggesting that PR emerged independently within these isolates from the same genomic background. However, 6 clusters had possible transmission with PR isolates from different patients who had the same genetic combination leading to PR, implicating 16 patients and 24 isolates in total, with a median pairwise SNV distance of 6 (IQR 1–9). Of these patients, 9/16 (56%) had polymyxin exposure prior to culture of the PR isolate.

Figure 3.

Phylogenetic analysis of polymyxin resistance in key multi-locus sequence types. Polymyxin-susceptible isolates were denoted with a light blue tip point, whereas polymyxin-resistant isolates had dark blue tip points. Other metadata were shown with 3 different bands. Band 1 was used to indicate different clusters of closely related isolates that contained polymyxin-resistant isolates. Band 2 used different colors to represent patients these isolates came from, with each color representing an individual patient. Band 3 showed isolates that were associated with the likely transmission of polymyxin resistance.

Figure 4.

Variants in canonical polymyxin resistance locus according to clusters of closely related isolates. Most clusters of closely related isolates had multiple combinations of genetic alterations leading to polymyxin resistance. Each color indicates an individual variant in the noted canonical polymyxin resistance gene.

DISCUSSION

Through the integration of genomic and clinical data, our study provides novel insights into how PR develops and spreads in a setting where carbapenem-resistant K. pneumoniae is endemic. Genomics allowed us to detect a multitude of genetic alterations leading to PR. This diversity occurred even within closely related clusters of isolates or individual patients, where we found evidence for the frequent de novo emergence of PR. Polymyxin exposure was detected in the majority of patients and may have been a driver of this phenomenon. The transmission of PR was uncommon, suggesting that, in most cases, PR arises either after polymyxin exposure or sporadically, rather than through clonal spread. These data provide a comprehensive overview of the clinical genomics of PR and highlight its heterogeneity, with important implications for diagnostics and infection prevention and for control of these extensively drug-resistant organisms.

While many forms of antimicrobial resistance result from alterations in or the acquisition of single genes [26], we noted not only a remarkable diversity of variants in PR canonical genes, but also complex interplay between them. In at least one-quarter of PR isolates, multiple hits may have contributed to PR, with the acquisition of variants in different PR canonical genes having an additive effect to increase polymyxin MICs. The duration of polymyxin exposure may have contributed to this and weakly correlated both with increasing MICs and the number of PR canonical genes with variants present. Disruptions of mgrB play an important role in PR [12]; however, more than 40% of isolates have no alterations in mgrB, suggesting additional pathways to PR.

Despite comprehensive genomic screening, no putative genetic cause of PR was identified in approximately 10% of isolates. To address this, we examined genes that have been proposed to constitute a secondary resistome [1] and found potential candidate variants in ramA and ompW, which require functional validation. Although beyond the scope of this study, future avenues for identifying novel PR determinants include analyzing the plasmid content of these isolates and conducting transcriptomic analyses. Previous studies noted the increased expression of transcripts when traditional WGS did not find mutations present, particularly in heteroresistant Enterobacter isolates [27, 28].

Conversely, we noted variants in PR canonical genes in polymyxin-suceptible isolates. Although functional validation is necessary for each variant, a PROVEAN analysis predicted many to not have a deleterious effect on function. In addition, there were instances where the same variants were seen both in PR and polymyxin-susceptible isolates, raising the possibility of additional changes that suppressed PR, as noted previously in phoPQ [14].

While our findings confirmed previous work regarding the disruption of mgrB by IS [12, 14, 29], it was striking how infrequently other PR genes were affected by IS disruption. The underlying reasons for this predilection of IS for mgrB remain to be determined, but may be related to mgrB’s role as a constitutive suppressor, with inactivation leading to the upregulation of downstream targets leading to PR [1]. Some of this disruption by IS may be due to the transfer of IS from plasmids, and could offer a novel path to plasmid-mediated PR, in addition to mcr genes [30].

While the transmission of PR isolates has been invoked as a possible cause of PR in patients without polymyxin exposure [4, 6], this was an uncommon occurrence in our study: it accounted for PR in only 16/88 patients (18%). This suggests that in the majority of patients, PR arose after polymyxin exposure in isolates from a similar genomic background. The diversity of combinations of PR canonical gene alterations we found within clusters (Figure 4) provides further evidence of this.

Polymyxin exposure was likely a major contributor to the emergence of PR and was noted in the majority of our patients, in contrast to previous reports [4, 6]. Ongoing, selective pressure from polymyxin use may also explain why we saw distinct variants in PR canonical genes arise within clusters and individual patients. Interestingly, non-IV formulations may play a role in PR developing or being maintained. We found only exposure to topical polymyxin in 3 patients prior to the development of PR, and nebulized colistin was commonly used, although usually in conjunction with IV therapy.

There are several limitations to this work. Firstly, it was observational and based at a single center. Polymyxin-susceptibility testing remains a challenge, even when performing broth microdilution [31]. This is of particular concern for isolates with MICs close to the breakpoint, as this impacts susceptibility classification in downstream analyses. Despite our analyses, in 28/88 (32%) patients we did not find evidence of polymyxin exposure or transmission. This may reflect polymyxin exposure outside our institution or, alternately, the undetected transmission of PR due to silent carriers of PR. There are also inherent limitations in genomic analyses: variant calling may have been affected by the selection and quality of our references. IS detection is difficult and our approach relied upon the draft de novo assembly of each isolate.

Taken together, our findings reveal PR to be a remarkable challenge on multiple fronts. The detection of PR is a crucial first step, and our study should alert physicians that PR may emerge rapidly or exist concurrently with polymyxin-susceptible isolates, particularly in the setting of polymyxin exposure. The phenotypic diagnosis of PR remains difficult and, while there is interest in predicting antimicrobial resistance phenotypes from genotypes [32], the polygenic nature of PR noted in our study may temper that enthusiasm. This underscores the importance of developing reliable phenotypic assays. Several are in development, including polymyxin broth disk elution, rapid colorimetric tests, and screening agars [33, 34].

Our study also provides a roadmap for preventing the emergence of PR in K. pneumoniae: polymyxin exposure and de novo emergence of resistance may be the major contributors in a nonoutbreak setting, rather than transmission. We can therefore redouble antimicrobial stewardship efforts. The inappropriate use of polymyxins should obviously be the primary target, including non-IV formulations of polymyxins. This may be of particular relevance in the intensive care unit, where colistin is the most frequently used nebulized antibiotic and topical polymyxins are frequently used for decolonization [35, 36]. Low-dose polymyxin exposure has also previously been linked to the emergence of resistance [20]; therefore, therapeutic drug monitoring of polymyxins offers another opportunity, which is being investigated [37]. While novel β-lactam/β-lactamase inhibitor combinations are a welcome addition in the treatment of CRE infections, they are also limited by the emergence of resistance during treatment [21]. Polymyxins will, therefore, remain essential components of our CRE armamentarium and need to be preserved.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. The authors thank Mia Scholl for assistance in performing mgrB polymerase chain reaction.

Disclaimer. The funders had no role in the study design, data collection and interpretation, or decision to submit the work for publication.

Financial support. This work was supported by the National Institute of Allergy and Infectious Diseases, National Institutes of Health (grant numbers AI116939 to A.-C. U., S1-AI116939 to A.-C. U. and A. G.-S., and 5T32AI100852 and K23 AI137316 to A. G.-S.) and the Irving Scholarship from Columbia University (to A.-C. U.).

Potential conflicts of interest. N. M. has received research support from GlaxoSmithKline, unrelated to the current study. A.-C. U. has received research funding from Merck, GlaxoSmithKline, and Allergan, unrelated to the current study. All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.