Carbapenem-Resistant Klebsiella pneumoniae Exhibiting Clinically Undetected Colistin Heteroresistance Leads to Treatment Failure in a Murine Model of Infection

ABSTRACT Antibiotic resistance is a growing crisis and a grave threat to human health. It is projected that antibiotic-resistant infections will lead to 10 million annual deaths worldwide by the year 2050. Among the most significant threats are carbapenem-resistant Enterobacteriaceae (CRE), including carbapenem-resistant Klebsiella pneumoniae (CRKP), which lead to mortality rates as high as 40 to 50%. Few treatment options are available to treat CRKP, and the polymyxin antibiotic colistin is often the “last-line” therapy. However, resistance to colistin is increasing. Here, we identify multidrug-resistant, carbapenemase-positive CRKP isolates that were classified as susceptible to colistin by clinical diagnostics yet harbored a minor subpopulation of phenotypically resistant cells. Within these isolates, the resistant subpopulation became predominant after growth in the presence of colistin but returned to baseline levels after subsequent culture in antibiotic-free media. This indicates that the resistance was phenotypic, rather than due to a genetic mutation, consistent with heteroresistance. Importantly, colistin therapy was unable to rescue mice infected with the heteroresistant strains. These findings demonstrate that colistin heteroresistance may cause in vivo treatment failure during K. pneumoniae infection, threatening the use of colistin as a last-line treatment for CRKP. Furthermore, these data sound the alarm for use of caution in interpreting colistin susceptibility test results, as isolates identified as susceptible may in fact resist antibiotic therapy and lead to unexplained treatment failures.

including urinary tract, bloodstream, and soft tissue infections (3). Carbapenemresistant Klebsiella pneumoniae (CRKP) is one of the carbapenem-resistant Enterobacteriaceae (CRE), an emerging cause of antibiotic-resistant, health care-associated infections. CRE were listed as one of the most urgent antibiotic resistance threats by the CDC and WHO (4,5). In part due to the difficulty of effectively treating infections with CRE, mortality rates can be as high as 40 to 50% (6). These infections are a worldwide problem, with recent reports indicating that CRE are widespread in the United States (7), Europe (8), and China (9). Unfortunately, resistance to "last-line" drugs, such as colistin, is emerging in CRKP strains, and in some cases, isolates are resistant to all antibiotics tested (10). Here, we describe the identification of two multidrug-resistant CRKP isolates exhibiting colistin heteroresistance, a phenomenon in which only a minor subpopulation of genetically identical cells is phenotypically resistant. Since the frequency of the resistant subpopulation is exceedingly low in these isolates, they are not detected as being colistin resistant by clinical diagnostic tests.
Two CRKP urine isolates (Mu9 and Mu156) were collected from different patients in Atlanta, GA, area hospitals as part of the Multi-site Gram-Negative Surveillance Initiative (MuGSI), a nationwide surveillance network for CRE hospital isolates. Isolates were grown from single colonies and frozen at Ϫ80°C prior to the study. Mu9 and Mu156 were confirmed as being genetically distinct by pulsed-field gel electrophoresis (using XbaI-digested total DNA and separated by electrophoresis on a Chef-DR III apparatus [Bio-Rad Laboratories, Hercules, CA] at 200 V [6 V/cm] with a 90-s switch time for 23 h) (data not shown). Both isolates were resistant to nearly all antibiotics tested, including all carbapenems and some aminoglycosides (see Table S1 in the supplemental material). PCR for various resistance genes revealed several beta-lactamases in each isolate, including Klebsiella pneumoniae carbapenemase (KPC) in both strains (Table S2). Colistin susceptibility testing by broth microdilution (11) in cation-adjusted Mueller-Hinton (MH) broth (BD Biosciences, Franklin Lakes, NJ) using colistin sulfate (Sigma-Aldrich, St. Louis, MO) (MIC of 0.5 g/ml) and the colistin Etest (BioMérieux, Marcy-l'Étoile, France), performed on MH agar (Remel, San Diego, CA) (MIC ϭ 0.125), classified both strains as susceptible to colistin (Fig. 1A). Subsequent examination for susceptibility was performed via population analysis profile (PAP) by plating serial dilutions of bacteria on MH agar (BD Biosciences, Franklin Lakes, NJ) containing various concentrations of colistin. PAP revealed the presence of a minor colistin-resistant subpopulation in each isolate that actively grew on antibiotic up to a concentration of 100 g/ml. The colistinresistant subpopulation was present at between 1 in 1,000 and 1 in 1,000,000 CFU at doses of colistin ranging from 2 to 100 g/ml (Fig. 1B). In contrast, PAP demonstrated that all the cells of a colistin-susceptible control isolate, GA65146, were killed by 2 g/ml of colistin (Fig. 1B).
It was concerning that broth microdilution and the Etest were unable to detect the colistin-resistant subpopulations in these isolates. The recommended incubation time for both tests is 24 h (11,12). Extension of the incubation time to 48 h resulted in the accurate identification of resistance by broth microdilution (Fig. S1A), likely because the minor resistant subpopulation had more time to grow out. In contrast, the increased incubation time had no effect on Etest results, which remained negative (Fig. S1B).
We next studied the dynamics of the resistant subpopulation following colistin treatment. After treatment with 100 g/ml colistin, the frequency of the resistant subpopulation was significantly increased in each isolate. Subsequent passage in an antibiotic-free medium greatly decreased the frequency of the resistant subpopulation ( Fig. S2), suggesting that this population was phenotypically resistant and not the result of a stable genetic mutation. Additionally, this suggests that there is some disadvantage to maintaining a majority colistin-resistant subpopulation. Indeed, it has been previously shown that colistin resistance in K. pneumoniae confers a fitness defect (13). To directly assess whether the resistant and susceptible subpopulations were genetically homogenous, we isolated cultures with majority resistant or susceptible subpopulations by subculturing them in medium containing 16 g/ml colistin or drug-free medium, respectively (Fig. 1C). This resulted in cultures containing Ͼ95% colistin- Undetected Colistin Heteroresistance in CRKP ® resistant cells or Ͼ95% colistin-susceptible cells (Fig. 1C). We then performed genomic sequencing on both populations using an Illumina HiSeq 4000 sequencer for a depth of coverage of Ͼ1,000ϫ, revealing that the resistant and susceptible subpopulations were indeed genetically identical, consistent with heteroresistance.
To investigate the phenotypic differences between the resistant and susceptible subpopulations, we quantified the expression of two genes in the PhoPQ twocomponent system pathway, which is known to mediate colistin resistance in K. pneumoniae. MgrB is a negative regulator of PhoPQ signaling (14), and mgrB expression was lower in resistant cells cultured in colistin than in susceptible cells grown in drug-free medium (Fig. 1D). Additionally, expression of phoP, which is autoinduced when the PhoPQ system is active (15), was increased in colistin-resistant cells of Mu9 compared to its expression in susceptible cells, although this was not observed in Mu156 (Fig. 1E). Taken together, these data are consistent with involvement of the PhoPQ pathway in the resistant subpopulations of both Mu9 and Mu156.
It was unclear whether the minor colistin-resistant subpopulations present in these isolates would have an effect on the outcome of colistin treatment during an in vivo infection. To assess the in vivo relevance of the colistin-resistant subpopulations, we used a mouse model of peritonitis (done in accordance with IACUC protocol #4000046). We infected mice (C57BL/6J; Jackson Laboratory, Bar Harbor, ME) intraperitoneally with a lethal dose (3 ϫ 10 8 CFU) of either of the heteroresistant K. pneumoniae isolates and subsequently left the mice untreated or treated them with colistin after 12 h (20 mg colistin methanesulfonate/kg of body weight [Chem Impex, Wood Dale, IL], given intraperitoneally every 6 h) to simulate infection and subsequent treatment upon clinical presentation. Interestingly, even in the absence of colistin, the frequency of the resistant subpopulations of both heteroresistant isolates increased following 24 h of in vivo infection compared to the frequency produced by the inoculum (Fig. 2A). This may be due to cross-resistance of these cells to host innate immune antimicrobials, such as antimicrobial peptides and reactive oxygen species, as has previously been demonstrated (16). We next assessed the impact of heteroresistance on colistin treatment outcome. Mice infected with the colistin-susceptible strain (GA65146) succumbed to infection in the absence of antibiotic but were rescued by colistin treatment (Fig. 2B). In contrast, mice infected with either of the heteroresistant isolates (Mu9 or Mu156) were unable to survive the infection, even in the presence of colistin ( Fig. 2C and D). These data strikingly demonstrate that colistin heteroresistance can lead to in vivo colistin treatment failure for CRKP.
Concluding remarks. This is the first report of colistin-heteroresistant K. pneumoniae in the United States. In highly resistant CRE isolates, colistin is a vital last-line treatment option. We show here that in a mouse model of infection, colistinheteroresistant CRKP isolates fail colistin therapy. This stresses the need to assess the relevance of colistin heteroresistance on the outcome of colistin therapy in human infection, which has yet to be determined.
When highly resistant CRKP strains are isolated in the clinic, testing of last-line antibiotics identifies crucial treatment options. Colistin-heteroresistant isolates, such as the ones reported here, can be misclassified as colistin susceptible, a "very major discrepancy" according to FDA susceptibility testing guidelines (17). Subsequent treatment of these isolates with colistin may then lead to unexplained treatment failure, as was demonstrated in our in vivo mouse model. Thus, the misclassification of colistin susceptibility status wastes critical time and resources and may lead to further infection complications and patient mortality. Clinical laboratories should consider testing for heteroresistance to colistin if this last-line antibiotic is required for treatment. Unfortunately, the current standard test for heteroresistance, the population analysis profile, is time-and labor-intensive, and it is cumbersome for most clinical laboratories to implement. Our findings suggest that broth microdilution with an increased incubation time (48 h) may detect colistin heteroresistance. However, the increased incubation time is a downside in itself, and there is also an increased chance that a culture of susceptible bacteria will become contaminated or that de novo mutant cells will have the time necessary to grow out, leading to an inaccurate identification as resistant. Therefore, novel diagnostics that rapidly and accurately detect colistin heteroresistance are needed.
Taken together, these findings serve to sound the alarm about a worrisome and underappreciated phenomenon in CRKP infections and highlight the need for more sensitive and accurate diagnostics. We suggest that clinical microbiologists and clinicians alike use caution when treating CRKP infections with colistin.