Biotyping of Multidrug-Resistant Klebsiella pneumoniae Clinical Isolates from France and Algeria Using MALDI-TOF MS

Background Klebsiella pneumoniae is one of the most important pathogens responsible for nosocomial outbreaks worldwide. Epidemiological analyses are useful in determining the extent of an outbreak and in elucidating the sources and the spread of infections. The aim of this study was to investigate the epidemiological spread of K. pneumoniae strains using a MALDI-TOF MS approach. Methods Five hundred and thirty-five strains of K. pneumoniae were collected between January 2008 and March 2011 from hospitals in France and Algeria and were identified using MALDI-TOF. Antibiotic resistance patterns were investigated. Clinical and epidemiological data were recorded in an Excel file, including clustering obtained from the MSP dendrogram, and were analyzed using PASW Statistics software. Results Antibiotic susceptibility and phenotypic tests of the 535 isolates showed the presence of six resistance profiles distributed unequally between the two countries. The MSP dendrogram revealed five distinct clusters according to an arbitrary cut-off at the distance level of 500. Data mining analysis of the five clusters showed that K. pneumoniae strains isolated in Algerian hospitals were significantly associated with respiratory infections and the ESBL phenotype, whereas those from French hospitals were significantly associated with urinary tract infections and the wild-type phenotype. Conclusions MALDI-TOF was found to be a promising tool to identify and differentiate between K. pneumoniae strains according to their phenotypic properties and their epidemiological distribution. This is the first time that MALDI-TOF has been used as a rapid tool for typing K. pneumoniae clinical isolates.


Introduction
Klebsiella pneumoniae are ubiquitous in nature and have two common habitats; one is the environment, including surface water, sewage, soils and plants [1], and the other is mammalian mucosal surfaces [2]. In humans, K. pneumoniae can be present in the intestinal tract, nasopharynx, and on the skin [3]. It is one of the most common Gram-negative bacteria encountered by clinicians worldwide as a cause of infections in humans [4] and is responsible for outbreaks due to the propagation of different clones associated with opportunistic infections in individuals with impaired immune defenses, such as diabetics, alcoholics and hospitalized patients with indwelling devices [3]. In the hospital environment, the principal reservoirs for K. pneumoniae transmission are blood products, contaminated medical equipment, the gastrointestinal and respiratory tracts of patients and the hands of hospital personnel [5]. The hospital-acquired infections caused by this organism mainly include pneumonia, septicemia, urinary tract infections and soft tissue infections [6].
Increased K. pneumoniae infections are also associated with an increase in multidrug-resistant (MDR) strains, especially those producing extended-spectrum beta-lactamases (ESBLs) [7] associated with the prior use of antibiotics, particularly the cephalosporins [8]. Furthermore, several carbapenemase-encoding genes have been described in K. pneumoniae species, including class A betalactamase KPC, class B beta-lactamases NDM, IMP and VIM, and class D beta-lactamase OXA-48 [9]. The hospital epidemiology of these infections is often complex because multiple clonal strains causing focal outbreaks may co-exist with sporadic strains that also have a reservoir in the community [10]. Infections caused by multidrug-resistant K. pneumoniae strains have been associated with adverse clinical outcomes, including increased mortality, prolonged hospital stays and increased economic costs [11].
Therefore, epidemiological typing is useful in determining the extent of an outbreak and in investigating the sources, the reservoir and the spread of bacterial infections. Various methods, including protein profiling by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and DNA profiling by multilocus sequence typing (MLST), restriction fragment length polymorphism (RFLP) and pulsed field gel electrophoresis (PFGE), have been used for the epidemiological typing of K. pneumoniae isolates [12][13][14]. However, most of these methodologies are time consuming, laborious, require special skills and are unsuitable for use in routine clinical laboratories [15,16]. In recent years, several reports have shown the feasibility of using matrix-assisted laser desorption ionization (MALDI)-time of flight (TOF) mass spectrometry (MS) to rapidly identify microorganisms [17]. There are only a few studies that have evaluated this method as a rapid tool to classify bacterial species at the strain level [18,19].
However, there are some recent examples of the use of MALDI-TOF MS for the rapid identification and typing of a limited number of clinical strains, such as Streptococcus pyogenes [20] or Klebsiella pneumoniae [19].
Here, we report the evaluation of MALDI-TOF MS as a rapid and powerful tool for determining the epidemiological distribution of a large series of K. pneumoniae clinical strains of different origins from patients with various infectious syndromes and the correlation between the pathotypes, geographic locations and clonalities of these strains using MALDI-TOF MS and data-mining approaches.

Clinical Data
The mean age of the infected patients was 53 years and was similar when comparing patients hospitalized in Algerian hospitals (53 years: range 12 days to 84 years) and those from French hospitals (53.1 years: range 1 day to 97 years). The male/female

Bacterial Identification
Using MALDI-TOF MS, all K. pneumoniae strains (100%) were identified with score values .1.9 using the Bruker Biotyper software.

Data Mining Analysis of Klebsiella pneumoniae MSP Dendrogram
The MSP dendrogram revealed five distinct clusters according to an arbitrary cut-off at the distance level of 500 ( Figure 1). Data mining analysis of the five clusters using PASW 17.0 software showed that K. pneumoniae strains isolated in Algerian hospitals (Tlemcen, Sidi Bel Abbes, Oran, Annaba) were significantly associated with respiratory tract infections and ESBL phenotype in the fifth cluster (p,0.0001), whereas K. pneumoniae strains isolated in Marseille hospitals were significantly associated with urinary tract infections and wild type phenotype in the first, the second and the fourth clusters (p,0.0001). K. pneumoniae strains isolated in Angers hospitals were associated with urinary tract infections and wild type phenotype in the fourth cluster (p,0.0001). Conversely, K. pneumoniae strains isolated in Nice hospital were associated with blood cultures and pus samples and wild type phenotype in the third cluster (p,0.0001) ( Table 3). All the details of the distribution of K. pneumoniae strains into the five clusters according to the   dendrogram are given in supplementary Table S1. Interestingly, clustering the strains according to the arbitrary distance levels of 180 and 100 significantly clustered strains from the same hospital into the same cluster. At the distance level of 180, 34 distinct clusters were identified, and at the distance level of 100, 52 distinct clusters were identified (Table S2). For example, at the distance level of 100, one cluster containing 15 strains showed that all of them were from Marseille hospital, a second cluster contained three strains (all from Nice hospital), and a third cluster contained ten strains (eight out of ten were from Angers hospital (p,0.0001)).

Multilocus Sequence Typing
The MLST allelic profile of all strains distributed in the five clusters is presented in Table 4. From the 28 tested strains, we found 24 different sequence types (ST) (Figure 2). No relationship was observed between ST and the clinical and epidemiological data.

Discussion
K. pneumoniae is an important pathogen with a complex pangenome responsible for serious nosocomial infections, especially in intensive care units (ICUs) and in wards for surgery, emergency, neurology, pediatrics, and neonatology [5]. More concerning has been the emergence and increase in the isolation rate of ESBLproducing K. pneumoniae worldwide [4], which frequently possess resistance factors to other classes of antibiotics, notably aminoglycosides, fluoroquinolones and trimethoprim/sulfamethoxazole [21]. In our study, we confirmed that there is an increase in the number of multidrug-resistant K. pneumoniae strains with a considerable variation between the two countries studied. The antibiotic resistance rate in France (26.8%) was lower than that in Algeria (88.6%). In comparing our results with those of the Mystic study (1997)(1998)(1999)(2000)(2001)(2002)(2003) [7] and another surveillance trial study (2004)(2005)(2006)(2007)(2008)(2009) [22], we notice a worldwide north-south gradient evolution of ESBL production rate in K. pneumoniae strains with 12.3-12.8% in North America, 16 The high infection occurrence of K. pneumoniae, both ESBLproducing and non-producing strains, in Algerian and French hospitals pushed us to examine the epidemiology of these strains, which are considered to have a complex pan-genome containing plastic genome repertoires that differentiate strains according to their geographical locations, pathotypes, ecotypes and resistance phenotypes [23]. MALDI-TOF MS was successfully used as a tool for biotyping because we found specific clusters that were significantly associated with particular phenotypes from different clinical and geographical sources and from different seasons. This result is seemingly supported by Trevino et al., with a series of only 13 clinical isolates of K. pneumoniae [24]. By increasing the number of strains collected, our dendrogram became more refined, not only by country but also by hospital. This result could be of clinical importance for unknown pathogenic isolates, whose geographical sources could be detected rapidly using MALDI-TOF MS. It is recognized that the distribution of bacteria may be related to geographical patterns, such as climatic zones and movement of human populations, and can provide information about pathogen evolution and transmission [25,26]. The spatial distribution of bacterial pathogens can be considered at different levels; in a hospital setting, this may indicate nosocomial transmission, but in a community setting, the simultaneous appearance of an identical phenotype in widely dispersed locations may be a warning of an outbreak [15]. Our data also suggested that rates of K. pneumoniae infections varied seasonally and were significantly associated with periods of isolation that were due to changes in temperature and humidity. Several characteristics of this species that were previously described support our findings, which show that temperature and dew point were both linearly predictive of increasing rates of K. pneumoniae clinical isolates [27,28]. Furthermore, most of the MALDI-TOF MS spectra are composed of wellconserved proteins with housekeeping functions that are minimally affected by environmental conditions and thus are considered to be optimal for the proteomic typing of bacteria, which is considerably different from genetic typing [29,30].
By comparing the clusters obtained from the MSP dendrogram ( Figure 1) with the ST distribution in the minimal spanning tree (MST) of K. pneumoniae strains (Figure 2), we noted that no correlation was observed for these two types of analysis. Moreover, no relationship was observed between ST, clinical and epidemiological data. These results suggest that the two approaches, comprising the MSP and MLST analyses, highlight two different bacterial aspects. Indeed, MLST analysis based on the conserved bacterial genes (the house-keeping genes) classifies bacteria according to their core genome, which represents less than 10% of the genome [31], while 90% of the genome is composed of ''accessory genes'' (mobile genetic elements) that can be lost or acquired by lateral gene transfer (LGT), and that is mainly responsible for the bacterial phenotype [23,32]. Therefore, this core genome does not represent the majority of expressed proteins, in contrast to the MSP dendrogram analysis, which is based on the functional and expressed proteins of whole cells that is more representative of the global phenotype [33]. Therefore, these two approaches could be complementary to the extent that the MST approach, based on the analysis of the core genome, allows us to classify bacteria according to their conserved genes independent of their ''accessory genes,'' which can affect phenotypes and bacterial classification. Many researchers have used molecular methods, particularly PFGE and MLST, to distinguish K. pneumoniae clinical isolates in order to understand transmission patterns and to aid in the management of these infections [34]. In 2008, a comparative study between these two methods found that PFGE is appropriate to discriminate among epidemiologically unrelated strains and appears more suitable for short-term epidemiology, while MLST is appropriate for strain phylogeny and large-scale epidemiology [35]. Furthermore, these molecular methods are costly and timeconsuming to obtain results and introduce delays when attempting to limit the spread of K. pneumoniae clones [36]. For instance, only a few hours are required to obtain results by MALDI-TOF MS, whereas several days are necessary to collect MLST data [37]. In addition, the cost of MALDI-TOF MS instrumentation is comparable to that of a sequencing machine, but running costs and consumables are considerably lower than for these methods [38]. Barbuddhe [20].
Thus, MALDI-TOF MS combined with a statistical classification strategy is appropriate for studies of local epidemiology and global population structure when compared to a local database from clinical strains. It is a powerful epidemiological method and is sufficiently reproducible and sensitive enough to rapidly survey the evolution of existing or emerging phenotypes with reduced financial and human costs [39].

Conclusion
We believe that our preliminary results should be expanded and confirmed with more strains obtained from different countries. To the best of our knowledge, this is the first study that analyzes the epidemiology of a large series of K. pneumoniae clinical isolates using MALDI-TOF MS application. We suggest the creation of a local database that is updated regularly to survey for the presence of abnormal phenotypes at the strain level. If a particular phenotype is detected, a real time genome sequencing approach could then be performed to investigate the origin and specific features of the strain. We believe that this methodology could be used routinely in clinical microbiology laboratories as a surveillance tool for hospital epidemiology studies to prevent outbreaks and dissemination of pathogens in hospital settings.  Table 5.

Klebsiella pneumoniae Identification using MALDI-TOF MS
Isolates were plated on Trypticase Soy Agar (BioMerieux) and incubated for 24 h at 37uC. One single colony from each isolate was deposited on a MALDI-TOF MTP 384 target plate (Bruker Daltonics, Bremen, Germany) in four replicates to minimize random effects. Two microliters of matrix solution (saturated acyano-4-hydroxycinnamic acid, 50% acetonitrile, 2.5% trifluoroacetic acid) were then added and allowed to co-crystallize with the sample. Analysis was performed in a MALDI-TOF MS spectrometer (337 nm) (Autoflex; Bruker Daltonics) with FLEX control software (Bruker Daltonics). Ions were accelerated in the positive ion mode with an accelerating voltage of 20 kV. The pulsed extraction of ions was optimized for 1000 Da. The software employed, Bruker Biotyper 2.0 (Bruker Daltonics), automatically acquired spectra with fuzzy control of the laser intensity and analyzed them by standard pattern matching against the spectra of 2881 species used as reference data. After comparing the unknown spectra with all reference spectra in the database, the log scores were ranked. Values of .1.9 were required for secure identification at the species level, and values between 1.9 and 1.7 were required for secure identification at the genus level.

Clustering of MALDI-TOF Spectra
A consensus spectrum was produced, and an MSP dendrogram was constructed using the correlation distance measure with the average linkage algorithm setting of the Biotyper 2.0 software. Clusters were then detailed and analyzed according to arbitrary distance levels at 500, 180 and 100.

Antibiotic Resistance Phenotypic Classification of K. pneumoniae Strains based on Beta-Lactamine Compounds
We considered a wild type phenotype as a strain that confer resistance to aminopenicillins, carboxypenicillins and to ureidopenicillins. The high level penicillinase phenotype was presented by a high penicillinase activity responsible for resistance to aminopenicillins and their inhibitors, to carboxypenicillins, to ureidopenicillins, and to first generation cephalosporins (1 GC). The inhibitor-resistant TEM penicillinase phenotype included resistance to aminopenicillins, carboxypenicillins, and ureidopenicillins. It was distinguished by resistance to aminopenicillins and carbocxypenicillins associated to the beta-lactam inhibitors. 1 GC generally retain their activity. The cephalosporinase phenotype corresponded to a marked resistance to penicillins, 1 GC, 2 GC, and to at least one 3 GC. The extended spectrum B-lactamase phenotype includes resistance to penicillins and cephalosporins except cephamycins. The resistance to 3 GC and 4 GC was more or less pronounced depending on the enzymes and the strains.

Statistical Analysis
Clinical and epidemiological data were recorded in an Excel file (Microsoft, Redmond, WA, USA), including the clustering obtained using the MSP dendrogram generated by the Biotyper software (version 2; Bruker Daltonics), and were analyzed using PASW Statistics software version 17.0 (SPSS Inc., Chicago, IL, USA). Dependent variable series were analyzed using Expert Modeler, which automatically generates the best-fitting model. The chi-square analysis was used also to compare proportions using the same software, and P values ,0.05 were considered to be statistically significant. Statistical analyses were conducted using Epi Info version 6 (Centers for Disease Control and Prevention, Atlanta, GA, USA).

Multilocus Sequence Typing
Randomly, 28 strains distributed in the five clusters were selected to perform the full MLST as described previously using seven housekeeping genes, including gapA, infB, mdh, pgi, phoE, rpoB and tonB [34]. The MLST database for K. pneumoniae can be found at http://www.pasteur.fr. Based on the allelic profiles, the evolutionary relationship between isolates was assessed by the minimal spanning tree (MST) algorithm in Bionumerics (Applied Maths, NV St-Martens-Latem, Belgium). A stringent definition of 6/7 shared alleles was used to define clonal complexes (single locus variants only).