Klebsiella pneumoniae bacteria are normal gastrointestinal flora but, particularly in people with weakened immune systems and/or debilitating diseases, can also cause severe infections whose management has been complicated by increasing antimicrobial resistance over the last 20 years. Global surveillance studies carried out during the 2000s showed that 20–80% of K. pneumoniae isolates were resistant to first-line antibiotics, including the cephalosporins, fluoroquinolones and aminoglycosides [1–3]. However, the recent worldwide spread of strains that are resistant to carbapenems is even more worrying because these antibiotics are often the last line of effective treatment of the infections caused by multidrug-resistant K. pneumoniae [4]. Various carbapenem-resistance mechanisms have so far been identified, but the most frequent is related to the production of K. pneumoniae carbapenemase (KPC), the expression of which is largely due to the plasmid-derived blaKPC gene, although some cases of chromosomally integrated blaKPC genes have been described in the literature [5–7].

The currently few treatment options for carbapenem-resistant K. pneumoniae (CRKP) infections include polymyxins, tigecycline, aminoglycosides and fosfomycin, but their use is limited by concerns about their efficacy and safety [8], and there are increasing reports of resistance to colistin and tigecycline [9, 10]. In the absence of aggressive infection control, CRKP can easily and rapidly spread, which explains why they have been responsible for outbreaks of infection within and between healthcare facilities [11]. Since 2010, Italy has seen a remarkable increase in the occurrence of CRKP, which greatly contributed to the epidemic spread of carbapenem-resistant Enterobacteriaceae (CRE) recorded by a countrywide cross-sectional survey, the results of which were published in 2013 [12].

Methods of discriminating and characterising different K. pneumoniae isolates are essential in order to be able to direct the targeting of infection control resources. Traditional systems based on phenotypes (serotypes, biotypes or antibiograms), and molecular methods such as multi-locus sequence typing (MLST) and repetitive extra-genic palindromic polymerase chain reaction (REP-PCR) have long been used, but may not provide sufficient resolution because of the high degree of clonality of K. pneumoniae [13]. Following improvements in sequencing technologies, whole-genome sequencing (WGS) is now well placed to become the gold standard for bacterial typing as it can define the complete genomic structure of a pathogen [14, 15] which, in the case of K. pneumoniae, is one circular chromosome and a variable number of small and large plasmids, the primary sources of virulence and resistance [16–18].

We fully sequenced 68 K. pneumoniae strains isolated at Luigi Sacco University Hospital-ASST Fatebenefratelli Sacco (Milan, Italy) that had previously been tested for drug susceptibility and carbapenemase phenotyping, and investigated by means of REP-PCR. In order to compare and integrate the microbiological, phenotypic and genotypic data with the in silico and whole-genome analyses, we used web tools and bioinformatics software to characterise the genotypic, resistance and virulence traits of K. pneumoniae molecularly. Starting from the WGS data, we defined the sequence type (ST) strains using in silico MLST (the gold standard for Klebsiella genotyping), determined correlations among the samples by analysing the core single nucleotide polymorphisms (SNPs), and extracted molecular data that are important for clinical and epidemiological purposes.

Klebsiella pneumoniae infection is an imminent threat to human health because of its virulence, its increasing resistance to antibiotics, and its ability to spread silently in human hosts who can act as carriers [34].

CRKP bacteria are some of the most challenging antibiotic-resistant pathogens involved in nosocomial infections worldwide because they greatly complicate the therapeutic management of hospitalised patients, thus leading to high morbidity and mortality rates, and also increasing hospital costs [35]. The global CRKP-related risk is even more worrying because of the growing number of strains that are resistant to last-resort drugs such as colistin and tigecycline [9, 10].

CRKP have rapidly become endemic in Italy, and there is a considerable likelihood of infectious outbreaks [36–38]. The relatedness of isolates during hospital outbreaks has traditionally been deduced on the basis of antimicrobial susceptibility patterns, crude microbiological methods, or molecular typing techniques such as REP-PCR and MLST, which has limited our knowledge of their genetic diversity within a single hospital [13].

WGS is becoming the gold standard in bacterial typing as it provides fine resolution and is highly discriminant in characterising isolates. Continuous technological advances, improvements in turnaround times, and the accessibility of DNA sequencing techniques are now approaching a stage at which genomic data can be rapidly generated in order to facilitate our understanding of the spread and evolution of infectious agents within a clinically useful time frame [39, 40].

The aim of this study was to analyse the alarming and prolonged outbreak of KRCP infection that occurred between January 2012 and December 2014 in different wards of Luigi Sacco University Hospital-ASST Fatebenefratelli Sacco, Milan, Italy. Traditional methods were used to investigate the antimicrobial susceptibility and the clonal relationships of the CRKP strains, and this was complemented by WGS data and the results of an in silico analysis that allowed MLST, the characterisation of resistome, virulome, plasmid content, and core SNP genotyping.

Antibiotic susceptibility testing of the CRKP strains circulating in our hospital showed the emergence of multi-drug resistance characterised by a worrying increase in tigecycline and colistin resistance. These data were confirmed by the in silico detection of the presence of genes conferring resistance to β-lactam antibiotics, aminoglycosides, fluoroquinolones, phenicol, sulfonamide, tetracyclines, trimethoprim and macrolide-lincosamide-streptogramin. In line with the findings of an Italian epidemiological survey, β-lactamase characterisation of our CRKP isolates showed the predominance of carbapenemase production, and that the most frequently detected variants of the blaKPC gene were blaKPC-2 and blaKPC-3. The blaSHV, blaTEM, blaCTX-M and blaOXA genes were also detected. However, we were unable to establish any clear correlations between the genetic background and the most complicated multifactorial tigecycline and colistin resistance mechanisms. The observed SNPs and gene losses, mutations and deletions were not exclusive to the tigecycline-resistant strains as defined by microbiological methods, and determining whether the resistance was mutation related would have required complementary experiments and the evaluation of expression levels. Moreover, although comparative genomic analyses in Italy and Spain have demonstrated that colistin resistance in K. pneumoniae clinical isolates following exposure to colistin may be related to the insertional inactivation of the mgrB gene [17, 37], this was found in only two of our colistin-resistant isolates (Kp19 and Kp37).

The co-existence or co-evolution of antibiotic resistance and virulence factors is one of the most worrying possibilities as it could lead to the emergence of untreatable invasive K. pneumoniae infections. K. pneumoniae strains can produce virulence factors such as fimbrial adhesions, and capsule and iron uptake systems that enhance colonisation. The expression of adhesins is particularly important during the colonisation stage, when mechanical forces such as peristalsis and salivary secretions hamper bacterial invasion of the host [41]. Type 3 fimbriae have been identified as accessories mediating the formation of biofilms on biotic and abiotic surfaces (e.g. catheters), and antimicrobial drug resistance can increase the likelihood of bacterial cells existing in the biofilm by up to one thousand times [42]. The presence of type 3 fimbriae encoded by the mrk cluster in all of our clinical isolates (39.7% involving complicated urinary infections) supports the importance of this virulence trait in the pathogenesis and spread of CRKP.

Capsules can also play an important role in Klebsiella persistence inside and outside human hosts by resisting complement-mediated lysis or phagocytosis, and offering some protection against environmental desiccation; they may also have a neutralising effect on antibodies as a result of the release of excessive capsular material [41]. However, we did not detect the presence of any of the well-known hypervirulent capsular serotypes (i.e. K1, K2 and K5) in our isolates, which supports the view that multidrug-resistant strains and hypervirulent clonal complexes do not currently overlap [43].

Iron scavenging is important during infections because hosts have little free iron under physiological conditions [44]. Interestingly, our finding of the Yersinia pestis HPI in all but one of the ST258 isolates from the 12 patients with complicated urinary tract infections, and in one ST512 isolate from a patient with septicemia, suggests that these pathogens had adapted to their hosts very well. It should be noted that the chromosomal rather than plasmidic location of the virulence genes is a pathogenic strategy because chromosomal genes are retained by replication even in the absence of selective pressure, and this reduces the likelihood of the gene being lost to the bacterial population [18]. The pathogenicity of K. pneumoniae is further increased by the additional and easy acquisition of β-lactamase encoding genes although, ultimately, the successful evolution of the infection also greatly depends on a number of host-dependent factors [45].

Together with some specific characteristics of the bacterial strain, plasmid replicon content is currently used as a marker for bacterial typing during epidemiological investigations because it is the primary source of gene variability [46]. K. pneumoniae can have multiple replicon types, which have been classified into discrete incompatibility (inc) groups or families on the basis of the inability of closely related plasmids to stably propagate within the same bacterial strain [47]. Whole-genome sequencing using short-read technologies (e.g. Illumina) has become cheap and accessible, but some limitations relating to de novo assembly prevent the accurate reconstruction of the genomic context surrounding repeated sequences in plasmids. However, comparison of our 68 Klebsiella genomic sequences with those in the PlasmidFinder database allowed us to detect the plasmids often associated with antimicrobial resistance in clinically relevant bacterial pathogens and relate them to the sequences currently included in the database. The variability in plasmid structure and content among strains isolated in the same hospital suggests that additional plasmid types and arrangements may occur, and close surveillance is required.

The genotyping results of the in silico MLST analysis showed that, as in a number of other countries, the epidemic spread of CRKP in Italy is mainly due to hyper-epidemic clonal complex 258. However, increasing clone diversity has been reported in the literature [48], and our samples included three ST307 strains and one ST101, in addition to the ST512, ST258, ST745 and ST1519 strains.

The identification of the whole-genome core SNP set allowed us to obtain a higher-resolution strain tracking and discrimination than those provided by automated REP-PCR, which is currently used in our hospital to investigate bacteria clonal relationships. The REP-PCR and MLST results did not completely correlate: the REP-PCR group 2 and larger group 3 contained all of the ST258 strain and its variants ST745 and ST1519, but also ST512 (Kp7), ST307 (Kp56) and ST101 (Kp27). The phylogenetic core SNP tree showed that the four major lineages correlated with the MLST results. Among the isolates with the same ST, it was often possible to identify common genetic traits associated with virulence (the Yersinia pestis HPI in all but one ST258 strain) and antibiotic resistance (ST258_blaKPC-2 and ST512_blaKPC-3 isolates, and the common mgrB gene disruption in two ST512 colistin- resistant strains). However, the phylogenetic reconstruction obtained from the core genome SNPs, which correlated well with the genome background of related isolates, was not fully corroborated by conventional epidemiological data.

The effective tracing and tracking of outbreak isolates in order to control the spread of K. pneumoniae requires a detailed understanding of possible transmission routes. In healthcare settings, Klebsiella bacteria can spread as a result of person-to-person contact or, less frequently, environmental contamination. Unfortunately, the retrospective nature of this study means that it has some limitations in terms of the acquisition of the patients’ epidemiological and clinical data. For example, it was often impossible to find details concerning previous hospitalisations in other facilities or previous antibiotic treatments in the patients’ medical records, or establish the degree of environmental contamination during a patient’s stay in our hospital. Consequently, our analysis hypothesised possible patient-to-patient transmission events only on the basis of blaKPC gene correlations between the isolates of patients who were in the same ward for an overlapping period. In the A2a sub-lineage, the Kp1 and Kp5 samples were isolated from two patients in the infectious diseases ward within a period of just a few days. In the A2b sub-lineage, two ST512 isolates (Kp8 and Kp10) from patients simultaneously admitted to the general surgery ward, two isolates (Kp24 and Kp28) from patients in the medicine ward, and two isolates (Kp61 and Kp62) from patients in the infectious diseases ward were considered to be involved in person-to-person transmission. Two further strains (Kp3 and Kp7), which grouped closely in the B branch of the core SNP tree, were recovered from patients simultaneously staying in the surgery ward. These results only partially correlated with the REP-PCR data, which showed a very close genetic correlation only in the case of two isolate pairs (Kp8/Kp10 and Kp24/28).

In conclusion, our findings show that whole-genome analysis can accurately reconstruct the molecular characterisation of isolates, including resistome, virulome and plasmid replicon content, and determine the phylogenetic relationships among strains circulating in a single hospital. The future clinical implementation of real-time sequencing could transform the surveillance and management of problematic nosocomial pathogens by overcoming the resolution limitations of analysing only a small portion of the genome. The latest techniques may also reduce working times and allow the prompt administration of the best treatment, thus avoiding the emergence of new antibiotic-resistant strains. Finally, it may be possible to identify the genetic variations underlying changes in bacterial virulence or tissue specificity, and genomic surveillance of these key phenotypic attributes may improve our understanding of the pathogenicity of multi-resistant agents and control its evolution.