Forensic genomics of a novel Klebsiella quasipneumoniae type from a neonatal intensive care unit in China reveals patterns of colonization, evolution and epidemiology

Abstract

During March 2017, a neonatal patient with severe diarrhoea subsequently developed septicaemia and died, with Klebsiella isolated as the causative microorganism. In keeping with infection control protocols, the coincident illness of an attending staff member and three other neonates with Klebsiella infection triggered an outbreak response, leading to microbiological assessment of isolates collected from the staff member and all 21 co-housed neonates. Multilocus sequence typing and genomic sequencing identified that the isolates from the 21 neonates were of a new Klebsiella sequence type, ST2727, and taxonomically belonged to K. quasipneumoniae subsp. similipneumoniae (formerly referred to as KpIIB). Genomic characterization showed that the isolated ST2727 strains had diverged from other K. quasipneumoniae subsp. similipneumoniae strains at least 90 years ago, whereas the neonatal samples were highly similar with a genomic divergence of 3.6 months. There was no relationship to the Klebsiella isolate from the staff member. This demonstrates that no transmission occurred from staff to patient or between patients. Rather, the data suggest that ST2727 colonized each neonate from a common hospital source. Sequence-based analysis of the genomes revealed several genes for antimicrobial resistance and some virulence features, but suggest that ST2727 is neither extremely-drug resistant nor hypervirulent. Our results highlight the clinical significance and genomic properties of ST2727 and urge genome-based measures be implemented for diagnostics and surveillance within hospital environments. Additionally, the present study demonstrates the need to scale the power of genomic analysis in retrospective studies where relatively few samples are available.

Data Summary

This Whole Genome Shotgun project has been deposited at DDBJ/ENA/GenBank under accession numbers JAAOBC000000000 to JAAOBX000000000, BioProject ID PRJNA610124. All bioinformatic protocols used to process the genomic data are available at https://github.com/vjlab/KpIIB_ST2727.

Impact Statement.

Colonization of the neonatal gut is a complex process with contributions from the mother’s microbiome and environmental sources. In this study, 21 neonates housed in a single room were sampled across 22 days for the carriage of Klebsiella in response to an outbreak-status event. To forensically dissect the details of this case, we generated whole genome data for all bacterial isolates and carried out a detailed genomic characterization of virulence and antimicrobial resistance. A major outcome from this retrospective, deep, sequence analysis is the detection of colonization of the neonates by an emergent multilocus sequence type, ST2727, strain of K. quasipneumoniae subsp. similipneumoniae from a neonatal intensive care unit (NICU) environment, underscoring the need for both genomic surveillance and preventative measures in NICU environments.

Introduction

Species of Klebsiella are widespread in the environment, found in soil and ground-water, as commensal organisms on plants and are carried widely in animal hosts [1]. Klebsiella pneumoniae is recognized as a leading cause of healthcare-associated urinary tract infections, surgical site infections and pneumonia, increasingly threatening neonates and the immunocompromised [2]. However, in recent years, an increasing diversity of Klebsiella species have been identified to cause infection in humans [3, 4]. Genome sequencing has revealed that bacteria historically classified as K. pneumoniae belonged to the K. pneumoniae species complex containing multiple phylogenetically distinct but closely related bacterial species. These include K. pneumoniae (previously classified as phylogroup KpI), K. quasipneumoniae subsp. quasipneumoniae (previously KpIIA) and subsp. similipneumoniae (previously KPIIB), K. variicola (previously KPIII) [5], as well as two strains forming divergent lineages indicating potentially novel species within this species complex [6]. Whole genome sequencing (WGS) has shown that the predominant cause of clinical infection worldwide has been K. pneumoniae (KpI) [5] but there is an increasing trend, and increasing concern, in the detection of severe cases due to the newly described species [7]. Different studies report K. quasipneumoniae isolates causing infections resistant to treatments with ceftazidime and other oxyimino-betalactam antibiotics, and widespread carriage of genes encoding carbapenemase (KPC) [3, 4, 8–11].

With the emergence of a greater diversity of Klebsiella causing severe infections, their ability to colonize environmental niches such as sinks and ventilators in hospitals [12], increased detection of hybrid strains with drug-resistant and hypervirulent (hv) phenotypes, especially in regions such as China where the earliest reports of hvKp originated about three decades ago [13], an early warning system for the detection of Klebsiella species is being promoted in hospitals in China [14]. Routine surveillance is conducted using multilocus sequence typing (MLST) and followed by antimicrobial and virulence testing to provide an avenue to investigate potential outbreaks and develop better health control measures.

Here we report a retrospective study of an incident that occurred in March 2017, in the neonatal intensive care unit (NICU) of a major hospital in China with >3000 beds. A neonate with severe diarrhoea subsequently developed septicaemia and died, with Klebsiella isolated as the causative microorganism from a blood sample. During the same period, severe symptoms of diarrhoea were found in three other neonatal patients as well as an attending NICU staff member, triggering a response to quell the apparent outbreak. In addition to intensified hygiene procedures, the response included MLST and microbiological antimicrobial resistance (AMR) screening of stool samples from patients (21 neonates) and the staff member. Preliminary microbiological characterization of the samples revealed that 20 of the isolates taken from the infants belonged to a new MLST designation, ST2727, and one infant was infected with ST477. The sample from the staff member was of ST23 type, a well-defined hypervirulent clonal group frequently detected in severe cases of infection, whereas little was known of ST2727 or ST477. To forensically dissect the details of this case, we generated whole genome data for all bacterial isolates and carried out a detailed genomic characterization of virulence and AMR. Using a globally sampled dataset comprising 3611 genomes we elucidate the overall genetic diversity, evolution and epidemiology to illuminate elements of nosocomial Klebsiella transmission and ecological colonization that impact generally on understanding hospital-acquired infections.

Methods

Collection of clinical isolates and clinical data

Klebsiella was isolated from the blood sample taken prior to death of the neonate, and the stool samples collected from all 21 infants in the NICU and a staff member who suffered an episode of enteritis. These isolates were characterized using a Vitek II system (bioMérieux). Relevant clinical characteristics of the patients and staff were taken from their medical records, including demographic characteristics, underlying medical conditions, clinical manifestations, treatments and outcomes. The study was approved by the Ethics Committee of the First Affiliated Hospital of Wenzhou Medical University (China).

Molecular genotyping and sequence typing of isolates

Bacterial isolates were characterized by PFGE after digestion of genomic DNA samples with XbaI under the following conditions: temperature 14 °C, voltage 6 V cm^–1, pulse angle 120°, and pulse duration of 5–35 s for 18 h, using a well-characterized XbaI digest (Takara Bio) of Salmonella enterica serotype H9812 as a molecular marker. DNA digest patterns were analysed and interpreted according to Tenover et al. [15]. MLST was carried out by PCR-amplifying seven housekeeping genes: gapA, infB, mdh, pgi, phoE, rpoB and tonB [16], and the data were used to assign sequence types (STs) using the Institute Pasteur MLST database (https://bigsdb.pasteur.fr/). Through this process, the new ST2727 was established. Additionally, a standard multiplex-PCR was used to confirm that the isolates did not contain any of the five virulent extracellular polysaccharide capsules (CPS) (K antigen status; K-type) K1, K2, K20, K54 and K57.

Bacterial genome sequencing and assembly

DNA was extracted using a Biospin Bacterial Genomic DNA extraction kit (Hangzhou Bioer Technology). Libraries were prepared using the TruePrepTM DNA Library Prep Kit V2 for Illumina (Vazyme). Briefly, the DNA sample was fragmented and tagged with adapters using a single transposase enzymatic reaction, followed by amplification using an optimized, limited-cycle PCR protocol and indexing. Individual libraries were assessed on the QIAxcel Advanced Automatic nucleic acid analyser, and quantified through quantitative PCR using a KAPA SYBR FAST qPCR Kit. Illumina Next Generation Sequencing technology (Hiseq PE150) was used to generate a paired-end library of short read sequences (150 nt) for each isolate. Sequence quality controls were performed using FastQC version 10-01-18 and trimmed and refined using Fastx toolkit version 0.0.13. De novo assembly was performed using SPAdes version 3.13.1 [17]. Scaffold generation and ordering were performed using the MeDuSa version available in July 2018 and Mauve version 2.4.0, respectively [18, 19]. Closely related genomes [NTUH-K2044 (Kp9), CP027602 (Kp10), HKUOPLC (Kp1–8, Kp11–22)] from NCBI were identified based on genome identity using MagicBlast v1.5 [20]. Upon assembly and alignment of all ST2727 genomes, single nucleotide variations (SNVs) were inferred using the Parsnp tool from Harvest Suite version 1.1.2 using a reference Kp1 isolate genome [NCBI accession numbers JAAOBC000000000 to JAAOBX000000000; Table S1 (available in the online version of this article)] [21].

Genomic analysis

Genes contributing to drug resistance and virulence were detected using Kaptive version available in July 2018 and Kleborate version 0.3.0 [22, 23]. VRprofile version 2.0 was used to detect gene clusters characteristic of protein secretion systems of type 3 (T3SS), type 4 (T4SS) or type 6 (T6SS) [24]. The presence of genes encoding a type 2 secretion system (T2SS) was detected by blast version 2.7.1, using as queries known T2SS genes [25]. Phylogenetic analysis was conducted using 3611 Klebsiella genomes downloaded from the NCBI database (downloaded on 20 June 2018) to find the closest related genomes to the novel strains (Data S1). Core genome alignment of Klebsiella genomes (including ST2727 isolates) was generated using Roary version 3.11.2. Excluding poorly assembled genomes, the final dataset constituted 3611 genomes comprising 3251 genes that were in at least 99 % of the sequences [26]. Phylogenetic relationships were estimated using the maximum-likelihood (ML) method in RAxML version 8.2.12 [27] using the General Time Reversible (GTR) nucleotide substitution model with a gamma (Γ) distribution of among-site rates. Branch support was estimated using an ML bootstrap analysis with from 10 to >1000 replicates for the different datasets analysed. Additionally, the SNVs found in the core genome of all ST2727 isolates were concatenated to reconstruct a phylogenetic tree using the program SplitsTree v.4. and a BioNJ algorithm [28].

K. quasipneumoniae subsp. similipneumoniae (KpIIB) genomes were used to determine the evolutionary rate of this clade. To confirm that the genome datasets contained sufficient genetic change between sampling times, necessary for the reconstruction of the timescale of evolution, we used a root-to-tip regression of sampling years against genetic diversity in TempEst v1.5 [29], optimizing the best fit for the root to maximize the determination coefficient, R ². The slope of the regression was positive, showing that the genomic data reflect temporal signal, so a molecular clock was estimated using a least-squares dating (LSD version 0.3) method [30] with 1000 samples for the confidence interval. The reliability of the analysis was confirmed using a permutation test in which the sampling years were randomized for 100 replications, with a Z-score test [31].

To detect signatures of natural selection in the ST2727 outbreak samples, we used a McDonald–Kreitman Test (MKT) implemented in the R package PopGenome [32]. The MKT compares the ratio of non-synonymous to synonymous mutations within strains (i.e. polymorphisms) (P _n/P _s) to the ratio of variation between strains (i.e. divergence) (D _n/D _s). A neutrality index (P _n/P _s)/(D _n/D _s)>1 is indicative of negative selection and >1 is characteristic of positive selection [33]. The MKT was performed for the gene and the whole genome datasets independently.

Results

Clinical scenario and diagnosis

This study was initiated when outbreak status was called in an NICU due to a fatal case of a neonatal patient presenting with symptoms of respiratory and gastrointestinal infections. Despite treatment with meropenem, vancomycin, cefoperazone and sulbactam, the neonate died from early-/late-onset neonatal septicaemia, septic shock, gastrointestinal haemorrhage and multiple organ dysfunction syndrome. A blood sample from the neonate tested positive for K. pneumoniae . In the days after the fatal case, 20 more neonates were sampled along with an NICU staff member who presented with symptoms of infection (Fig. 1c). According to a Vitek II system (see Methods, Table S2), stool samples collected from the 21 neonates co-housed in the single-room NICU and the staff member tested positive for K. pneumoniae . The neonates presented with a wide range of symptoms: three with respiratory and digestive symptoms, 13 with respiratory disease such as dyspnoea, severe asphyxia, tachypnoea and/or severe cyanosis, one with digestive symptoms alone, while three presented with normal health (Table 1). The staff member presented with digestive symptoms including vomiting and severe diarrhoea. MLST classification using the genes gapA, infB, mdh, pgi, phoE and rpoB showed that 20/21 neonate isolates were novel, revealing a new MLST designation, ST2727, whereas one neonate was of ST477 and the staff member was infected with a distinct and well-characterized hypervirulent K. pneumoniae ST23. This suggested that the fatality was caused by the ST2727, and demonstrated that the staff member was not the cause of the event. Here, we sought to understand the carriage and genomic characteristics of ST2727 in the neonates and test the hypothesis of transmission in this closed scenario.

Fig. 1.

Genomic epidemiology and timeline of K. quasipneumoniae subsp. similipneumoniae. (a) Evolutionary relationships and timescale of evolution of 54 K. quasipneumoniae subsp. similipneumoniae (KpIIB) strains. (b) Phylogenetic tree of the ST2727 isolates collected in the NICU during February and March 2017. The phylogenetic tree was built using the BioNJ algorithm on SplitsTree. Scale bar represents nucleotide substitutions per site. (c) Timeline of sample collection in the NICU, where calendar dates have been replaced by a timeline where the first sample was collected on day 0.

Table 1.

Patient and Klebsiella pneumonia strains

Sample ID	MLST*	Patient	Birth	Symptoms†	Antibiotics	Invasive procedures	Clinical diagnosis
KP1	ST2727	Neonate	Vaginal	Digestive, respiratory	Yes	Indwelling gastric tube (IGT)	Premature, low birth weight (LBW), neonatal septicaemia, septic shock, gastrointestinal bleeding, multiple-organ failure
KP2	ST2727	Neonate	Caesarean	Digestive, respiratory	Yes	Venous catheter (VC)	Premature, LBW
KP3	ST2727	Neonate	Caesarean	Digestive, respiratory	Yes	Intubation	Premature, LBW, hyaline membrane disease, hypocalcaemia
KP4	ST2727	Neonate	Vaginal	Respiratory	No	IGT	Premature, LBW
KP5	ST2727	Neonate	Vaginal	Respiratory	No	IGT	Premature, LBW
KP6	ST2727	Neonate	Caesarean	Respiratory	Yes	IGT	Premature, LBW, congenital infection
KP7	ST2727	Neonate	Caesarean	Respiratory	Yes	VC, artery intubation (AI), artificial respirator, intubation	Premature, LBW, pneumonia
KP8	ST2727	Neonate	Caesarean	Respiratory	No	No	Premature, aspiration of amniotic fluid and meconium syndrome
KP9	ST23	Staff	n/a	Digestive	–	No data	Healthy
KP10	ST477	Neonate	Vaginal	Digestive	Yes	No	Swallowing syndrome of newborn, intestinal infection
KP11	ST2727	Neonate	Caesarean	Respiratory	Yes	VC, artificial respirator intubation	LBW, pneumonia
KP12	ST2727	Neonate	Vaginal	Respiratory	Yes	No	Premature, wet lung of the newborn
KP13	ST2727	Neonate	Caesarean	Asymptomatic	No	No	Premature, LBW, polydactyly, patent foramen ovale
KP14	ST2727	Neonate	Caesarean	Respiratory	Yes	Intubation	Premature, LBW, severe asphyxia, congenital infection
KP15	ST2727	Neonate	Caesarean	Respiratory	No	No	Premature, LBW, wet lung of the newborn
KP16	ST2727	Neonate	Vaginal	Respiratory	No	IGT, nasal catheter, VC	Premature, LBW, neonatal asphyxia, bronchial dysplasia
KP17	ST2727	Neonate	Caesarean	Respiratory	Yes	IGT	Premature, LBW, congenital infection
KP18	ST2727	Neonate	Vaginal	Non-infection	No	IGT	Premature, LBW
KP19	ST2727	Neonate	Vaginal	Asymptomatic	No	No	Premature, LBW
KP20	ST2727	Neonate	Vaginal	Respiratory	Yes	AI, venipuncture	Premature, LBW, intestinal infection
KP21	ST2727	Neonate	Caesarean	Respiratory	Yes	VC	Premature, LBW
KP22	ST2727	Neonate	Caesarean	Asymptomatic	No	No	Premature, polydactyly, patent foramen ovale

*Multilocus sequence typing.

†Symptoms are classified depending of what type of infection they are related to; ‘Non-infections’ suggests the patient has symptoms related to other diseases that are non-infectious, for example symptoms related to cardio-respiratory chronic diseases; and ‘Asymptomatic’ refers to no symptoms at all. ‘Respiratory’ and ‘Digestive’ refer to symptoms related to respiratory infections and digestive infections, respectively.

Genomics-based evaluation of the NICU strains

To determine the genomic characteristics of the bacteria recovered in this study, the complete genomes of all 20 ST2727 isolates were sequenced using a paired-end short read sequencing methodology. The lengths of the assembled genomes ranged from 5 192 865 to 5 202 717 bp that encoded between 4706 and 5191 annotated genes (Table S1). The GC content of all genomes was 57 %. Genomic analysis showed the isolates exhibited high genomic similarity (0.018 % sequence divergence), with 1066 SNVs along their genomes corresponding to an average of one SNV every 2392 bases. The variation observed among the ST2727 samples was at the lower range to the median divergence of 0–0.08 % observed among the same MLST types [5]. The high similarity of the ST2727 strains suggests a clonal population. Phylogenetic analysis using the SNV sites of the core genome identified weak clustering between isolates (Fig. 1b). A branch containing five isolates (Kp1, 3, 14, 16 and 20) could be distinguished from the remaining isolates, but potential transmission patterns could not be deduced due to low branch lengths between the clusters. The tip branches of Kp15 and Kp20 were considerably longer, suggesting evolution subsequent to divergence from other isolates (Fig. 1b, Table S3). Our investigation of SNVs with potential importance for resistance identified two unique polymorphisms (R124Q and T142A) in the carbohydrate porin scrY gene of Kp1 that require further investigation.

Phylogenies generated from a core genome of 3611 globally sampled Klebsiella strains (Fig. S1, Data S1), including isolates from the present study, revealed that the ST2727 strains belong to K. quasipneumoniae subsp. similipneumoniae [34] (Figs 1a and S2). A root-to-tip regression of sampling dates and genetic distance of 54 K. quasipneumoniae subsp. similipneumoniae genomes sampled during 1994–2017, including the ST2727 genomes, showed a positive correlation (correlation coefficient: 0.64) between time and mutation, indicating that the dataset contained sufficient signal for the estimation of a molecular clock. We used the least-squares dating method to estimate an evolutionary rate for the K. quasipneumoniae subsp. similipneumoniae clade of 4.596×10⁻⁵ nucleotide substitutions per site per year [95 % confidence interval (95 % CI) 4.180–4.980×10⁻⁵]. Whilst these estimates are faster than those estimated for representative bacterial orders or families (approximately 10⁻⁷) [35], our estimates are consistent with estimates made independently for Klebsiella clones and other bacterial clades (10⁻³ to 10⁻⁵) [36, 37]. The ST2727 group exhibited a genomic divergence of >90 years [mean time to the most recent common ancestor (TMRCA) 1926; 95 % CI, 1917–1933] to the most closely related isolate collected in 2016 from a patient from Thailand (KPPSTH03; MLST type ST929-1LV) (Fig. 1a).

The TMRCA of all K. quasipneumoniae subsp. similipneumoniae lineages was 1874 (95 % CI, 1861–1886), whereas the mean TMRCA of ST2727 collected within the hospital was approximately 3 months prior to the first case Kp1 (mean TMRCA 2016.94, 95% CI 2016.88–2016.97). This showed a divergence time much greater than the duration of sampling across 22 days in February–March 2017 (Fig. 1c). These results suggest that what was designated as an outbreak was instead the colonization of infants by a community of bacteria that were transmitted to the neonates during this period. Due to the variation present between ST2727 isolates, it is unlikely that the colonization occurred through direct transmission between one neonate and another. We suggest instead that the isolates were introduced to the hospital or the ward in the months prior to the outbreak and were sustained within the hospital during this period. The precise mechanism of colonization of the neonates is not clear without comprehensive sampling of the built environment and the equipment used in the treatment and care of the neonates.

The core genome of the ST2727 clade exhibited positive selection in comparison with strains in the K. quasipneumoniae subsp. similipneumoniae clade [MKT, neutrality index (NI): 0.35, NI<1], indicating that ST2727 isolates are under selection pressure. Gene-wise estimates of selection pressure in the ST2727 clade in comparison with K. quasipneumoniae subsp. similipneumoniae strains showed that 21 genes were under negative selection (NI>1) (Table S4) and 10 were under positive selection (Table S5), whereas mutations in the remaining genes were either strongly deleterious or were under neutral selection pressure [33]. Many of these genes are fundamental to bacterial cell biology and have been functionally characterized in various K. pneumoniae strains, and these functions can be expected to be conserved in K. quasipneumoniae subsp. similipneumoniae . Genes under purifying or negative selection included genes coding for transcription factors, metabolic enzymes, a structural protein and proteins for active transport of solutes (Table S4). These transporter proteins that were found to be under purifying selection conform to a related set of protein families: the ATP-Binding Cassette (ABC) transporters and Major Facilitator Superfamily (MFS) transporters, as well as antiporters and symporters. ABC transporters under negative selection were those designated as responsible for the transport of thiamine, nickel and carbohydrates, which are required for bacterial survival in nutrient-rich environments [38]. MFS transporters uptake nutrients from rich environments using chemiosmotic ion gradients [38], while antiporters and symporters maintain ion homeostasis for these and other functions [38]. Taken together, this suggests that the ST2727 population has adapted for growth in a nutrient-poor environment. Conversely, while the genes under positive selection also include some for active transport of solutes (Table S5), these are transporters responsible for the export of toxins and antibiotics as efflux systems: for example, genes coding for efflux RND transporter permease subunits (drug efflux) and HlyD toxin secretion [38]. This reliance on drug efflux is also consistent with the observation that the structural protein peptidoglycan glycosyltransferase/peptidoglycan dd-transpeptidase (PBP1A) encoded by the gene mrcA was under negative selection: PBP1A catalyses the transglycosylation and transpeptidation of murein, which is a major component of the cell wall, and PBP1A is also the target of beta-lactam antibiotics [38].

O- and K-antigen serogroups of NICU isolates

WGS analysis using Kaptive from Kleborate [22, 39] identified that ST2727 strains have a KL55 capsular type (wzi56) (Table S6). O-typing of the surface lipopolysaccharide revealed all ST2727 were of the O3/O3a serotype, which has a strong adjuvant effect compared with other lipopolysaccharide types [40], but is a rarely reported serotype with no known clinical impact [41]. Genome sequence data revealed that all isolates were wcaG– (Fig. 2a).

Fig. 2.

Phylogenetic distribution of virulence and resistance genes among KPII and KPIII genomes. The presence of virulent and resistance genes is shown in black. Virulent genes are grouped in categories: T2SS (pulA, pulB, pulD, pulE, pulF, pulG, pulI, pulJ, pulK, pulS) adhesion genes (uge, fimH, mrkD), Yers (yersiniabactin), Coli (colibactin), Aerob (aerobactin), Salmo (Salmochelin) and rmpA, rmpA_2 (hypermucoid). Resistant genes are grouped in categories: Gly (glycopeptides), MLS (macrolides), Phe (phenicols), Rif (rifampin), Sul (sulphonamides), Tet (tetracyclines), Tmt (trimethoprim), Bla (beta-lactamases), Bla_Carb (carbapenemase), ESBL (extended spectrum beta-lactamases), inhR (extended spectrum beta-lactamases with resistance to beta-lactamase inhibitors), broad (broad spectrum beta-lactamases), broad.inhR (broad spectrum beta-lactamases with resistance to beta-lactamase inhibitors). Virulent MLST types are highlighted in red.

Genes associated with virulence and AMR

All ST2727 isolates shared three genes considered diagnostic as virulence factors (Fig. 3a, Table S7): uge (uridine diphosphate galacturonate 4-epimerase), fimH (type 1 fimbrial adhesion) and mrkD (type 3 fimbrial adhesion). The gene uge is responsible for the conversion of UDP-GlcA to UDP-GalA and is key for the production of polysaccharides containing GalA residues, a feature which has been shown to promote colonization of human tissues [42]. Fimbriae are an important feature of Klebsiella , and the gene fimH encodes a well-conserved adhesin subunit that is found in around 90 % of Klebsiella strains [43]. This gene has high mobility within clones through horizontal transfer [43]. The gene encoding mrkD is associated with ‘type III’ fimbriae, important for adhesion to promote establishment of bacterial biofilms in harsh environments [44].

Fig. 3.

Major resistance, virulence and capsular genes detected in ST2727. (a) Summary of genes contributing to drug resistance, iron acquisition and other virulence-enhancing features. MIC, minimal inhibitory concentration. (b) Genomic architecture of the T2SS cassette in ST2727 andother K. quasipneumoniae subsp. similipneumoniae . Gene predictions arecolor-coded according to the legend, with black representing genes detected insilico with greater than 80 % sequence identity to the corresponding genesin K. pneumoniae .

Antimicrobial susceptibility testing using minimal inhibitory concentration (MIC) analysis indicated that all isolates were sensitive to all third-generation cephalosporins tested (Table S2). In terms of other AMR, two genes encoding beta-lactamases were detected (Fig. 2a, Table S7): the ubiquitous ampH, responsible for resistance to penicillin G, cefoxitin and cephalosporin C; and bla _OKP-B-10, which is a diagnostic feature of K. quasipneumoniae subsp. similipneumoniae [45].

Secretion systems

Protein secretion systems are key components of virulence in many bacterial pathogens [46], and in Klebsiella the T2SS is the best characterized of these virulence determinants [47]. We identified a unique T2SS architecture in K. quasipneumoniae subsp. similipneumoniae. All ST2727 isolates collected have a similar T2SS architecture, typically composed of pulS and pulA-pulM [48] (Fig. 3b), however a clear homologue of GspN was absent in K. quasipneumoniae subsp . similipneumoniae genomes; GspN is a bitopic, integral membrane protein present in the inner membrane embedded section of the core T2SS in K. pneumoniae and K. oxytoca [49], essential for protein secretion in K. pneumoniae but not in K. oxytoca [50]. Other common secretion systems (T3SS, T4SS, T5SS and T6SS) were not found in any of the ST2727 isolates, while HlyD, a diagnostic component of the type 1 secretion system (T1SS), was detected in the ST2727 strains (Table S5).

Evolutionary relationship of ST2727 to globally isolated Klebsiella strains

Klebsiella quasipneumoniae was previously considered of low clinical risk [3–5]. Mapping of virulence and resistance phenotypes observed among K. quasipneumoniae subsp. similipneumoniae and globally collected Klebsiella genomes showed that two K. quasipneumoniae subsp. similipneumoniae MLSTs, ST1901 (GCF_002248055.1) and ST2320 (GCF_001729665.1), and a K. variicola type (ST981) (GCF_001463685.1) [51] contained virulence markers that were previously only observed among K. pneumoniae strains [22, 23]. The ST1910 genome collected in Thailand in 2016 contained the virulence marker aerobactin (iuc3), whereas the ST2320 strain collected in India during 2015 can express aerobactin (iuc1), yerseniabactin (ybt 9; ICEKp3) and salmochelin (iro1), as well as producing a hypermucoidy phenotype (rmpA_4, rmpA2_3). The emergence of virulence markers among K. quasipneumoniae genomes raises concerns as there has been a parallel increase in hospital-acquired infections caused by K. quasipneumoniae [5, 7]. Figure 4 shows that ST2727 is closely related to ST498, a group that has been found to have numerous AMR genes including aadB, sul1 and blaOKP-B-7, and the genes blaOXA-2 and blaSHV-18 that encode extended-spectrum beta-lactamases [8] (Fig. 2).

Discussion

Colonization of the neonate gut is a complex process with contributions from the mother’s microbiome and environmental sources, which cannot easily be disentangled [52–54]. In this study, 21 neonates housed in a single room were sampled across 22 days for the carriage of Klebsiella in response to an outbreak-status event. The infection control procedures required that an outbreak be called in response to multiple cases of infection being reported in the same ward within a short period. These infections were diagnosed on clinical grounds, suspected to be hospital-acquired, and triggered the need for sampling and detailed diagnosis of other potentially at-risk neonatal patients. However, it is now clear from genomics data that the hyper-virulent ST23 isolated from the staff member was an irrelevant coincidence to the sepsis event in the neonates. A major outcome from this retrospective, deep, sequence analysis is the data that it provides on the colonization of the neonates by a strain of K. quasipneumoniae subsp. similipneumoniae from the NICU environment.

The Klebsiella isolated from the sick neonates carried a previously unreported sequence type (ST2727), with core-genome phylogenies showing they belonged to K. quasipneumoniae subsp. similipneumoniae [34]. Despite the availability of few K. quasipneumoniae genomes, our analysis of 54 available genomes including the 20 new ST2727 samples, showed positive correlation between genetic distances and date of sampling [29]. This indicates that our dataset was adequate to show that the NICU samples had diverged months prior to the detection of the first case, and showing that ST2727 was not transmitted between patients during the outbreak. These results were further supported by a permutation test where the sampling dates were randomized [31, 55]. Furthermore, the lower sampling of K. quasipneumoniae genomes, as compared to the greatly sampled K. pneumoniae genomes, would not have an effect on the estimation of the molecular clock for the outbreak samples because bacterial phylogenies in general show a large genetic distance between MLST types – signified by the long phylogenetic branch lengths between MLST types (star-shaped species-level phylogeny), indicating that adequate sampling of closely related bacteria would improve estimates rather than greater sampling of highly divergent genomes within a bacterial species. Without a detailed analysis focused on the built environments of the NICU and surrounding community it is difficult to infer with more confidence from where they originated. Overall, however, this study suggests that a population of ST2727 with some sequence variability is probably endemic within the hospital environment, and that distinct members within this population have infected the neonates in their first days of life.

It is probable that ST2727 is under antibiotic selective pressure due to their prevalence in the hospital environment for the months prior to the events analysed here. This is further confirmed with an MKT indicating positive selective pressure ongoing for genes related to efflux systems that are a key part for the transport of toxic substances as antibiotics. However, further investigation on the evolutionary mechanism selecting for resistant genes and genes related to resistance mechanisms would need to be done in order to assert these findings. Previously, estimates of selection pressure on the genes encoding beta-lactamases in the clades KpI, KpII and KpIII have shown neutral selection of these genes [51]. In agreement with this, we did not find either of the genes encoding beta-lactamases in ST2727 to be under natural selective pressure.

Although it is now highly unlikely, the data in this study cannot rule out a maternal contribution to Klebsiella carriage in the neonates, but we note that (i) 12/20 neonates were delivered via Caesarian section (Table 1) where the maternal contribution to the microflora should be inconsistent and minimal, and that (ii) ST2727 is probably not a dominant sequence type in the community as it has not been sampled previously, despite the hospital having more than 1500 Klebsiella isolates collected from adult patients over 15 years. We suggest that a substantial contribution to carriage comes from the endemic microflora in the NICU, with this specific built environment colonized by a population corresponding to ST2727. A suggestion for future analyses would be inclusion of more detailed samplings of the NICU environment, including the built environment itself and systematic sampling of staff members and visitors. Our study also urges the continual monitoring of hospital environments to improve understanding of the dynamics of bacterial communities in hospitals as reservoirs of resistance and virulent genes [56].

While K. quasipneumoniae subsp. similipneumoniae has been less widely reported and discussed than K. pneumoniae , similar to our investigation, there has been an increased recognition of K. quasipneumoniae subspecies in hospital surveys and their association with resistance genes [57, 58], indicating that a comprehensive assessment of their role in AMR ecology and contribution to disease burden is required. Their under-recognition is also due to the mis-assignment of non- K. pneumoniae strains to the type species.

Previous analysis estimated that at least half of Klebsiella infections result from the patients’ own microbiota [59]. That the neonate who died probably had a bowel perforation to initiate sepsis from a strain carried in the gut supports the contention that K. quasipneumoniae carriage contributes to the risk of infection in hospital environments. Thus, carriage of drug-resistant or hypervirulent clones of Klebsiella , including K. quasipneumoniae, in the gut would be of great concern in people with lowered immune systems – such as neonates – or those with conditions requiring surgical treatment where bowel perforation is a risk factor [52]. Our study highlights the importance of WGS for elucidating the transmission and carriage of hospital-resident bacterial populations, and urges for the early identification and eradication of persistent strains that increase the rate of hospital-acquired infections.