Introduction

Klebsiella pneumoniae is a major cause of hospital acquired infections including pneumonia, bloodstream and new-borns infection [1,2,3,4]. Treating infections caused by K. pneumoniae and other Enterobacteriaceae has been challenging worldwide due to emergence and rapid spread of resistant strains [5,6,7,8,9]. Initiatives towards monitoring of AMR should nonetheless go hand-to-hand with identification strategies of hypervirulent K. pneumoniae (hvKP) strains. Because roles played by virulence factors (VFs) in hvKP like yersiniabactin [10], colibactin [11], aerobactin and salmochelin [12] in enhancing severity of infections and increase their survival is significant. The global dissemination of hvKP clones [13,14,15] do pose a serious public health threat, thus underscoring a necessity to characterise VFs in K. pneumoniae. Treatment failures and possibly many deaths could now be attributed to K. pneumoniae infections [16,17,18]. Reports have generally described that treating bacterial infections in low- and middle-income countries (LMICs) is adding more burden on the already disease-burdened communities [19,20,21] both psychosocially and economically. In some countries, it is believed that even carbapenems will no longer treat more than a half of K. pneumoniae infections [22]. Having a bacterial infection will soon mean death if measures against irrational prescription and misuse of antibiotics are not taken seriously [23,24,25]. Antibiotics have been over prescribed and over-consumed especially in LMICs [26, 27]. These practices have been implicated to be fuelling up drug resistance selection pressure [28, 29].

In LMICs like Tanzania, lack of efficient clinical laboratory diagnostic systems is one of the factors often leading to empirical treatment. We previously reported the huge potential for whole-genome sequencing (WGS) to improve clinical diagnostics and infection control at a tertiary hospital in Tanzania where clinical laboratories lack access to molecular-based methodologies for regular typing of bacterial isolates [30, 31]. In this report, we used WGS to determine molecular relatedness, antimicrobial resistance genes, virulence genes and plasmids diversity in K. pneumoniae isolates from patients at KCMC, which is a tertiary care hospital in Kilimanjaro, Tanzania.

Materials and methods

Study design, participants and specimen collection

A hospital-based prospective cross-sectional study was conducted at KCMC hospital from 2013 to 2015. Part of the study’s methods has been described in details by Kumburu et al. [32]. Geographically, KCMC is located in Moshi municipality in Kilimanjaro and it is one of the biggest referral hospitals in Tanzania. It serves as a zonal referral hospital for a catchment area of around 15 million people. The hospital has a bed capacity of 650 with approximately 500 outpatients seeking medical services daily. This study was granted ethical approval by the KCMUCo Research Ethics Committee and the National Institute for Medical Research in Tanzania. A written informed consent was obtained from each participant or from parents or guardians of children before enrolment into the study. A convenient sampling method was used to recruit the study participants. It included participants suspected to have bacterial infection and admitted in medical and surgical wards. Specimens collected for bacterial culture included sputum, wound or pus swab and stool samples. Bacteria culture, isolation and identification were performed following in-house standard operating procedures as well as the Clinical and Laboratory Standards Institute (CLSI) guidelines. Sequentially, all K. pneumoniae isolates recovered over the study period were included for whole-genome sequencing and analysis. Over a 2-year period, 590 samples were collected without apriori knowledge of the infecting agent. A total of 377 bacterial strains were isolated, and whole genome sequenced. A number of isolates from this collection were randomly selected for antimicrobial susceptibility testing. A total of 34 K. pneumoniae collected sequentially were included in this study; amongst which, 16 K. pneumoniae isolates had phenotype-based antimicrobial susceptibility results.

Genomic DNA isolation, whole genome sequencing and analysis

For all K. pneumoniae isolates, genomic DNA (gDNA) was purified and its concentration was determined using the Easy-DNA Extraction Kit (Invitrogen®) and the Qubit dsDNA Assay Kit (Invitrogen®), respectively. The gDNA library preparation was performed following Nextera® XT DNA Sample Preparation Guide [33]. In brief, each gDNA was tagmented (tagged and fragmented) by the Nextera® XT transposome. The transposome simultaneously fragments the input DNA and adds adapter sequences to the fragment ends. Then, a limited-cycle PCR amplification followed, whereby indexes required for cluster formation were added to each DNA piece. Then, each gDNA library was normalised to ensure equal representation during sequencing. Equal volumes of the normalised library were combined, diluted in hybridization buffer and heat denatured prior to sequencing on the Illumina MiSeq platform (Illumina Inc.). The sequencer output was analysed using the standard WGS pipeline at KCRI, which is based on local implementations of the bioinformatics services available at https://cge.cbs.dtu.dk/services/. Quality control of the reads was performed using FastQC 0.11.4 [34]. De novo assembly was performed with SPAdes 3.11.1 [35], and quality assessed using QUAST 4.5 [36]. For this article’s purpose, the analyses included resistance gene identification using ResFinder 2.1 [37], multi-locus sequence typing (MLST) determination using MLST 1.8 [38], plasmid and plasmid MLST determination using PlasmidFinder 1.3 and pMLST 1.4 [39] and virulence gene determination using VirulenceFinder 1.4 [40]. Phylogeny reconstruction was done using CSI Phylogeny [41] (reference NTUH-K2044). The 34 assembled K. pneumoniae genomes of the present study have been submitted to the European Nucleotide Archive (ENA) with project accession number PRJEB26616. Stata 13 (College Station, TX, 77845, USA) was used to determine Cohen’s kappa coefficient of agreement between the phenotype- and whole-genome sequence-based antimicrobial resistance results.

Results

Study participants and Klebsiella isolates

A total of 34 K. pneumoniae isolates were recovered: 9 (26.5%) in 2013, 17 (50.0%) in 2014 and 8 (23.5%) in 2015. Out of 34 K. pneumoniae, 25 (73.5%) isolates were from wound or pus swabs, 5 (14.7%) from sputum, 3 (8.8%) from stool and 1 (2.9%) from throat swab. Sixteen (47.1%) of K. pneumoniae were isolated from surgical wards, 3 (8.8%) from surgical ICU, 12 (35.3%) were isolated from patients admitted in medical wards, 1 (2.9%) from medical ICU and 2 (5.9%) were isolated from outpatients. A total of 11 (32.4%) K. pneumoniae were isolated from participants with infected wounds. The proportion of K. pneumoniae from participants with cough was 6 (17.6%), burn 6 (17.6%), wounds 6 (17.6%) and diabetes 6 (17.6%) (Table 1).

Table 1 Characteristics of participants from which K. pneumoniae were isolated
Full size table

MLST and capsular (K) typing

A total of 16 (47.1%) STs were identified in 30 (88.2%) of the analysed isolates whilst the remaining isolates could not be typed (unknown STs, 4 (11.8%)). A total of 6 (17.6%) were K. pneumoniae ST17, of which 4 were recovered from patients in surgical and 2 in medical wards. A total of 4 (10.8%) were K. pneumoniae ST392, of which 2 were recovered from patients in medical and 2 in surgical wards. Three (8.8%) were K. pneumoniae ST348 and all were recovered from patients in surgical wards. K. pneumoniae ST15, ST25, ST299 and ST1562 each was observed in 2 (5.4%) of the isolates (Table 2). A total of 10 (29.4%) different K types were identified including K2, K7, K10, K19, K23, K28, K34, K41, K60 and K80 whilst isolates with unknown K types were 17 (50.0%). All 4 K. pneumoniae ST392 were of unknown K types, whereas 5 out of 6 K. pneumoniae ST17 were unknown K types (Table 2).

Table 2 Beta-lactam-resistance determinants in K. pneumoniae
Full size table

Beta-lactam-resistance determinants

All isolates were carrying at least one beta-lactam-resistance determinant. The commonest detected determinants were blaCTX-M-15 in 16 (47.1%) isolates, blaSHV in 30 (88.2%), blaOXA-1 in 8 (23.5%) and blaTEM-1 in 18 (52.9%) isolates. A variable population of blaSHV genes was found, whereby blaSHV-11 was found in 15 (44.1%) and blaSHV-1 in 7 (20.6%) isolates. Other infrequently blaSHV-detected genes included blaSHV-12, blaSHV-28, blaSHV-61, blaSHV-83, blaSHV-99 and blaSHV-133. At least three beta-lactam- resistance genes were found in 15 (44.1%) isolates. Carriages of four beta-lactam-resistance genes, namely blaCTX-M-15, blaSHV-11, blaOXA-1 and blaTEM-1B, were frequently observed in K. pneumoniae ST392 isolates (Table 2).

Aminoglycoside, fluoroquinolone and quinolone determinants

Almost all isolates were carrying at least one resistance determinant for aminoglycosides in 21 (61.8%) isolates, fluoroquinolones in 13 (38.2%) and quinolones 34 (100%). Fluoroquinolones gene, aac(6’)Ib-cr, was identified in 13 (38.2%) isolates. Determinants for aminoglycoside resistance included aph(3″)-Ib and aph(6)-Id in 16 (47.1%) and 17 (50.0%) isolates, respectively. Other infrequent determinants for aminoglycosides were aac(3)-IIa, aac(3)-IId, aph(3′)-Ia, aadA16, aadA1, aadA5, aadA24 and aadB (Table 3). The quinolone resistance determinants identified were oqxAB in all 34 (100%) and qnrB in 11 (32.4%) isolates. The variants for qnrB gene included qnrB1, qnrB6, qnrB49 and qnrB66.

Table 3 Antibiotic resistance determinants other than beta-lactamase genes in K. pneumoniae
Full size table

Fosfomycin, macrolide and phenicol

The determinants FosA and mph(A) for fosfomycin and macrolide resistance were identified in 34 (100%) and 4 (11.8%) isolates, respectively. Several gene families for phenicol resistance were detected: catA2 in 10 (29.4%), catA1 in 2 (5.9%), catB4 in 8 (10.8%) and cmlA1 in 2 (5.9%) isolates.

Rifampicin, sulphonamide tetracycline and trimethoprim

Seven (20.6%) isolates were carrying arr-3, a rifampicin resistance gene. Sulphonamide resistance determinants, sul1 and sul2, were detected in 12 (35.3%) and in 22 (64.7%) isolates, respectively. Two tetracycline resistance genes detected were tet(A) in 7 (20.6%) and tet(D) in 4 (11.8%) isolates. Trimethoprim resistance genes were identified in 23 (67.6%) isolates. A total of 11 (32.4%) isolates were carrying dfrA14 and 5 (13.5%) carrying dfrA27. Other infrequently observed trimethoprim resistance genes detected included dfrA(1/5/7/15/17/25/30) and dfrG (Table 3).

Comparison of phenotype- and whole-genome sequence-based antimicrobial resistance

Agreement between phenotype- and whole-genome sequence-based antimicrobial resistance was done for 16 out of 34 K. pneumoniae isolates (Table 4). On average, agreement across all antibiotics tested was 77.4%. Overall, the phenotypically determined resistance was higher than the whole-genome sequence-based resistance. Nevertheless, all antibiotics but ampicillin showed substantial (61–80%) or strong agreement (81–100%) between phenotype- and sequence-based resistance results. Ampicillin showed moderate agreement: 56.3%, kappa = 0.13 and p = 0.1508. The sequence-based analysis predicted that resistance to ampicillin was in 19 (55.9%) isolates, whereas phenotypic testing revealed 15 (93.8%) of isolates to be resistant. Ceftazidime and ceftriaxone showed the strongest agreement: 93.8%, kappa = 0.87 and p = 0.0002. Sequence-based analysis predicted resistance to both ceftazidime and ceftriaxone in 16 (47.1%) of isolates, whereas 6 (37.5%) of isolates were resistant phenotypically.

Table 4 Agreement between phenotypically tested and whole-genome sequence-predicted antimicrobial resistance
Full size table

Virulence determinants in K. pneumoniae

Further, we analysed virulence determinants in all K. pneumoniae strains. Yersiniabactin was detected in a significant proportion, in 12 (35.3%) of isolates out of which, all (n = 3) K. pneumoniae ST348 were carrying yersiniabactin genes. The ferric uptake operon system (kfuABC) was found in 10 (29.4%) of isolates. Different from the rest, isolate 315 (K. pneumoniae ST2042) was carrying genes coding for aerobactin, salmochelin and yersiniabactin but lacked the ferric uptake operon system (Table 5).

Table 5 Virulence factors in K. pneumoniae
Full size table

Plasmid multi-locus sequence typing

Plasmid analysis revealed a high diversity of incompatibility groups (Inc) or plasmid replicons (Table 6). The most frequent replicon was IncFIB(K) that was found in 30 (88.2%) K. pneumoniae isolates. Other IncFI members included IncFIB(pKPHS1), IncFIA(HI1), IncFIB(pECLA) and IncFIB(pENTAS01). Another prevalent replicon IncR was identified in 18 (52.9%) of K. pneumoniae isolates. The IncFII was another frequently detected replicons. In this group, IncFII(K) was found in 14 (41.2%), and IncFII in 7 (20.6%). Others were IncFII(pCRY), IncFII(pECLA) and IncFII(Y). Plasmid replicons that were infrequently identified include IncHI1B, IncHI2, IncHI2A and IncN3. A high diversity of plasmid multi-locus sequence types (pMLSTs) was identified (Table 6). IncF[F-:A16:B-] was detected in 5 (14.7%) of K. pneumoniae isolates, IncF[F-:A-:B10 detected in 4 (11.8%) of K. pneumoniae isolates and 9 (26.5%) and 5 (14.7%) of K. pneumoniae isolates had unknown pMLSTs belonging to IncF[Unknown ST] and IncHI1[Unknown ST], respectively. Other pMLSTs identified are shown in Table 6.

Table 6 Plasmid multi-locus sequence typing (pMLST) in K. pneumoniae
Full size table

Phylogenetic analysis

For epidemiological tracking of nosocomial infections, SNP difference was calculated to show how closely or distantly related the isolates are. The observed minimum and maximum number of SNP difference between K. pneumoniae isolates were 57 and 42,893, respectively (Supplemental Table 1). The minimum SNP difference was observed between isolates 134 and 131E (both ST392), showing a possibility of nosocomial infections. The maximum SNP difference was observed between K. pneumoniae isolates ADE ST37 and 16 ST231, showing that it is unlikely these isolates closely related. The tree topology showed two isolates (79C ST 297 and 302 with undetermined ST) segregating very distinctly from others. The heatmap showed clear patterns of high beta-lactam, aminoglycoside and quinolone resistance gene proportions spreading almost universally across all isolates. Furthermore, an apparent pattern was observed indicating an inverse correlation between yersiniabactin genes across the isolates (Fig. 1).

Fig. 1
figure 1

Phylogenetic analysis of 31 K. pneumoniae isolates showing STs, resistance and virulence genes. The heatmap shows the frequency of AMR and virulence genes present in an isolate. The stronger the red colour is, the higher the number of genes identified across antibiotic classes is. The green colour stands for isolates from which a virulence gene was present

Full size image

Discussion

The present study used whole-genome sequence-based approach in characterising clinical K. pneumoniae isolated from hospitalised patients at KCMC hospital in Moshi, Tanzania. All isolates were analysed to determine (1) K. pneumoniae subtypes and molecular relatedness for establishing existence of nosocomial transmissions or outbreaks, (2) virulence and antibiotic resistance determinants and (3) types of plasmids. The present study reveals high diversity of K. pneumoniae in the hospital. The observed K. pneumoniae diversity is plausibly attributed to the fact that specimens were collected from a diverse population as this is a consultant hospital that is serving the northern, eastern and central zones of Tanzania. Nonetheless through MLST, the majority of K. pneumoniae that were clonally related were actually isolates from patients admitted to the same wards. For instance, K. pneumoniae ST17 with number 17, 29 and 320 were from surgical wards. Also, instance K. pneumoniae ST17 with number 41 and 284 were from medical wards. Although few numbers of strains were identified within distinct ST groups (clusters), this may be an indication of nosocomial transmissions or outbreaks within the hospital. Similar to the present report, polyclonal existence of K. pneumoniae with predominance of K. pneumoniae ST17 in hospital settings was reported in the Netherlands by Souverein et al. [42]. Identification of K. pneumoniae clones particularly ST17 and ST348 within surgical wards in the present study compares with the findings in Norway [43] and Mwanza, Tanzania [6]. In both reports, it was shown that K. pneumoniae ST17 and ST348 strains were the likely causes of neonatal sepsis and outbreaks in neonatal ICU. Given the superiority of WGS over classical approaches in microbial identification, typing and tracing of outbreak sources [44,45,46], the possibility that there were sporadic nosocomial transmissions of K. pneumoniae in this hospital becomes highly likely.

Our data further suggests that K. pneumoniae circulating in the hospital are carrying high proportions of antimicrobial resistance determinants. These findings are in line with findings of study done in Kenya on K. pneumoniae isolates from stool [47]. This study identifies multiple carriages of resistance determinants including those for beta-lactams: blaSHV, blaCTX-M-15 and blaTEM-1. Despite the fact that we noted blaSHV being the most prevalent determinant, Tellevik et al. [48] and Mshana et al. [6] had earlier reported blaCTX-M-15 as the most prevalent determinant in Dar es Salaam and Mwanza, respectively. On average, a strong agreement was observed between phenotype- and sequence-based resistance to all antibiotics tested, findings that are consistent with the previous report on E. coli that was conducted in the same settings [49]. Nonetheless, ampicillin revealed the lowest but moderate agreement between the two methods. The phenotypically determined resistance to ampicillin was higher than sequence-based resistance. Plausibly, the observed difference could be due to the fact that WGS analysis uses only known resistance genes and it is also true that not all genes involved in resistance mechanisms have been included in these databases.

The observed multiple carriage of beta-lactam- resistance determinants in this study, particularly amongst K. pneumoniae ST17 and ST392 isolates, might substantially be a reason for their persistence in this hospital as also noted elsewhere [6]. Apart from being prevalent in this study, K. pneumoniae ST392 appeared to carry multi-resistance determinants. Findings are similar to those reported in a hospital-based study at an Italian hospital [50], which showed that K. pneumoniae ST392 strain might become very aggressive. Although we could not identify a single resistance gene for carbapenem in the present study’s isolates, K. pneumoniae ST15 and ST348 have been reported in Portugal to be the cause of KPC outbreaks [51]. The identification of this aggressive ST348 strain in this hospital should at least signal for the emergence and spread of MDR bacteria and that no sooner than later common infections caused by K. pneumoniae and other bacteria will become untreatable.

The co-carriage of aminoglycoside, fluoroquinolone and quinolone resistance determinants was very common in almost all K. pneumoniae. The fluoroquinolone, aminoglycoside and quinolone resistance genes: aac(6’)Ib-cr, oqxA and oqxB, appeared to be associated with the carriage of blaCTX-M-15 and other beta-lactam- resistance determinants, findings that are consistent with a report [52] on K. pneumoniae strains from urban settings in Barcelona.

The abundance and variability of resistance determinants to sulphonamides, tetracyclines, fluoroquinolones and trimethoprims was found to be high, findings that are in line with those by Taitt et al. [47]. In the current study, we further found a relatively higher proportion of K. pneumoniae carrying arr-3 for rifampicin resistance than the proportion documented by Taitt et al. [47]. Uncontrolled and excessive use of first- and second-line antibiotics for both clinical and veterinary purposes is a plausible explanation to the emergence and spread of these determinants in Enterobacteriaceae [53].

We further observed a significant proportion of yersiniabactin, colibactin, aerobactin and salmochelin in these K. pneumoniae strains. However, yersiniabactin was the most prevalent VF and it has been associated with K. pneumoniae infection rather than carriage, findings consistent with Holt et al. [54]. Based on the molecular characteristics proposed by Li et al. [55] for a hypervirulent K. pneumoniae, isolate 315 (K. pneumoniae ST2042) was the likely candidate. It was carrying genes coding for regulators of mucoid phenotype (rmpA), aerobactin (iucABCD and iutA), salmochelin (iroBCDN) and yersiniabactin (ybt, fyuA and irp1/2) but lacking the ferric uptake operon (kfuABC). Interestingly, we observe that this strain 315 (K. pneumoniae ST2042) with high virulence potential has low AMR. Apart from fosA and oqxAB, which it shares with all other isolates, its only beta-lactam gene is blaSHV-99, which notably none of the other isolates possess. This combination of high virulence and low AMR has been observed elsewhere [54, 56]. Amongst our isolates with many AMR genes, virulence determinants tend to be reduced.

Further analyses revealed that the most frequent plasmid replicon identified was IncF (I/II). Other replicons that were infrequently identified included IncHI1 and IncN3. The plasmids carried by K. pneumoniae appeared to be highly diverse. However, the IncFII plasmids seem to be common and correlated with the observed carriage of blaCTX-M-15, similar to a Moroccan study [57] that identified IncFII plasmid as a carrier of blaCTX-M-15 amongst K. pneumoniae ST466 strains. For instance K. pneumoniae ST15 appeared to carry plasmid ST IncF[K9:A13:B-], and K. pneumoniae ST17 was carrying plasmid ST IncF[K2*:A13:B-]. Plasmid ST IncF[K8:A-:B-] was identified in K. pneumoniae ST348 and ST231 and plasmid ST IncF[K7:A-:B-] was found in K. pneumoniae ST392 and ST29.

We acknowledge the presence of several limitations to this study. First, due to small numbers of isolates, the study lacked epidemiological analysis that might have shown correlation between AMR and virulence genes with patients’ demographics (gender, age) and clinical characteristics including admission outcomes, antibiotics use, hospitalisation history and comorbidities. Second, phenotype-based resistance results were available for small numbers of bacterial isolates; this may have impacted on the agreement between phenotype- and sequence-predicted resistance results. Third, WGS analysis relied on resistance and virulence databases that at time of analysis might comprise of known genes and not all genes involved in resistance or virulence mechanisms have been documented or included in those databases. Further, the existence of genes encoding different resistance and virulence factors do not necessarily indicate gene activity in the isolates. There is therefore a need for future genomics studies to focus on quantifying expression levels of genes encoding different resistance and virulence factors.

Conclusions

In this study, the amount of antimicrobial resistance and virulence determinants detected in K. pneumoniae is alarming. Besides its application for research purposes, in resource-limited settings, WGS-based diagnostic approach has showed promising potentials in clinical microbiology, hospital outbreak source tracing, virulence and AMR detection. Having been implemented successfully in Kilimanjaro, WGS can be used as a surveillance tool for infectious agent and AMR detection nationwide. It has the potential of accelerating informed decisions in formulation of pragmatic antimicrobial stewardships, and other infection prevention and control initiatives.