Pathogenomics and Evolutionary Epidemiology of Multi-Drug Resistant Clinical Klebsiella pneumoniae Isolated from Pretoria, South Africa

Antibiotic-resistant Klebsiella pneumoniae is increasingly being implicated in invasive infections worldwide with high mortalities. Forty-two multidrug resistant (MDR) K. pneumoniae isolates were collected over a 4-month period. Antimicrobial susceptibility was determined using Microscan. The evolutionary epidemiology, resistome, virulome and mobilome of the isolates were characterised using whole-genome sequencing and bioinformatics analysis. All isolates contained the blaCTX-M gene, whilst 41/42(97%) contained blaTEM, 36/42(86%) contained blaOXA and 35/42(83%) harboured blaSHV genes. Other resistance genes found included blaLEN, aac(6′)-lb-cr, qnrA, qnrB, qnrS, oqxAB, aad, aph, dfr, sul1, sul2, fosA, and cat genes. Fluoroquinolone and colistin resistance-conferring mutations in parC, gyrAB, pmrAB, phoPQ and kpnEF were identified. The blaLEN gene, rarely described worldwide, was identified in four isolates. The isolates comprised diverse sequence types, the most common being ST152 in 7/42(17%) isolates; clone-specific O and K capsule types were identified. Diverse virulence genes that were not clone-specific were identified in all but one isolate. IncF, IncH and IncI plasmid replicons and two novel integrons were present. The blaCTX-M-15 and blaTEM-1 genes were bracketed by Tn3 transposons, ISEc9, a resolvase and IS91 insertion sequence. There were 20 gene cassettes in 14 different cassette arrays, with the dfrA and aadA gene cassettes being the most frequent. Phylogenetic analysis demonstrated that the isolates were evolutionarily associated with strains from both South Africa and abroad. These findings depict the rich resistome, mobilome and virulome repertoire in clinical K. pneumoniae strains, which are mainly transmitted by clonal, multiclonal and horizontal means in South Africa.

. Demographic characteristics of patients from whom the clinical specimen were obtained. The patients were of almost equal distribution in terms of sex although their ages varied widely. The samples were obtained mainly from Steve Biko Academic Hospital and Kalafong.
Antibiotic susceptibility of isolates. The isolates obtained from the samples were identified phenotypically and genomically as K. pneumoniae. All the isolates were phenotypically positive for extended-spectrum β-lactamase (ESBLs) production. They were mostly resistant or non-susceptible to all the β-lactams except the carbapenems and cephem-β-lactamase inhibitors viz., sulbactam, tazobactam and clavulanic acid. Specifically, all isolates were resistant to aztreonam, cephalothin, cefuroxime, cefotaxime, ceftazidime and cefepime. Notably, resistance to amoxicillin-clavulanate (AUG) and piperacillin-tazobactam was very common compared to cephem-β-lactamase inhibitor combinations. Sixteen isolates had cefoxitin MICs ≥ 16 mg/L. There was variable susceptibility to carbapenems with all isolates demonstrating MICs ≤ 1 mg/L for imipenem and meropenem. Overall, 41 isolates had a doripenem MIC ≤ 1 mg/L, and four isolates, namely K181, K145, K059 and K091, had ertapenem MICs of >1 mg/L (Supplementary Table S2). The phenotypic resistance data tallied largely with the genomic results in that there was no carbapenemase gene found, confirming the absence of carbapenem resistance. As well, the phenotypic ESBL results were confirmed by the presence of bla TEM , bla CTX-M-15 , bla SHV , bla OXA and bla LEN genes (Tables 1-2, Figs. 2-7 and supplementary Table S1).
Except for amikacin, the isolates were resistant to all aminoglycosides. Further, all the isolates were resistant to ciprofloxacin but susceptible to norfloxacin; twenty and two were respectively resistant to levofloxacin and nalidixic acid (Supplementary Table S2). Resistance to minocycline (n = 18 isolates) and tetracycline (n = 25 isolates) was common than tigecycline (n = 7 isolates). Resistance to chloramphenicol (n = 33), colistin (n = 10), nitrofurantoin (n = 16) and sulphamethoxazole-trimethoprim (SXT) (n = 42) was identified, with none being resistant to Fosfomycin (Supplementary Table S2). Contrary to the β-lactams, there were substantial discrepancies between the phenotypic and genomic results for the non-β-lactam antibiotics. For instance, the isolates were mostly susceptible to Fosfomycin although fosA genes were ubiquitous among the isolates. Further, the presence of aac(6′)-Ib, OqxAB, Qnr and aac(3′)-Ib genes and mutations in parC and gyrAB only conferred resistance to ciprofloxacin and levofloxacin but not to norfloxacin. A similar observation was made with regards to amikacin and gentamicin/tobramycin (aadA4, aph(3′), strAB, aac(6′)-Ib-cr) and between minocycline, tetracycline and tigecycline (tet). However, the presence of chloramphenicol and SXT resistance was mostly corroborated by the presence of the appropriate resistance genes i.e., cml/Cat and sul/dfrA. For some strains and antibiotics, no known resistance determinant was found to explain the observed phenotypic resistance (Supplementary Tables S1 and S2). ARGs in Klebsiella pneumoniae. Several resistance genes were present in the isolates (Table 2 and   Supplementary Table S1). Many isolates had at least two β-lactamase genes, with bla CTX-M-15 (n = 42), bla OXA and bla TEM being commonest (Figs. 2-7). The bla TEM gene was present in 41/42 (97.6%) isolates, bla OXA was present in 36/42 (85.7%) isolates and bla SHV in 35/42 (83.3%). We identified three bla TEM genes, of which the most frequent was bla TEM-1B. Isolate K021 did not have a TEM β-lactamase but had the bla OXA and bla CTX-M-15 genes. WGS revealed several bla SHV and bla OXA genes, with the most diverse being found in the bla SHV -containing isolates. The bla LEN gene, first identified by Arakawa et al. in 1986 30 , was detected in four isolates, namely K053, K126, K137 and K146.
No plasmid-mediated colistin resistance gene was identified in the isolates having increased colistin MICs. Chromosomally encoded mutations in the pmr, pho and kpn genes were, however, identified in 9/11 (82%) of these isolates (Table 1). No novel putative colistin ARGs were identified on any of the available databases. Isolates with ccrB mutations, however, were susceptible to colistin and no truncation were observed in the mgrB in all the isolates (Table 3).

Sample code (MLST) Integron
Cassette arrays www.nature.com/scientificreports www.nature.com/scientificreports/ source of the isolates as isolates from blood, urine; in fact, in some cases isolates from sputum had more ARGs than those from urine or blood (Supplementary Table S1).
Sequence types and the genetic environment of the ARGs. We detected 11 different sequence types in the isolates ( Table 3). The most prevalent sequence types were the ST152 (n = 9, 33%) and ST1552 (n = 6, 22%). Both sequence types were associated with integron ln369 ( Table 1). The globally distributed ST15 was only detected in four isolates (Tables 1-2).
As well, we identified the IncF, IncN and IncH incompatibility plasmid replicons (Supplementary Table 1), with nine isolates carrying multiple plasmid replicons simultaneously. The IncF plasmid group was most frequently identified. Nine of the isolates with this IncF incompatibility group also harboured the IncH group. Several unknown plasmid sequence types were also identified ( Table 3). The IncF group was also associated with most of the STs.  Table S2). Two novel class 1 integrons were identified in two isolates, K021 and K145, which were given new numbers, In1481 and ln1482, respectively. The isolates were registered in the GeneBank database with specific accession numbers NXIU000108 (In1481) and NXKB01000066 (ln1482).

Isolate ID
The most frequent integron on the IncF plasmid was In369, which captured the dfrA1b -aadA1b cassette array and has been described in other Enterobacteriaceae. Two isolates also harboured the aacA4cr cassette gene. This integron was more frequently associated with ST1552 and ST152 but was also found in two ST101 isolates and one ST234 isolate. In27, a narrow spectrum integron that has been described in the literature was only identified in ST234 isolates capturing the dfrA12-gcuF-aadA2 cassette array, confirming its narrow spectrum status. In191 was, however, identified in four different sequence types, indicating that it is not a narrow spectrum integron.
Virulome and capsular characteristics. A total of 62 virulence genes were identified in all the strains, with EC588_3547, ecpABCR, entB, fepC, fimABCDEFGHK, mrkABC, pulBCED and rpoS occurring in almost all the isolates except for K094, which had no virulence gene (Fig. 8). Among the isolates, EC0103_3368, EC55989_3335, APEC01_3698, ECP_2822 and cah were the least occurring, with cah being only present in K090. As shown in Fig. 8c, the virulome was not clone-specific in that isolates of the same clone had different virulence genes. Although most of these virulence genes occurred in isolates obtained from urine (n = 653) and blood (n = 605), their distribution does not suggest their association with these sources (Fig. 8). The highest number of virulence genes to occur in a single isolate was 57 (K031) whilst all but K094 had more than 30 virulence genes in a single isolate (Fig. 8c). hypervirulence genes were however absent.
The O and K capsules types in the strains were highly clone specific, with same clones having the same O and K capsule types. However, minor discrepancies were observed as some strains of the same clone had O and K capsule types that were different from those of members of the same clone. A case in point is that of K031 of ST152, which had OL102 and KL107 while other members of the same clone had O4 and KL149 (Figs. 9-12; Supplementary data S3). As can be observed, the O1v1 capsule type was more dominant across the clones whilst KL149 was the most dominant K capsule type. In all, there were eight different O capsule types whilst 12K capsule types were identified, evincing the higher diversity of the K capsule types. It is worthy of mention that isolate K094 had no O or K capsule, just as it also had no virulence genes (Figs. 9-12; Supplementary data 3), which is a very interesting finding evolutionary phylogenomics and epidemiology. The K. pneumoniae isolates showed significant phylogenetic diversity (Fig. 13), with the whole-genome phylogenetics showing higher resolution than the MLST typing scheme. For instance, K080 (ST234) was phylogenetically closer to K038 (ST1552) than other ST1552 strains, which were themselves found on different branches, albeit of the same clade. Further, K129 (ST643) and K161 (ST25) were of very close evolutionary distance, albeit of the different clones and capsular types. However, strains of the same MLST clustered within the same clade, with some single STs clusteing closely. Examples include K118 (ST14) within the ST15 clade, K137 (ST182) within the ST101 clade, K014 (ST607) within the ST152 clade, K123 within the ST234 clade, and K090 (ST323) within the ST1414 clade. Interestingly, K120 (ST17), K001 (ST179) and K094 were phylogenetically distant from every strain within the collection (Fig. 13A).
Within the African context, the isolates (coloured with red branches and labelled in blue) were largely phylogenetically related to other K. pneumoniae strains of the same clones (STs) and clades such that strains of the  www.nature.com/scientificreports www.nature.com/scientificreports/ same clones were clustered together within the same clade. This can be seen with K120 (ST17), which clustered with same clinical clones from Nigeria and South Africa; K0179 and K145 (ST39) with a ST38 strain from Nigeria, a ST39 strain from Cameroon and a strain from Uganda. This pattern is observed around the tree under the respective STs (Fig. 13B). However, as observed above, strains of different STs were also found clustered together; this can be seen with K137 (ST182) and K129 (ST643) and PR042E3 (ST31) from pigs in Cameroon ( Fig. 13B; Supplementary data S4).

Discussion
The molecular mechanisms of resistance and virulence dissemination in clinical K. pneumoniae circulating in two referral hospitals in South Africa were characterised and found to be richly endowed with diverse determinants of resistance, virulence and mobile-genetic elements. Notably, the isolates were MDR to several clinically important  www.nature.com/scientificreports www.nature.com/scientificreports/ antibiotics except for reserved ones such as the carbapenems, colistin and tigecycline. The presence of these MDR strains in specimens from this diverse patient demographics in two important referral hospitals make this a very worrying finding. Particularly, same STs were identified in both referral hospitals, suggesting their circulation in both health centres. As expected, the strains' phenotypic resistance characteristics correlated with known genetic mediators of resistance except for amikacin, nalidixic acid, minocycline, and fosfomycin for which the presence of resistance genes such as aac, aad and aph, tet(A/D/J) and fosA led to no phenotypic resistance. Whereas we could not undertake expression analyses to determine the expression state of these genes, we suspect that the lack of resistance in their presence could be due to little or no expression.
The presence of the bla CTX-M , bla TEM , bla OXA , and bla SHV ESBL genes in similar genetic contexts have been described previously in same and different species in South African and international isolates 7,8,12,19,20,[31][32][33][34][35][36] . The presence of these ARGs within the same genetic context and on the same plasmid replicons across same and different species around the globe strongly suggests the clonal and plasmid-mediated spread of these ARGs. Specifically, ISEc9 and IncF plasmids have been shown to mobilize and facilitate the global spread of bla CTX-M-15 , Figure 8. (a-d) Virulence genes distribution frequency and their association with specific K. pneumoniae clones and specimen sources. The virulence genes were not clone specific as same clones had different virulence genes (a). The virulence genes distribution and frequency were also not determined the specimen sources (a). K094 had no virulence gene, capping the highest frequency of virulence genes per isolate to 41 with the lowest (cah) being 1 (b). The highest number of virulence gene per clone was 57, with K094 having none (c). Strains from urine and blood had the highest number of virulence genes, which could obviously be due to the relatively larger number of strains isolated from urine and blood (d). www.nature.com/scientificreports www.nature.com/scientificreports/ alongside aac(6′)Ib-cr, bla OXA-10 and bla TEM , across species 6,7,12,37 . Thus, it is not surprising to have the ISEc9 and IncF plasmid replicons dominating in these strains that all harboured bla CTX-M-15 alongside aac(6′)Ib-cr, bla OXA-10 and bla TEM. Interestingly, the same genetic context around the aac(6′)Ib-cr, bla OXA-10 and bla TEM genes were also observed in E. coli strains from the same hospitals, suggesting plasmid-mediated circulation of these genes within these academic hospitals 12 .
bla CTX-M-15 being present in all the isolates is worth noting, but is not new as an earlier study also found this gene in all the Enterobacteriaceae species studied 7 . Other studies from South Africa have described the presence of the bla CTX-M-15 gene in E. coli isolates and more recently, in K. pneumoniae 7,12,33 . We also report a higher prevalence (87.5%) of OXA β-lactamase genes in these K. pneumoniae isolates as well as bla SHV in 83.3% of isolates, confirming that this supposedly chromosomally encoded gene, is not universally found in K. pneumoniae species 38 . Interestingly, four isolates also contained the narrow spectrum, chromosomally encoded bla LEN gene, comprising of bla LEN9 and bla LEN12 , which are rare in South Africa and Africa although it has been previously described in Kenya 39 . The bla LEN β-lactamase gene was first identified by Arakawa 30 .
The global dominance of the IncF plasmid, a MGE associated with HGT, is thus herein confirmed 6,7,40 . As well, an association between IncF replicons and multi-drug resistance (MDR) was observed, as reported globally 6,10,11 . The presence of several plasmids in which integrons that capture cassette genes are located has been shown to cause MDR 6,10,11 . The rich repertoire of ARGs in these isolates suggest the presence of one or multiple plasmids, corroborated by the plasmid replicons (Supplementary Table 1).
The integrons identified herein contained diverse gene cassettes, which are novel in the South African context. Particularly, the dfrA and aadA gene cassettes corroborates the global spread of these MGEs 41,42 . Moreover, aadA1 types, aadA1b and aadA1a, which are different from results reported in K. pneumoniae from Korea where aadA2 types were more frequently identified, were observed. These differences signify subtle changes in genetic composition at a local level 11,13,31 . Overall, 20 gene cassettes in 14 different cassette arrays, which have not been described previously, were observed. Significantly, the most prevalent cassette arrays viz., dfrA1b-aadA1a and dfrA1b-aadA1b, differed from those described by Partridge et al., which included aadA1a, aadA2 and aadB cassettes 11,13 .
Integron In369, which captured the dfrA1b -aadA1b cassette array was identified for the first time in South African isolates. This integron and cassette array were also reported in a Portuguese environmental study 43 . In the current study, this integron was more frequently associated with ST1552 and ST152 but was also found in two ST101 isolates, suggesting its broad host range and promiscuity. Interestingly, integron 27, capturing the dfrA12-gcuF-aadA2 cassette array was only identified in ST234 isolates. Integron 191 was however identified in four different STs suggesting a diversity of clones with this MGE. The dominance of the class 1 integron in this study is also consistent with genomes found in Africa, Europe and South America 8,31,41,44 . Two novel integrons identified had the aadA16 cassette variant in K021 as well as dfr30b and dfrA14b cassette variants in K145, confirming the ongoing evolutionary processes in these genomes resulting in the diversity of gene cassettes in K. pneumoniae.
Our findings suggest plasmid and chromosomally mediated quinolone resistance genes in the clinical K. pneumoniae isolates. The simultaneous presence of qnr, oqx and aac(6″)-lb-cr PMQR genes in these quinolone-resistant isolates and the dominance of the latter is consistent with the literature [49][50][51][52][53] . This is the second report of the oqxAB gene being found in Africa, although previously described in carbapenemase-producing Enterobacteriaceae 51,54 . Significantly, we did not detect the qep efflux pump gene which is not frequently reported 55,56 . In this study, 90% of isolates contained the aac(6′)-lb-cr gene, which is of higher prevalence than studies reported from Spain, Uruguay and Sweden 53,57-59 , which could be explained by the localised spread of related sequence types carrying this gene. Mutations such as S83A and D87A/G in gyrA, D553V and Q/L657G/M in gyrB and S80I and N304S in parC were also found in K. pneumoniae and other Enterobacteriaceae in Durban, South Africa 51 . Coupled with the PMQR genes, these mutations underlies the resistance to fluoroquinolones to some of the strains (Table 3), albeit no resistance was expressed towards norfloxacin 60 .
Hospital-acquired Gram-negative infections are usually caused by multi-drug resistant organisms, limiting options available for treating such patients [1][2][3]61 . Herein, these isolates from hospitalised patients simultaneously contained ESBL β-lactamase genes and PMQR genes, as well as co-resistance to other antibiotic classes, implying that significant antibiotic use causes resistance, co-selection of resistance genes and more significantly, HGT (horizontal gene transfer) 3,54,62 . These genes have been shown to be co-transmitted on plasmids and other MGEs 6,11 .
All the isolates in this study were trimethoprim-resistant, although trimethoprim-sulfamethoxazole is no longer recommended for treatment of outpatient conditions such as urinary tract infections in South Africa and internationally because of the high prevalence of resistance [63][64][65] . Similarly, fosfomycin, used in the outpatient  www.nature.com/scientificreports www.nature.com/scientificreports/ setting for urinary tract infections may not be efficacious since most isolates had the fosA gene, a finding only recently described in Enterobacteriaceae in South Africa 7,64,65 . The presence of chromosomal colistin resistance in nine out of the 11 ESBL-containing isolates, with an MIC ≥ 4 g/mL, is of concern (Table 3). However, we were unable to confirm these with the broth microdilution, which is the recommended method for colistin resistance determination 66 . No mcr gene was found, suggesting that these were vertically acquired or engendered de novo.
The diversity and complexity of the virulome and capsule types identified among the strains are concerning as they are implicated in virulence. Fortunately, no hypervirulence genes were found albeit capsule type K2, identified in some strains (Figs. [9][10][11][12], are associated with increased virulence and resistance [67][68][69] . It is interesting to note that the K2 serotype was also identified in a ST14 strain (K118) in this study as was reported in China recently 69 . The highly clone-specific nature of the O and K capsule types suggest their conserved nature within the genome, contrary to the virulence genes, which differed even within clones (Figs. 3 and 4). The diversity of the virulence genes within clones from the same hospital setting suggest that they were mostly acquired horizontally rather than vertically and could be associated with plasmids. We were unfortunately unable to determine their mobility and presence on plasmids. Further, no association could be established between the specimen sources Isolates of the same clone (ST) clustered together, although some strains of different STs were found within or close to strains of same STs (A). Isolates from South Africa are coloured in blue letters while those from this study have red branches. Strains from South Africa were largely clustering together albeit the isolates also clustered with strains from Nigeria, Cameroon, Uganda and Sudan that had the same or closely related ST. Notably, strains from Durban (South Africa) were more closely related to our strains (B). Globally, the strains were related to clones from Belgium, Brazil, China, Ghana, India, Lebanon, Thailand, UK, and USA (C-E). RAXmL and Parsnp were used to draw the trees, which were subsequently annotated with Figtree. and the virulome or capsule types. K094 presents a very interesting observation in that it contained no virulence gene or capsules, had no mutations in genes conferring resistance to fluoroquinolones and colistin, contained very few resistance genes (n = 8) and was phylogenetically distant from all STs and clades (Figs. [3][4][5].
The diversity of sequence types identified in these MDR K. pneumoniae isolates is consistent with findings reported from South, Central and North America, Europe, Asia and North Africa [70][71][72][73][74] . While ST152 and ST1552 were dominant in this study, and related to Chinese, Ghanaian and Thai genomes, they are not often described in the literature, compared to ST238, an ST responsible for many CRE outbreaks globally 6,75,76 . The only globally reported ST identified was ST15, confirming the global diversity of this clonal group that was identified in 15% of K. pneumoniae CRE isolates in an international multicentre study encompassing Morocco, Cameroon, Senegal, Madagascar and Vietnam 77 . ST152 has also been reported in Durban 14 , as a major clone in Cuba 77 and as a major career of NDM-1 in Saudi Arabia 78 . However, ST147 and ST258, which are global STs associated with antimicrobial resistance 79-81 , were not detected in these isolates. Most sequence types had global spread, being more frequently related to Ghanaian, Thai and Chinese isolates. Four sequence types, viz., ST1552, ST234, ST1414 and ST152 also demonstrated local spread. Local and international outbreaks, in both South Africa and abroad, were also observed in the trees under distinct clades.
An analysis of the molecular epidemiology of these isolates confirmed the international dissemination of specific STs between South Africa and the world. Clonal similarity of isolates from South Africa, Thailand, Nigeria and China was evident although there was a diversity of sequence types associated with other countries. There was also similarity in clones within South Africa; specifically, between Durban and Pretoria. The resolution power of whole-genome sequencing over MLST is demonstrated herein by the clustering of different STs on the same branch and clade, supporting the need to shift to genomic epidemiology for better epidemiological surveillance and infection control.

conclusion
The burden of ARGs and virulence genes in these isolates from hospitalised patients in two major referral hospitals within Pretoria confirm the global threat of ABR, mediated by MGEs. The findings demonstrate the centrality of MGEs in defining the resistome of MDR strains. Phylogenetic analysis confirmed this global spread including evolutionary relationships of the different STs. IncF plasmid replicons and class1 integrons, both of which have been globally reported, were also dominant in these isolates. Significantly, two novel integrons were identified. The presence of the rare chromosomal bla LEN gene in four isolates is also notable.

ethics. Ethical approval was provided by the Human Research Ethics Committee of the University of
Witwatersrand (Ref M1710100). All protocols and consent forms were executed according to the agreed ethical approval terms and conditions. All clinical samples were obtained from a reference laboratory and not directly from patients, who agreed to our using their specimens for this research. The guidelines stated by the Declaration of Helsinki for involving human participants were followed in the study.