Genomic and phenotypic characterisation of fluoroquinolone resistance mechanisms in Enterobacteriaceae in Durban, South Africa

Resistance to fluoroquinolones (FQ) is being increasingly reported and found to be mediated by efflux pumps, plasmid-mediated quinolone resistance genes (PMQR) and mutations in gyrA, gyrB, parC and parE. However, studies reporting on FQ resistance mechanisms (FQRM), particularly in Africa, are focused mostly on Salmonella. This study used a whole-genome-based approach to describe FQRM in forty-eight clinical Enterobacteriaceae isolates comprising of Klebsiella pneumoniae (n = 21), Serratia marcescens (n = 12), Enterobacter spp. (n = 10), Citrobacter freundii (n = 3), Escherichia coli (n = 1), and Klebsiella michiganensis (n = 1) with reduced susceptibility to FQ in Enterobacteriaceae. All the isolates exhibited exceptionally high-level resistance (MIC of 4-512mg/L) to all three FQs, which could not be reversed by carbonyl cyanide m-chlorophenyl hydrazine (CCCP), verapamil (VRP) or reserpine (RSP). PMQR genes such as oqxAB (n = 43), aac(6’)-Ib-cr (n = 28), and qnr(S1, B1, B2, B9, B49, B66) (n = 23) were identified without transposons or integrons in their immediate environments. Multiple and diverse mutations were found in gyrA (including S83I/Y and T/I83I/T), gyrB, parC and parE, which were clonally specific. There were vertical and horizontal transmission of high-level FQ resistance in Enterobacteriaceae in hospitals in Durban, South Africa, which are mediated by efflux, PMQR genes, and gyrA, gyrB, parC and parE mutations.


Introduction
Until recently, fluoroquinolones (FQ), which were made from the first quinolone called nalidixic acid (NAL), (a by-product of chloroquine synthesis) were very potent and were the most prescribed broad-spectrum antibiotic class for treating fatal bacterial infections [1,2]. They are the only antibiotic class that directly inhibits DNA synthesis/replication by inhibition of DNA gyrase (encoded by gyrA and gyrB) and topoisomerase IV (encoded by parC and parE) [1,2]. However, their increased overuse has led to increased resistance among both Gram-negative and Gram-positive bacteria, making them less effective [1,2]. Hence, they are used more in a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 combination therapies with last line antibiotics such as carbapenems, colistin and tigecycline than as monotherapy [3,4].
Although there are numerous studies characterising the prevalence and molecular epidemiology of FQRM, such studies are largely focused on Salmonella enterica and to a lesser extent, Escherichia coli and Vibrio cholerae through the use of PCR and pulsed field gel electrophoresis (PFGE)-based typing, particularly in Africa and South Africa [1,[11][12][13]. Thus the ability to compare FQ-resistant strains between different countries in Africa is limited and a true genomic characterisation and epidemiology of FQRM, specifically among Enterobacteriaceae is scarce [1]. Due to the ability of enteric bacteria to associate in biofilms and share plasmids among themselves, it is imperative to broaden the scope of research beyond Salmonella to identify other Enterobacteriaceae that are reservoirs of FQ resistance. To our knowledge, there is no study using whole-genome sequencing to (WGS) describe FQRM in Enterobacteriaceae in South Africa and to a large extent, in Africa.
To provide a comprehensive description of FQRM among Enterobacteriaceae in Durban, South Africa, this study was undertaken using a large collection of diverse Enterobacteriaceae species with the view of providing a bedrock to facilitate comparative analysis in future studies and enhance meaningful epidemiological conclusions and resolutions. Moreover, an in-depth description of the transfer mechanisms of FQ resistance is important for the arrest and control of FQ-resistant strains in hospitals.

MICs of CIP, NOR, and NAL with and without the inhibitors
The MICS of ciprofloxacin (CIP), norfloxacin (NOR) and nalidixic acid (NAL) were determined for all the isolates and controls both in the absence and presence of efflux pump inhibitors (EPIs) to assess the role of efflux pumps in FQs resistance. The MICs of NAL was very high (>512mg/L) for all the isolates whilst that of CIP and NOR ranged from 4 to 512mg/L, which makes all the isolates very resistant per the EUCAST (2016) breakpoints; MIC of >1 mg/L is defined as resistant (Table 1 and S1 Table) [14]. Most of the isolates had CIP and NOR MICs above 128mg/L.
An MIC fold change of !8 was adopted as significant [15]. Hence, none of the inhibitors viz., carbonyl cyanide-m-chlorophenylhydrazine (CCCP), verapamil (VRP) and reserpine (RSP) could affect the MICs of NAL (Table 1 and S1 Table) significantly i.e. had a fold change of !8, albeit fold changes of 2-4 were recorded in S. marcescens, Enterobacter spp., C. freundii and K. michiganensis. CCCP, followed by VRP, had the least effect on the MICs of the antibiotics in all the species whilst RSP had the most effect; majority of the MIC fold changes effected by RSP were significant (Table 1 and S1 Table). RSP resulted in more significant MIC fold changes in NOR (n = 24) than in CIP (n = 16) and was the only EPI that effected a consistently significant MIC fold-change reduction in all the isolates except for NAL (S1 Table).
Among the species, E. coli had no significant MIC fold change and only RSP resulted in significant MIC fold changes in K. pneumoniae; however, both RSP and VRP resulted in significant

Species frequency and distribution of plasmid-mediated quinolone resistance (PMQR) genes
The presence and frequency distribution of each PMQR gene was assessed in each isolate' genome sequence. No qepA gene was found in any of the isolates. Moreover, no PMQR gene  PMQR genes occurred in all isolates except E. coli. oqxAB genes were the most common occurring PMQR gene(s) followed by aac(6')-Ib-cr and qnr genes. oqxAB was common in K. pneumoniae whilst aac(6')-Ib-cr and qnr were dominant in Enterobacter spp.
No transposon or integron was found in the immediate environment of the qnr and aac (6')-Ib-cr genes. The truncated nature of the plasmid contigs made it impossible to link the PMQR genes to particular plasmids or replicon/incompatibility types.

Multi locus sequence typing (MLST) and S. marcescens typing
The MLST of the isolates are shown in Table 2 and it can be seen that the isolates were very multiclonal, albeit many K. pneumoniae strains (n = 14) were of sequence type (ST) 101. The MLST of K. michiganensis was determined using that of K. oxytoca due to misidentification of the former as K. oxytoca.
Alignment of the dnaA, fusA, gyrB, leuS, pyrG, rplB, and rpoB housekeeping genes of all the 12 S. marcescens isolates resulted in 10 of the isolates being classified into the same clone (i.e. SA1) and two of the remaining being grouped into SA2 clone. Isolates of each clone showed no variation from the reference sequences of housekeeping genes of their respective clones. The SA2 clone only differed from SA1 clone at the nucleotide level in dnaA and leuS by a single (T106C) and double (A7G and A91G) silent single nucleotide polymorphisms (SNPs) respectively

Mutations in gyrA, gyrB, parC and parE
The sequences of gyrA, gyrB, parC and parE were analysed to determine the presence of mutations that could account for the high-level FQ resistance as mutations in these four genes have been incriminated in high-level FQ resistance. The mutations in gyrA, gyrB, parC and parE mostly occurred in a manner that reflected the clonality of the strains i.e. the mutations in all the genes were mostly clonally related. For instance, K. pneumoniae ST101 and ST2017 had the same mutations in all the genes except for parC at which only three ST101 strains (J, D and C) differed. The mutations in K. pneumoniae ST1478, ST323 and ST14 were unique. The clonal nature of the mutations was more obvious with S. marcescens in which all SA1 and all SA2 clones had the same mutations for these genes within their members. The close relation between SA1 and SA2 clones were further underscored by the similarity and few differences in the mutations observed in their gyrA, gyrB, parC and parE genes. This observation corroborated the correctness of our typing system. Within the Enterobacter spp. and C. freundii isolates, the uniqueness of their gyrA, gyrB, parC and parE mutations was but a reflection of their different species or clones. However, C. freundii isolates 51_S25 and 48_S23, except isolate 14_S7, had same mutations in gyrB, parC and parE. Positions 83 in gyrA was a common mutation position in many isolates and species whilst mutations at position 80 in parC was common in C. freundi, E. coli, Enterobacter spp,. K. pneumoniae and K. michiganensis. No parE mutations were found in K. pneumoniae (except in 47_S22), S. marcescens, C. freundii and K. michiganensis (Table 2).

Discussion
The importance of this study is underscored by the discovery that there is a substantial number of high-level fluoroquinolone resistance among clonal and multiclonal Enterobacteriaceae circulating in hospitals in Durban, South Africa. This is evinced by the exceptionally high-level of MICs, mostly from 64mg/L to >512mg/L that was recorded among isolates of all species (Table 1 and S3 Table). To our knowledge, no study in South Africa or Africa has discovered such a number of high-level FQ-resistant Enterobacteriaceae with multiple mechanisms of FQ resistance, albeit not all the collected 82 isolates were analysed. This is a rather disturbing observation given the importance of FQ in the management of several bacterial infections in South Africa and the importance of South Africa as a destination for many medical tourists from Africa and Europe [1,4].
To determine the role of efflux pumps as mechanisms underlying this high-level FQ resistance, the isolates were grown in the presence of sub-MICs of CCCP, RSP and VRP. Interestingly, CCCP, a protonophore that indirectly reduce efflux pumps activity by reducing ATP production and unbalancing the electrochemical gradient could not significantly affect the isolates' MICs. Gram-negative bacteria, including Enterobacteriaceae, are known to have ABCtype efflux pumps that are directly powered/energised by ATP hydrolysis and secondary transporters (sodium-proton symporters) such a MATE-, SMR-, MFS-and RND-type efflux pumps that are powered by the proton motive force (PMF) [16][17][18]. Hence, the inability of CCCP to reduce the FQ MICs is rather surprising but it shows that efflux pumps affected by CCCP were not involved in resistance to CIP, NOR and NAL.
VRP has been shown recently to affect MATE-type efflux pumps by directly blocking the pumps channels [17,15]. Subsequently, we are convinced that MATE-type efflux pumps were involved in conferring resistance to CIP (n = 5) and NOR (n = 6) in a few of the isolates affected by VRP (Table 1). Moreover, it can be concluded that RSP, which is believed to affect RND-type efflux pumps in Gram-negative bacteria [15,19,20], was mostly involved in reducing the MICs of CIP (n = 16) and NOR (n = 24) significantly in many of the isolates. Nevertheless, it must be borne in mind that in all a total of 25 isolates, which is approximately half of the total sample size, were affected by the inhibitors, suggesting that efflux as a mechanism of FQ resistance, was present in one out of every two isolates.
This implicating of efflux as a FQRM in approximately half of the isolates is comparable to five studies involving FQ resistance, three of which involved Salmonella [9,10,21], one involved Vibrio cholerae [13]and the last involved Aeromonas veronii and A. hydrophila [12]. Out of these five studies, efflux was implicated as a FQRM in three [9,10,13] whilst they were found not to be involved in two studies [12,21]. Thus, efflux is not always a mechanism of resistance to FQ, albeit an important one. That the inhibitors were unable to reverse FQ resistance in any of the isolates is a testimony that other mechanisms other than efflux was involved [21]. Annotation of the WGS data showed that the most predominant PMQR was oqxAB, which has never been reported as a FQRM in any study in South Africa or Africa as far as we are concerned [1]. This is possibly because PCR has been used in gene-targeted amplification and sequencing, which has not involved the search for oqxAB genes; this stresses the need to incorporate WGS into antibiotic resistance research in Africa [1]. Interestingly, oqxA were not as prevalent as oqxB in the isolates and a pattern or correlation could not be established between their presence and the MICs of the isolates. However, as suggested by other authors [22,23], we also hypothesise that oqxB can work through acrA in the absence of oqxA as oqxAB has been shown to be a supplementary resistance mechanism to acrAB and its increased expression is dependent on a functional acrAB [22,23]. Hence, oqxAB is adding up to other mechanisms to increase the FQ resistance in the isolates and cannot be a major mechanism, particularly when equally or much higher MICs were recorded in strains (e.g. E. coli and C. freundii vis-à-vis S. marcescens and Enterobacter spp.) in which they were absent ( Table 2).
The second most dominant PMQR gene was aac-(6')-Ib-cr, which makes this the first study to report on the presence of this gene as a FQRM in Enterobacteriaceae in South Africa. However, this gene has been implicated in FQ resistance in E. coli in Nigeria and a few other African countries [1]. The qnr genes were the least prevalent of the PMQR genes although they are the only PMQR genes to have been reported already in S. Africa and are commonly identified in other African countries [1,12,21]. The commonest variants of the qnr gene identified in this study were the qnrS1, qnrB1 and qnrB49, which agrees with a study by Smith et al. [10] involving Salmonella and by Chenia [12] involving A. veronii and A. hydrophila that found only qnrS1 and qnrB and qnrS respectively. However, Chenia found a higher percentage of qnrB (41%) than qnrS (24%) from freshwater fish whilst we found qnrS1 to be higher than qnrB in clinical samples. There is the need to undertake further surveillance and molecular epidemiology to assess the possibility of these qnr genes being transferred from freshwater foods to humans or of human activities being the source of these resistance genes in freshwater and the environment as has been recently reported in China where PMQR genes in high numbers were found in Aeromonas spp. from rivers contaminated by hospital and aquaculture effluents [2,12].
On the whole, more studies have implicated mutations in gyrA, gyrB, parC and parE as the major mechanism of FQ resistance as the PMQR genes only mediate low-level FQ resistance [9][10][11]13]. This can be seen in our findings in which mutations in these four genes (except in parE in most isolates) were found in all isolates as compared to the PMQR genes that were not found in all strains. The numerous nature of the mutations observed in the individual genes as well as collectively in all four genes will surely add up to make the strains highly resistant as recorded. The common mutations seen per clone testifies to the clonality of the isolates and authenticates our typing scheme, particularly in S. marcescens. One of the commonest gyrA mutation recorded in literature is Ser83Ile or Ser83Tyr, which was also found in our strains ( Table 2). Chenia [12] found Ser80Ile in the parC of her A. hydrophila/veronii strains and this was also found among our isolates. As the mechanism of action of FQ involves the DNA gyrase and topoisomerase IV, it is reasonable that mutations in genes encoding these FQ targets should mediate high-level FQ resistance [1,2].
Hence, we conclude that high-level FQ resistance in Enterobacteriaceae in South Africa is mainly mediated by numerous and diverse mutations in gyrA, gyrB, parC and parE, in synergy with efflux upregulation and PMQR genes, and that these highly resistant strains are being disseminated vertically by clonal and multiclonal expansion and horizontally via plasmids in hospitals. Given that the use of other antibiotics can select for PMQR genes [8], the need for antibiotic stewardship is exceptionally important to prevent further escalation of this menace.

Ethical approval
The Biomedical Research Ethics Committee of the University of KwaZulu-Natal approved this study under the reference number BE040/14.

Bacterial strains
From a collection of 82 Enterobacteriaceae isolates with reduced susceptibility to FQ, which were collected from a private pathology laboratory in Durban, South Africa between 2012 and 2013, 48t isolates comprising of Klebsiella pneumoniae (n = 21), Serratia marcescens (n = 12), Enterobacter spp. (n = 10), Citrobacter freundii (n = 3), Klebsiella michiganensis (n = 1), and Escherichia coli (n = 1) that were fully FQ-resistant per further micro-broth dilution (MIC) testing, were selected. Disc diffusion with ciprofloxacin (CIP), norfloxacin (NOR), and nalidixic acid (NAL) was used to identify and collate these 82 isolates in the pathology laboratory, where all isolates with reduced susceptibility to FQ were included; however, further confirmatory testing with MIC showed that only 48 were fully FQ-resistant per EUCAST (2016) breakpoints [14,24]. Hence, only the 48 isolates were included. The isolates were presented by 10 different hospitals to the private pathology laboratory and were obtained from patients of both sex between the ages of 2 months and 83 years. E. coli ATCC 25922 and K. oxytoca ATCC 13178 were used as controls in the antibiotic sensitivity testing/screening (S2 Table).
Identification of the isolates were initially done with Vitek II and confirmed by NCBI BLAST and NCBI's ANI report of the whole genome sequence of the isolates. Thus, the Enterobacter spp. were identified as Enterobacter cloacae (n = 2), E. asburiae (n = 3), E. kobei (n = 2), Enterobacter cloacae complex "Hoffman cluster IV" (n = 1), Enterobacter cloacae complex "Hoffman cluster III" (n = 1), and Enterobacter sp. (n = 1). and used for the broth micro dilution assays. MIC determination and results interpretation for CIP, NOR and NAL were done according to EUCAST guidelines and breakpoints (2016) [14]. Deionized water was used to make VRP solutions whilst RSP was prepared in dimethyl sulfoxide (DMSO) and CCCP in 50% methanol (v/v) [25]. Cation-adjusted Mueller Hinton broth was used to determine the minimum inhibitory concentrations (MICs) of CIP, NOR, NAL, and VRP. All solutions were prepared on the day of the experiment and kept protected from the light.

Minimum inhibitory concentrations (MICs
In determining the effect of CCCP, RSP and VRP on CIP, NOR and NAL MICs, a sub-MIC (i.e. 0.5 × MIC) of CCCP, RSP and VRP viz., 8, 256 and 256mg/L were used respectively whilst serially increasing the concentrations of the antibiotics to determine the change in MICs (Table 1 and S1 Table). A sub-MIC was used to reduce the possibility of CCCP, RSP and VRP killing the cells and interfering with the true MIC of the antibiotic-inhibitor combination.
CCCP was used as a protonophore to assess its effects on the antibiotics' MICs through a reduction in efflux activity and disruption of the transmembrane electrochemical potential/ gradient. It acts by unbalancing the transmembrane proton gradient, reducing ATP production and indirectly inhibiting efflux activity [16]. RSP and VRP are broad spectrum efflux inhibitors that are commonly used in efflux inhibition experiments. They were used in this study to assess their effect on the antibiotics' MIC through inhibition of RND and MATE efflux pumps respectively [16,17,19,20].

Genomic DNA extraction, library preparation and whole genome sequencing
Genomic DNA (gDNA) of the isolates were extracted using the GenElute Bacterial Genomic DNA kit (Sigma Aldrich, St. Louis, MO, USA) per the manufacturer's instructions. The Qubit and Nanodrop were used to determine the concentrations and purity of extracted gDNA. The quality and integrity of the purified gDNA were confirmed by performing gel electrophoresis before proceeding to library generation. Libraries were quantified on the Bioanalyzer (Agilent Technologies) and combined in an equimolar mixture. A nanogram of gDNA was used as input for the Nextera XT kit (Illumina) to generate 300 bp paired-end libraries followed by sequencing on an Illumina MiSeq platform; a genome coverage of 20-90x was generated for all the isolates [6,26,27] (S3 Table).

Bioinformatic analysis
MLST of the isolates were determined using MLST 1.7 pipeline at the Center for Genomic Epidemiology (https://cge.cbs.dtu.dk/services/MLST/). The PMQR genes (aac(6')-Ib-cr, qepA, qnr and oqxAB) were tabulated and their genetic support/environment, namely transposons and integrons, were searched for using the annotations provided by PGAP. Translated nucleotide Basic Local Alignment Search Tool [tBLASTn] was used to search for genetic elements in the immediate environment of the aac(6')-Ib-cr, and qnr genes.
Typing of the S. marcescens isolates was done in-house using seven house-keeping genes that had been used for typing Enterobacter cloacae.: dnaA, fusA, gyrB, leuS, pyrG, rplB, and rpoB [30]. The same housekeeping genes in each S. marcescens isolate were aligned to each other in a gene-gene format using nucleotide BLAST [BLASTn] and SNPs were called for all seven genes. A single SNP in any of the genes was used as a cut-off for categorising the isolates into different clones.
Mutations in the chromosomal-borne gyrA, gyrB, parC and parE genes were determined using tBLASTn to call SNPs in these genes. Fluoro/-quinolone susceptible reference/type strains that were used for each species were as follows: K. pneumoniae ATCC 13883 (PRJNA244567) for K. pneumoniae; S. marcescens strain CDC_813-60 (ATCC 13880)

Data analysis
An MIC fold change was defined as the ratio of the MIC of antibiotic and inhibitor to that of the antibiotic alone. A fold change of !8 was deemed as significant and indicative of efflux activity [15]. Where the MIC was >512, an absolute value of 512 was used, specifically for NAL (S1 Table). The frequency and distribution of the PMQR genes per species were tabulated and translated into a graph (Fig 1).