Salmonella isolates (126) used in this study were isolated from pork samples in a large-scale slaughterhouse between 2013 and 2014 (Yang et al., 2017). The ciprofloxacin MICs ranged from <0.015 to 64 mg/L although none of these isolates were inhibited by ciprofloxacin concentrations between 0.125 and 2 mg/L. In order to generate a consecutive ciprofloxacin MIC distribution range for each MIC value, we selected 50 additional strains from food-producing animals and raw meat (chicken and pork) as described elsewhere (Zhang et al., 2016; Zhang C.Z. et al., 2019; Yang et al., 2017). This included 11 isolates from swine, 9 from chickens, 17 from ducks and 13 from meat samples (10 chicken and 3 pork) (Table 1). Although efforts were made to screen for strains with a ciprofloxacin MIC between 0.25 and 2 mg/L, these strains were rare. We only identified 3, 3, and 4 isolates in the entire strain collection that possessed ciprofloxacin MICs of 0.5, 1, and 2 mg/L, respectively.

The majority of the strains isolated from pork samples were serovars Derby, Rissen and Indiana as previously described (Yang et al., 2017). The 50 additional isolates included Typhimurium (15), Indiana (12), Enteritidis (7), Derby (5), and 11 others.

Salmonella enterica subsp. enterica serovar Typhimurium SL1344 (MIC_CIP = 0.015 mg/L) was kindly provided by Prof. Laura Piddock (Institute of Microbiology and Infection, University of Birmingham) and used as a standard reference strain in this study. The MICs of ciprofloxacin were determined by the agar dilution method following CLSI guidelines (CLSI, 2019).

Identification of mutations in the quinolone resistance determining regions (QRDR) of gyrA, gyrB, parC, and parE and identification of qnr, aac(6′)-Ib-cr, qepA and oqxABR genes were performed using PCR amplification, primers used are listed in Supplementary Table 1. All PCR amplicons were sequenced by the BGI (Shenzhen, China), and the wild-type Salmonella Typhimurium LT2 strain was used as a comparison to identify mutations.

All oqxABR-positive Salmonella isolates were screened for the presence of the IS26 insertion sequence (IS), and cyclization of the IS26-oqxABR region was determined using primers oqx-IF and oqx-IR. Sequence identity of all PCR amplicons was confirmed by DNA sequence analysis.

The location of PMQR genes and the sizes of PMQR gene-positive plasmids were estimated by S1-PFGE and/or I-CeuI-PFGE followed by Southern blot hybridization with digoxigenin-dUTP labeled probes of oqxAB, qnrS, and/or 23S rDNA genes. Primers used are listed in Supplementary Table 1. Briefly, agarose-embedded DNA for each Salmonella strain was digested with 1 U S1 nuclease (Takara Biotechnology, Dalian, China) at 37°C for 45 min or I-CeuI (NEB, Ipswich, MA, United States) for 3 h. The restriction fragments were separated by electrophoresis in 0.5 × Tris-borate-EDTA buffer at 14°C for 19 h with pulse times of 2.2 to 54.2 s using a CHEF-MAPPER System (Bio-Rad Laboratories, Hercules, CA, United States). Agarose-embedded DNA from strain H9812 that had been in situ digested with XbaI (Takara Biotechnology, Dalian, China) at 37°C for 3 h was used as a DNA size marker. The gel was stained with ethidium bromide and DNA bands were visualized under UV light. DNA was transferred to a Hybond-N + membrane for Southern hybridization using conditions suggested by the manufacturer (GE Healthcare, Little Chalfont, United Kingdom). Blots were probed with PMQR genes containing digoxigenin labels using a commercial DIG High Prime DNA Labeling and Detection Starter Kit I (Roche Applied Science, Mannheim, Germany).

Bacterial strains were grown in minimal media to an OD₆₀₀ nm of 0.6 and total RNA was isolated using the RNAiso Plus Kit (Takara); mRNAs were reverse transcribed using PrimeScript RT with gDNA Eraser (Takara). The cDNA was quantified using an iQ5 multicolor real-time PCR system (Bio-Rad) using gene-specific primers (Supplementary Table 1), and an iQ SYBR Green Supermix (Bio-Rad). The 16S rRNA gene was used as an internal control gene and relative expression levels of each gene were calculated using the ΔΔCt method and the software provided with the instrument. Data are presented as mean ± SD from three independent assays, in which each RNA sample was tested in triplicate.

To directly measure fitness impacts of different resistance mutations we used competition assays between pairs of strains selected to carry different combinations of resistance mechanisms, and with different susceptibility levels to ciprofloxacin. The selected pairs competed against each other were; S(−)/RS(−), RS(c +)/I(p +), I(p +)/R(c +), and I(p +)/HR(p +). In addition, to eradicate any possible strain and serovar specific bias, competition assays were also completed between selected serovar Indiana strains with different resistance mechanisms, RS(−)/HR(p +) (strains HB137/K46) and I(p +)/HR(p +) (strains SP80/K46, and SP80/CL108).

Fitness was analyzed as previously described with some modifications (Zhang C.Z. et al., 2017). Growth competition was determined by pyrosequencing the single nucleotide variations in ParC which distinguished each pair of competed strains. Briefly, the competitive paired strains were mix and co-cultured (1:1 ratio) in antibiotic-free LB broth for 18 h and DNA was extracted using a standard phenol-chloroform-heat method. The DNA from the competitive growth assays was amplified by PCR in triplicate using biotinylated primer pairs targeting the region containing the single nucleotide polymorphism to distinguish the two organisms in the assay by mutations in ParC. All PCR amplifications were visualized on 1% agarose gels prior to pyrosequencing. The purified PCR products were pyrosequenced at BGI. Competition coefficients were defined as the ratio of the strain pairs subjected to different incubation conditions and calculated by measuring percentage yield of the single nucleotide mutations in parC. A competition coefficient <1 indicated that the less susceptible strain was more abundant after competition.

A primary aim of our study was to identify Salmonella isolates with a broad range of ciprofloxacin MICs. We selected 176 Salmonella strains from 1,280 isolates, amongst which 91 were susceptible to ciprofloxacin, 20 were deemed intermediate and 65 were resistant according to CLSI breakpoints. However, epidemiologically there were clearly more than three levels of susceptibility in the populations, likely to reflect distinct combinations of resistance mechanisms. Therefore, we applied a 5-level interpretation system using “epidemiological breakpoints.” This system relies on overlapping normal distributions of MIC values (Kronvall, 2010; Zayed et al., 2015). The resulting categories were designated as: highly resistant (HR), resistant (R), intermediate (I), reduced susceptibility (RS), and susceptible (S) (Figure 1). We then randomly selected a total of 32 strains, 5∼7 strains from each group, for detailed study of gene expression and carriage of specific mutations (Table 2 and Supplementary Figure 1).

Among the isolates that were analyzed, none contained QRDR mutations in gyrB and parE. In contrast, parC mutations that resulted in the ParC T57S substitution were present in S and RS strains with an MIC_CIP of 0.06 mg/L. ParC T57S was not seen in RS strains when the MIC_CIP increased to 0.25 and 0.5 mg/L. Isolates with these MICs carried single GyrA S83Y or D87Y substitutions. The “I” strains tested only possessed GyrA D87N single substitutions; interestingly, the ParC T57S substitution was again seen in some R strains, but these lacked GyrA mutations (although had other mechanisms as outlined below). The HR strains all possessed multiple substitutions in both GyrA (S83F/D87G) and ParC (T57S/S80R) (Table 2). The identification of these mutations was positively correlated with their assignment into the 5-level interpretation system.

Reduced susceptibility strains (ParC T57S) that possessed MIC_CIP values of 0.06 mg/L also harbored chromosomal copies of oqxABR. PMQR genes were not present in the RS strains that contained the single GyrA substitutions S83Y or D87Y. When the MIC_CIP levels reached 1–2 mg/L (I strains), the presence of plasmid-borne copies of both oqxAB and aac(6′)-Ib-cr were detected on ∼180-kb plasmids. In the R strains, oqxAB, aac(6′)-Ib-cr and qnrS were detected and all were located on the chromosome. In HR strains, aac(6′)-Ib-cr was found on either ∼240-kb plasmids or was present with oqxAB and qnrS2 on the chromosome (Table 2).

We next analyzed whether increases in efflux pump expression correlated with the MIC categories of our isolates. The expression of efflux complex (acrA, acrB, and tolC) and regulatory (ramA, marA, and soxS) genes for the subset of 32 Salmonella isolates were measured (Table 2 and Supplementary Figure 2). Compared with S. Typhimurium SL1344, the RS strains exhibited higher expression of acrAB-tolC and the regulator marA but not ramA. Furthermore, the majority of I strains exhibited increased expression of efflux pump genes with concurrent overexpression of marA. Surprisingly, the efflux pump genes were not expressed more in R strains compared with S. Typhimurium SL1344, whereas there was increased expression of acrA, acrB, ramA, and marA in HR strains compared with S. Typhimurium SL1344. A two-way ANOVA test of expression values for the regulator genes indicated the differences between groups were not likely to be observed by chance (p 0.007).

All 16 oqxABR-positive strains possessed IS26 sequences flanking the oqxABR operon (Figure 2A). In addition, all these strains, regardless of whether the genes were located on a plasmid or chromosome, were positive for the presence of circular intermediates containing IS26 and the oqxABR genes (Figures 2B,C). Sequencing of this circular intermediate confirmed that the length of the intermediate was 6,026 bp and that oqxABR harbored a complete copy of IS26. This intermediate structure was like the oqxABR operon present in the widespread composite transposon Tn6010 found in plasmid pOLA52 from E. coli (Norman et al., 2008). These results indicate that the Tn6010 element was unstable and prone to excision in our Salmonella isolates.

We compared the competitive fitness of closely related pairs of isolates selected to represent the different combinations of resistance mechanisms and susceptibility levels observed in the whole panel. This included the presence of target gene mutations and presence of PMQR genes on the chromosome (c +) or on plasmids (p +). The competitive growth experiments included the pairs S(−)/RS(−), RS(c +)/I(p +), I(p +)/R(c +), and I(p +)/HR(p +) and Indiana strains RS(−)/HR(p +) (strain HB137/K46) and I(p +)/HR(p +) (strain SP80/K46, and SP80/CL108). All strains used in competition assays are shown in Table 3.

The competition coefficient for the S (−)/RS (−) pair was 1.13, suggesting that the presence of a substitution of GyrA D87Y presented a burden to strain HB137 (RS), while ParC T57S did not impair strain M36 (S). The competition coefficient for RS(c +)/I(p +) counterparts was 1.53, indicating that carriage of both GyrA D87N and the two plasmid PMQR genes produced a burden on strain SP79 (I). The I(p +)/R(c +) competition yielded a competition coefficient of 0.76, indicating the three chromosomal PMQR genes in strain F64 (R) produced less of a burden than when these genes were plasmid-borne. The HR strain K46 carried 2 mutations in GyrA and 2 in ParC. The presence of the plasmid-borne PMQR gene aac(6′)-Ib-cr resulted in a competitive disadvantage when compared with the I strain SP79 that possessed a GyrA single mutation; this resulted in a competition coefficient of 1.95 for I(p +)/HR(p +) (Figure 3). The same patterns were seen from the corresponding experiments using identical Indianan strains [I(p +)/HR(p +)], SP80/K46, and SP80/CL108. However, RS strain HB137, harboring a substitution of GyrA D87Y and overexpressing AcrAB, presented a competitive disadvantage when compared with HR strain K46.

The development of FQ resistance is understood to often require accumulation of multiple mutations in a stepwise process, with mutations that alter GyrA at codon 83 often being considered the “first step” to selection of high-level resistance. Here, we observed a high prevalence of the ParC T57S substitution in strains with lowest MICs of ciprofloxacin (S and RS strains). Whether the ParC T57S substitution contributes to quinolone resistance, is still controversial (Baucheron et al., 2002; Ling et al., 2003; Eaves et al., 2004; Gunell et al., 2009; Zhang W.H. et al., 2017). We found that the presence of ParC T57S was altered as the ciprofloxacin MIC increased; this mutation only appeared alone in strains with MIC_CIP ≤ 0.06 mg/L. Therefore, we suggest that any impact of this mutation alone of quinolone susceptibility must be low and can only provide a low degree of protection against ciprofloxacin. This change alone may carry little fitness cost which would explain its prevalence in the S and RS strains, however, it may incur a different impact on fitness when associated with GyrA substitutions as no strain carried a single substitution within ParC and any GyrA change in any of the groups. Most resistant strains did carry multiple substitutions within both GyrA and ParC and the specific nature of these changes is likely to impact fitness.

In the intermediate isolates where the MIC_CIP was higher (I; 4–8 mg/L), strains carried a QRDR mutation, such as S83Y or D87Y/N but the ParC T57S mutation was not present (suggesting this is not a precursor required for development of resistance). ParC T57S was seen in R and HR strains, along with multiple substitutions in both GyrA and ParC, this demonstrates the ParC T57S substitution is viable when in combination with multiple other topoisomerase mutations.

Expression of acrAB-tolC was elevated in strains carrying the ParC T57S substitution, but only those where the MIC_CIP was below the resistance breakpoint. This again suggests that a combination of the ParC T57S change with increased active efflux activity may be important in the early stages of resistance development (where MICs are relatively low). The growth competition assays revealed that the ParC T57S mutation was associated with enhanced bacterial fitness again supporting a low-impact but high-fitness role for this substitution.

Plasmid-mediated quinolone resistance determinants cause decreased susceptibility to FQs while also facilitating the selection of higher levels of FQ resistance (Rodriguez-Martinez et al., 2016). In the present study, oqxAB was the most prevalent PMQR gene detected. It was not present in S strains but was first observed in RS strains with an MIC_CIP 0.06 mg/L and the gene was chromosomally located in all the positive RS strains. The occurrence of oqxAB in combination with existing mechanisms, such as ParC T57S and elevated expression of acrAB-tolC, were linked to significant increases in MIC_CIP, compared to strains without oqxAB but with these other changes. The chromosomal location of oqxAB did not appear to incur a fitness cost in strains which also increased acrAB-tolC expression and may represent a successful combination of these two mechanisms in terms of susceptibility and fitness. Isolates with an MIC_CIP of 0.25–0.5 mg/L, also carried mutations in gyrA, which imposed a fitness cost.

Some isolates where the MIC_CIP achieved clinical resistance levels (although without additional mutations in the target genes) carried >2 PMQR genes (oqxAB, aac(6′)-Ib-cr, and qnrS) combined with mutation in gyrA. These genes were found in plasmid locations in intermediate strains (MIC_CIP 1–2 mg/L) while being observed at chromosomal locations in R strains (MIC_CIP 4–8 mg/L). Interestingly, the R group did not exhibit any known chromosomal mechanisms, such as target gene mutations or overexpression of acrAB-tolC for increased resistance, but did carry three plasmid-borne PMQR genes. This phenomenon is consistent with previous observations where the presence of 2–3 PMQR genes was observed in ciprofloxacin resistant Salmonella isolates from pork samples that lacked any target gene mutations (Lin et al., 2015). Our results confirmed that the presence of multiple PMQR genes or multiple copies of PMQR genes can confer resistance to ciprofloxacin.

OqxAB is an efflux pump from the RND family that mediates resistance to olaquindox, chloramphenicol and nalidixic acid. The presence of this pump elevates MICs for other antimicrobial reagents including ampicillin and gentamicin (Hansen et al., 2007). The pump is encoded by an operon (oqxABR) flanked by IS26 sequences, constituting a composite transposon (Tn6010) (Norman et al., 2008).

The oqxABR operon can form a circular structure to facilitate mobilization (He et al., 2015). In this study, we observed that oqxABR shifted from chromosomal to plasmid locations in isolates with relatively high MIC_CIP. We performed inverse PCR to determine whether an IS26-oqxAB circular intermediate was involved in this process. We found that all 16 oqxAB-positive strains possessed the circular intermediate regardless of genomic location. This indicated that the circular intermediate promotes mobilization of this PMQR between chromosome and plasmid. Interestingly, we detected qnrS in R strains located on the chromosome with aac(6′)-Ib-cr and oqxAB. This is consistent with other several studies, where qnrS co-existing with aac(6′)-Ib-cr and oqxAB were frequently detected in Salmonella spp. of food animal origin in China (Zhang et al., 2016; Wang et al., 2017; Kuang et al., 2018), and qnrS were also detected on the chromosome along with other PMQR genes (Lin et al., 2015). Co-existence of qnrS in the chromosome with other genes may be a strategy that confers a relatively high degree of tolerance to FQs without a prohibitive fitness cost which may explain the frequent observation of this combination.

The switch from plasmid to chromosomal location might represent a bacterial strategy for maintaining a resistance trait while adapting to antimicrobial stress (Machuca et al., 2014). Our growth competition assays confirmed that strains carrying PMQR genes on the chromosome had a growth advantage and could outcompete strains possessing PMQR genes on a plasmid. Interestingly, in the majority of our HR strains, only aac(6′)-Ib-cr was present. The latter gene was located on a plasmid and its presence coincided with multiple target gene mutations and overexpression of acrAB. aac(6′)-Ib-cr was associated with a fitness cost, in consistent with Machuca et al. (2014) observation in E. coli. Growth competition assays demonstrated that transfer of the aac(6′)-Ib-cr gene back to a plasmid location resulted in a fitness cost. Whole genome sequencing data identified a broad host range plasmid IncQ1 carrying no resistance genes in this HR strain. Thus, it is likely that the presence of this plasmid played a role in aac(6′)-Ib-cr gene movement from chromosome to plasmid, which has been previously suggested (Loftie-Eaton and Rawlings, 2012; Barry et al., 2019; Chen et al., 2019).

Increased drug efflux primarily occurs via overexpression of AcrAB-TolC, the primary efflux pump, in Enterobacteriaceae and drug efflux is a primary mechanism for quinolone resistance in Salmonella (Baucheron et al., 2004; Blair et al., 2009, 2014). In accordance with studies that analyzed experimentally derived ciprofloxacin mutants (Redgrave et al., 2014; Hooper and Jacoby, 2015), we observed elevated expression of AcrAB in the early stages of resistance development prior to the occurrence of gyrA mutations (Vidovic et al., 2019). Significant levels of acrAB-tolC were observed in the RS strains.

Multidrug efflux pumps are expressed under precise transcriptional control. In Salmonella, RamA is often considered the master regulator of acrAB (Ricci et al., 2014). We did not detect the overexpression of ramA in strains that expressed high levels of acrAB but had relatively low ciprofloxacin MICs (Groups S and RS). Instead, this overexpression was accompanied by increases in the expression of marA. A similar observation was made with laboratory selected ciprofloxacin-resistant mutants where mutation of ramR was mainly seen in mutants with very high ciprofloxacin MICs (Vidovic et al., 2019). Interestingly, efflux pump genes were not expressed more in R strains compared with S. Typhimurium SL1344, in agreement with earlier observations made with MDR Klebsiella and MDR E. coli that strains capable of evolving multiple energetically favorable QRDR mutations will use less efflux than isolates with fewer favorable QRDR alterations (Toth et al., 2014; Johnson et al., 2015). The overexpression of ramA was only detected in highly resistant strains. Our data suggests that whilst both MarA and RamA can up-regulate acrAB, there is a preference for utilizing MarA in less resistant strains and RamA in the most resistant strains. Both MarA and RamA are pleiotropic regulators and the overexpression of ramA may play additional roles as well as regulating acrAB in highly resistant strains more compatible with this phenotype than over-expression of marA. This is in line with the previous research in which a deletion in the RamR-binding sites in the ramA promoter correlated with highest expression levels of efflux pumps (Fabrega et al., 2016).

In conclusion, we observed a complex interplay between a pool of quinolone resistance mechanisms, their contribution to susceptibility and fitness costs (Figure 4). In general, we found no significant differences between different serovars of Salmonella suggesting common impacts from these genotypes across the genus although only S. Indiana isolates were found in the HR group suggesting they may be more able to develop high level resistance. Whilst natural variations in genes related to phylogeny may mask impacts on resistance, the topoisomerase genes and other known quinolone associated resistance pathways are highly conserved across the Salmonella genus and studying a large and diverse panel of real-world isolates can be informative for mutations circulating in practice which have been selected in real conditions.

We suggest an important role for substitutions within ParC and acrAB expression in the early stages of ciprofloxacin resistance development prior to the occurrence of gyrA mutations. Highly resistant isolates acquire multiple target site mutations and PMQR determinants although these are often integrated in the chromosome to minimize fitness costs and expression of acrAB-tolC is returned to wild-type levels, probably for the same reason. Whilst we are able here to identify genotypes which appear to be fit in conferring different levels of ciprofloxacin tolerance, it is still not possible to precisely assign a universal pathway in relation to the selection of resistance. Understanding the relative fitness of different genotypes in different selective conditions will help predict the evolutionary trajectory of AMR while also helping us to understand how resistance emerges in complex situations where multiple genes collaborate to confer resistance.