AcrB drug-binding pocket substitution confers clinically relevant resistance and altered substrate specificity

To identify mutations associated with the emergence of MDR, Salmonella enterica serovar Typhimurium L3 (pretherapy) and L18 (5 mo posttherapy) were sequenced by Illumina sequencing and compared with the published genome sequence of Salmonella enterica serovar Typhimurium LT2. Five SNPs relative to LT2 were identified in L18 that were not present in L3, suggesting that they arose during the time course of the infection (Table S1). Targeted PCR and Sanger sequencing of specific genes confirmed all mutations. One SNP occurred in an intergenic region, and the remaining four caused changes in the protein sequence of affected genes (nonsynonymous SNPs). These included a mutation at residue 464 of gyrB, which encodes a mutant B subunit of the topoisomerase II enzyme; the effect of this mutation has been previously studied in detail and confers quinolone resistance but not MDR (17). The genome sequencing also revealed a novel substitution from a glycine to an aspartic acid at residue 288 of AcrB.

Antibiotic resistance, particularly to ciprofloxacin [minimum inhibitory concentration (MIC) ≥0.5 µg/mL], occurred and increased in the isolates along the time course of infection (Table 1; susceptibility data for other antibiotics have been previously published and are shown in Table S2) (14, 15, 18). Previously, the MDR was attributed to increased expression of the acrB transcript (18). Indeed, measurement of acrB expression using an acrAB promoter–GFP fusion confirmed increased acrB expression in all strains isolated after the onset of therapy but the expression level did not correlate with the MDR phenotype. For example, acrB expression in L10 was not significantly different from that of L18, but L18 was more resistant. However, the level of MDR did correlate with which residue was present at position 288 of AcrB; the isolates with the highest level of MDR all carried the G288D substitution in AcrB. Isolates with an aspartic acid at residue 288 were two- to fourfold less susceptible to antibiotics compared with those with glycine at the same residue (Table 1). The changes in susceptibility were sufficient to confer clinically relevant resistance to the drugs used to treat the patient.

To confirm the importance of AcrB to the phenotype, acrB was inactivated in the MDR isolate, L18, to give strain L1299 and susceptibility to known substrates of AcrB was determined. Decreased MICs of ciprofloxacin, nalidixic acid, chloramphenicol, and tetracycline, to within one dilution of those for the pretherapy isolate L3, were obtained (Table 2).

To further investigate the effect of the G288D substitution, the wild-type and mutant acrB sequences were cloned into pWKS30 and transformed into the multidrug-susceptible L1299 (L18 with acrB deleted; Table S3). Similar levels of wild-type (L1351) and mutant (L1315) acrB/AcrB were expressed (as shown by RT-PCR and Western blotting; Table S4 and Fig. S2). Apart from ciprofloxacin, where only a modest decrease in susceptibility was seen (due to the mutation in gyrB) (19), L18 ΔacrB with wild-type AcrB (L1351) had susceptibility to antibiotics similar to the pretherapy isolate L3. However, L18 ΔacrB with mutant G288D AcrB (L1315) was MDR and the MIC values were within one dilution of those for L18 for ciprofloxacin, nalidixic acid, chloramphenicol, and tetracycline. Interestingly, L1315 was more susceptible to minocycline and doxorubicin than L1351, which had wild-type AcrB. This was not seen in L18. This may be due to the level of AcrB being greater in L18 or due to one of the SNPs in another gene confounding the effect of the G288D substitution.

Drug binding within AcrB is complex (20, 21) and is proposed to include several possible sites before extrusion including a vestibule and proximal and distal binding pocket. It has been suggested that low-molecular-mass compounds could bind directly to the phenylalanine-rich distal binding pocket (5, 20, 22), bypassing the proximal binding pocket and vestibule, which is (23) implicated in the recognition of larger compounds such as erythromycin (4, 5).

Structural alignment of the S. Typhimurium AcrB with the corresponding E. coli sequence shows high conservation of the distal pocket residues, suggesting a conserved mechanism of drug binding, and justifying the use of the E. coli structure as a guide for rationalization of the effect of the G288D substitution in L18 AcrB (Fig. S3). Residue 288 is also highly conserved in AcrB of other Enterobacteriaceae, suggesting an important structural role.

In the available crystal structures of AcrB in its apo- (PDB ID code 2J8S) (24) and in doxorubicin- (PDB ID code 4DX7) and minocycline-bound forms (PDB ID code 4DX5) (22), G288 is located in the middle of a beta strand, with the neighboring side chains pointing away from the pocket, suggesting that the substituted side chain will protrude into the pocket cavity. Hence, this bulky and charged aspartate residue will have a drastic impact on structure and dynamics of this primarily hydrophobic pocket, not only by changing the electrostatic profile and polar environment of the cavity but also by causing steric clashes with several key residues involved in the hydrophobic stacking interactions with the substrate (Fig. 1). This in turn could affect the affinity to various drugs such as doxorubicin and minocycline whose binding positions have been identified by cocrystallization experiments (22) (Fig. 1). Indeed, previous studies have shown that several drug recognition determinants are deeply perturbed by a single substitution in a key region of the transporter (25, 26). To analyze possible changes in the distal binding pocket as a result of the mutation we built homology models of wild-type and G288D variant of S. Typhimurium AcrB using the crystal structure of wild-type E. coli AcrB (PDB ID code 2J8S) as a template. Subsequently, MD simulations in explicit solvent were carried out to identify major differences in structure and dynamics of the distal binding pocket. A tested, reduced model of the S. Typhimurium AcrB containing only the periplasmic domains was used as in previous computational studies on E. coli AcrB (27–29).

Comparison of representative structures (i.e., the conformations with the lowest rmsd from the averages) of the wild-type and mutant proteins revealed that the introduction of the aspartate induces local structural changes in the distal binding pocket (rmsd 3.1 Å). In particular, residues F178 and Q176 assume alternative conformation to avoid steric clashes with the aspartate and F136 shows a large displacement (Fig. 2). As a result, the radius of gyration calculated for the residues lining the distal pocket increased by ∼1 Å in the G288D variant (from ∼10.5 Å in wild type to ∼11.5 Å, Fig. S4). The radius of gyration is indicative of the level of compaction in the pocket and the measured variation indicates an increased size. However, these changes have only limited effect on the secondary structure of the distal binding pocket (Fig. S5A).

In addition to these local structural changes, the introduction of the aspartate increases the polarity of the otherwise highly hydrophobic distal binding pocket (Fig. 1) and hence may affect its hydration. To test this, radial distribution function (RDF) profiles, which show the probability of finding water molecules at a certain distance from the G288/D288 residues, were calculated. As anticipated in the mutant protein the D288 residue was found to have a higher probability of being surrounded by water molecules during the course of the simulation (Fig. 2, solid lines). The sharp peak at around 2.5 Å from D288 is due to the presence of the first hydration shell. To the contrary, water molecules are on average absent in the close vicinity of G288, consistent with the more hydrophobic environment in the wild-type protein (Fig. 2, dashed lines). In addition, the mutant exhibits a notable change in residence time for water molecules within the distal pocket (Table S5).

We then analyzed the effect of the G288D substitution on doxorubicin, minocycline, and ciprofloxacin efflux. Doxorubicin and minocycline, although not clinically relevant, have been cocrystallized with E. coli AcrB, making phenotypic and structural data easier to correlate and interpret. We have compared these cocrystal structures with the results from the MD runs of the G288D variant. Consistent with MIC and drug accumulation data below, and based on the assumption that the binding of tetra-ring structures follows that in the cocrystal structures, the G288D substitution would result in pronounced steric clashes with both doxorubicin and minocycline, due also to displacement of amino acid F178 discussed above. Although an experimental structure of the ciprofloxacin–AcrB complex is lacking, a recent computational study (27) suggested that it binds to the bottom of the distal pocket, in a different position than doxorubicin (Fig. 3). In contrast to doxorubicin and minocycline, the G288D substitution should not impair the binding of ciprofloxacin; instead, the additional hydroxyl groups present in this compound may form hydrogen bonds with the introduced aspartate and resident waters, improving its recognition. In agreement with these hypotheses, G288D in L1315 (L18 ΔacrB pG288DacrB) not only led to increased susceptibility to doxorubicin and minocycline, likely due to decreased efflux, but also conferred increased efflux of, and decreased susceptibility to, ciprofloxacin and other fluoroquinolones (Table 2; MICs of more agents are shown in Table S6).

Although the differential effect of the mutation on the different classes of antibacterial drugs is not immediately obvious, it is notable that the efflux of more hydrophobic substrates such as doxorubicin and minocycline is correlated with the observed changes in the hydrophobicity of the distal binding pocket. Accordingly, minocycline and tetracycline seem differentially affected (Table 2), consistent with the former being 10-fold more lipophilic than the latter (Table S7).

Given the above considerations about structure, dynamics, and hydration of G288D AcrB, along with our experimental MIC data, we predicted that the susceptibility to ciprofloxacin would be decreased in AcrB G288D strains owing to an increase in its efflux efficiency.

Both L3 and L18 have a mutation in ramR (16) and L18 has a mutation in gyrB (17); both of these are known to influence susceptibility to fluoroquinolones. Because SL1344 ΔramR has MICs of doxorubicin and minocycline similar to those for L18 for these agents we hypothesize that the impact of the mutation in ramR in L18 masked the impact of the G288D substitution for these drugs. Therefore, to further investigate the effect of G288D, the mutant AcrB protein was expressed in two well-recognized standard antibiotic-susceptible reference strains, S. Typhimurium SL1344 and E. coli MG1655. In this way, unlike in L18, there were no other mutations that could affect antibacterial drug susceptibility and drug efflux. This allowed study of the G288D substitution in the absence of other mutations that could confound the phenotype. To explore whether efflux activity was altered due to the AcrB G288D substitution alone, the accumulation of ciprofloxacin, doxorubicin, and minocycline was measured. In both Salmonella and E. coli the strain expressing mutant AcrB (L1352 and I972, respectively) accumulated less ciprofloxacin than the strain expressing wild-type AcrB (L1365 and I971, respectively), suggesting more efficient efflux of ciprofloxacin by the mutated protein (Fig. 4). Conversely, strains of both species expressing mutant AcrB accumulated more doxorubicin and minocycline, indicating reduced efflux of these compounds (Fig. 4 A and C). In addition, Salmonella and E. coli with G288D AcrB extruded doxorubicin less than strains with wild-type AcrB (Fig. 4 B and D). These data support the hypothesis that the G288D mutation in AcrB improves efflux of ciprofloxacin and so confers resistance to ciprofloxacin (and other fluoroquinolones) while perturbing doxorubicin and minocycline binding and reducing efflux, so increasing susceptibility to these agents. This discrepancy in efflux of different drug classes is explained by our computational and structural analysis of drug binding into the distal pocket, which revealed distinct and nonoverlapping modes of binding for these drug classes (Fig. 3). Other substitutions at residue 288 were also constructed by site-directed mutagenesis. These too affected susceptibility to and accumulation of ciprofloxacin, further supporting the importance of this residue in the transport function of AcrB (Fig. S6).

Because we have previously shown that AcrB is important in the virulence of S. Typhimurium, we also examined whether the G288D substitution affected any other Salmonella attributes that could have contributed to the disseminated salmonellosis and the failure to effectively treat the patient. The G288D mutation did not alter fitness when grown in competition with a strain with wild-type AcrB (Fig. S7A) and did not affect invasion in the tissue culture model of infection (Fig. S7B).

The clinical isolates were obtained over 5 mo (Fig. S1 and Table S3) (15). S. enterica serovar Typhimurium SL1344 and E. coli MG1655 with corresponding isogenic mutants were used for comparison (39, 40). The wild-type and G288D mutant acrB genes were expressed by cloning into pWKS30 as described previously (41). Further site-directed mutations at residue 288 were constructed using the Stratagene QuikChange Site Directed Mutagenesis kit.

DNA was extracted using the Invisorb Spin Cell Mini Kit according to manufacturer’s instructions (Stratec). Library preparation and whole genome sequencing was carried out by Gene Pool, Edinburgh using the Illumina GAIIx platform to produce 50-bp paired-end reads. Reads were aligned against the published S. Typhimurium LT2 genome sequence (accession no. AE006468) using Bowtie2 (42). Read pileups were generated using Samtools mpileup, and SNPs were identified using bcftools (43). The sets of SNPs identified in L18 relative to LT2 and in L3 relative to LT2 were compared with each other using BEDtools (44) to find differences between L18 and L3. Putative mutations identified were filtered to remove those with a phred-scaled quality score of <100.

MIC values were determined using standard methods (45). Resistance to ciprofloxacin was determined with reference to the European Committee on Antimicrobial Susceptibility Testing and Clinical and Laboratory Standards Institute recommended breakpoint concentrations. Measurement of the accumulation of ciprofloxacin and other fluoroquinolones has been previously described and is based upon the original method for norfloxacin described by Chapman and Georgopapadakou (46, 47). Doxorubicin efflux was measured directly as described previously with the following modifications (48). Cells were incubated for 1 h with 10 µM doxorubicin and 40 µM carbonyl cyanide m-chlorophenylhydrazone. Cells were washed and resuspended in 20 mM potassium phosphate buffer (pH 7.5) with 5 mM MgSO₄ and loaded into a 96-well plate. Fluorescence was measured using a Fluostar OPTIMA with excitation at 450 nm and emission at 600 nm; 25 mM glucose was injected after the first reading to energize the cells.

A GFP transcriptional reporter fusion was created with the promoter region of acrAB fused to GFP in the plasmid pMW82 as described previously (49). Expression of acrAB was measured after growth of strains to mid-logarithmic phase (OD₆₀₀ 0.6) in MOPs minimal medium. Differences in expression were determined using t tests with a Bonferroni correction.

To build the wild-type and the G288D AcrB structure for S. enterica we used the program Modeler (50) and the crystal structures of AcrB for E. coli (24) as reference. For each system we generated 20 different models, which were then sorted by their discrete optimized protein energy score (51). The best structures according to this criterion were selected for further MD simulations. These latter were performed using the NAMD 2.9 simulation package (52) and the ff14SB force field (53). The reduced model of AcrB (validated in ref. 27) was used for MD runs, and the protein was embedded in a 0.1 M KCl water solution [∼41,000 water molecules were used, modeled with the modified TIP3P parameters (54)]. After structural optimization, the system was equilibrated by linearly increasing the temperature from 0 to 310 K in 1 ns of NPT MD simulation and equilibrated further for additional 5 ns. Then, multicopy simulations (three for each system: one of 400 ns and two of 200 ns for the wild type and three of 400 ns for the G288D system) in the NPT ensemble were performed both on the wild-type and on the G288D systems. Structural representations have been rendered using PyMol (PyMOL Molecular Graphics System; Schrödinger, LLC.) and VMD (55), and structural alignment was performed using ESPript 2.2 (56) based on a T-coffee input file (57).

Water residence times and average numbers of water molecules were calculated as previously described (58, 59) with the following alterations. First, we defined fast-, medium-, and long-time waters as those having survival probabilities τ of less than 100 ps, between 100 ps and 1 ns, and higher than 1 ns, respectively. Second, because we saved trajectory frames every 10 ps, there was no need to use a stretched exponential to fit survival probability profiles. Radii of gyration were calculated for residues lining the distal pocket (27) using the cpptraj module of Amber 14 (60).