Bacterial Resistance to Leucyl-tRNA Synthetase Inhibitor GSK2251052 Develops during Treatment of Complicated Urinary Tract Infections

Abstract

GSK2251052, a novel leucyl-tRNA synthetase (LeuRS) inhibitor, was in development for the treatment of infections caused by multidrug-resistant Gram-negative pathogens. In a phase II study (study LRS114688) evaluating the efficacy of GSK2251052 in complicated urinary tract infections, resistance developed very rapidly in 3 of 14 subjects enrolled, with ≥32-fold increases in the GSK2251052 MIC of the infecting pathogen being detected. A fourth subject did not exhibit the development of resistance in the baseline pathogen but posttherapy did present with a different pathogen resistant to GSK2251052. Whole-genome DNA sequencing of Escherichia coli isolates collected longitudinally from two study LRS114688 subjects confirmed that GSK2251052 resistance was due to specific mutations, selected on the first day of therapy, in the LeuRS editing domain. Phylogenetic analysis strongly suggested that resistant Escherichia coli isolates resulted from clonal expansion of baseline susceptible strains. This resistance development likely resulted from the confluence of multiple factors, of which only some can be assessed preclinically. Our study shows the challenges of developing antibiotics and the importance of clinical studies to evaluate their effect on disease pathogenesis. (These studies have been registered at ClinicalTrials.gov under registration no. NCT01381549 for the study of complicated urinary tract infections and registration no. NCT01381562 for the study of complicated intra-abdominal infections.)

INTRODUCTION

Approximately 5% of patients admitted to hospitals in the United States develop nosocomial infections that increase not only patient mortality (1) but also hospitalization time and cost of treatment (2, 3). In the United States, Gram-negative bacterial pathogens, such as Escherichia coli, Klebsiella pneumoniae/Klebsiella oxytoca, Pseudomonas aeruginosa, and Acinetobacter baumannii, are responsible for 35% of the most common hospital-acquired infections (HAIs) or conditions, including urinary tract infections (UTIs), pneumonia, and surgical site and bloodstream infections. Furthermore, >70% of the bacteria causing HAIs are resistant to at least one of the most commonly used antibiotics (4–6). The continuing emergence of resistance has compromised treatment options for Gram-negative bacterial pathogens and forced the use of polymyxins (7), an old antibiotic class with nephrotoxicity issues. Despite the clear need for alternatives to treat these multidrug-resistant life-threatening pathogens, a review of the antibacterial pipeline reveals a scarcity of new antibiotic candidates (8).

Aminoacyl-tRNA synthetases (AaRSs) play an essential role in protein synthesis (9) and have been clinically validated to be antibacterial targets of mupirocin (10), an inhibitor of isoleucyl-tRNA synthetase that has been successfully used since 1985 in the topical treatment of Gram-positive bacterial skin infections (11). Their ubiquitous nature, high degree of conservation within a broad spectrum of bacterial species, and considerable divergence between prokaryotic and eukaryotic enzymes have made AaRSs the subject of multiple antibacterial programs (12, 13), although no other AaRS inhibitor has reached the market to date.

The demonstration that the antifungal tavaborole could specifically inhibit yeast cytoplasmic leucyl-tRNA synthetase (LeuRS) (14) triggered the search for additional benzoxaboroles capable of inhibiting the editing domain of the bacterial counterpart. A structure-based design effort focused on the production of potent Gram-negative bacterial LeuRS inhibitors led to the discovery of GSK2251052 (AN3365) (15). This bacteriostatic compound demonstrates in vitro activity against a wide range of anaerobic (16) and Gram-negative bacteria, including multidrug-resistant pathogens, such as extended-spectrum-β-lactamase-, metallo-β-lactamase-, and carbapenemase-producing organisms (17), and has a low propensity to be effluxed (15). A phase I study in healthy volunteers showed that GSK2251052 had linear pharmacokinetics (18) and caused no serious or dose-limiting adverse events (15, 19). These data from the phase I study, combined with the impressive in vitro microbiology profile of GSK2251052 against multidrug-resistant Gram-negative organisms, led to the initiation of two prospective, multinational, randomized, double-blind phase IIa clinical trials. These studies were conducted to evaluate the safety, tolerability, and efficacy of two twice-daily (BID) doses (750 and 1,500 mg) of intravenous (i.v.) GSK2251052 in adult subjects with complicated urinary tract infections (cUTIs; study LRS114688, ClinicalTrials.gov registration no. NCT01381549) or complicated intra-abdominal infections (cIAIs; study LRS114689, ClinicalTrials.gov registration no. NCT01381562).

(Part of this work was presented at the 53rd Interscience Conference on Antimicrobial Agents and Chemotherapy, Denver, CO, September 2013 [20].)

MATERIALS AND METHODS

Study design.

Study LRS114688 involved eight centers in three countries (the United States, Spain, and Greece) and was conducted between 28 June 2011 and 6 March 2012. Study LRS114689 involved five centers in four countries (the United States, Spain, France, and Russia) and was conducted between 3 October 2011 and 5 March 2012.

In both studies, 210 subjects at least 18 years old were to be treated for ≥5 to ≤14 days. Subjects in cUTI study LRS114688 were randomized (1:1:1) to receive GSK2251052 at 750 mg or 1,500 mg BID or imipenem-cilastatin at 500 mg every 6 h. Twenty subjects had been enrolled when this study was terminated, and 14 of them had received GSK2251052 at doses of 750 mg (n = 6) or 1,500 mg (n = 8) BID. Nineteen of the 20 subjects had a qualifying Gram-negative bacterial uropathogen at the baseline, with E. coli being the most frequently isolated pathogen (18 of 26 isolates). Subjects in cIAI study LRS114689 were randomized (1:1:1) to receive GSK2251052 at 750 mg or 1,500 mg BID or meropenem at 1 g three times a day. Fifteen subjects had been enrolled when this study was terminated, with nine of them having received GSK2251052 at doses of 750 mg (n = 5) or 1,500 mg (n = 4) BID. Ten of the 15 subjects had a confirmed Gram-negative bacterial pathogen at the baseline, with E. coli being the most frequently isolated pathogen (9 of 28 isolates).

The institutional review boards of the institutions involved in the clinical trials reviewed and approved the studies prior to subject enrollment. Written informed consent was obtained from each subject prior to the performance of any study-specific procedures.

Susceptibility testing.

All bacterial strains were cultured at 35°C on Trypticase soy agar (TSA) with 5% sheep blood or in cation-adjusted Mueller-Hinton broth (MHB). E. coli ATCC 25922, 1162222, and 1161435 were obtained from the GlaxoSmithKline Microbiology Culture Collection. E. coli, Proteus mirabilis, Enterococcus faecalis, Pseudomonas aeruginosa, Klebsiella pneumoniae, and Staphylococcus epidermidis isolates were obtained from four subjects participating in the LRS114688 cUTI clinical trial.

MICs of antibacterial agents were determined using the broth microdilution methodology (21). Validated dry-form custom-made broth microdilution panels were used in cross-resistance studies with GSK2251052, amikacin, ampicillin, aztreonam, cefepime, ceftazidime, ceftriaxone, ciprofloxacin, doripenem, ertapenem, imipenem, levofloxacin, meropenem, nitrofurantoin, piperacillin-tazobactam, tigecycline, and trimethoprim-sulfamethoxazole.

The frequency of spontaneous resistance to GSK2251052 was determined by spreading bacterial suspensions of the different E. coli strains in MHB (∼2.5 × 10⁸ CFU) onto TSA plates containing 4× or 10× MIC of GSK2251052 and a control plate with no compound. Colonies were counted after 48 h incubation at 35°C. To confirm the resistance phenotypes, the susceptibility to GSK2251052 of a representative number of colonies from each antibiotic concentration-strain combination was tested. Colonies were defined as resistant if their MIC was ≥4-fold that of the parent strain. The total number of resistant colonies was extrapolated from this representative set. The frequency of resistance was calculated by dividing the total number of resistant colonies by the total number of CFU plated.

Genetic characterization of isolates with reduced susceptibility to GSK2251052.

The PCR primers used for DNA amplification were designed from the appropriate regions of publically available genomic sequences. To prepare the PCR templates, cells from overnight culture plates were resuspended in 50 μl Tris-EDTA (TE) buffer, boiled for 10 min, and placed on ice for 5 min before a 2-min high-speed centrifugation. Supernatants (2 μl) were used as the templates in the PCRs, carried out on an Applied Biosystems GeneAmp PCR system 9700. The PCR products were sequenced using a BigDye Terminator cycle sequencing kit (v3.1; Applied Biosystems) and analyzed in a genetic analyzer (3730XL; Applied Biosystems).

Fragments of the leuS gene were amplified by PCR, sequenced, and aligned with the sequences from their corresponding parent organisms. The K. pneumoniae DNA sequence from subject 751 was aligned with sequences from the published genomes of strains MGH78578 and JH1 and the proprietary genome of strain 1161486. A well-validated PCR-based allelic replacement method utilizing lambda Red recombinase (22) was used to introduce a LeuRS T247I mutation in E. coli 1162222 and E. coli 1161435.

Genomic DNA from both strains was used as the template for the amplification of their leuS gene by a 3-step PCR protocol, which allowed the introduction of a nucleotide change (ACT to ATT) in the codon corresponding to LeuRS residue 247. Subsequently, these PCR fragments were introduced by electroporation into the corresponding strains expressing the lambda Red recombination system in plasmid pKD46 (22). Isolates carrying the LeuRS T247I mutation were selected on plates containing GSK2251052. Mutations were confirmed by PCR amplification and sequencing of the leuS gene.

Illumina library preparation and whole-genome sequencing.

DNA samples were prepared from monocultures of single bacterial colonies. Fragment libraries (mode length, 350 bp) were constructed using an Illumina paired-end DNA sample preparation kit (v2) according to the instructions of the manufacturer, with some modifications. Genomic DNA (1 μg in 50 μl TE buffer) was sheared for 180 s on a Covaris S220 sonicator (duty factor, 10; peak power, 175 W; 200 cycles per burst). For accurate quantification, quantitative PCR was performed in triplicate using a KAPA library quantification kit on a 7900HT real-time PCR instrument (Life Technologies). Samples (2 nmol) were normalized and pooled into 7 batches containing 2 samples each. A unique adapter index sequence was used for each sample to allow independent samples to be pooled for sequencing and subsequent bioinformatic segregation of the data output. Fragment libraries were separated and size selected (300- to 400-bp insert range), and primer dimers were removed using an Agilent high-sensitivity DNA kit on an Agilent 2100 bioanalyzer. The Illumina GAII sequencer instrument, reagents, and flow cell were prepared according to Illumina TruSeq SBS kit (v5) protocols. Libraries were diluted to a final 12-pmol dilution. The instructions of the TruSeq PE Cluster kit (v2) for clustering using the Illumina cBot system were followed.

Genomic analysis.

Sequencing was performed in paired-end multiplex 2× 72-bp mode. Base calling and demultiplexing were performed using the CASAVA program (v1.7) on default parameters. Sequence analyses were performed with the CLC Genomics Workbench (v5.1.5) sequencing module using default parameters. Raw reads were processed to remove adapter sequences and low-sequence-quality reads or ambiguous base calls.

The uropathogenic E. coli strain CFT073 (23) genome was selected as the assembly scaffold for the genomes from subjects 202 (baseline and blood samples) and subject 252 (baseline and urine samples) on the basis of the overall highest BLASTN (24) homology to selected conserved housekeeping genes. Whole-genome sequences of E. coli isolates recovered on days 2, 4, 8, and 13 and at follow-up (subject 252) or recovered from blood on days 2, 3, and 4 and urine on day 1 (subject 202) were mapped to the reference sequences of isolates recovered on day 1 to identify mutations relative to the sequences of those genomes. On average, all samples were sequenced to approximately 250-fold coverage.

Day 1 reference sequence assembly required at least a 20-fold coverage of the single nucleotide variants (SNVs) identified in the resistant isolates. Only those SNVs that occurred in protein-coding genes in 100% of the reads are reported.

Evolutionary relationships were determined from concatenated nucleotide alignments on the basis of (i) E. coli conserved multilocus sequence typing (MLST) genes (9,018 nucleotides [nt]) or (ii) complete genes with one or more SNVs (8,637 nt). Phylogenetic trees were reconstructed by neighbor-joining and Bayesian methods. A neighbor-joining tree with 1,000 bootstrap replicates was reconstructed using the programs DNADIST (default F84 matrix) and NEIGHBOR from the PHYLIP (v3.6) package (25). A Bayesian posterior probabilities phylogenetic tree method, as implemented by the software MrBayes (v3.0B4) (26, 27), was also used. A gamma-distributed rate (GTR) model with 6 discrete rate categories was chosen. Markov chains were run for 1 × 10⁶ generations, burn-in values were set for 1 × 10⁴ generations, and trees were sampled every 100 generations. All trees were visualized with the program TREEVIEW (28). Venn diagrams were generated using the VennDiagram R package (29).

Nucleotide sequence accession numbers.

This whole-genome shotgun project has been deposited at DDBJ/EMBL/GenBank under accession numbers AXOA00000000, AXOB00000000, AXOC00000000, AXOD00000000, AXOE00000000, JTDS00000000, JTDT00000000, JTDU00000000, JTDX00000000, JSYR00000000, and JSYS00000000.

RESULTS

Phase II clinical trials with GSK2251052.

Twenty subjects, 14 of whom received i.v. BID doses of GSK2251052, were enrolled in study LRS114688 when Gram-negative bacterial isolates with reduced susceptibility to GSK2251052 were recovered from four subjects (subjects 202, 203, 252, and 751) (Table 1), and, as a consequence, the study was terminated. The isolates from three of these four subjects demonstrated a significant increase (≥32-fold) in GSK2251052 MIC between the baseline and day 2. One subject (subject 202) received doses of 1,500 mg for 4 days, and two subjects (subjects 203 and 252) received doses of 750 mg for 6 and 14 days, respectively. The fourth subject (subject 751) had different pathogens isolated at the baseline and postbaseline and had received 1,500-mg doses for 6 days.

TABLE 1.

Characterization of isolates from clinical trial LRS114688, their susceptibility to GSK2251052, and mutations in LeuRS identified by Sanger sequencing^a

Subject no. (BID treatment), source of pathogen	Day 1 (baseline)			Day 2 (on therapy)			Day 3 (on therapy)			Day 4 (on therapy)			Day 5 (on therapy)			Day 8 (on therapy)			Day 13 (on therapy)			End of therapy			Test of curec,d			Late follow-up			Microbiological responsed
	Isolate	MIC (μg/ml)	LeuRSb	Isolate	MIC (μg/ml)	LeuRS	Isolate	MIC (μg/ml)	LeuRS	Isolate	MIC (μg/ml)	LeuRS	Isolate	MIC (μg/ml)	LeuRS	Isolate	MIC (μg/ml)	LeuRS	Isolate	MIC (μg/ml)	LeuRS	Isolate	MIC (μg/ml)	LeuRS	Isolate	MIC (μg/ml)	LeuRS	Isolate	MIC (μg/ml)	LeuRS
202 (1,500 mg)
Urine	E. coli	0.25	None	No growth	No growth	No growth	No growth	No growth	No growth	ND	ND	ND	NAe	NAe	NAe	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	Failure
Blood	E. coli	0.5	None	E. coli	16	P339L	E. coli	16	P339L	E. coli	16	P339L	NAe	NAe	NAe	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	NA	Failure
252 (750 mg), urine	E. coli	0.25	None	E. coli	512	T247I	E. coli	512	T247I	E. coli	512	T247I	E. coli	512	T247I	E. coli	1,024	T247I + A323V	E. coli	1,024	T247I + D251N	E. coli	1,024	T247I + A323V	NA	NA	NA	E. coli	1,024	R344H + D345A	Failure
203 (750 mg), urine	P. mirabilis	0.5	None	P. mirabilis	1,024	T247I	ND	ND	ND	ND	ND	ND	ND	ND	ND	NAe	NAe	NAe	NA	NA	NA	P. mirabilis	1,024	T247I	NA	NA	NA	NA	NA	NA	Failure
751 (1,500 mg)
Urine	P. aeruginosa	2	NT	No growth	No growth	No growth	No growth	No growth	No growth	No growth	No growth	No growth	ND	ND	ND	ND	ND	ND	ND	ND	ND	No growth	No growth	No growth	No growth	No growth	No growth	K. pneumoniae	1,024	R344S	Success
Blood	E. faecalis	32	NT	E. faecalis	32	NT	E. faecalis	8, 32	NT	E. faecalis	32	NT	E. faecalis	16	NT	ND	ND	ND	ND	ND	ND	E. faecalis	16	NT	E. faecalis	32	NT	S. epidermidis	0.5	NT	Success

^aNT, not tested; ND, not done, sample not taken; NA, not applicable due to a change in antibacterial therapy. Isolates selected for whole-genome sequencing are in bold font.

^bleuS and promoter regions sequenced.

^cThe primary efficacy objective was microbiological, and the clinical response was evaluated at the test-of-cure visit (5 to 9 days posttherapy).

^dThe responses at the test of cure, end of therapy, and follow-up were the same.

^eDay that therapy was changed.

E. coli isolates with GSK2251052 MICs of 0.25 and 0.5 μg/ml were isolated, respectively, from baseline cultures of urine and blood samples from subject 202. While there was no growth from subsequent urine cultures on days 2 and 3, blood cultures continued to grow E. coli isolates with a GSK2251052 MIC of 16 μg/ml. Proteus mirabilis with a GSK2251052 MIC of 0.5 μg/ml was isolated from the baseline culture of the urine sample from subject 203, whereas there was no growth from cultures of blood samples. On day 2, the culture of the urine sample from subject 203 grew a P. mirabilis isolate with a GSK2251052 MIC of 1,024 μg/ml. An E. coli isolate with a GSK2251052 MIC of 0.25 μg/ml was isolated from the baseline culture of the urine sample from subject 252, whereas there was no growth from cultures of blood samples. All subsequent urine samples recovered up to the late follow-up visit from subject 252 grew E. coli isolates with GSK2251052 MICs of ≥512 μg/ml. The baseline culture of the urine sample from subject 751 grew a P. aeruginosa isolate with a GSK2251052 MIC of 2 μg/ml, but subsequent cultures of urine samples collected on days 2 to 4 showed no growth; however, the culture of the urine sample collected at the late follow-up visit grew a K. pneumoniae isolate with a GSK2251052 MIC of 1,024 μg/ml.

Due to the emergence of resistance in study LRS114688, study LRS114689, which had 15 subjects enrolled, 9 of whom had received GSK2251052, was terminated. In this cIAI study, for isolates from three subjects for whom postbaseline samples were available, no significant increase (>4-fold) in the GSK2251052 MIC between pathogens recovered at baseline and postbaseline visits was demonstrated.

Definitive conclusions regarding the safety and/or efficacy of GSK2251052 could not be made in either study due to the limited number of subjects enrolled at the time of early termination.

Characterization of clinical isolates with reduced GSK2251052 susceptibility.

Sequencing of the promoter and coding regions of the leuS gene, encoding LeuRS, from the E. coli, P. mirabilis, and K. pneumoniae isolates recovered from study LRS114688 subjects showed that while the sequences of the baseline isolates were identical to those of the published genomes, all organisms recovered postbaseline contained either single or double mutations in the editing domain of LeuRS (Table 1) that were presumably responsible for their reduced susceptibility to GSK2251052. Amino acid substitution T247I, the most commonly encountered mutation, conferred high-level resistance, with the MICs for the isolates being 512 or 1,024 μg/ml. Equivalent resistance levels were associated with changes in residue R344 (GSK2251052 MIC, 1,024 μg/ml), whereas mutation P339L increased GSK2251052 MICs to 16 μg/ml (Table 1).

In cross-resistance studies, no differences in the MICs of antibacterials typically used to treat Gram-negative bacterial infections were observed between baseline and postbaseline isolates. All E. coli isolates recovered from subjects 202 and 252 were susceptible to the comparator agents with activity against Gram-negative bacteria tested. The P. mirabilis isolates from subject 203 were susceptible to all comparators except imipenem (intermediate), ampicillin, nitrofurantoin, and trimethoprim-sulfamethoxazole. The K. pneumoniae isolate recovered from subject 751 was ampicillin resistant.

Susceptibility of E. coli LeuRS T247I isogenic mutants to GSK2251052.

In order to confirm that specific changes in the editing domain of LeuRS were solely responsible for the resistance phenotype, a mutation was introduced in the leuS gene of two E. coli strains, 1162222 and 1161435, to render LeuRS T247I. Susceptibility testing of the isogenic pairs demonstrated that this mutation causes high-level resistance to GSK2251052. In E. coli 1161435, MICs increased from 1 μg/ml (parent) to 512 μg/ml (mutant), and similarly, in strain 1162222, they increased from 0.5 μg/ml to 256 μg/ml.

Whole-genome DNA sequencing of E. coli isolates from subjects 202 and 252.

E. coli isolates from subjects 202 and 252 (Table 1) were selected for whole-genome DNA sequencing on the basis of their broad time course of sampling and the availability of day 1 baseline isolates to be used as reference genomes to identify SNVs.

All mutations previously identified in the LeuRS editing domain of these E. coli isolates by Sanger DNA sequencing were confirmed by whole-genome sequencing. The numbering of specific LeuRS amino acid residues differs slightly between methods, because a different reference E. coli strain was used in each analysis: E. coli K-12 (Sanger DNA sequencing) and CFT073, a known uropathogenic E. coli (UPEC) strain (Illumina DNA sequencing) (Tables 2 and 3).

TABLE 2.

Amino acid residue variations between GSK2251052-resistant E. coli urine isolates and the baseline urine isolate from subject 252 determined by whole-bacterial-genome DNA sequencing

Time(s) of sample collection	Gene	Description	Mutation(s)a
Day 13	leuS	Leucyl-tRNA synthetase	D298N (D251N)
Days 2, 4, 8, and 13	leuS	Leucyl-tRNA synthetase	T294I (T247I)
Day 8	leuS	Leucyl-tRNA synthetase	A370V (A323V)
Follow-up	leuS	Leucyl-tRNA synthetase	D392A, R391H (D345A, R344H)
Days 2, 4, 8, and 13 and follow-up	acnB	Bifunctional aconitate hydratase	V705I
Days 2, 4, and 8 and follow-up	c0279	Hypothetical protein	T21A
Day 4	c1464	Tail fiber component K of prophage	G41b
Follow-up	c1472	Hypothetical protein	T15A
Follow-up	c1473	Hypothetical protein	R8G, 42Eb
Follow-up	c1474	Hypothetical protein	K209Q
Day 2	c3157	Hypothetical protein	T152A
Days 2 and 4 and follow-up	c3157	Hypothetical protein	A158V, W159S, P218L
Day 2 and follow-up	c3157	Hypothetical protein	R136L, D142G
Day 8	c3157	Hypothetical protein	W159S
Days 8 and 13	c3157	Hypothetical protein	T152A, A158V, P218L
Days 2, 8, and 13	c3158	Tail component of prophage	Y85H
Follow-up	c3670	Hypothetical protein	E148G
Follow-up	c3671	RadC-like protein YeeS	L13P
Days 2, 4, 8, and 13	c3674	Hypothetical protein	A160P
Days 2 and 4 and follow-up	c3679	Hypothetical protein	C13Y, L17S
Day 4	c4492	ShiA-like protein	K302Q, E305K
Day 2 and 8	c4572	Hypothetical protein	P98H
Days 2, 4, 8, and 13 and follow-up	c5158	Hypothetical protein	S30,b I22V
Day 8	c5213	Transposase IS629	207K,b H203Y
Days 4, 8, and 13 and follow-up	ompR	Osmolarity response regulator	R207H
Days 2, 4, 8, and 13 follow-up	ycbS	Outer membrane usher protein YcbS	M315I
Follow-up	ydfR	Hypothetical protein	M59R

^aAmino acid residue according to E. coli CFT073 numbering system, with E. coli K12 equivalents in parentheses.

^bNonsense mutation.

TABLE 3.

Amino acid residue variations between GSK2251052-resistant E. coli blood and urine isolates from baseline blood isolate from subject 202 determined by whole-bacterial-genome DNA sequencing

Time(s) of sample collection or sample	Gene	Description	Mutation(s)a
Days 2, 3, and 4 blood	leuS	Leucyl-tRNA synthetase	P386L (P339L)
Baseline urine	atpF	ATP synthase FoF1 subunit B	A5E
Days 2, 3, and 4 blood and baseline urine	c1471	Hypothetical protein	D59V
Days 2, 3, and 4 blood and baseline urine	c1472	Hypothetical protein	I75N
Baseline urine	c1527	Hypothetical protein	T45S
Days 2, 3, and 4 blood and baseline urine	c1590	Tail component of prophage	A803S
Baseline urine	c2021	Electron transport complex protein RnfC	E624A, E624D
Days 2, 3, and 4 blood and baseline urine	c2295	Hypothetical protein	38Sb
Days 2, 3, and 4 blood and baseline urine	c2473	Transposase	I113V
Day 3 blood	c4898	Hypothetical protein	I45S

^aAmino acid residue according to E. coli CFT073 numbering system. E. coli K12 equivalents in parentheses.

^bNonsense mutation.

In both subjects, nonsynonymous (NS) codon changes mostly occurred by day 2 and were retained over subsequent time points (Fig. 1A and B), with the exception of those in E. coli LeuRS from subject 252, where mutations T247I (days 2, 4, 8, and 13), A323V (day 8), and D251N (day 13) were absent in the posttherapy follow-up isolate (Tables 2 and 3), which had, instead, two different mutations, D345A and R344H.

FIG 1.

Venn diagram showing the variation in nonsynonymous SNVs in protein-coding genes relative to the sequence of the genome at baseline day 1 for subject 252 (A) and subject 202 (B). Variants in prophage, transposase, and hypothetical genes were excluded from the figure to focus on SNVs in genes with a functional annotation.

Isolates from subjects 202 and 252 did not share any NS mutations, and only a hypothetical protein-encoding gene (c1472) was mutated, at different positions, in some E. coli isolates from both subjects. Among isolates from subject 252, amino acid changes occurred in various hypothetical proteins and prophage elements, as well as in several known proteins, such as OmpR, a DNA-binding response regulator; YcbS, an outer membrane usher protein; and AcnB, a bifunctional aconitate hydratase 2/2-methylisocitrate dehydratase (Fig. 1A; Table 2).

Fewer amino acid changes in the sequences of isolates from subject 202 relative to the sequence of the baseline organism from blood were observed (Fig. 1B; Table 3). In addition to the NS changes in LeuRS, mutations were seen in several hypothetical genes, a transposase, and a prophage gene. Some changes, such as those in AtpF (ATP synthase F_oF₁ subunit B) and RnfC (an electron transport protein), were specific to the baseline urine isolate.

Phylogenetic trees based on the nucleotide sequences of either multilocus sequencing typing (MLST) or SNV genes showed that E. coli isolates were more similar within subjects than between subjects (Fig. 2A and B), suggesting that they were highly clonal and no new reinfections had occurred, at least over the course of the study period and for the bacterial genomes sequenced. Furthermore, the sequences of these genomes were similar but not identical to those of any published UPEC genomes.

FIG 2.

Phylogenetic trees of E. coli strains based on published concatenated nucleotide sequences of representative E. coli isolates causing urinary tract infection. (A) Phylogeny of MLST genes based on the alignment of 9,018 nt. (B) Phylogeny of loci with SNVs based on the alignment of 8,637 nt derived from whole-genome comparisons of day 1 isolates from subjects 202 and 252 to those genomes from the respective subsequent days. Trees were reconstructed using both bootstrap neighbor-joining (shown) and Bayesian methods. Branch points marked with an asterisk were supported in more than 80% of 1,000 bootstrap replications and by a Bayesian posterior probability of ≤0.01. The location of E. coli CFT073, used for gene annotation mapping, is also indicated.

Characterization of laboratory-isolated mutants resistant to GSK2251052.

To ascertain if these clinical isolates developed resistance at a higher rate than other E. coli strains, the in vitro frequency of spontaneous resistance to GSK2251052 was determined in two standard strains, E. coli ATCC 25922 and E. coli 1161435, and in the two baseline isolates from subjects 202 and 252, E. coli 998584 and 671925, respectively.

GSK2251052-resistant mutants appeared at similarly high rates, 1 × 10⁻⁷ to 4 × 10⁻⁷, in both sets of strains, E. coli laboratory standards and clinical isolates (see Table S1 in the supplemental material), and had no obvious in vitro growth defect in standard media. Equally similar was the percentage of mutants that showed high-level resistance to GSK2251052, as the majority of isolates of E. coli ATCC 25922 (83%), 1161435 (79%), 998584 (88%), and 671925 (78%) had GSK2251052 MIC increases of ≥32-fold with respect to those for their parent strains.

To identify the mutations responsible for this resistance phenotype, a fragment of leuS encoding the editing domain of LeuRS was sequenced from a set of ∼40 representative colonies of each of the different E. coli strains. In 119 mutants, 43 single base changes were identified in 20 different amino acids of the LeuRS editing domain (Table 4). One mutant with a 27-bp insertion in this region, one mutant with a 12-bp deletion in this region, and one mutant with double mutations in this region were also identified. The additional 36 isolates did not have any sequence change in the editing domain of LeuRS. Nearly 50% of the mutants carried mutations in one of three residues: G229 (n = 20), D345 (n = 20), and D251 (n = 18). Comparable numbers of isolates with mutations in a specific residue were obtained from the standard laboratory strains and from the two clinical isolates, except in one case. Mutations in D345 were more commonly encountered in the clinical strains, specifically, in E. coli 671925, where 14 of the 39 mutants analyzed carried substitutions in that residue (see Table S2 in the supplemental material).

TABLE 4.

Summary of characterization of laboratory-isolated mutants with reduced susceptibility to GSK2251052

LeuRS amino acid residuea	Amino acid change(s)	No. of mutants		Fold increase in MIC
		Standard strains	Clinical strains
S227	F/Y	1	1	>256/16
G229	D/S/V	12	8	>256
Y246	C/H/N	4	4	64–>256/>256/64–>256
T247	N	0	1	>256
T248	P	4	0	>256
R249	C/H/L/P	1	3	16/>256/64/>256
D251	G/H/N/V/Y	11	7	8–64
C256	R	0	1	32
A260	P	1	0	16
W321	WAANFVLME ×2	1	0	>256
V326	G	0	1	256
Y330	C/D/L/N	0	4	8/>256/32/>256
G331	D	2	3	64–128
G333	C/D/V	5	2	8/128–256/256
A334	P/T/V	3	3	>256/4–16/8
A337	P	1	0	>256
V338	F/ΔVPGH	2	0	>256
P339	L/Q	3	2	64/16–64
D342	G/Y	4	1	8/64
D342+R344	G	0	1	>256
R344	H	2	3	>256
D345	G/N/Y	3	17	>256/16–32/>256
No change in editing domain		19	17	4–>128

^aAmino acid numbers based on E. coli strain K-12 LeuRS.

Mutations in the single residues associated with clinical resistance, P339, T247, and R344, were also found among the laboratory-isolated mutants. Although some carried identical substitutions (P339L, R344H), mutation T247I, found in isolates from two of the four subjects in study LRS114688, was not detected. Substitutions in amino acids D251 and D345, which appeared as secondary mutations in isolates from subject 252, were also represented amid the in vitro-obtained mutants, but isolates with changes in A323 were not found.

DISCUSSION

Prior to initiating clinical trials, in vitro studies showed that resistance to the novel LeuRS inhibitor GSK2251052 developed in strains of E. coli, P. aeruginosa, and K. pneumoniae at a rate of 1 × 10⁻⁷ to 8 × 10⁻⁷, similar to that observed with other antibacterials, such as ceftazidime or ciprofloxacin (15). The mutants recovered from these studies were susceptible to norvaline and, therefore, likely to be editing deficient (15). Although we recognize the limitations of animal models when investigating resistance, given that total bacterial loads can be significantly below those in the clinic, multiple attempts were made in our laboratories to select resistant organisms in rodent models of infection with only a single mutant organism isolated. Despite this, resistance to GSK2251052 developed very rapidly during therapy in 3 of 14 subjects enrolled in the phase II cUTI study LRS114688. In fact, isolates with GSK2251052 MIC increases of ≥32-fold with respect to the MIC for the baseline pathogen were recovered from the infections on day 2, indicating that resistance either was present in the bacterial cell population at the start of therapy or emerged within the first 24 h of treatment. The rapid isolation of these mutants also suggests that there was no in vivo fitness cost associated with these mutations. Cultures of blood from subject 202, treated with 1,500 mg BID, and cultures of urine from subjects 203 and 252, each receiving 750 mg BID, were positive for GSK2251052-resistant Gram-negative bacterial pathogens. All cases were microbiological failures; however, subject 252 showed clinical resolution of signs and symptoms. Neither of the two dosing regimens used could yield the GSK2251052 blood levels necessary to prevent the growth of these resistant isolates, but in this very limited data set, the rapid emergence of resistance seemed to occur more frequently in subjects exposed to lower drug concentrations (2/6 in the 750-mg group versus 1/8 in the 1,500-mg group). Treatment of subject 751 with 1,500 mg GSK2251052 BID did not result in the development of resistance in the baseline isolate, but a different pathogen resistant to GSK2251052 was recovered from a urine sample taken at the late follow-up visit. A possible hypothesis is that for this subject the emergence of resistance occurred at an alternate infection site, as the MIC for this K. pneumoniae isolate was ≥512-fold higher than that for the expected wild-type distribution (17). Without a corresponding pretreatment isolate, the exact origin of resistance cannot be fully elucidated. Resistance was not observed in the additional subjects treated with GSK2251052, and at the end of therapy, most (3 of 4 receiving 750 mg and 5 of 6 receiving 1,500 mg) had both clinical and microbiological successes.

Development of resistance during antibacterial therapy has previously been documented for other agents (30). A review of 173 clinical trials involving ∼14,000 subjects and eight antibacterial classes found that resistance emerged in 4% of the organisms and was more frequent with penicillin (6.4%) and aminoglycoside (5.5%) monotherapies (30). The type of infection and causative agent and underlying diseases also seem to play a critical role in the development of resistance during therapy.

Analysis of leuS sequences from the study LRS114688 clinical isolates shows that GSK2251052 resistance is due to specific target-based mutations and not to the existence of strains with an additional refractory LeuRS, acquired by gene duplication or horizontal gene transfer, as has been previously described with resistance to other AaRS inhibitors (31–33). Cocrystallization studies have shown that benzoxaboroles make key interactions with LeuRS editing domain residues, including T247 (15), and, therefore, the mutations identified in these isolates, such as P339L, T247I, and R344S, might be expected to affect the binding of GSK2251052 and induce the resistance observed. In fact, E. coli isogenic mutants carrying LeuRS substitution T247I showed a 512-fold increase in the GSK2251052 MIC with respect to that of their parent strains.

In vitro frequency-of-resistance studies showed that baseline isolates from subjects 202 and 252 were not more prone to the development of resistance than standard laboratory strains. Rates of resistance were high in all cases, and 82% of the resistant isolates had GSK2251052 MIC increases of ≥32-fold. Sequence analysis of the LeuRS editing domains of 158 lab-generated resistant strains revealed 45 individual mutations at 22 different amino acid positions, which were very likely responsible for the resistance phenotype observed. The large variety of mutations associated with reduced susceptibility to GSK2251052 and the lack of any obvious fitness cost clearly contribute to the relatively high in vitro frequency of resistance observed with this compound.

To reveal new information about the evolution of pathogens at the level of individual patients and to confirm the resistance mechanisms, several E. coli isolates from subjects 202 and 252 were selected for next-generation DNA sequencing (34), a technology recently used to characterize whole genomes of bacterial pathogens isolated from clinical outbreaks (35, 36). Phylogenetic analysis showed that E. coli isolates from subjects 202 and 252 were very similar within subjects, suggesting that they resulted from clonal expansion of their corresponding baseline pathogen and were not the result of new infections. Mutations conferring resistance to GSK2251052 occurred by day 2 in both cases and were retained in subsequent days. Few amino acid substitutions were observed in isolates from subject 202, but changes in some interesting proteins were identified in isolates from subject 252 from days 2/4 to the follow-up visit. OmpR, the transcriptional regulator in the EnvZ/OmpR two-component signal transduction system, involved in osmoregulation, plays a pivotal role in UPEC survival and pathogenicity in the urinary tract (37, 38). Mutation R207H is located in an area of the protein that has been directly implicated in the expression of OmpF even in high-osmolarity medium (39). Amino acid changes also occur in YcbS, an outer membrane usher protein, and AcnB, an aconitase that acts as an iron and oxidative stress-responsive posttranscriptional regulator and modulates levels of the flagellum protein FliC, impacting E. coli motility and colonization in the urinary tract (40, 41). These mutations could play a role in the maintenance of UPEC in the urinary tract, as they appear early and are maintained throughout the infection but are not present in the blood isolates from subject 202.

This report demonstrates the importance of performing thorough preclinical in vitro resistance studies and the potential role of genome-wide sequencing of bacteria in the clinical development of novel antibiotics to better understand their efficacy in patients and evaluate drug resistance mechanisms.

The development of resistance to the LeuRS inhibitor GSK2251052 in subjects enrolled in study LRS114688 is likely the result of a perfect storm involving a high spontaneous frequency of resistance across bacterial species, high-level resistant mutants in which resistance has no obvious in vivo fitness cost, the high bacterial burden of the disease, the genetic plasticity of the target gene, and, even possibly, the bacteriostatic nature of GSK2251052. Although some of these factors can be assessed preclinically in in vitro and in vivo studies, the clinical relevance of such work can be unclear, as the clinical efficacy of an antibacterial agent is not solely dependent on its inherent properties and those of the specific pathogen involved in the infection. The pathogenesis of the disease also plays a critical role in the final outcome, which, ultimately, can be evaluated only in clinical studies. This work highlights the impressive adaptation of bacterial pathogens during human infection and further emphasizes the challenges of antibacterial drug discovery.