ABSTRACT
Pseudomonas aeruginosa is an opportunistic pathogen that causes a wide range of acute and chronic infections. An increasing number of isolates have mutations that make them antibiotic resistant, making treatment difficult. To identify resistance-associated mutations, we experimentally evolved the antibiotic-sensitive strain P. aeruginosa PAO1 to become resistant to three widely used antipseudomonal antibiotics, namely, ciprofloxacin, meropenem, and tobramycin. Mutants could tolerate up to 2,048-fold higher concentrations of antibiotics than strain PAO1. Genome sequences were determined for 13 mutants for each antibiotic. Each mutant had between 2 and 8 mutations. For each antibiotic, at least 8 genes were mutated in multiple mutants, demonstrating the genetic complexity of resistance. For all three antibiotics, mutations arose in genes known to be associated with resistance but also in genes not previously associated with resistance. To determine the clinical relevance of mutations uncovered in this study, we analyzed the corresponding genes in 558 isolates of P. aeruginosa from patients with chronic lung disease and in 172 isolates from the general environment. Many genes identified through experimental evolution had predicted function-altering changes in clinical isolates but not in environmental isolates, showing that mutated genes in experimentally evolved bacteria can predict those that undergo mutation during infection. Additionally, large deletions of up to 479 kb arose in experimentally evolved meropenem-resistant mutants, and large deletions were present in 87 of the clinical isolates. These findings significantly advance understanding of antibiotic resistance in P. aeruginosa and demonstrate the validity of experimental evolution in identifying clinically relevant resistance-associated mutations.
INTRODUCTION
Pseudomonas aeruginosa is an opportunistic pathogen responsible for a wide range of acute and chronic infections. It is a frequent cause of hospital-acquired infections and chronically infects the lungs of most adults with cystic fibrosis (CF), bronchiectasis, or chronic obstructive pulmonary disease (1, 2). Although a number of antibiotics are available for treating P. aeruginosa infections, many isolates are now resistant to one or more of these (3). The failure of antibiotic therapy to eradicate P. aeruginosa infections is a contributing factor to the development of resistance, as bacteria are likely to experience subinhibitory concentrations of antibiotics with selection for more highly resistant mutants (4). Treatment of infections with antibiotics that are ineffective due to resistance increases the length of hospital stays, increases patient morbidity, and can also increase the emergence of resistance (5, 6). Consequently, there is a need for improved tools for determining optimal antibiotic treatment. One attractive option is the use of whole-genome sequencing of infecting bacteria to determine which antibiotics are likely to be effective, an approach that has been developed successfully for other species (7, 8). This approach requires a thorough understanding of the genetic basis of antibiotic resistance.
The basis of antibiotic resistance in P. aeruginosa has been investigated using genetic and biochemical approaches and is multifactorial. Mechanisms of resistance can be classified into the following four groups: reduced antibiotic uptake, enhanced antibiotic efflux, reduced affinity of antibiotics to their cellular targets, and inactivation of antibiotics (9, 10). Resistance mainly occurs through mutations, although genes obtained via horizontal gene transfer also confer a resistant phenotype.
The emergence of whole-genome sequencing (WGS) technologies has provided new approaches to understanding the molecular mechanisms driving antibiotic resistance (11). Experimental evolution of antibiotic-resistant mutants from sensitive parent strains, followed by whole-genome sequencing to identify resistance-conferring mutations, is an approach that has been applied to a number of species, including P. aeruginosa (12–18). Studies to date have confirmed the involvement in resistance of genes previously proposed to be associated with resistance in P. aeruginosa. These include the fusA1 gene that encodes the EFG-1 protein and is associated with aminoglycoside resistance (19) and the ftsI gene that encodes a penicillin binding protein and is associated with carbapenem resistance (20). This approach also has a high potential for identifying previously unknown antibiotic resistance-associated mutations and genes, but some key issues remain to be addressed. Experimental evolution of small numbers of mutants may overlook mutations that can contribute to resistance but do not always arise, making it more difficult to draw robust conclusions. The mutations obtained may be influenced by the selection method used—for example, continuous exposure to increasing amounts of antibiotic in broth culture may give different outcomes to intermittent antibiotic exposure, as occurs during infection in patients. Last and perhaps most importantly, a rigorous comparison of experimentally evolved bacteria and those that have evolved naturally during infection is lacking despite the possibility of resistance mutations in experimentally evolved bacteria differing from those that arise in bacteria during infection. For example, mutations that increase resistance in the laboratory setting may not be tolerated in the complex environment of an infection. It is also of clinical importance to determine whether experimentally evolved mutants have cross-resistance or increased susceptibility (i.e., collateral sensitivity) to other antibiotics. The overall aim of this research was to address these issues and to determine the relevance of mutations arising during experimental evolution to the evolution of antibiotic resistance during infection.
RESULTS
Antibiotic resistance of experimentally evolved mutants.Our research strategy for identifying mutations causing antibiotic resistance is illustrated in Fig. S1 in the supplemental material. First, highly resistant mutants of P. aeruginosa PAO1 were evolved in 13 parallel experiments for each of 3 antibiotics, namely, tobramycin, meropenem, and ciprofloxacin. Each mutant was selected through serial passage on antibiotic gradient agar plates containing increasing amounts of antibiotic, interspersed with growth in antibiotic-free broth. Mutants were considered to have reached maximum resistance when selection failed to give rise to any further increase in resistance (typically between 6 and 8 serial passages). A control culture was passaged 6 times in the absence of antibiotic.
Figure 1 shows the MIC values for each of the 39 experimentally evolved mutants when tested against 6 different antibiotics belonging to 3 different classes. Compared with the parental strain P. aeruginosa PAO1, the evolved mutants had a minimum of 64-fold and a maximum of 2,048-fold increase in their MIC values for the selecting antibiotic. All mutants had similarly increased resistance to a second antibiotic of the same class (levofloxacin, imipenem, and gentamicin for ciprofloxacin-, meropenem-, and tobramycin-selected mutants, respectively).
MICs of experimentally evolved antibiotic-resistant mutants to fluoroquinolones, carbapenems, and aminoglycosides. MIC values are shown for the parental strain P. aeruginosa PAO1 and for mutants selected for resistance to ciprofloxacin (C1 to C13), meropenem (M1 to M13), or tobramycin (T1 to T13). MIC values are shown in mg/liter and colored based on log2-fold change from strain PAO1, ranging from −4 (gray) to 11 (red). Clinical breakpoints as defined by EUCAST (www.eucast.org) in mg/liter are ≥0.5 (ciprofloxacin [Cip]), ≥1 (levofloxacin [Lev]), ≥8 (meropenem [Mer]), ≥8 (imipenem [Imi]), ≥4 (tobramycin [Tob]), and ≥4 (gentamicin [Gen]). A control culture that underwent six serial passages in the absence of any antibiotic selection had the same MIC for all antibiotics as the parental P. aeruginosa PAO1 strain.
The mutants were also tested for cross-resistance to antibiotics of different classes. Most of the meropenem-selected mutants had increased resistance to fluoroquinolone antibiotics and were at or above the EUCAST clinical breakpoint for resistance in many cases (Fig. 1). Similarly, most of the tobramycin-selected mutants had increased resistance to the carbapenems tested, in particular imipenem. Conversely, most of the ciprofloxacin-selected mutants had increased (collateral) sensitivity to at least one carbapenem or aminoglycoside, and some of the meropenem-selected mutants also had increased aminoglycoside sensitivity.
Growth of antibiotic-resistant mutants.To assess the impact of antibiotic resistance on bacterial growth, we undertook growth analysis for each of the experimentally evolved mutants (Fig. 2; see Fig. S2 and S3 in the supplemental material). Growth was measured as area under the growth curve (AUC) in order to capture differences in both growth rate and bacterial cell density in stationary phase. In the absence of antibiotics, mutants in all three resistance groups grew significantly less well than the parental strain P. aeruginosa PAO1, indicating a reduction in fitness. Tobramycin-selected mutants showing the greatest reduction in growth (median values of 1,147.5, 747.5, 610.3, and 468.0 AUC units for strain PAO1 and for ciprofloxacin-, meropenem-, and tobramycin-resistant mutants, respectively). However, there was a large degree of heterogeneity in growth within each group, with some mutants in each group having at least twice the growth of others in the same group.
Growth analysis of experimentally evolved mutants in the absence of antibiotics. Growth of each experimentally evolved mutant was measured during 18 h of incubation at 37°C and area under the concentration-time curve (AUC) values were determined. At least 3 biological replicates were carried out for each mutant, with mean AUC values shown. A one-way ANOVA with post hoc Dunnett’s test was carried out on each antibiotic selection using PAO1 as a comparison (n = 11). Bonferroni-corrected P values for Cip-, Mer-, and Tob-evolved mutants are 3.98 × 10−6, 3.36 × 10−7, and 2.11 × 10−10, respectively. Cip, ciprofloxacin; Mer, meropenem; Tob, tobramycin.
Identification of antibiotic resistance-associated mutations.Whole-genome sequencing (WGS) and variant calling of the 39 antibiotic-resistant mutants showed that each mutant contained between 2 and 8 mutations. Mutations were present across 78 genes; 27 genes were mutated in 2 or more mutants (Table 1; see Table S1 in the supplemental material). In addition, 4 intergenic mutations were present in meropenem-resistant mutants and 5 of the 13 meropenem-resistant mutants contained large deletions ranging in size from 225 to 479 kb (Table S1; see Fig. S4 in the supplemental material). Putative large duplications up to 600 kb were identified in 2 tobramycin-resistant isolates and 1 meropenem-resistant isolate (see Fig. S5 in the supplemental material). No mutations were present in bacteria serially passaged in the absence of antibiotic selection.
Genes mutated in more than one experimentally evolved antibiotic-resistant mutant
Ciprofloxacin-selected mutants.Twenty-nine mutated genes were identified following WGS of the 13 ciprofloxacin-evolved mutants. All of the mutants had a mutation in the gyrA gene and 9 of the mutants also had mutations in the parC or parE genes that encode DNA topoisomerase, known targets of ciprofloxacin (21). Twelve of the mutants also had mutations in nfxB that encodes an efflux pump regulator, with mutations in nfxB being known contributors to ciprofloxacin resistance (21). Ten of the mutants had mutations in pilin-encoding pil genes (Table 1; Table S1). The relationship between pil mutations and ciprofloxacin resistance is not clear, although very recently, other researchers also reported an association between pil mutations and ciprofloxacin resistance (22). Eleven of the mutants had a mutation in the gene PA3491 that, to the best of our knowledge, has not previously been associated with antibiotic resistance.
Meropenem-selected mutants.A total of 26 genes were mutated in the 13 meropenem-selected mutants, with 8 genes mutated in more than 1 mutant (Table 1; Table S1). Many of the mutations were in genes previously associated with meropenem resistance. Eleven of the mutants contained mutations in the porin-encoding oprD gene, with mutations in this gene representing a primary mechanism for carbapenem resistance (13, 23–25). Ten of the mutants had mutations in nalC, nalD, or mexR, with mutations in these genes causing upregulation of the efflux pumps and being associated with β-lactam resistance (26–29). Mutations in ftsI, which encodes the meropenem-binding protein PBP3, were present in 3 mutants, consistent with the known role of such mutations in resistance (13, 16). Mutations were also present in genes not commonly associated with meropenem resistance. Three of the evolved mutants contained mutations within the aroB gene that encodes an enzyme, dehydroquinate synthase, required for the synthesis of aromatic amino acids. Five of the mutants had mutations in genes encoding tRNA ligases, with a different gene being mutated in each case.
Five of the 13 mutants had large deletions ranging in size from 225 kb up to 480 kb. These deletions all overlapped, with the genome region between PA2022 and PA2208 being deleted in all the five mutants. A smaller deletion (1,986 bp) spanning PA2022 to PA2024 was present in one mutant. The deleted region contains PA2023 (galU). Insertion mutations in galU lead to an increase in resistance to meropenem, cephalosporin, and aminoglycosides (24, 25, 30).
Tobramycin-selected mutants.A total of 24 individual genes were mutated across the 13 tobramycin-selected mutants, with 10 genes mutated in more than 1 mutant (Table 1; Table S1). All mutants had at least one mutation in fusA1 that encodes elongation factor G. Mutations in fusA1 have recently been reported to confer tobramycin resistance in laboratory-evolved mutants (19). Nine mutants also had mutations in the wbpL or PA5001 genes that are associated with lipopolysaccharide synthesis, and 10 mutants had mutations in genes directly involved in oxidative phosphorylation and generation of proton motive force (cco, nuoG, PA1549, and PA4429). Mutations affecting either of these processes increased aminoglycoside resistance in a whole-genome screen for resistance genes (31). Several other genes were mutated in a smaller number of tobramycin-selected mutants (Table S1). Many of these, including PA1767, mexY, amgS, and nalC have previously been implicated in aminoglycoside resistance (10, 31–33).
Frequencies of mutations in clinical isolates of P. aeruginosa.It is common for mutations that confer antibiotic resistance in P. aeruginosa to arise during infection in cystic fibrosis patients, with the rate of mutation being increased in some patients by the presence of hypermutator strains (34, 35). We, therefore, addressed the following question: do mutations that arose in the experimental evolution experiments reflect those that have arisen in clinical isolates of P. aeruginosa? To do so, we analyzed the genomes of 558 P. aeruginosa clinical isolates. For comparison, we analyzed the genomes of 172 P. aeruginosa isolates obtained from the general environment and which are, therefore, unlikely to have had antibiotic exposure. All genes that were mutated in two or more of our experimentally evolved mutants were analyzed in each genome to determine whether likely resistance-associated mutations had occurred. The absence of genome sequences of ancestral strains prevented the direct identification of mutations. Instead, genetic variants that are likely to alter protein function by contributing to antibiotic resistance were inferred using PROVEAN, a widely used tool for predicting the likelihood that amino acid differences affect protein function (36). The results are summarized in Table 2.
Frequencies of predicted change-of-function mutations in clinical and environmental isolates of P. aeruginosa, for genes identified through experimental evolutiona
Many of the genes that were mutated in our experimentally evolved mutants had higher frequencies of predicted function-altering variants in clinical isolates than in environmental isolates. For example, the antibiotic target genes gyrA (ciprofloxacin resistance), ftsI (meropenem resistance), and fusA1 (tobramycin resistance) all had predicted function-altering variants in over 15% of the clinical isolates. In many cases, the variations in the clinical isolates were identical to those in experimentally evolved mutants. For example, the variant T83I in GyrA was present in 96 (51.3% of isolates containing predicted functional variants) clinical isolates. Predicted function-altering variants in genes encoding regulatory proteins that are associated with antibiotic resistance, such as nalD, mexR, and amgS, were also common in the clinical isolates. Predicted differences in mexY, which encodes an efflux pump component known to be associated with tobramycin resistance (32), were also frequent in the clinical isolates. Furthermore, the aroB gene, which does not have a characterized role in antibiotic resistance, was identified through whole-genome sequencing of the experimentally evolved meropenem-resistant mutants and was predicted to have function-affecting differences in over 10% of the clinical isolates. Some genes (pilT, pilF, nalC, and PA1767) that were mutated in our experimentally evolved mutants had only low (<5%) frequencies of predicted function-altering differences in the clinical isolates, implying that mutations in these genes may not be advantageous during infection. However, overall, predicted change-of-function variants in the genes analyzed had significantly less impact on the genomes of the environmental isolates than on the genomes of clinical isolates (P = 2.0 × 10−16) (Fig. 3).
Comparison of change-of-function variants in clinical and environmental isolates. PROVEAN values were determined for clinical and environmental (ENV) isolates for all variants in genes that were mutated in at least two experimentally evolved antibiotic-resistant mutants. More negative values indicate variants that are more likely to affect function. The dashed line represents a cutoff value of −2.5, with variants below this value being highly likely to affect protein function. The medians and first and third quartiles are shown in the overlaid boxplots (whiskers showing 1.5 times the interquartile range above the 75th percentile and below the 25th percentile). An ANOVA test using Tukey honestly significant difference (HSD) correction showed that there was a significant difference in the PROVEAN values in the clinical group compared with the ENV group (P = 2 × 10−16).
In addition to point mutations, five of the experimentally evolved meropenem-resistant mutants had large (over 200 kb) deletions. Large deletions of comparable size were also present in 87 of the clinical isolates and only 1 of the environmental isolates (see Fig. S6 in the supplemental material; Table 2).
DISCUSSION
In this study, we have extended the understanding of the genetics of antibiotic resistance in P. aeruginosa. In contrast to previous experimental evolution studies, we used an agar-based selection method interspersed with growth in antibiotic-free broth. This methodology, in conjunction with an analysis of a large number of resistant mutants for each of three clinically relevant antibiotics, identified a number of resistance-associated genes not found in other studies. Importantly, comparison with the genomes of P. aeruginosa isolated from chronically infected patients as well as isolates from the general environment has confirmed that many mutations in experimentally evolved mutants are clinically relevant, while indicating that some appear to be restricted to the laboratory situation.
Our approach of carrying out antibiotic selection on agar plates interspersed with periods of antibiotic-free growth has some parallels with the conditions faced by P. aeruginosa when it colonizes the lungs of cystic fibrosis patients. In both circumstances, bacterial growth is on a semisolid surface with intermittent exposure to antibiotics and so selects for mutations that are stably inherited in the absence of antibiotics. Many of the mutations identified here were in genes that were identified in other experimental evolution studies. For example, mutations altering the target site proteins GyrA and ParC as well as the efflux pump regulator NfxB are well-established contributors to fluoroquinolone resistance (21). Our selection protocol and the large number of experimentally evolved mutants that we analyzed also allowed us to identify some genes that were known to contribute to clinical resistance but were not identified in previous experimental evolution studies. These included parE (ciprofloxacin resistance) (37) and amgS and wbpL (tobramycin resistance) (38).
Importantly, experimentally evolved mutants in this study also had mutations in genes not usually associated with resistance. These include PA3491 and pilin-encoding genes in ciprofloxacin-selected mutants; aroB and tRNA ligases in meropenem-selected mutants (16, 25); and PA1767, cco, and wbpL in tobramycin-selected mutants. The occurrence of mutations in these genes in multiple independently evolved lines (Table 1; Table S1) strongly suggests that the mutations increase antibiotic tolerance in our selection system. Entry of tobramycin into P. aeruginosa can be affected by changes to the cell surface or to membrane potential (31, 39), and mutations in cytochrome c oxidase or wbpL may result in such changes. Cytochrome mutations may also reduce the formation of radical oxygen species that have been proposed to contribute to antibiotic-induced cell death (40). How the other genes may affect resistance is not clear. Further research will be required to determine how the mutations in these genes increase resistance as well as their role (if any) in the resistance of clinical isolates.
Another noteworthy observation was the prevalence of large deletions (up to 8% of the genome) in mutants selected for resistance to meropenem, something that has also been observed previously (13). All the deletions overlapped, with PA2023 (galU) that encodes an enzyme required for LPS core synthesis being deleted in all cases. In a genome-wide screen, a mutation in galU increased meropenem tolerance (24, 25). Deletion of galU may, therefore, contribute to increased tolerance in the deletion-containing mutants obtained here, perhaps in conjunction with other deleted genes.
We used PROVEAN to assess the frequency of likely change-of-function mutations in isolates of P. aeruginosa from patients with CF. For comparison, we analyzed a panel of isolates from the general environment (Table 2). A significant proportion of isolates from CF patients are antibiotic resistant (2, 3), whereas environmental isolates are typically sensitive to antibiotics (41). A high proportion of clinical isolates contained likely function-altering differences in genes that are established as contributing to clinical resistance to fluoroquinolones (gyrA and nfxB), carbapenems (mexR), and/or aminoglycosides (mexY, amgS) (Table 2) (9, 10). These genes contained few or no predicted function-altering differences in isolates from the general environment, emphasizing their clinical relevance as well as validating our approach. Three genes, namely, fusA1, ftsI and aroB, that have only recently been identified as affecting antibiotic resistance (aroB in this study) also had predicted function-altering differences in clinical isolates but few or no environmental isolates (13, 19). These findings demonstrate the utility of experimental evolution for identifying clinically relevant genes associated with antibiotic resistance.
The oprD gene that is well established as being associated with carbapenem resistance had a high frequency of predicted function-altering differences in clinical isolates, as expected. It also had a high frequency of such differences in environmental isolates of P. aeruginosa consistent with earlier findings (23), a finding that was related to its role in the environmental adaptation of P. aeruginosa. Nonetheless the mean PROVEAN scores for OprD were lower in the clinical isolates (−2.8 clinical, −0.8 environmental), likely reflecting the need for more severe loss-of-function changes to OprD in contributing to carbapenem resistance.
Conversely, the parE, parC, and nalC genes that were previously shown to influence the resistance phenotype of clinical isolates (26, 37, 42) and in experimental evolution studies (Table 2) (12, 13) had only low frequencies of predicted function-altering differences in the clinical isolates in this study. This may indicate that mutations in these genes are associated with higher levels of antibiotic resistance than that of the isolates in our study. Mutations in these genes may also have a high fitness cost in the clinical environment or their effects may be influenced by epistatic interactions with other genes (43).
A number of genes that were mutated in multiple experimentally evolved mutants are not commonly associated with antibiotic resistance. Many of these, such as pil genes, tpiA, and PA1767, had only low frequencies of predicted function-changing differences in the clinical isolates. This finding suggests that mutations in these genes contribute to increased antibiotic tolerance in laboratory culture but likely do not do so during infection and further emphasizes the importance of comparing experimentally evolved mutants with clinical isolates. In contrast, mutations in genes such as mexZ are well-characterized in clinical isolates (44–48) but were not identified in our experiments or in other experimental evolution studies, indicating that experimental evolution studies alone are not necessarily sufficient to identify all resistance-associated mutations.
The experimentally evolved mutants had increased tolerance to antibiotics of the same class (Fig. 1), as found previously (14). Many of the mutants also had altered MICs for antibiotics of different classes, and some of these changes could be related to the mutations that were present. For example, many of the meropenem-selected mutants had mutations in nalC, and such mutations lead to the overexpression of mexAB-oprM (29), which may reduce susceptibility to fluoroquinolones. Conversely, several of the ciprofloxacin-selected mutants had increased susceptibility to aminoglycosides and to imipenem, likely due to the presence of nfxB mutations (49). The occurrence of collateral changes in antibiotic sensitivity as a result of mutations selected in response to antibiotic exposure has also been observed in isolates of P. aeruginosa from patients (50–53). This has important clinical implications because such outcomes during infection may affect treatment options for patients with CF or other chronic infections.
In the absence of antibiotics, the majority of the experimentally evolved mutants showed significant reductions in growth relative to the wild-type control, consistent with previous studies (Fig. 2) (13, 54–56). It is noteworthy that isolates of P. aeruginosa from patients are often slow growing (57), although the extent to which this phenotype is related to antibiotic resistance is not clear.
The analysis of whole-genome sequences of infecting bacteria, in order to predict which antibiotics will be effective in treatment, has the potential to significantly improve the quality of patient care, and this approach is advanced for some species of bacteria (7, 8). The genetic complexity of antibiotic resistance in P. aeruginosa makes it more challenging to develop accurate tools for the prediction of resistance phenotypes from genome sequences. However, cataloguing the gene set that contributes to resistance phenotypes through experimental evolution, coupled to our analysis of clinical isolates, will advance efforts to develop reliable genome-based prediction tools.
In conclusion, our study demonstrates the power of experimentally evolving multiple mutants for identifying genes that, when mutated, contribute to increased antibiotic resistance. The evolution of multiple independent mutants reveals the frequencies at which mutations arise in individual genes, which is likely to be related to the extent to which they contribute to increased resistance in our selection system. Use of an agar-based method instead of the broth-based methods of earlier studies identified genes not previously associated with resistance and showed that the spectrum of resistance mutations is influenced by the selection protocol. Crucially, the analysis of resistance-associated genes in clinical isolates of P. aeruginosa validated the clinical relevance of genes identified only through experimental evolution approaches, while demonstrating that some genes that are mutated in laboratory-based experiments are unlikely to be relevant to resistance during infection. The extension of our approach to other antibiotics and, indeed, other bacterial species will greatly strengthen our understanding of the genetic basis of antibiotic resistance in infectious bacteria.
MATERIALS AND METHODS
In vitro evolution of antibiotic-resistant mutants.Antibiotic gradient plates were prepared, according to the methodology described by Bryson and Szybalski (58), to evolve meropenem- (Penembact, Venus Remedies Limited), tobramycin- (Mylan New Zealand ltd), and ciprofloxacin- (Cipflox, Mylan New Zealand Ltd.) resistant P. aeruginosa mutants. Briefly, 15 ml aliquots of molten Difco Muller-Hinton (MH) agar were poured into a petri dish tilted at an approximate incline of 15° to ensure a slanted slope of media. When cooled, the petri dish was placed on a flat surface, and an additional 15 ml of molten MH agar, supplemented with the antibiotic treatment, was poured onto the slanted MH agar to create an antibiotic gradient across the plate. Overnight broth culture prepared from a single colony subculture of reference strain P. aeruginosa PAO1 was adjusted to an optical density at 600 nm (OD600) of 0.01 (1.5 × 106 CFU/ml) and a 2-ml aliquot poured onto the antibiotic gradient plate. Excess liquid was removed by pipette following a 10-minute incubation at room temperature. Inoculated plates were incubated under aerobic conditions for 24 hours at 37°C. An isolated single colony, growing furthest up the antibiotic concentration gradient, was selected and cultured overnight in Luria broth at 37°C/200 rpm. This inoculum was then adjusted as described above and used to inoculate a subsequent gradient plate containing antibiotics of one doubling concentration greater than the previous plate. This technique was repeated with increasing antibiotic concentrations until it was no longer possible to identify mutants resistant to higher amounts of antibiotic. Thirteen replicate experiments were carried out for each antibiotic, with one mutant from experiment being used for further study. As a control, strain PAO1 was passaged six times on agar without antibiotic. An overview of methods used is shown in Fig. S6.
MIC determination.MIC determination was carried out in accordance with the protocol described by Wiegand and colleagues (59). Briefly, overnight broth cultures of the evolved mutants and the laboratory strain PAO1 were adjusted to OD600 of 0.01 (1.5 × 106 CFU/ml), and 5-μl aliquots were spread onto MH agar plates containing doubling antibiotic concentrations. Control plates, MH agar without antibiotic supplementation, were used as a growth comparison. All MIC plates were incubated in aerobic conditions at 37°C for 24 hours. The MIC for each isolate was determined as the lowest concentration that inhibited visible growth, excluding single colonies or faint haze. Antibiotics selected for testing included each of the target antibiotics along with one other antibiotic from the same class (aminoglycosides [gentamicin], carbapenems [imipenem], and fluoroquinolones [levofloxacin]). The EUCAST guidelines (www.eucast.org) were used to interpret antimicrobial susceptibility patterns.
Growth analysis.Overnight broth cultures of the antibiotic-resistant mutants and laboratory strain PAO1 were adjusted to OD600 of 0.01 (1.5 × 106 CFU/ml), and 200-μl aliquots were dispensed into JETbiofil 96-well tissue culture plates. The microtiter plates were incubated in a BMG FLUOstar Omega microplate reader at 37°C/200 rpm for 18 hours. Optical density (OD600) was recorded every 30 minutes to measure the growth of the isolates. Area under the growth curve (AUC) was used to give a measure of growth that included lag phase, rate of growth during log phase, and final cell density. Growth dynamics were calculated by using the R package GrowthCurver version 0.2.1 (60). Logistic AUC was used as the metric for quantifying growth.
Clinical and environmental isolates of P. aeruginosa.A cohort of 558 clinical and 172 environmental genome assemblies of P. aeruginosa isolates from multiple countries were used in this study (see Table S2 in the supplemental material) (61, 62). Clinical isolates were from patients with cystic fibrosis, bronchiectasis, or chronic obstructive pulmonary disease. Six isolates from patients with cystic fibrosis were unique to this study.
Statistical analysis.All statistical analyses were carried out using R version 3.5.0 (63). Post hoc Dunnett tests were done using R package multcomp version 1.4-8 (64). Plots were created using R package ggplot2 (65).
Whole-genome sequencing.Genome assemblies of clinical and environmental isolates have been described previously (61, 62) (Table S2). For experimentally evolved mutants and clinical isolates sequenced in this study, genomic DNA was extracted from overnight cultures of endpoint-resistant mutants using the MoBio UltraClean microbial DNA isolation kit. Library preparation and sequencing were carried out by New Zealand Genomics Limited using the Illumina HiSeq 2000 and Illumina MiSeq platforms.
Analysis of genomic data and mutation detection.Raw sequencing reads were examined for quality pre- and posttrimming using FastQC version 0.11.5. Trimming of raw reads was done using Trimmomatic version 0.36 (66). Draft genomes of clinical isolates sequenced in this study were assembled using SPAdes version 3.12.0 (67). The genome sequence of the P. aeruginosa PAO1 used in this study (PAO1-Otago) was assembled using a combination of in-house code and GDtools part of the BreSeq package (68), with mapping to the P. aeruginosa PAO1 genome sequence at https://www.pseudomonas.com (69). A full list of differences between the genomes is shown in Table S3 in the supplemental material. Mutations in the experimentally evolved resistant isolates were identified through comparison to PAO1-Otago using BreSeq version 0.30.0 (68).
Prediction of duplications.For the prediction of duplicated regions within the experimentally evolved mutants, CNOGpro (70) was used. Hits files were generated from bam file output in the breseq analysis and parsed into CNOGpro, G+C content was normalized, and 1,000 bootstrap replicates were performed using quartiles of 0.025 and 0.975. Hidden Markov model (HMM) correction was performed, allowing for a maximum of 3 possible states and allowing for an error rate of 0.01. Tables were output containing the predicted copy number for each gene and intergenic region. Regions were called as putative duplications when there was a sustained increase in predicted copy number across more than 5 genes. Known multicopy operons were removed after comparison to the wild-type P. aeruginosa PAO1-Otago coverage.
Mutant comparisons to clinical and environmental P. aeruginosa isolates.Protein sequences of genes found to be mutated within this study were obtained from https://www.pseudomonas.com (69). These were used as query sequences for a tblastn search against compiled databases containing either clinical or environmental P. aeruginosa isolates (Table S2). BLAST outputs were aligned to the reference PAO1 sequence using Clustal-Omega version 1.2.4 (71). Polymorphic sites were called using a modified version of snp-sites version 2.3.2 (72). Nonsynonymous polymorphic sites (variants) were extracted from vcf files. The effects of variants were predicted using PROVEAN version 1.1 (36) and compiled using BLAST version 2.2.28+ (73), psiBLAST version 2.2.28+ (74), cd-hit version 4.7 (75), and the NCBI nonredundant database (version retrieved 1 May 2019) (76). Variations with a score of −2.5 or below were considered being likely to affect the biological functions of proteins (36).
Data availability.Raw sequence reads (fastq format) of experimentally evolved mutants are available under BioProject accession number PRJNA542028 in the NCBI Sequence Read Archive (SRA). Accession numbers for the genomes of other isolates of P. aeruginosa are listed in Table S2.
ACKNOWLEDGMENTS
This research was supported by research grants to I.L.L. and W.M.P. from the University of Otago and the New Zealand Health Research Council (17/372). C.W. gratefully acknowledges support from the Cystic Fibrosis Trust (UK). S.J.T.W. was supported by a Postgraduate Scholarship from the University of Otago and A.R. by a New Zealand International Doctoral Research Scholarship.
We are very grateful to Roger Levesque and coworkers for making genome assemblies available prior to publication. Additionally, we are grateful to all members of the International Pseudomonas Consortium Database (IPCD), who donated strains utilized in this study. We are grateful to Kay Ramsay and Scott Beatson for their comments on an earlier version of the manuscript.
FOOTNOTES
- Received 11 August 2019.
- Returned for modification 9 September 2019.
- Accepted 23 September 2019.
- Accepted manuscript posted online 30 September 2019.
Supplemental material for this article may be found at https://doi.org/10.1128/AAC.01619-19.
- Copyright © 2019 American Society for Microbiology.
