Fitness cost of antibiotic susceptibility during bacterial infection

See allHide authors and affiliations

Science Translational Medicine  22 Jul 2015:
Vol. 7, Issue 297, pp. 297ra114
DOI: 10.1126/scitranslmed.aab1621


Advances in high-throughput DNA sequencing allow for a comprehensive analysis of bacterial genes that contribute to virulence in a specific infectious setting. Such information can yield new insights that affect decisions on how to best manage major public health issues such as the threat posed by increasing antimicrobial drug resistance. Much of the focus has been on the consequences of the selective advantage conferred on drug-resistant strains during antibiotic therapy. It is thought that the genetic and phenotypic changes that confer resistance also result in concomitant reductions in in vivo fitness, virulence, and transmission. However, experimental validation of this accepted paradigm is modest. Using a saturated transposon library of Pseudomonas aeruginosa, we identified genes across many functional categories and operons that contributed to maximal in vivo fitness during lung infections in animal models. Genes that bestowed both intrinsic and acquired antibiotic resistance provided a positive in vivo fitness advantage to P. aeruginosa during infection. We confirmed these findings in the pathogenic bacteria Acinetobacter baumannii and Vibrio cholerae using murine and rabbit infection models, respectively. Our results show that efforts to confront the worldwide increase in antibiotic resistance might be exacerbated by fitness advantages that enhance virulence in drug-resistant microbes.


The challenges presented by the continued increase in antibiotic resistance among microbial pathogens are confounded by the complexity of the interplay among the microbes, drugs, and infected hosts. For many years, it has been assumed that acquisition of resistance comes with a negative fitness cost for most microorganisms (1). However, using a bank of ~300,000 different transposon (Tn) insertion mutants of Pseudomonas aeruginosa strain PA14, we showed recently (2, 3) that gastrointestinal (GI) tract colonization and bacterial dissemination into the spleen after induction of neutropenia strongly selects for mutant bacteria that are resistant to the antibiotic carbapenem and are unable to produce the porin OprD, an outer membrane channel required for drug entry into the bacterial cell (4, 5). Both carbapenem-resistant, OprD-deficient laboratory strains and comparable clinical isolates were uniformly found to be more fit for in vivo persistence than matched, isogenic, drug-susceptible strains (3). In the clinic, carbapenem-resistant P. aeruginosa strains are commonly isolated from patients, and OprD deficiency is the main mechanism responsible for this resistance in patients (6). Therefore, a potentially overlooked consequence of the acquisition of antimicrobial resistance could be enhanced fitness and virulence of pathogens. This hypothesis suggests that a reduction in antibiotic use might not lead to the expected beneficial effects of having fewer drug-resistant strains that cause infections. Instead, in the absence of selective pressure from antibiotics, extant drug-resistant strains might still outcompete the less resistant but also less fit strains for environmental persistence, host infectivity, and transmission.

Severe and difficult to manage manifestations of microbial infections often occur in the respiratory tract. Examples include P. aeruginosa infections in individuals with cystic fibrosis (7, 8) as well as immunocompromised, neutropenic patients (9) and those hospitalized in intensive care units (ICUs) who also require mechanically assisted ventilation (10). One of the main obstacles to successful therapy for a substantial fraction of P. aeruginosa strains is their high level of intrinsic and acquired resistance to antibiotics that are active against other Gram-negative bacteria such as Escherichia coli and Klebsiella pneumoniae, including amoxicillin–clavulanic acid (β-lactams), kanamycin (an aminoglycoside), nalidixic acid (a quinolone), and sulfonamide-trimethoprim (6). Because these drugs are not used clinically for the treatment of P. aeruginosa, this pathogen most likely retains the chromosomally encoded resistance mechanisms in the absence of antibiotic selective pressures.

Thus, to obtain a more comprehensive view of the genetic basis of P. aeruginosa virulence during lung infections and to analyze the contribution of intrinsic and acquired antibiotic resistance to fitness in this setting, we used a saturated Tn insertion bank in P. aeruginosa strain PA14 to ascertain the roles of all nonessential gene products for fitness in murine lungs. Our results indicated that the loss of genes that encode intrinsic antibiotic resistance factors is associated with reduced fitness during pulmonary infection, whereas the acquisition of drug resistance enhanced fitness in the same setting. To ascertain whether this principle also applied to other bacterial pathogens, we analyzed the effect on fitness associated with the loss or acquisition of antibiotic resistance in Acinetobacter baumannii, a problematic, multidrug-resistant organism commonly associated with hospital infections (11), and in Vibrio cholerae, an organism that is responsible for a major disease in various parts of the world and does not routinely encounter antibiotic selective pressure during treatment. Our findings consistently indicate that for each of these very different pathogens, maintenance of intrinsic drug resistance and acquisition of new resistances promote fitness and survival in an infected host, a finding counter to the currently prevailing view in the field (1) that increased antibiotic resistance has a negative fitness cost.


Transposon sequencing analysis in an animal model of P. aeruginosa acute pneumonia

To establish lung infection, C3H/HeN female mice were inoculated with a bank of ~300,000 random Tn insertion mutants of P. aeruginosa PA14 grown overnight in LB as previously described (2, 3). This inoculum assured that a bacterial burden at least 10 times greater than the size of the bank [that is, >3 × 106 colony-forming units (CFU)] was recovered from the lung for each time point examined (1, 6, and 24 hours, or until the mice were moribund, between 36 and 48 hours after infection for the last time point). Bacterial DNA was recovered, and Tn-chromosome junctions were sequenced and quantified using the high-throughput techniques described previously (2, 3). A comprehensive analysis of the results was carried out using a rigorous strategy to determine the gain or loss of strains with Tn insertions in different genes (2). This strategy involved first grouping the disrupted genes into functional classes, then into operons, and finally conducting a specific gene-by-gene analysis to indicate the likely role of each individual gene product in microbial biology in vivo. Circular representations of the genome at several scales were used to depict the specific genes of interest, their locations within the P. aeruginosa genome, their size, and their associated phenotypes. The sequencing reads per kilobase per million [RPKM; defined as (number of reads)/(kilobase length of gene) × (millions of reads in the data set)] were used to account for differences in gene sizes and sequencing efficiencies between different DNA samples (2, 3).

Analysis by functional classes

The changes in in vivo fitness of mutants within each of 27 functional genomic classes (12, 13) ( are shown in fig. S1 (see data files S1 and S2 for a more complete description of the analysis by functional classes). In the bacterial population recovered from mouse lungs 1 hour after infection, we observed minor changes in the distribution of Tn mutants compared to those in the original inoculum grown in culture in LB (fig. S1A). This result confirmed the absence of an initial bottleneck after intranasal installation of the library in a P. aeruginosa population. Loss of gene functionality after Tn insertion that leads to enhanced fitness (that is, more reads from Tn-interrupted genes in the output compared with the input) occurs rarely in virulence studies. Only Tn insertions in genes in one functional class, motility and attachment, were found to be significantly increased at all time points analyzed (fig. S1). A marked (>679%) increase in RPKM for this group was observed at 24 hours after infection, whereas RPKM recovered for all other functional classes were decreased (fig. S1C). Within the motility and attachment functional class, only strains with Tn insertions in genes associated with the production of flagella (see data file S2) and type IVa pili (T4aP) displayed an increased in vivo fitness during murine lung infection (fig. S2). This pattern for the T4aP Tn insertion mutants was similar to that previously found in bacterial strains recovered from the mouse cecum during P. aeruginosa GI colonization (fig. S3). Strains with Tn insertions that lead to a phenotype of preserved or even increased production of pili (hyperfimbrial) (2) did not display increased in vivo fitness. In contrast, P. aeruginosa that carried a Tn insertion in most of the genes needed to produce T4aP displayed a reduction in fitness after 48 hours in water, a natural habitat for P. aeruginosa (fig. S4) (14), indicating that these structures are beneficial for survival in the environment but are detrimental to survival in an infected lung, perhaps because T4aP is recognized by the host’s immune defenses.

Analysis by operons

Significant changes in fitness of strains with Tn insertions in virulence factor (VF) genes defined in the VF database (15) ( were observed only after 24 hours of infection (Fig. 1 and fig. S5). Tn insertions in genes that encode protein components of the quorum-sensing system and of the type 1, 2, 3, 5, and 6 secretion systems (which permit the secretion of proteins to the extracellular space) as well as proteins that participate in the synthesis of the exopolysaccharides alginate and Pel, lipopolysaccharide (LPS), the rhamnolipids, pyochelin, pyoverdine, and pyocyanin were all under-represented after 24 hours of infection compared to those in the strains recovered at the 6-hour time point. The only exceptions were Tn insertions in the algR gene, which encodes a regulator of the production of both alginate and T4aP; the fitness of strains that carried these mutations was increased after 24 hours in the lung (Fig. 1), likely because of the negative impact of T4aP expression in vivo. These observations indicate that alginate is a bona fide VF, but the absence of T4aP is dominant for survival in the host.

Fig. 1. In vivo fitness of P. aeruginosa mutants with Tn insertions in four classes of genes annotated as VFs after 1, 6, or 24 hours of infection in the murine lung.

The inner green circle represents RPKM changes between LB-cultured P. aeruginosa and 1 hour of lung infection; pink circle depicts changes from 1 to 6 hours of infection; and yellow circle represents changes from 6 to 24 hours of infection. The outer circle is the entire P. aeruginosa chromosome with each gene represented by one of six different shades of blue organized into a repetitive pattern. The genes associated with the identified VFs are represented at ×20 magnification in the outer circle. The VFs are depicted in four color-coded categories (green, orange, gray, and purple) in the green, pink, and yellow inner circles, and across a gradient of lighter to darker bars to differentiate the different gene clusters. A decrease in RPKM (bars pointing to circle’s center indicate a decreased fitness) of Tn insertions in genes encoding for all the known major VFs of P. aeruginosa was observed after 24 hours of lung infection (yellow circle) but not after 1 hour (green circle) or 6 hours (pink circle) of infection, except for Tn insertions in genes encoding LPS O-antigens (darker green).

Impact of lung infection duration on P. aeruginosa fitness

After 24 hours in the lung, only one nonessential P. aeruginosa gene cluster, PA14_23360-23470, which contains 12 genes associated with LPS O-antigen production, had a >10-fold decrease in RPKM (fig. S6). By comparison, in the GI tract model, after 5 days of infection, >75% of the Tn insertion mutations in genes important for bacterial colonization had at least a 10-fold decrease in RPKM (Fig. 2A) (2). This notable difference suggests that either the shorter duration of selection or conditions in the respiratory tract have a differential selective effect on P. aeruginosa fitness. To discriminate between these interpretations, we compared the complete RPKM data sets generated after 5 days of GI tract colonization with those obtained from acute lung infection at various times after inoculation. Examination of these data sets revealed very few changes in the composition of the Tn insertion mutants after 1 or 6 hours of lung infection (Fig. 2, B and C). Notably, more pronounced differences in fitness were seen after 24 hours of infection (Fig. 2D) but still not close to the magnitude of the RPKM changes found in the GI tract after 5 days. However, when we analyzed the impact of lung infection at a point at which the mice were moribund (36 to 48 hours; Fig. 2E), we found that the RPKM changes were very similar to those found after 5 days of GI tract colonization (Fig. 2, A and E, and table S1). This finding supports the hypothesis that P. aeruginosa uses similar strategies to establish infection and overcome host defenses in the GI tract and lung but that detection of maximal differences is dependent on having sufficient time for selection.

Fig. 2. In vivo fitness of P. aeruginosa PA14 Tn library.

(A) RPKM changes that occurred after five days of GI tract colonization by P. aeruginosa, as previously reported (3). (B to E) Relative ranking and absolute number of RPKM that changed for 5977 genes of P. aeruginosa PA14 during lung infection, comparing the RPKM in the LB input with those obtained 1 hour after infection (B) or comparing RPKM obtained 1 hour after infection with those obtained at 6 hours after infection, and 24 hours after infection with 32 to 48 hours after infection (labeled “Lung 48 h”) (C to E). Dots above the input lines indicate Tn insertions in genes with a positive fitness (increase in in vivo RPKM), whereas dots below the input line indicate those with a negative fitness (decrease in in vivo RPKM).

Fitness cost of antibiotic susceptibility—Analysis of selected genes

By comparing the RPKM data for each nonessential P. aeruginosa gene at each time point of bacterial sampling during murine lung infection, we identified a total of 116 genes with Tn insertions that had at least 100 RPKM after 24 hours of infection and with a twofold or greater increase in RPKM after 36 to 48 hours of infection (table S2). These findings strongly suggest that these 116 genes are ones that, when disrupted, are likely to lead to enhanced fitness in an in vivo infection. Within this collection of 116 genes, in addition to those that encode products involved in motility and attachment, we measured fitness enhancement of P. aeruginosa strains with Tn insertions in the oprD gene (PA14_Tn-oprD), which also imparts a carbapenem resistance phenotype, and glpT (PA14_Tn-glpT), which is responsible for fosfomycin resistance (Fig. 3A). The OprD protein is a channel for entry of carbapenem antibiotics, and the glpT gene encodes a glycerol 3-phosphate transporter that brings fosfomycin into E. coli and P. aeruginosa (16).

Fig. 3. Increased fitness of carbapenem (PA14_Tn-oprD)– and fosfomycin (PA14_Tn-glpT)–resistant P. aeruginosa Tn insertion mutants in murine lung infection.

(A) Increases in RPKM in Tn-glpT or Tn-oprD P. aeruginosa strains from the TnSeq analysis in the murine model of pneumonia. (B) Virulence of P. aeruginosa PA14 glpT or oprD Tn insertion mutants in lung infections compared to wild-type (WT) and complemented strains (n = 12 mice per group; WT versus Tn-oprD: P < 0.0001, WT versus Tn-glpT: P < 0.0001, Tn-oprD versus Tn-oprD::PoprD: P = 0.0002, and Tn-glpT versus Tn-glpT::PglpT: P < 0.0001, log-rank test). (C) Killing of J774 macrophages by the fosfomycin-resistant glpT mutant compared to the WT or glpT-complemented strains. (D) Multiplication of P. aeruginosa PA14 WT, ΔglpT, and glpT-complemented mutant inside J774 macrophages. (E) Multiplication of the internalized ΔglpT compared with WT and glpT-complemented strains in MH-S alveolar macrophages after 24 hours. For (C) to (F), bars represent means of triplicate determinations, and error bars indicate the SD. *P < 0.05, Tukey’s post hoc test versus control. P < 0.05, overall analysis of variance (ANOVA) for each data set. PA14 background lacking the exoU gene was used in the in vitro studies to avoid cytotoxic effects of this effector of the type 3 secretion system.

We attempted to corroborate the transposon sequencing (TnSeq) findings by using individual mutants in the murine pneumonia model. Here, we used strains with Tn insertions in the oprD and glpT genes obtained from the ordered Tn library of P. aeruginosa PA14 (17). More of the mice infected with the Tn-oprD or the Tn-glpT strains became moribund or died when compared with those infected with the wild-type PA14 strain or the complemented strains Tn-oprD::oprD or Tn-glpT::glpT (Fig. 3B; wild type versus Tn-oprD: P <0.0001, wild type versus Tn-glpT: P < 0.0001, Tn-oprD versus Tn-oprD::oprD: P = 0.0002, and Tn-glpT versus Tn-glpT::glpT: P < 0.0001, log-rank test). These observations indicate a specific increased virulence after loss of the OprD or GlpT proteins.

To extend our results to clinical P. aeruginosa strains, and particularly to carbapenem resistance, which is a major issue in the treatment of P. aeruginosa infections, we used our murine pneumonia model to evaluate virulence of isogenic clinical strains recovered from two patients at different time points. For each patient, the clinical P. aeruginosa isolates taken at earlier time points (strains 48.1 and 51.1) had intact oprD genes and were carbapenem susceptible [imipenem minimum inhibitory concentration (MIC) <1 mg/liter] (3); in contrast, both isolates taken from the patients at later time points (strains 48.2 and 51.2) carried mutations in the oprD genes that ceased production of OprD and were carbapenem-resistant (imipenem MIC ≥ 32 mg/liter) (3). As shown in fig. S7, carbapenem-resistant and OprD-deficient clinical strains 48.2 and 51.2 were significantly more virulent in the murine pneumonia model than their corresponding carbapenem-susceptible strains (P = 0.0039, log-rank test) and also were more virulent compared to the oprD mutant strains complemented with an intact oprD gene (48.2 versus 48.2::PoprD: P = 0.0016 and 51.2 versus 51.2::PoprD: P = 0.0001, log-rank test).

We previously showed that the increased in vivo fitness of OprD-deficient P. aeruginosa is associated with an enhanced killing of murine macrophages (3). Therefore, we compared the survival of activated murine J774 macrophages after infection with wild-type or glpT-deleted (ΔglpT) P. aeruginosa constructed in a PA14 background that lacked the exoU gene (to avoid the cytotoxic effects of this effector of the bacteria’s type 3 secretion system). The ΔglpT strain was able to kill macrophages, whereas there was a 40% increase in macrophage numbers infected with the glpT-complemented strain 24 hours after infection (Fig. 3C). To assess the role of GlpT in the resistance of P. aeruginosa to phagocytic killing, we tested wild-type, ΔglpT, and glpT-complemented strains for survival in murine J774 macrophages (Fig. 3D) and murine MH-S alveolar macrophages (Fig. 3E). Both of these cell types were able to either limit replication or survival of the wild-type and glpT-complemented strains, whereas the ΔglpT strain was able to replicate intracellularly during the 24-hour experimental infection period. Thus, the increased fitness and virulence of the fosfomycin-resistant Tn-glpT mutant in the pneumonia model may be linked to its enhanced killing of and survival within macrophages and possibly other cells of the immune system.

P. aeruginosa is intrinsically resistant to multiple antibiotics from several different classes, and consequently, these drugs cannot be used to treat infections. We wondered whether—similar to glpT- and oprD-deficient, antibiotic-resistant strains generated by Tn mutagenesis (Fig. 3)—chromosomal genes that encode determinants of intrinsic antibiotic resistance are associated with an increased in vivo fitness (and thus evolutionarily selected) as a co-result of the expression of resistance-promoting gene products.

To this end, we evaluated the fitness of strains with Tn insertions in genes coding for natural resistance of P. aeruginosa to several different antibiotic classes: ampC (encodes a cephalosporinase for resistance to amoxicillin–clavulanic acid), aph (encodes an aminoglycoside phosphotransferase for high-level resistance to kanamycin), and the mexAB-oprM operon (encodes the components of an efflux pump for the resistance to both nalidixic acid and trimethoprim-sulfonamide) (tables S3 to S5). Analysis of the TnSeq data generated to evaluate fitness for murine GI colonization (3) and the current data set from murine lung infections showed that Tn insertions in genes within these three distinct antibiotic resistance loci resulted in drug-susceptible strains with a decreased in vivo fitness for both GI colonization and lung infection (fig. S8, A and B). Therefore, we tested the in vitro and in vivo fitness of specific mutants with Tn insertions in the selected genes. First, we confirmed the loss of the appropriate antibiotic resistance for each mutant (tables S3 to S5). Furthermore, we showed that there were no growth differences between wild-type P. aeruginosa PA14 strain and mutated strains with Tn insertions in the ampC, aph, mexA, or oprM genes when cultured separately in vitro in LB (fig. S9A). To measure the in vitro competition index of the mutants, we mixed the various Tn insertion mutants with an equal amount of the wild-type PA14 strain in LB and grew the cells on agar plates to count colonies after 24 hours of competition in liquid culture. Wild-type PA14 was able to outgrow only the Tn-oprM strain (fig. S9B), indicating that the loss of OprM decreases fitness in LB when in competition with wild-type cells (2). In contrast, we observed no comparable fitness cost associated with the loss of OrpM or any of the other antibiotic resistance genes when cocultured in water with the wild-type PA14 strain for up to 5 days (fig. S9C). Thus, genes that encode intrinsic antibiotic resistance factors in P. aeruginosa are not essential for in vitro survival in a natural environment such as water.

Next, we tested the ampC, aph, mexA, and oprM Tn insertion mutants for fitness in the murine GI colonization and pneumonia models (Figs. 4 and 5). The in vivo fitness of all four Tn insertion mutants was reduced when compared to wild-type PA14 in the GI tract (Fig. 4) and the lung infection model (Fig. 5A). The magnitude of the decrease in fitness in the GI colonization model ranged from near total—as determined by the rapid clearance of the PA14_Tn-ampC and PA14_Tn-mexA mutants 24 hours after establishing colonization (Fig. 4, A and B)—to partial, as shown by a ~40 to 50% decrease in in vivo fitness of the PA14_Tn-aph strain (Fig. 4C, left panel) and the >85% decrease in recovery of the PA14_Tn-oprM mutant (Fig. 4D, left panel). In the lung infection model (Fig. 5A), fewer mice became moribund or died by 48 hours after challenge with P. aeruginosa strains that carried Tn insertions in the ampC (P = 0.0162), mexA (P = 0.0288), oprM (P = 0.0036), or aph (P = 0.0050) genes (by log-rank test). Together, these data suggest that loss of intrinsic antibiotic resistance in P. aeruginosa is associated with a reduced in vivo fitness.

Fig. 4. Effect of the loss of intrinsic antibiotic resistance genes in P. aeruginosa on fitness during GI tract colonization.

The ability to colonize the murine GI tract was measured by the competitive index (CI). The CI is calculated by dividing the proportion of mutant cells at the end of the competition by the proportion at the start. (A to D) A CI ratio <1 in the left-hand panels indicates that the Tn insertion mutant is less fit. Right-hand panels depict the qRT-PCR analysis of the expression of the transcript for each indicated gene when P. aeruginosa was grown in LB, drinking water used for colonization, or recovered from the murine GI tract. (A) Resistance to amoxicillin–clavulanic acid (Tn-ampC). (B and C) Resistance to nalidixic acid, chloramphenicol, and cotrimoxazole (Tn-mexA and Tn-oprM). (D) Resistance to kanamycin (Tn-aph). Bars for CI represent means from four mice, and error bars represent the SD. Bars for qRT-PCR represent means of three individual experiments, and error bars represent the SD. *P < 0.05, one-sample t test (default = 1).

Fig. 5. Effect of the loss of intrinsic antibiotic resistance genes in P. aeruginosa on fitness in the murine lung infection model.

(A) Survival curves of mice after lung infection with WT P. aeruginosa or strains carrying Tn insertions in oprM, mexA, ampC, or aph genes. A significant decrease in virulence was observed between WT and the various insertion mutants (oprM: P = 0.0036, ampC: P = 0.0162, mexA: P = 0.0288, and aph: P = 0.0050, log-rank test). (B to E) mRNA transcript expression levels of ampC, aph, mexA, and oprM as determined by qRT-PCR for the indicated P. aeruginosa mutant strain during infection of J774 macrophages for 1, 6, or 24 hours or in the murine lung for 1 hour. For each sample, transcript levels of ampC, aph, mexA, and oprM were assessed by relative quantification using the 2−ΔΔCt method. Expression of the rpsL gene was used as a housekeeping control gene.

To validate the observed differences between in vitro and in vivo fitness of Tn mutants in intrinsic antibiotic resistance genes, we performed quantitative reverse transcription polymerase chain reaction (qRT-PCR) experiments to measure expression levels of the four genes in LB, water, and the ceca of colonized mice. mRNA transcripts for the ampC, mexA, oprM, and aph genes were found at higher levels in bacterial cells recovered from the GI tract compared with that in bacterial cells recovered from the two in vitro conditions, indicative of an important in vivo role for these genes and consistent with the absence of an in vitro growth defect of the corresponding Tn insertion mutants (Fig. 4, A to D, right panels). Furthermore, the transcript level for the oprM gene was higher in LB than in water (Fig. 4D, right panel), consistent with the in vitro defect of PA14_Tn-oprM observed in competition with wild type in LB but not in water (fig. S9, B and C). In addition, we found a progressive increase in the expression of all four intrinsic resistance genes in wild-type P. aeruginosa after internalization by macrophages (Fig. 5, B to E), parallel to the observed effects on bacterial survival determined after macrophage internalization for all strains except ΔampC (fig. S10). Together, ampC, mexA, oprM, and aph genes exhibited increased mRNA transcript levels in vitro (macrophages) and in vivo (GI tract and lung), supporting the conclusion that a functional loss of these gene products and the resultant change from antibiotic-resistant to antibiotic-sensitive phenotypes results in decreased fitness.

Fitness cost of antibiotic susceptibility in other pathogenic bacteria

A. baumannii is another major Gram-negative pathogen, commonly isolated in the ICU, that has numerous intrinsic and acquired antimicrobial resistance factors (11). A recent report (18) used the TnSeq approach to describe the A. baumannii genes necessary for persistence in the murine lung and found that A. baumannii mutants with a Tn insertion in the A1S_1649 and A1S_1801 genes—which are predicted to encode resistance–nodulation–cell division (RND)–type efflux systems—were significantly less fit in mice. The RNDs play a major role in the intrinsic multidrug resistance phenotypes of both A. baumannii and P. aeruginosa (6, 19, 20). To test our hypothesis that intrinsic and acquired antibiotic resistance enhances fitness during infection, in other clinically relevant organisms, we assessed the fitness of Tn insertions in the A1S_1649 and A1S_1801 genes prepared in a highly virulent A. baumannii strain AB5075 (21) using a murine model of lethality after intraperitoneal challenge (22). As expected, mutations in both genes led to increased susceptibility to antibiotics, with the MIC for tobramycin decreasing from 48 mg/liter for AB5075 wild type to 12 and 16 mg/liter for AB5075 with Tn insertions in homologs of A1S_1649 and A1S_1801, respectively. Furthermore, both of the more antibiotic-susceptible AB5075 strains were also less virulent, with significantly fewer mice (P = 0.01, log-rank test) entering a moribund state than when infected with the wild-type parental strain (Fig. 6A).

Fig. 6. Fitness cost of antibiotic susceptibility in A. baumannii and V. cholerae.

(A) Effect of Tn insertions in genes associated with constitutive antibiotic resistance on the virulence of A. baumannii in a murine lethal peritonitis infection setting. WT A. baumannii (AB) strain 5075 and isogenic strains with Tn insertions in genes homologous to A1S_1649 and A1S_1801 were used at a challenge dose of 5 × 109 CFU per mouse, administrated intraperitoneally. A significant decrease in virulence was observed when comparing the WT with either of the two Tn insertion mutants (AB5075-A1S_1649 and AB5075-A1S_A1S_1801; P = 0.01, log-rank test). (B) Effect of Tn insertions in genes tolC and lpp (tolC::Tn and lpp::Tn, respectively) on V. cholerae susceptibility to antibiotic polymyxin. V. cholerae C6706 WT, tolC::Tn, and lpp::Tn were grown in liquid culture (LB) in the presence of polymyxin (0.5 mg/liter) for 5 hours, and 10-fold serial dilutions were plated on agarose plates (LB). (C and D) Infant rabbit competition assays using WT, lpp::Tn, and tolC::Tn mutant V. cholerae strains. The in vivo CIs were determined phenotypically. V. cholerae C6706 WT strain had an insertion of TnFGL3 into lacZ (lacZ::Tn), whereas tolC::Tn and lpp::Tn produced β-galactosidase for differentiation when competing against the parental ΔlacZ WT strain. Inoculum used was 109 CFU per rabbit (C) or 107 CFU per rabbit (D). Significance was determined with the Student’s t test by comparing the colonization ratios of C6706 WT lacZ::Tn versus the tolC::Tn (P < 0.0001) or lpp::Tn (P < 0.01) mutant strains. Mean plus SEM is shown.

Last, we decided to extend our finding that an increase in fitness could be associated with antibiotic resistance by evaluating whether antibiotic resistance might provide an increase in in vivo fitness using V. cholerae, the causative agent of cholera (23). Care for cholera patients is generally focused on treating signs and symptoms, and assuring adequate oral or intravenous hydration of the patients. Antibiotics are not usually a major factor in the treatment of cholera, although their use can decrease the volume and duration of diarrhea by 50% and are recommended for patients with moderate to severe dehydration (23). Further, antibiotic resistance is not considered a major issue for V. cholerae, despite some reports indicating more frequent isolation of antibiotic-resistant strains (2426). On the basis of data from a TnSeq analysis that identified a number of genes required for V. cholerae GI tract colonization (27), we tested for polymyxin sensitivity as well as fitness in an infant rabbit model of cholera, V. cholerae strains, with either a Tn insertion in tolC—a porin homolog of OprM involved in bacterial multidrug resistance and survival of pathogens during infection (28)—or a Tn insertion in lpp—which is predicted to encode the major outer membrane lipoprotein Lpp (29) conferring susceptibility to polymyxin B. As shown in Fig. 6 (B to D), the tolC::Tn mutant was more susceptible to polymyxin B and had a decreased ability to colonize the GI tract of infant rabbits (P < 0.0001), whereas the polymyxin B–resistant lpp::Tn strain was able to significantly outcompete the wild-type V. cholerae C6706 strain in vivo (P < 0.01, as determined by t test).


A number of studies have suggested that phenotypic changes that confer antibiotic resistance are associated with a fitness cost in vivo and a reduction in infectivity (for example, decreased virulence and microbial transmission) in pathogenic bacteria such as E. coli (30), Salmonella enterica serovar Typhimurium (31), P. aeruginosa (32), Mycobacterium tuberculosis (33), and Staphylococcus aureus (34). However, the data reported here show that a fitness cost associated with microbial antibiotic resistance might not be a general feature of all infections in which therapy-associated resistance is encountered. With the use of a variety of clinically relevant bacterial strains (P. aeruginosa, A. baumannii, and V. cholerae) and in vivo models of infection, a consistent picture emerged: acquisition of antibiotic resistance increased in vivo fitness, and intrinsic resistance to different classes of antibiotics also enhanced in vivo virulence and fitness. This conclusion was supported not only by results generated with a Tn library in P. aeruginosa but also by analysis of naturally occurring, antibiotic-resistant strains isolated from infected patients. Together, these results indicate the utility of Tn insertions to accurately identify phenotypes in experimental analyses even though other mutations, which do not result in total loss of a gene product or its function, could plausibly produce less-pronounced or qualitatively different results. Increased fitness of antibiotic-resistant strains was not specific to P. aeruginosa, as there was also a measurable in vivo fitness cost when genes encoding antibiotic resistance were interrupted in both A. baumannii and V. cholerae. Increases in in vivo fitness for antibiotic-resistant strains occurred in mouse and rabbit infection models, thus providing broad-based support for challenging the concept that drug-resistant bacteria are less fit and thus less virulent during infections.

The increased virulence of antibiotic-resistant strains in experimental settings raises a serious concern that drug-resistant strains might be better fit to cause serious, more difficult to treat infections beyond just the issues raised by the complexity of antibiotic treatment. These findings also could have an impact on the use of drugs such as fosfomycin, which is dispensed in combination with other antibiotics and which has been proposed as a means to avoid the emergence of drug-resistant P. aeruginosa (35). Using the same P. aeruginosa strain with the PA14 background, Rodríguez-Rojas et al. (36) reported that there was no fitness cost of the Tn-glpT fosfomycin-resistant mutant strain when tested in a mouse model of pneumonia that used a mouse strain different from the one used here, the C57BL/6J mouse. Although these findings did not demonstrate the marked increased fitness that we measured for the P. aeruginosa Tn-glpT strain, both studies imply that fosfomycin treatments that lead to the emergence of resistant strains will not result in an in vivo fitness cost for bacterial survival and conceivably could lead to increased in vivo virulence, which would be problematic in the ICU setting (37). Overall, infections caused by microbial strains with natural or acquired resistance to drugs such as fosfomycin or carbapenems could have serious clinical consequences beyond the difficulties in choosing an effective therapeutic treatment. It is already established that acquired carbapenem resistance in P. aeruginosa leads to increased patient mortality (38).

The basis for the increased transcription of genes that encode antibiotic resistance determinants in P. aeruginosa in the absence of selective pressures from the drugs might be related to the need for organisms to coexist with the natural microbiota of animals and humans as well as with environmental organisms. In these settings, pathogenic organisms encounter multiple species-level phylotypes that produce antimicrobial compounds (39). A recent report showed that methicillin-resistant S. aureus selected to resist bacteriocins secreted by other microbial strains also became resistant to intermediate levels of vancomycin (40), one of the only antibiotics available to treat multidrug-resistant S. aureus infections. Furthermore, the recent identification of a wide distribution of biosynthetic gene clusters for synthesis of the thiopeptide antibiotic lactocillin within the organisms that make up the human microbiome suggests that microbial interactions in the host play a major role in the selection of general drug resistance factors that are also present in pathogenic bacteria (41). Most of the identified antibiotic biosynthetic clusters present in the normal microbiome (3118 total clusters were found in 2430 reference genomes) still cannot be assigned a biological function. This vast potential for the production of antimicrobials by the microbiota represents a potential strategy that allows human commensal bacteria to outcompete pathogens, with concomitant pressure placed on pathogens to become resistant to natural antimicrobial factors that then confer resistance to modern antimicrobial drugs.

Overall, genes in pathogenic microorganisms that confer intrinsic and acquired antibiotic resistance could provide a survival advantage when in competition with other microbes within a particular niche. Alternatively, gene products that mediate antibiotic resistance might play a role in defense against bactericidal factors produced by the host, including antimicrobial peptides (AMPs). At least several AMPs, including human cathelicidin LL-37, human β-defensins 1, 2, and 3, and CRAMP, are able to kill P. aeruginosa (42, 43). Overexpression of the MexAB-OprM efflux system in P. aeruginosa confers resistance to the synthetic AMP polymyxin B (44), and we found that polymyxin B–resistant V. cholerae also has an increased in vivo fitness compared with the wild-type parental strain. Other factors, such as the host’s diet, could also play a role in maintaining intrinsic antibiotic resistance genes, because foods (45) such as sweet potatoes and coffee contain caffeic acid, a compound that has an antimicrobial activity toward P. aeruginosa (46). In conclusion, the findings that intrinsic and acquired antibiotic resistance genes are associated with increased in vivo fitness of P. aeruginosa, A. baumannii, and V. cholerae in four different experimental infection settings, along with recent reports about the lack of fitness costs associated with antibiotic resistance in S. enterica serovar Typhimurium (47) and E. coli (48), emphasize the necessity to effectively control the emergence of antibiotic-resistant pathogens as well as the development of alternative approaches to prevent and treat infections. Last, our findings point to additional potential consequences, wherein virulent strains of serious microbial pathogens that are both more drug-resistant and more pathogenic may be establishing themselves as the predominant organisms able to infect at-risk humans.


A full description of the experimental methods can be found in Supplementary Materials and Methods.

Study design

The aim of this study was to explore the concept that antibiotic resistance could be associated with increased in vivo fitness and increased virulence during bacterial infections. We did an initial screening by using a next- generation, high-throughput sequencing approach, termed TnSeq, on three different saturated banks of mutants of the pathogenic bacteria: P. aeruginosa, A. baumannii, and V. cholera. The TnSeq was applied in vivo using mouse and rabbit models of bacterial infections. A comprehensive approach was used to manage the millions of sequencing reads generated by the TnSeq: an initial analysis by functional classes, then an analysis by operon, and finally a gene-by-gene analysis. The signals detected by TnSeq in favor of the in vivo association between antibiotic resistance and increased fitness and virulence were confirmed using individual strains of each bacterial species. Individual mutants were tested in vivo to confirm the results of the TnSeq screening. In vitro experiments were designed to remove potential bias, complete the in vivo finding, raise hypotheses, and reveal mechanisms.

Predefined study components: The primary pathogenic agent selected was P. aeruginosa because of the severity of P. aeruginosa infections and the major intrinsic and acquired antibiotic resistance of this species. The second pathogen selected was A. baumannii to confirm the initial results in another antibiotic-resistant pathogenic bacterium. The last pathogen, V. cholerae, was chosen for its medical impact. The main mouse model selected was a pneumonia model because of the clinical relevance of P. aeruginosa lung infections. The antibiotics selected belonged to the main classes used to treat P. aeruginosa infections: β-lactams, aminoglycosides, and fluoroquinolones. The cell line selected was macrophages, to explore the role of the host immune defenses, including alveolar macrophages, to gain a more complete understanding of the interplay of antibiotic resistance and in vivo fitness in a murine model of pneumonia. Mice were housed under specific pathogen–free conditions, and all animal experiments complied with institutional and federal guidelines regarding the use of animals in research.

Statistical methods

Two-sample comparisons used t tests, either unpaired for normally distributed data or the Mann-Whitney test for nonparametric data. Survival was analyzed by log-rank tests. For analysis of TnSeq results, the fold changes under all conditions were determined, and the results for all genes were analyzed for statistically significant differences in their occurrence using the On Proportions function of CLC with corrected P values calculated by the Bonferroni false discovery rate method.


Materials and Methods

Fig. S1. Evolution of the RPKM sequencing reads for Tn insertions in genes in each of 27 functional classes from LB to the lung after 1 hour (A), from 1 to 6 hours in the lung (B), and from 6 to 24 hours in the lung (C).

Fig. S2. In vivo fitness of P. aeruginosa mutants with Tn insertions in genes from the functional class “motility and attachment” after 1, 6, or 24 hours of infection in the lung.

Fig. S3. Comparative in vivo fitness (in the GI tract and lung) of bacterial strains with Tn insertions in genes needed to produce T4aP components.

Fig. S4. Comparative in vitro fitness in LB and water of bacterial strains with Tn insertion mutants in genes needed to produce T4aP components.

Fig. S5. Comparative in vivo fitness of bacterial strains with Tn insertions in genes that encode all of the annotated VFs of P. aeruginosa (after 1, 6, or 24 hours in the lung).

Fig. S6. Evolution over time of the RPKM for Tn-interrupted genes in the LPS O-antigen locus.

Fig. S7. Increased virulence of oprD mutant carbapenem-resistant clinical strains of P. aeruginosa (48.2 and 51.2) in lung infection compared to isogenic (48.1 and 51.1) and complemented strains (48.2::PoprD and 51.2::PoprD).

Fig. S8. Evolution over time of the changes in the RPKM for Tn-interrupted genes associated with constitutive antibiotic resistance in P. aeruginosa in the GI tract, spleen, and lung.

Fig. S9. Analysis of in vitro growth and survival of Tn mutants deficient in genes associated with constitutive antibiotic resistance in P. aeruginosa.

Fig. S10. Effect of deletion of intrinsic antibiotic resistance genes on survival of P. aeruginosa in J774 macrophages.

Table S1. Analysis of selective pressures detected by RPKM reads during P. aeruginosa lung infection.

Table S2. Genes (116) whose loss shows an increased fitness for lung infection based on having Tn insertions with at least 100 reads after 24 hours and with reads further increasing more than twofold between 24 and 48 hours of infection.

Table S3. Antibiotic sensitivity results for wild-type P. aeruginosa PA14 and strains deleted for genes encoding intrinsic antibiotic resistance.

Table S4. Antibiotic sensitivity results for wild-type P. aeruginosa PA14 and strains deleted for genes encoding intrinsic antibiotic resistance (inhibition diameter).

Table S5. Antibiotic sensitivity results for wild-type P. aeruginosa PA14 and strains deleted for genes encoding intrinsic antibiotic resistance (MIC as measured by E test).

Table S6. Primers for genomic amplification used in this study.

Table S7. Bacterial strains used in this study.

Table S8. Plasmids used in this study.

Data file S1. Analysis of the TnSeq data by functional classes of P. aeruginosa PA14.

Data file S2. Fitness of flagellin mutants identified in the study.

References (4959)


  1. Acknowledgments: We thank C. Manoil at the University of Washington for providing the A. baumannii mutant strains. Funding: This work was supported by a grant from the Cystic Fibrosis Foundation (PIER14GO). D.R. received grants from the AXA Research Fund and the Société de Réanimation de Langue Française. T.G. received grants from the Conseil Régional de Champagne-Ardenne, the Philipp Fondation, and the Association pour le Développement de la Microbiologie et de l’Immunologie Rémoises. J.J.M. is supported by grant AI-026289 and O.D. is supported by grant AI-115962, both from the U.S. National Institute of Allergy and Infectious Diseases. D.S. was supported by a grant from the Seedlings Foundation. Author contributions: D.R. performed and was the leader of the in vivo experiments, analyzed data, worked on the manuscript, and contributed to the study concept. O.D. performed and was the leader of the in vitro experiments, analyzed data, and worked on the manuscript. T.G. and V.C. performed P. aeruginosa experiments. Y.F. performed V. cholerae experiments. J.M. performed P. aeruginosa experiments. T.G., V.C., Y.F., and J.J.M. analyzed data and edited the manuscript. H.A., F.A., and J.-D.R. analyzed data and performed statistical analyses. Seedlings Foundation and S.L. contributed to TnSeq and in vitro study design, discussion, and the manuscript. G.B.P supervised the project, developed the study concept, and edited the manuscript. D.S. supervised the project, performed experiments, analyzed data, wrote the manuscript, and contributed to the study concept. Competing interests: The authors declare that they have no competing interests.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article