Strong Environment-Genotype Interactions Determine the Fitness Costs of Antibiotic Resistance In Vitro and in an Insect Model of Infection

The acquisition of antibiotic resistance commonly imposes fitness costs, a reduction in the fitness of bacteria in the absence of drugs. These costs have been quantified primarily using in vitro experiments and a small number of in vivo studies in mice, and it is commonly assumed that these diverse methods are consistent. Here, we used an insect model of infection to compare the fitness costs of antibiotic resistance in vivo to those in vitro.

ABSTRACT

The acquisition of antibiotic resistance commonly imposes fitness costs, a reduction in the fitness of bacteria in the absence of drugs. These costs have been quantified primarily using in vitro experiments and a small number of in vivo studies in mice, and it is commonly assumed that these diverse methods are consistent. Here, we used an insect model of infection to compare the fitness costs of antibiotic resistance in vivo to those in vitro. Experiments explored diverse mechanisms of resistance in a Gram-positive pathogen, Bacillus thuringiensis, and a Gram-negative intestinal symbiont, Enterobacter cloacae. Rifampin resistance in B. thuringiensis showed fitness costs that were typically elevated in vivo, although these were modulated by genotype-environment interactions. In contrast, resistance to cefotaxime via derepression of AmpC β-lactamase in E. cloacae resulted in no detectable costs in vivo or in vitro, while spontaneous resistance to nalidixic acid, and carriage of the IncP plasmid RP4, imposed costs that increased in vivo. Overall, fitness costs in vitro were a poor predictor of fitness costs in vivo because of strong genotype-environment interactions throughout this study. Insect infections provide a cheap and accessible means of assessing the fitness consequences of resistance mutations, data that are important for understanding the evolution and spread of resistance. This study emphasizes that the fitness costs imposed by particular mutations or different modes of resistance are extremely variable and that only a subset of these mutations is likely to be prevalent outside the laboratory.

INTRODUCTION

Antibiotics target enzymes with essential cell functions, such as RNA polymerase, DNA gyrase, and the 30S or 50S subunit of the ribosome, while resistance mutations in binding sites can impair essential biochemical functions such as transcription (1, 2). Though mutations in drug targets confer a fitness advantage in the presence of an antibiotic over some range of doses, they also typically reduce fitness in the absence of antibiotics (3–5). This detrimental consequence of resistance is commonly described as a fitness cost (3–5). While target site mutations are obvious candidates for imposing high fitness costs, the acquisition of plasmids bearing antibiotic resistance genes can also reduce fitness (6–8). Moreover, while the individual cost of each additional gene coding for resistance can be small (4, 9, 10), the cumulative effect of increasing numbers of resistance genes is associated with impaired growth (11).

A greater understanding of the evolutionary biology of resistance, including fitness costs, can inform efforts to control the spread of resistance (3, 12–14). A range of approaches to antimicrobial stewardship increase the heterogeneity of selection pressure by prescribing different drugs to different parts of the population (“mixing”), or by using preferred drugs for short periods of time (“cycling”). With cycling strategies in particular, the fitness costs of resistance (12, 15) and patient turnover (16) are expected to be the main factors driving down the frequency of resistance when drugs are withdrawn. Modeling studies have argued that there are insufficient data to estimate some of the key parameters shaping resistance evolution (16, 17) or to predict which antibiotic stewardship interventions would be most effective (18).

Importantly, antibiotic resistance mechanisms are diverse, and not all will necessarily impose a cost: there will be a distribution of fitness effects (5). Most commonly, fitness costs are assessed in vitro, and the results are extrapolated to costs in vivo (4). While some have argued that fitness costs are broadly similar in vitro and in vivo (10), this is not universally accepted (4). Some resistance mutations in a range of microbes show enhanced or altered costs in vivo (4, 19–21). In addition, resistance mutations seen in clinical contexts often represent a low-cost subset of those detectable in vitro (20, 22, 23), the implication being that selection based on fitness costs might be particularly efficient in vivo. Arguably, in vivo conditions are likely to be harsher than in vitro conditions, and environmental stress can increase fitness costs in a wide range of organisms (24–26). Although this pattern is by no means universal (1), environmental conditions could substantially affect antibiotic resistance fitness costs (1, 4, 25), and there are still limited comparative data on costs in vivo and in vitro (10).

Previous experimental studies of in vivo costs have been almost entirely restricted to mouse models (10). In this study, we wanted, first, to explore the practicality of a cheaper, higher-throughput insect model. Second, previous in vivo studies have focused on spontaneous target site modifications in housekeeping genes, where pleiotropic fitness costs are likely to be high (4, 10). Here, we compared spontaneous target site resistance with metabolic resistance based on constitutive upregulation of β-lactamases. Finally, we incorporated a measure of the cost of carriage of a typical antibiotic resistance plasmid, because there are limited data on the costs of plasmid carriage in vivo (10).

To examine the difference in antibiotic resistance fitness costs in vitro and in vivo, this study used an insect model of infection and two different microbes: a Gram-positive pathogen, Bacillus thuringiensis, and a Gram-negative intestinal symbiont, Enterobacter cloacae. The greater wax moth (Galleria mellonella) is a commonly used insect model for bacterial infections, partly because it can be reared at 37°C (27). Insects also have well-developed innate immune systems that share several features with those of vertebrates (27, 28). However, wax moth larvae have weaker immunity than many insects, with reduced expression of antimicrobial peptides (29) and weak cellular immunity to macroparasites, such as nematodes (30, 31). Their status as permissive hosts for infection generally means that they are widely used as traps for the isolation of new insect parasites (32, 33).

Weak immunity may make the wax moth a questionable host for assessing in vitro fitness costs of antibiotics and studying antibiotic resistance. There is abundant circumstantial evidence suggesting that immunity is important for the fitness of resistant microbes. For instance, repressed immunity may facilitate the persistence of high-cost resistance mechanisms (34, 35), and active immunity is thought to be important for shaping the fitness of drug resistance mutants in HIV (36). Since interactions with the immune system may be an important factor for fitness costs, this study used larvae (caterpillars) of an alternative host, the diamondback moth (Plutella xylostella), which is a well-established infection model for both B. thuringiensis and E. cloacae (37–40). While this insect cannot be reared at 37°C, it has high fecundity and a short generation time (ca. 14 days at 25°C) and can be reared gnotobiotically on an artificial diet in petri dishes (41), making it a valuable alternative model host.

RESULTS

Characterization of mutants.

Sequencing of the rifampin-binding pocket of rpoB in the B. thuringiensis mutants gave 37 sequences that could be aligned with the wild-type (WT) template. This region is 216 bp long, starting at base 1054 in the template rpoB gene. All the mutants contained one of five single nucleotide polymorphisms (SNPs), each a single-base-pair substitution that resulted in a single amino acid change (Table 1). Amino acid alignments for these mutants are presented in Fig. S1 in the supplemental material. The Ser-to-Tyr mutants (genotype 9a) could not be reliably cultured and thus were excluded from fitness experiments; one mutant from each of the remaining genotypes was used in subsequent experiments. All mutants had MICs at least 100 times greater than that of the wild type (Table 1).

TABLE 1.

Characterization of spontaneous rifampin-resistant colonies of B. thuringiensis with mutations in rpoB

Mutant IDa	MIC (μg/ml)b	Position of mutation (bp)	Amino acid change	No. of mutants isolated
11c	100	18	Glu → Lys	1
4b	200	18	Glu → Arg	5
2d	100	31	His → Asp	10
10a	100	36	Ser → Phe	17
9a	100	36	Ser → Tyr	4

^aID, identification. All genotypes except 9a were used in competition experiments.

^bThe MIC for wild-type B. thuringiensis was 0.2 μg/ml.

After 48 h of incubation on cefotaxime-supplemented agar (4 μg/ml), two distinct phenotypes of resistant E. cloacae mutants were observed: satellite-colony-forming and non-satellite-colony-forming mutants (Fig. S2). Satellite colonies are susceptible cells that are able to grow in the zone around a resistant mutant colony on agar plates containing 4 μg/ml cefotaxime. Satellite colonies are phenotypically susceptible and unable to grow on cefotaxime in the absence of a mutant colony. Satellite colonies are produced when β-lactamases detoxify solid media and facilitate the growth of susceptible cells coming out of persister states (6).

Eight mutants of each resistant phenotype were sequenced at two potential target genes: ampD and ampR. Sequences were assembled and analyzed using a wild-type positive control, the annotated reference genome of E. cloacae KU6334 (GenBank accession number AY789446) (42), and a wild-type ancestor of experimental strains, E. cloacae isolate jjbc. No genetic changes were observed in sequences of the ampD gene. In contrast, sequence variation was found in the ampC transcriptional activator gene ampR, resulting a range of genetic variants, three of which were used in experiments (Table 2). Two genotypes produced satellite colonies (cef D-A and cef D-Y); the third (cef ins) did not. Genotypes cef D-A and cef D-Y had nonsynonymous single-base-pair mutations on the same codon, at positions 410 and 409, respectively. Fitness experiments were conducted with one mutant for each of these genotypes. Two additional mutants with substitutions affecting the same residue, resulting in a change to glycine or valine, were isolated but not used further. All three experimental genotypes had MICs at least 100-fold greater than that for the wild type, although the cef ins genotype, which did not produce satellite colonies, had the lowest MIC of all the strains used (Table 2).

TABLE 2.

Characterization of spontaneous cefotaxime-resistant colonies of E. cloacae with mutations in the transcriptional regulator gene ampR

Mutant IDa	MIC (μg/ml)b	Mutation	Position of mutation (bp)	Amino acid change	No. of mutants isolated
cef D-A*	256	A to C	410	Asp → Ala	2
cef D-Y*	256	G to T	409	Asp → Tyr	2
cef ins	64	25-bp insertion	Starts at 370	+ Asp, Leu, Gln, Leu, Ser, Thr, His, Asn	1

^aID, identification. Mutants marked with asterisks produced satellite colonies on LB agar plates containing 4 μg/ml cefotaxime.

^bThe MIC for wild-type E. cloacae was 0.25 μg/ml.

Nalidixic acid-resistant mutants of E. cloacae, capable of growth at 30 μg/ml, could be readily produced. However, all clones characterized had wild-type sequences in both gyrA and parC. MIC values (<0.1 mg/ml) were modest compared to those for fluoroquinolone-resistant clinical isolates of E. cloacae, which are commonly >100 mg/ml nalidixic acid (43, 44). We selected a single mutant (11.1B) for fitness experiments to act as a comparison for the cefotaxime-resistant mutants; this genotype was also the host for the RP4 transconjugants.

Relative fitness.

For the rifampin-resistant mutants of B. thuringiensis, we observed strong environment-genotype interaction effects on competitive fitness in the absence of antibiotics (glm F test with degrees of freedom: F_3,206 = 26.0 [P < 0.0001]) (Fig. 1) and strong differences between genotypes (F_3,201 = 59.6 [P < 0.0001]). As hypothesized, we saw lower mutant fitness overall in larvae and, therefore, enhanced fitness costs for the in vivo infections (F_1,209 = 25.9 [P < 0.0001]) (Fig. 1). In broth, B. thuringiensis typically underwent 11.9 doublings (standard error [SE], 0.35), and our estimate of the typical number of generations in vivo was 13.0 (SE, 0.08), although this assumes an initial infection bottleneck of 50 cells.

FIG 1.

Relative fitness of four rifampin-resistant mutants of B. thuringiensis in pairwise competition assays against the WT ancestor in vitro (in LB) and in oral in vivo infections of larvae of the diamondback moth, P. xylostella. Box plots show medians and interquartile distances; data points show results of independent infections of insects or replicated broth culture. Fitness was calculated according to the change in the proportions of resistant and wild-type genotypes from the initial inoculum to the end of culture/infection.

In E. cloacae, the fitness cost imposed by resistance depended on the particular resistance mechanism (F_4,139 = 72.4 [P < 0.0001]) (Fig. 2A). As hypothesized, the fitness costs of resistance could be enhanced in vivo, but this was true only for nalidixic acid resistance and carriage of the RP4 plasmid (F_4,135, 41.1 for genotype-treatment interaction [P < 0.000]) (Fig. 2A). Carriage of RP4 was the only resistance trait that imposed a detectable cost in vitro (Fig. 2A). The enhanced fitness cost of RP4 in vivo was not as clear in our supplementary analysis on the estimates of Malthusian parameters (Fig. S3). Note that both these analyses exclude nine replicates in which relative fitness could not be calculated because plasmid-carrying cells could not be detected in live hosts although these replicates support a picture of poor fitness in the larval environment. In the E. cloacae experiments, there was greater disparity in the number of generations between different environments, with 11.1 generations in broth (SE, 0.05) and approximately 17.5 generations in larvae (SE, 0.17).

FIG 2.

(A) Relative fitness of diverse antibiotic-resistant mutants of E. cloacae in pairwise competition assays against the WT ancestor in vitro (in LB) and in oral in vivo infections of larvae of the diamondback moth, P. xylostella. Fitness was calculated according to the change in the proportions of resistant and wild-type genotypes from the initial inoculum to the end of culture/infection. Cefotaxime-resistant genotypes (with the prefix cef) are listed in Table 2; nal and RP4 refer to a spontaneous nalidixic acid-resistant mutant and a carrier of the IncP plasmid RP4, respectively. (B) Total population sizes of cefotaxime-resistant mutants and wild-type E. cloacae in intestinal infections of live larvae. Zero counts reflect infections in which single genotypes were recovered from larvae. Box plots show medians and interquartile distances; data points show results of independent infections of insects or replicated broth culture.

The three cefotaxime-resistant genotypes of E. cloacae behaved similarly, and for these three genotypes, there no was no evidence that environment affected fitness (F_1,93, 1.21 for treatment-genotype interaction [P = 0.3]) (Fig. 2A), nor was there a significant impact of genotype (F_1,95 = 1.17 [P = 0.31]). There was some evidence that the competitive fitness of resistant mutants in larvae was actually slightly greater than that of the wild type (test for difference in fitness > 0 [mean, 0.186; SE, 0.08; t = 2.306; P = 0.0232]); for the cef D-Y mutant in particular, the mean fitness had confidence intervals (0.55, 0.15) that did not overlap with zero, indicating a fitness benefit for this mutation.

Using data from both E. cloacae and B. thuringiensis, we also assessed whether fitness costs in broth culture were correlated with costs in vivo, as investigated previously (10). We found that fitness costs in vitro were a poor predictor of costs in vivo, based on a simple linear model of mean log-transformed fitness values (R-squared [Rsq] = 0.035; F_1,7 = 0.25 [P = 0.63]) (Fig. 3).

FIG 3.

Competitive fitness in vitro was a poor predictor of the fitness costs of resistance in vivo during infections of live larvae. Data are mean log₁₀-transformed fitness values for the different resistance genotypes of both E. cloacae and B. thuringiensis in this study. The line represents the fitted linear model with a 95% confidence interval.

While growth rates or relative fitness in direct competition are the most commonly reported measures of fitness, there are other fitness components that likely impact pathogen transmission, such as the total population size of infectious propagules within a host, sometimes termed yield (4). For the three cefotaxime-resistant mutants, we had sufficient data to conduct a yield analysis. An important biological factor for naturalistic in vivo infection is the existence of a considerable bottleneck during establishment (45). The infection bottleneck in E. cloacae meant that 50 larvae carried strains consisting of single genotypes (either the wild type or a resistant mutant). With these infections, competitive fitness cannot be calculated, and yield offers an important complementary fitness parameter. The high number of zero yield counts for one partner or another (Fig. 2B) meant that the data were not normally distributed. However, a nonparametric analysis indicated that resistant mutants reached higher population sizes than wild-type bacteria within larvae (Wilcoxon rank sum test statistic = 1,783 [P < 0.0001]) (Fig. 2B). The slight excess of infections in which the wild-type strain failed to establish (37 wild-type infections out of the 50 single-genotype infections) contributed to this pattern (Fig. 2B). Again, there was no good evidence for differences in yield between infections with the three cefotaxime-resistant mutants as determined by Kruskal one way analysis of variance (ANOVA) (χ = 3.2; df = 2; P = 0.2).

DISCUSSION

The fitness costs of antibiotic resistance are well established. However, the literature directly comparing such costs in vitro and in vivo is very limited (10), probably because conducting in vivo studies in mice is both costly and laborious. As a result, tractable model systems present a valuable opportunity to close this knowledge gap and potentially inform antibiotic stewardship. We investigated the effects of genotype-environment interactions on the fitness costs of rifampin resistance in B. thuringiensis because of the clinical importance of rifampin resistance in other Gram-positive organisms (23), but also because rifampin has been used to develop a well-established model of the genetics of resistance (1, 4, 5, 46, 47). Comparisons of data from the insect model to those from prior studies can give us some idea of the value of this system. Encouragingly, we found that distinct mutations in the rifampin binding pocket produced a distribution of fitness costs, as in prior studies, although the range of mutants isolated is lower than that seen for Pseudomonas aeruginosa or UV-irradiated Bacillus anthracis (5, 48). Here, we were also able to demonstrate that fitness costs were consistently higher in vivo, although this effect varied with genotype. This result is also consistent with the apparent efficiency of selection in limiting the diversity of mutations seen in clinical studies (23).

Intestinal infection of moth larvae with E. cloacae also constituted an experimentally tractable model for the spontaneous evolution of resistance to third-generation cephalosporins. For instance, it was possible to readily isolate a number of mutants with independent mutations that are consistent with the constitutive upregulation of AmpC β-lactamases, as seen in clinical Enterobacteriaceae (42, 49, 50). In contrast to all other resistance mechanisms in this study, AmpC upregulation imposed undetectable fitness costs in medium. This result is consistent with previous in vitro work on the low costs of plasmid-encoded extended-spectrum β-lactamases (9) and AmpC β-lactamases (51). Comparative studies suggest that some classes of β-lactamases impose costs, but the dominant mechanisms found in the clinic are those that impose the lowest fitness costs (52, 53).

In this study, we found some evidence that AmpC upregulation conferred a slight fitness benefit in vivo, both in terms of competitive fitness and in terms of population size. This is difficult to explain and should be treated with caution, although a range of resistance mutations can provide context-dependent fitness benefits (54). The methods of calculating relative fitness used here are not without bias or limitations, particularly since wild-type, susceptible bacteria lack independent markers. Nevertheless, similar biases should pertain to in vivo and in vitro treatments, and the two treatments used identical methods for scoring genotypes, as well as common inoculation methods. The production of β-lactams by resident microbes is possible but unlikely, given that insects are reared under near-aseptic conditions (41). Finally, β-lactam production by arthropods has been reported but is thought to be extremely rare and must be considered a very remote possibility (55). Of speculative interest is the fact that the insect pathogen B. thuringiensis carries multiple β-lactamase genes (56), while its close relative, B. anthracis, is typically β-lactam sensitive and is a mammalian pathogen (57), suggesting that β-lactamases may have some fitness benefits in insect infections.

The insect infection model yielded valuable fitness cost data for both Gram-negative and Gram-positive bacteria. Not all drug classes or resistance modes were equally amenable to study. We believed it would be easy to isolate fluoroquinolone resistance-conferring mutations in gyrA and parC, but these proved elusive and may be much rarer than we anticipated, or not easily recoverable in this genetic background. Alternative sites for fluoroquinolone resistance are efflux pumps and marR, which are known to be important sources of mutations in other Enterobacteriaceae (58, 59), and these could account for the resistance phenotypes observed here. In addition, our isolate of insect-associated E. cloacae appears to be poorly adapted for the carriage of a classic IncP plasmid: the plasmid appears to be nonconjugative in this background (60), and this study shows that it imposes high fitness costs. IncP plasmids can conjugate into a broad range of bacteria (61). Nevertheless, this does not mean that these plasmids are able to persist and be transmitted in all recipient genetic backgrounds (62), a fact that can explain the restricted distribution of plasmids in the field (63).

Overall, there are several possible mechanisms that could account for elevated fitness costs in vivo. Since growth in a host requires the expression of a range of virulence factors not required in media, one hypothesis is that resistance mechanisms affecting gene expression impose greater costs in vivo (19). The fact that the costs of plasmid carriage commonly manifest through disrupted gene expression (64, 65) is one possible explanation for the elevated costs of plasmid carriage in vivo that merits further study. In addition to the fundamental differences in nutrient supply between in vitro and in vivo conditions that can alter fitness costs (54), host immunity may play an important role in altering or increasing fitness costs. For instance, interactions with macrophages can shape the environment-dependent fitness costs of rifampin resistance (66), while all growth rate-based fitness costs could confer increased susceptibility to immune cells. Moreover, a range of spontaneous resistance mutations can confer increased sensitivity to antimicrobial peptides, especially for mutations that affect the makeup of the cell wall (67, 68). One possibility that is unlikely to be important here is the presence of competing microbial species, since our experiments used gnotobiotic insects; other studies have shown that a community context can be important for shaping the evolution of resistance (69).

One consequence of the experimental techniques used here is that naturalistic oral infections lead to strong infection bottlenecks, which will increase the variation in measures of competitive fitness and produce a proportion of single-genotype or clonal infections (45). This can be thought of as either a methodological benefit or a drawback. Bottlenecks are commonly produced during the process of infection, and thus, hosts are frequently colonized by single genotypes (70–72). With a single genotype per host, selection on competitive fitness will be weak, since within-host competition occurs infrequently, so parameter estimates of competitive fitness may not reflect the transmission potential of resistant mutants. Instead, competition between hosts and selection on traits such as population size and efficiency of transmission will be more critical (73). In this case, experiments that incorporate the establishment of infection and can measure population size are valuable. If, on the other hand, it is important to measure competitive fitness precisely, experiments with insects can use injection rather than oral feeding to overcome bottlenecks (38) or can simply increase sample sizes to cope with noise (74).

We have argued previously that the fitness costs of resistance are not necessarily a reliable basis for antimicrobial stewardship (13). This is partly because compensatory mutations that reduce fitness can occur quickly and readily (64, 65). This study emphasizes that the fitness burdens of different mechanisms or mutations may be magnified in vivo but also that fitness costs differ considerably with genotype. This does not necessarily improve our prospects for exploiting fitness costs. Instead, one consequence is that elevated costs in vivo shape mutation supply by determining which mutations or resistance mechanisms are able to persist effectively and establish themselves in clinical contexts (14, 21, 23, 52). Similarly, while the horizontal mobility of resistance plasmids is often emphasized, clinically resistant microbes are often dominated by the clonal expansion of successful lineages with effective compensation in place. The success of the Escherichia coli lineage ST131, an unholy alliance of plasmid-borne β-lactamases, chromosomal fluoroquinolone resistance, and plasmid compensation, is a well-known example (75). If high-cost resistance is transient or exceptional (14), then it will be a poor basis for antibiotic resistance management. Antibiotic stewardship may be better served by studying in detail the fitness of resistance lineages in specific targets (12) than by assuming generally high fitness costs.

MATERIALS AND METHODS

Bacterial isolates.

Bacillus thuringiensis serovar kurstaki 7.1.o is a recently isolated wild-type (WT) strain that is highly pathogenic to many Lepidoptera (41). This isolate has sequence type 8, placing it within an abundant and successful clonal expansion in this species (76). Enterobacter cloacae jjbc was recently isolated from the midgut of diamondback moth (Plutella xylostella) larvae and forms a persistent symbiotic association with this host (38).

Independent resistant mutants were produced via a modified fluctuation test based on reference 5. In brief, overnight cultures were diluted into lysogeny broth (LB) in 24-well plates such that each well contained 1 ml of broth inoculated with approximately 10² cells. Plates were cultured overnight at 30°C and the resulting cultures centrifuged (5,000 × g, 8 min) and resuspended in 200 μl of saline (0.85% NaCl). Resuspended cells were spread over LB agar plates containing either 10 μg/ml rifampin (B. thuringiensis only), 4 μg/ml cefotaxime, or 30 μg/ml nalidixic acid. Colonies with strong growth were restreaked at least twice on a medium containing antibiotics before sequencing and storage of glycerol stocks. RP4 transconjugants of E. cloacae were produced by costreaking Escherichia coli MG1655 carrying the IncP plasmid RP4 (provided by Tatiana Dimitriu) with a nalidixic acid-resistant mutant (11.1B [Nal^r]) of E. cloacae on agar plates. Transconjugants were identified by replating cells on nalidixic acid and tetracycline. The RP4 plasmid is essentially nonconjugative in this system (60). For experiments, antibiotic-resistant mutants were cultivated by streaking wild-type ancestors from glycerol stocks onto selective antibiotic plates, followed by overnight growth in LB at 30°C.

Characterization of antibiotic-resistant mutants.

MICs for wild-type and resistant colonies were determined by broth dilution methods in microtiter plates using a 2-fold dilution series of antibacterial agents, according to EUCAST recommendations (77). MICs were defined as the lowest concentrations of drug that completely inhibited visible growth of the inoculum after incubation for 18 h at 30°C.

PCR and sequencing.

B. thuringiensis clones were sequenced to identify mutations in rpoB. The primers used to amplify the rpoB gene in B. thuringiensis were RIF-F1 (5′-CGTGAGAGAATGTCGATCC) and RIF-R1 (5′-CGCGAACGAAGATAATGA) for the cluster I region β-subunit in RNA polymerase (48). The ampD and ampR genes of cefotaxime-resistant E. cloacae clones were targeted for PCR using primers obtained from M. Hilty et al. (78) or designed in-house: for ampR, EcAmpR_74F (5′-TGTGCCTGACAAACGGTTAA-3′) and EcAmpR_1112R (5′-AGCGGTAAAGGGGTCTTCTA-3′); for ampD, ampD_F (5′-TATTAATACGTTCCAGAAGC-3′) and ampD_R (5′-CATGGTAAACAACGTCATGT-3′).

For the nalidixic acid-resistant mutants of E. cloacae, we amplified gene fragments at two loci typically associated with fluoroquinolone resistance in Enterobacteriaceae: the gyrase A gene and the gene encoding the ParC subunit of topoisomerase IV (parC) (43, 44). PCRs used the following primers: EcGyrA_1734F (5′-CGCATACCGTCTTTGTCAGA-3′), EcGryA_2672 (5′-TGCGAGAGAAATTACACCGG-3′), ParC_970F (5′-CAGAATCGCCTGAAGCTGAT-3′), and ParC_2088R (5′-GCCAAGTTCAAGAAATCCG-3′).

All PCRs used 25-μl reaction mixtures with 0.4 μl of 0.2 mM deoxynucleoside triphosphates (dNTPs), 0.4 μl of each primer required at 25 nM, 0.1 μl of 0.05 U AmpliTaq DNA polymerase, and 1.5 μl of the template DNA (overnight culture mixed with an equal volume of Tris-HCl [pH 8.0], boiled for 10 min). PCR conditions for rpoB amplification used initial denaturation at 95°C for 5 min, followed by 35 cycles of denaturation at 95°C for 30 s, annealing at 52°C for 30 s, and extension at 72°C for 30 s, with a final extension at 72°C for 7 min. Amplification of ampR used similar conditions, but with an annealing temperature of 54°C, while all other PCRs (ampD, gyrA, parC) used an annealing temperature of 56°C. PCRs were subject to ExoSAP cleanups using 0.2 μl Exonuclease I (20,000 U/ml; New England Biolabs [NEB]), 0.1 μl Antarctic phosphatase (NEB), and 1.5 μl ExoI buffer per 10 μl of PCR product. Products were incubated in a thermocycler for 30 min at 37°C and deactivated for 15 min at 80°C before being subjected to Sanger sequencing in both directions by Eurofins.

Competition experiments.

All mutants were competed against the wild-type ancestor in vitro and in vivo, with the exception of the RP4 transconjugant, which was competed against the relevant plasmid-null strain, the nalidixic acid-resistant E. cloacae mutant 11.1B. All comparisons of fitness in vitro and in vivo were assessed using a common inoculum, a master mix of wild-type and resistant bacteria. B. thuringiensis inocula used spores, while the E. cloacae experiment used stationary-phase cells. Each inoculum was plated onto selective LB agar containing antibiotics and onto antibiotic-free LB agar and was incubated overnight at 30°C to establish initial and final ratios of resistant bacteria. Spores of B. thuringiensis were produced, pasteurized, and enumerated via CFU counts as described previously (39). Stocks of spores for each genotype were prepared at 500 CFU/μl. Inoculants for in vitro treatments were prepared by mixing stocks in a 1:1 ratio with the wild-type ancestors, and 10 μl of each inoculant was added to 1 ml LB in 24-well plates. Mixtures of wild-type and resistant E. cloacae mutants were prepared from overnight cultures in LB (at 30°C with shaking at 180 rpm). Cultures were diluted 10-fold in sterile saline (0.85% NaCl) and, after adjustment for differences in optical density at 600 nm, were mixed with their ancestors in a 1:1 ratio. E. cloacae cultures were also inoculated with 10 μl of each mixture; the final dilutions were equivalent to a 1,000-fold dilution of overnight cultures. All experiments used a minimum of six replicates; each culture plate used six wells with uninoculated broth as checks for contamination; and in vitro competitions were run for 18 h at 30°C and were repeated twice.

For in vivo competition treatments, P. xylostella larvae, from the population VLSS, were raised under gnotobiotic conditions on a sterile diet as described previously (79). Ten third-instar larvae were inoculated in 50-mm petri dishes using 100 μl of each inoculant; all inoculation took place at 25°C. Each inoculum was added to the surface of a quarter of sterile caterpillar diet and was allowed to dry in a class 2 safety cabinet. B. thuringiensis stocks were used at a total dose of 1,000 CFU/μl, as described above; this dose confers approximately 100% mortality with this pathogen (39). E. cloacae mixtures were diluted a further 100-fold to give a final inoculum concentration equivalent to a 1,000-fold dilution of overnight cultures. Previous work has shown that this dose of E. cloacae ensures effective colonization of the midgut but minimizes pathogenic effects (38).

Cadavers killed by B. thuringiensis were recovered after 3 to 4 days, and final ratios of resistant and wild-type spores were assessed in homogenized cadavers as described previously (74). In brief, cadavers were incubated at 30°C for 5 days to ensure complete sporulation before being pasteurized, homogenized in a bead beater (Qiagen TissueLyser II), and dilution plated. Inoculation methods were adjusted for the gut symbiont E. cloacae. Approximately 50 eggs (just prior to hatching) were placed in petri dishes containing the inoculated diet; emergent larvae were allowed to feed for 48 h before being transferred to fresh, uninoculated diet for another 48 to 72 h and were then homogenized for bacterial enumeration. This procedure ensures that experimental larvae have persistent gut infections, not simply the initial inocula, and are approximately the same size as B. thuringiensis-killed insects (early third instars). Each master mix was inoculated into at least 32 larvae, although not all insects yielded data suitable for analysis of competitive fitness.

In vitro and in vivo fitness.

The proportions of each strain and the ancestral competitor were calculated using total colonies on LB agar and resistant colonies on selective agar plates. If the number of colonies on antibiotic plates was equal to, or exceeded, the LB agar counts, the proportion of wild-type bacteria was inferred to be zero. Initial and final proportions on both LB and antibiotic-supplemented LB plates were used to calculate relative fitness (V) according to the equation V = x₂(1 – x₁)/x₁(1 – x₂), where x₁ is the initial proportion of resistant mutants in the inoculum and x₂ is their final proportion (80). Fitness analyses were also conducted using the ratio of the Malthusian parameters, W. This method, however, relies on indirect estimation of the initial infection bottleneck in insects, while in vitro initial densities are calculated directly. We preferred to use a parameter that was less subject to this potential bias, although analyses using W produced qualitatively similar results (Fig. S3).

Data availability.

The BioProject accession number for the genome of the wild-type B. thuringiensis 7.1.o is PRJNA395643 (76). A draft genome of E. cloacae jjbc is available through the JGI Genome Portal, with the reference tag Anc C2. Raw experimental data for this project are publicly available from the University of Exeter institutional repository (https://doi.org/10.24378/exe.2503).