Choice of β-Lactam Resistance Pathway Depends Critically on Initial Antibiotic Concentration

ABSTRACT

Antibiotic resistance trajectories with different final resistance may critically depend on the first mutation, due to epistatic interactions. Here, we study the effect of mutation bias and the concentration-dependent effects on fitness of two clinically important mutations in TEM-1 β-lactamase in initiating alternative trajectories to cefotaxime resistance. We show that at low cefotaxime concentrations, the R164S mutation (a mutation of arginine to serine at position 164), which confers relatively low resistance, is competitively superior to the G238S mutation, conferring higher resistance, thus highlighting a critical influence of antibiotic concentration on long-term resistance evolution.

INTRODUCTION

Alternative trajectories of antibiotic resistance may lead to significantly different resistance levels (1–5). It is therefore informative to understand the factors that affect these alternative choices. One potentially important factor is the antibiotic concentration, since mutations affecting different resistance mechanisms vary in terms of resistance levels and fitness costs and, hence, may provide selective benefits at different concentrations (4–7). Another fundamental factor influencing the choice of alternative resistance trajectories is the population size. In sufficiently small populations, mutations are being fixed on a first-come-first-served basis and mutations with a high rate and those providing large fitness benefits have similar impacts (8, 9). In contrast, in larger populations, increased mutation supplies and longer fixation times provide disproportional benefits to mutations with large fitness benefits even if they occur at lower rates (10, 11).

The emergence and spread of extended-spectrum β-lactamases (ESBLs) pose a major risk to our health care system, as they can cause resistance to clinically important β-lactam antibiotics. Many ESBLs are derived from the narrow-spectrum β-lactamase TEM-1, which efficiently hydrolyses penicillins and early cephalosporins. However, broad resistance toward more modern β-lactam antibiotics can be achieved by many different substitutions (12). Mutations G238S (a mutation of glycine to serine at position 238) (TEM-19) and R164S (TEM-12) are among the most commonly observed mutations of TEM-type ESBLs in clinical and laboratory cefotaxime (CTX)-resistant strains (1, 12, 13), with G238S having a 4-fold-larger effect on the cefotaxime MIC than R164S (Table 1). Due to strong sign epistatic interactions between the two mutations, they represent alternative first-step mutations of distinct adaptive pathways leading to different cefotaxime resistance levels (1, 14).

TABLE 1.

Absolute frequencies of R164S and G238S mutations in TEM-1 in small and large populations

Mutation	No. of populations of indicated size in which mutation occurreda		Median fold increase in MICb	Relative growth rate without CTX (mean ± SD)c
	Small (n = 72)	Large (n = 24)
G238S (G→A)	0	13	64	0.90 ± 0.03
R164S (C→A)	3	0	16	0.98 ± 0.05

^aAs identified by Schenk et al. (15).

^bFold increases in MIC in populations expressing R164S and G238S relative to the MIC of populations expressing TEM-1 (median of three replicates).

^cGrowth compared to that of populations expressing TEM-1 was measured in LB medium without CTX (mean value ± standard deviation from 16 replicates).

In a recent study, Schenk et al. (15) compared the adaptation of a strain of Escherichia coli expressing TEM-1 β-lactamase to gradually increasing concentrations of cefotaxime in many parallel small (effective population size, ∼2 × 10⁶) and 100-fold-larger populations. Sequencing randomly evolved clones from each population revealed a remarkably divergent pattern relating to the mutations identified in this β-lactamase: whereas G238S occurred in 13 of the 24 large and none of the 72 small populations, R164S occurred only in three of the small and none of the large populations (Fisher’s P = 0.0018) (Table 1). Here, we examine two alternative hypotheses to investigate why R164S and G238S were preferentially selected in small and large populations, respectively, in Schenk et al. (15).

The first hypothesis posits that the small-effect substitution R164S has a higher mutation rate than the larger-effect substitution G238S, which outcompetes R164S in large populations where both mutations co-occur (Fig. 1A). To assess this hypothesis, we considered the relative mutation rates of the underlying base pair substitutions of R164S and G238S. All G238S substitutions were caused by the same G-to-A transition in the first position of the codon translating for G238 (GGT to AGT). The R164S substitutions were exclusively caused by a C-to-A transversion in codon position one (CGT to AGT). Mutation accumulation experiments show that G-to-A transitions occur on average at 2- to 4-fold-higher rates than C-to-A transversions in E. coli (16–19), including in the strain used by Schenk et al. (15), providing evidence against the mutation bias hypothesis. Furthermore, no evidence for context-dependent mutation rate differences, i.e., via methylation or neighboring nucleotides, was evident (see the supplemental material).

FIG 1.

(A) Cartoon illustrating the two proposed hypotheses. The mutation bias hypothesis (left) proposes that the small-effect substitution R164S has a higher mutation rate than the G238S substitution, enhancing its fixation in small populations. The selection bias hypothesis (right) states that R164S has a selective benefit over G238S at low cefotaxime concentrations, which were prevalent in small populations due to their slower adaptation. (B) Fitness of G238S relative to that of R164S as determined by pairwise competition assays at different cefotaxime concentrations. Error bars represent the standard errors of the means of 12 biological replicates. The blue line is the regression estimate for relative fitness as a function of cefotaxime concentration (R² = 0.39, P < 0.0001). The shaded area shows the 95% confidence interval of this regression. The black arrow indicates the initial concentration, 0.011 μg/ml, used in the evolution experiment by Schenk et al. (C) Fitness of R164S and G238S relative to TEM-1. Asterisks show a significant difference between the mean values for the groups (one-sample t test; ** P < 0.01; ***, P < 0.001). The P values are from Student’s t tests comparing R164S and G238S. Error bars represent the standard errors of the means of six biological replicates. (D) Boxplots showing the numbers of generations where cefotaxime was below 0.03 μg/ml in small and large populations that fixed R164S or G238S in the evolution experiment by Schenk et al. (15).

Our second hypothesis postulates that R164S is more beneficial in small than in large populations (Fig. 1A). During Schenk et al.’s evolution experiment, the cefotaxime concentrations were increased by a factor of 2^0.25 from an initial concentration of 0.011 μg/ml whenever populations reached a predetermined minimal cell density by adapting to cefotaxime. Since large populations exhibited faster adaptation, large populations were, on average, subjected to 4.6-fold-higher concentrations of cefotaxime than small populations and reached an average final concentration of 611.7 μg/ml, while small populations reached 24.0 μg/ml (15). Consequently, if the smaller resistance effect of R164S comes with a smaller fitness cost than G238S in the absence of antibiotic, R164S mutants may outcompete G238S mutants at low cefotaxime concentrations, while G238S mutants are competitively superior at higher concentrations. If so, the lower cefotaxime concentrations experienced in small populations relative to the concentrations reached in large populations may have provided a longer window of opportunity for R164S mutants to spread and fix.

To test the second hypothesis, we ran competition assays between isogenic strains expressing G238S and R164S at increasing concentrations of cefotaxime (see detailed methods in the supplemental material). These assays revealed that when the cefotaxime concentration was lower than 0.033 μg/ml, R164S outcompeted G238S (Fig. 1B). However, at increasing concentrations, G238S took over in a concentration-dependent manner (effect of CTX; P < 0.001). Next, we determined the fitness of R164S and G238S relative to TEM-1 in the absence of cefotaxime and at a cefotaxime concentration close to the initial concentration used in the study by Schenk et al. (0.01 μg/ml) (Fig. 1C) (15). In the absence of antibiotics, both R164S and G238S were significantly less fit than the ancestral allele, TEM-1 (Fig. 1C), but G238S was more so than R164S, indicating greater fitness costs for G238S than for R164S (Student’s t test for growth rates, P < 0.001) (Table 1). Similarly, at 0.01 μg/ml cefotaxime, R164S could outcompete its ancestral allele, while G238S could not.

These results suggest that the presence of R164S and G238S in small and large populations as observed in Schenk et al. (15) is caused by a shift in their relative selective advantages at increasing concentrations of cefotaxime. Indeed, the number of generations below the critical cefotaxime concentration (0.033 μg/ml) during the evolution experiment was almost double for small populations that fixed R164S compared to the number of generations in large populations with G238S (τ_large = 73; τ_small = 122; Mann-Whitney, U  = 36.5; two-tailed P = 0.02) (Fig. 1D). In large populations, the 100-fold-larger mutation supply allowed faster adaptation and, hence, a faster increase in tolerable cefotaxime concentrations, which limited the time window in which R164S mutation would outcompete G238S.

In conclusion, we show that the choice between two cefotaxime resistance-conferring mutations in TEM-1 β-lactamase, which are common in clinical isolates and laboratory evolution studies, critically depends on the antibiotic concentration. Due to their large effect on cefotaxime resistance and strongly negative epistatic interaction (13, 14), G238S and R164S have distinct consequences for cefotaxime resistance, as the MICs of TEM-1 variants containing G238S are considerably higher than those containing R164S (∼256 μg/ml versus ∼32 μg/ml, respectively) (1). Our findings show that the evolutionary accessibility of resistance maxima may critically depend on the initial antibiotic concentration.