Fig 1.
Fitness variability changes drastically in the presence of different antibiotics.
(A) Sample growth curves (optical density over time) of wild type (WT; black) and gene deletion mutants pdxJ (green) and iscS (red); yellow lines are exponential fits (Materials and Methods). (B) Histogram of growth rates (i.e., approximated DFE) of ~4,000 gene deletion strains in the absence of drug; histogram of 476 WT replicates is outlined in black. (C) DFE in the presence of the antibiotics trimethoprim, nitrofurantoin, tetracycline, chloramphenicol, ciprofloxacin, mecillinam, cefoxitin, and ampicillin (Table 1); vertical black lines show median of WT replicates; drugs were used at concentrations inhibiting WT growth by one-third. Growth rates are normalized to median of WT in the absence of a drug. The interquartile ranges (IQRs) of the DFEs are shown in Fig 2B. Numerical data is in S1 Data.
Table 1.
Antibiotics used in this study.
Fig 2.
The drug-specific dose-sensitivity is robust to genetic perturbations and correlates with fitness variability.
(A) Dose-response curves for eight antibiotics; circles (●) show trimethoprim; pluses (+), tetracycline; downward triangles (▼), chloramphenicol; stars (★), nitrofurantoin; squares (■), ciprofloxacin; leftward triangles (◄), cefoxitin; triangles (▲), mecillinam; rightward triangles (►), ampicillin. Dose-sensitivity n is shown (Materials and Methods). (B) Scatterplot of dose-sensitivity n and DFE width (IQR); Pearson’s ρ = 0.96, p = 1.3 × 10−4; n error bars show standard deviation of replicates; DFE width error bars show bootstrap 95% confidence interval (Materials and Methods). Horizontal dashed line shows DFE width in the absence of drug (cf. Fig 1B). Gray line shows a linear relation as a guide to the eye. (C) Mecillinam dose-response curves for 78 arbitrary deletion mutants (purple; see S1 Data) and 17 WT replicates (black). (D) Same data as in C with concentration rescaled to IC50 and growth rate response rescaled to g0 (Materials and Methods). See also S1 Fig Numerical data is in S1 Data.
Fig 3.
Resistance variability is similar for diverse antibiotics but extremely low for nitrofurantoin.
(A) Schematic: the effective drug concentration ceff experienced by each mutant is inferred from its response R via the WT dose-response curve (arrows). This transforms the DFE (y-axis) into the DEC of the drug (x-axis); the dose-sensitivity n determines the change in distribution width as shown. (B) DEC for different antibiotics; arrows show IQR; effective drug concentrations are normalized to the actual concentration. (C) DEC width (IQR) for different antibiotics. (D) Width of the distribution of relative IC50 changes determined directly from dose-response curves of 78 deletion mutants (S1 Fig). Error bars show bootstrap standard error (C) and 95% confidence interval from bootstrap (D), respectively; lighter bars show distribution width resulting from measurement noise alone (Materials and Methods). Note that the difference for chloramphenicol between panels C and D is not significant. Numerical data is in S1 Data.
Fig 4.
Resistance variability affects the dynamics of evolutionary adaptation to drugs.
(A) Simulation results from a theoretical model of resistance evolution in a morbidostat [11]: IC50 increase over time for a drug with narrow DEC and two available large-effect mutations (magenta) or none (gray); light lines are sample runs; dark lines are mean of 200 runs; inset: distribution of relative IC50 changes used in simulations (Materials and Methods). (B) Same as in panel A for wider DEC (Materials and Methods). (C, D) Results from morbidostat laboratory evolution experiments: IC50 increase over time for nitrofurantoin (C) and chloramphenicol (D); light lines are individual runs; dark lines are mean, error bars standard deviation; shaded region in C indicates early phase during which large-effect mutations fix (Materials and Methods). (E) Mutated loci in nitrofurantoin (left) and chloramphenicol (right)-resistant clones after 10 and 21 d, respectively. Filled pie segments show evolution replicates in which genes were mutated; (P) indicates promoter mutations. Bar chart shows diversity (entropy) of mutations under nitrofurantoin (magenta) and chloramphenicol (gray); p < 0.002 (**) and p < 0.0003 (***) from two-sample t test; error bars show jackknife standard error (Materials and Methods). Numerical data is in S1 Data. Whole genome sequencing results are in S2 Table and S3 Table.